xref: /linux/arch/x86/entry/entry_32.S (revision 55f3538c4923e9dfca132e99ebec370e8094afda)
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	/* Clobbers %ebx */
256	FILL_RETURN_BUFFER RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
257#endif
258
259	/* restore callee-saved registers */
260	popl	%esi
261	popl	%edi
262	popl	%ebx
263	popl	%ebp
264
265	jmp	__switch_to
266END(__switch_to_asm)
267
268/*
269 * The unwinder expects the last frame on the stack to always be at the same
270 * offset from the end of the page, which allows it to validate the stack.
271 * Calling schedule_tail() directly would break that convention because its an
272 * asmlinkage function so its argument has to be pushed on the stack.  This
273 * wrapper creates a proper "end of stack" frame header before the call.
274 */
275ENTRY(schedule_tail_wrapper)
276	FRAME_BEGIN
277
278	pushl	%eax
279	call	schedule_tail
280	popl	%eax
281
282	FRAME_END
283	ret
284ENDPROC(schedule_tail_wrapper)
285/*
286 * A newly forked process directly context switches into this address.
287 *
288 * eax: prev task we switched from
289 * ebx: kernel thread func (NULL for user thread)
290 * edi: kernel thread arg
291 */
292ENTRY(ret_from_fork)
293	call	schedule_tail_wrapper
294
295	testl	%ebx, %ebx
296	jnz	1f		/* kernel threads are uncommon */
297
2982:
299	/* When we fork, we trace the syscall return in the child, too. */
300	movl    %esp, %eax
301	call    syscall_return_slowpath
302	jmp     restore_all
303
304	/* kernel thread */
3051:	movl	%edi, %eax
306	CALL_NOSPEC %ebx
307	/*
308	 * A kernel thread is allowed to return here after successfully
309	 * calling do_execve().  Exit to userspace to complete the execve()
310	 * syscall.
311	 */
312	movl	$0, PT_EAX(%esp)
313	jmp	2b
314END(ret_from_fork)
315
316/*
317 * Return to user mode is not as complex as all this looks,
318 * but we want the default path for a system call return to
319 * go as quickly as possible which is why some of this is
320 * less clear than it otherwise should be.
321 */
322
323	# userspace resumption stub bypassing syscall exit tracing
324	ALIGN
325ret_from_exception:
326	preempt_stop(CLBR_ANY)
327ret_from_intr:
328#ifdef CONFIG_VM86
329	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
330	movb	PT_CS(%esp), %al
331	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
332#else
333	/*
334	 * We can be coming here from child spawned by kernel_thread().
335	 */
336	movl	PT_CS(%esp), %eax
337	andl	$SEGMENT_RPL_MASK, %eax
338#endif
339	cmpl	$USER_RPL, %eax
340	jb	resume_kernel			# not returning to v8086 or userspace
341
342ENTRY(resume_userspace)
343	DISABLE_INTERRUPTS(CLBR_ANY)
344	TRACE_IRQS_OFF
345	movl	%esp, %eax
346	call	prepare_exit_to_usermode
347	jmp	restore_all
348END(ret_from_exception)
349
350#ifdef CONFIG_PREEMPT
351ENTRY(resume_kernel)
352	DISABLE_INTERRUPTS(CLBR_ANY)
353.Lneed_resched:
354	cmpl	$0, PER_CPU_VAR(__preempt_count)
355	jnz	restore_all
356	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
357	jz	restore_all
358	call	preempt_schedule_irq
359	jmp	.Lneed_resched
360END(resume_kernel)
361#endif
362
363GLOBAL(__begin_SYSENTER_singlestep_region)
364/*
365 * All code from here through __end_SYSENTER_singlestep_region is subject
366 * to being single-stepped if a user program sets TF and executes SYSENTER.
367 * There is absolutely nothing that we can do to prevent this from happening
368 * (thanks Intel!).  To keep our handling of this situation as simple as
369 * possible, we handle TF just like AC and NT, except that our #DB handler
370 * will ignore all of the single-step traps generated in this range.
371 */
372
373#ifdef CONFIG_XEN
374/*
375 * Xen doesn't set %esp to be precisely what the normal SYSENTER
376 * entry point expects, so fix it up before using the normal path.
377 */
378ENTRY(xen_sysenter_target)
379	addl	$5*4, %esp			/* remove xen-provided frame */
380	jmp	.Lsysenter_past_esp
381#endif
382
383/*
384 * 32-bit SYSENTER entry.
385 *
386 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
387 * if X86_FEATURE_SEP is available.  This is the preferred system call
388 * entry on 32-bit systems.
389 *
390 * The SYSENTER instruction, in principle, should *only* occur in the
391 * vDSO.  In practice, a small number of Android devices were shipped
392 * with a copy of Bionic that inlined a SYSENTER instruction.  This
393 * never happened in any of Google's Bionic versions -- it only happened
394 * in a narrow range of Intel-provided versions.
395 *
396 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
397 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
398 * SYSENTER does not save anything on the stack,
399 * and does not save old EIP (!!!), ESP, or EFLAGS.
400 *
401 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
402 * user and/or vm86 state), we explicitly disable the SYSENTER
403 * instruction in vm86 mode by reprogramming the MSRs.
404 *
405 * Arguments:
406 * eax  system call number
407 * ebx  arg1
408 * ecx  arg2
409 * edx  arg3
410 * esi  arg4
411 * edi  arg5
412 * ebp  user stack
413 * 0(%ebp) arg6
414 */
415ENTRY(entry_SYSENTER_32)
416	movl	TSS_sysenter_sp0(%esp), %esp
417.Lsysenter_past_esp:
418	pushl	$__USER_DS		/* pt_regs->ss */
419	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
420	pushfl				/* pt_regs->flags (except IF = 0) */
421	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
422	pushl	$__USER_CS		/* pt_regs->cs */
423	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
424	pushl	%eax			/* pt_regs->orig_ax */
425	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
426
427	/*
428	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
429	 * and TF ourselves.  To save a few cycles, we can check whether
430	 * either was set instead of doing an unconditional popfq.
431	 * This needs to happen before enabling interrupts so that
432	 * we don't get preempted with NT set.
433	 *
434	 * If TF is set, we will single-step all the way to here -- do_debug
435	 * will ignore all the traps.  (Yes, this is slow, but so is
436	 * single-stepping in general.  This allows us to avoid having
437	 * a more complicated code to handle the case where a user program
438	 * forces us to single-step through the SYSENTER entry code.)
439	 *
440	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
441	 * out-of-line as an optimization: NT is unlikely to be set in the
442	 * majority of the cases and instead of polluting the I$ unnecessarily,
443	 * we're keeping that code behind a branch which will predict as
444	 * not-taken and therefore its instructions won't be fetched.
445	 */
446	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
447	jnz	.Lsysenter_fix_flags
448.Lsysenter_flags_fixed:
449
450	/*
451	 * User mode is traced as though IRQs are on, and SYSENTER
452	 * turned them off.
453	 */
454	TRACE_IRQS_OFF
455
456	movl	%esp, %eax
457	call	do_fast_syscall_32
458	/* XEN PV guests always use IRET path */
459	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
460		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
461
462/* Opportunistic SYSEXIT */
463	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
464	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
465	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
4661:	mov	PT_FS(%esp), %fs
467	PTGS_TO_GS
468	popl	%ebx			/* pt_regs->bx */
469	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
470	popl	%esi			/* pt_regs->si */
471	popl	%edi			/* pt_regs->di */
472	popl	%ebp			/* pt_regs->bp */
473	popl	%eax			/* pt_regs->ax */
474
475	/*
476	 * Restore all flags except IF. (We restore IF separately because
477	 * STI gives a one-instruction window in which we won't be interrupted,
478	 * whereas POPF does not.)
479	 */
480	addl	$PT_EFLAGS-PT_DS, %esp	/* point esp at pt_regs->flags */
481	btr	$X86_EFLAGS_IF_BIT, (%esp)
482	popfl
483
484	/*
485	 * Return back to the vDSO, which will pop ecx and edx.
486	 * Don't bother with DS and ES (they already contain __USER_DS).
487	 */
488	sti
489	sysexit
490
491.pushsection .fixup, "ax"
4922:	movl	$0, PT_FS(%esp)
493	jmp	1b
494.popsection
495	_ASM_EXTABLE(1b, 2b)
496	PTGS_TO_GS_EX
497
498.Lsysenter_fix_flags:
499	pushl	$X86_EFLAGS_FIXED
500	popfl
501	jmp	.Lsysenter_flags_fixed
502GLOBAL(__end_SYSENTER_singlestep_region)
503ENDPROC(entry_SYSENTER_32)
504
505/*
506 * 32-bit legacy system call entry.
507 *
508 * 32-bit x86 Linux system calls traditionally used the INT $0x80
509 * instruction.  INT $0x80 lands here.
510 *
511 * This entry point can be used by any 32-bit perform system calls.
512 * Instances of INT $0x80 can be found inline in various programs and
513 * libraries.  It is also used by the vDSO's __kernel_vsyscall
514 * fallback for hardware that doesn't support a faster entry method.
515 * Restarted 32-bit system calls also fall back to INT $0x80
516 * regardless of what instruction was originally used to do the system
517 * call.  (64-bit programs can use INT $0x80 as well, but they can
518 * only run on 64-bit kernels and therefore land in
519 * entry_INT80_compat.)
520 *
521 * This is considered a slow path.  It is not used by most libc
522 * implementations on modern hardware except during process startup.
523 *
524 * Arguments:
525 * eax  system call number
526 * ebx  arg1
527 * ecx  arg2
528 * edx  arg3
529 * esi  arg4
530 * edi  arg5
531 * ebp  arg6
532 */
533ENTRY(entry_INT80_32)
534	ASM_CLAC
535	pushl	%eax			/* pt_regs->orig_ax */
536	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest */
537
538	/*
539	 * User mode is traced as though IRQs are on, and the interrupt gate
540	 * turned them off.
541	 */
542	TRACE_IRQS_OFF
543
544	movl	%esp, %eax
545	call	do_int80_syscall_32
546.Lsyscall_32_done:
547
548restore_all:
549	TRACE_IRQS_IRET
550.Lrestore_all_notrace:
551#ifdef CONFIG_X86_ESPFIX32
552	ALTERNATIVE	"jmp .Lrestore_nocheck", "", X86_BUG_ESPFIX
553
554	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
555	/*
556	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
557	 * are returning to the kernel.
558	 * See comments in process.c:copy_thread() for details.
559	 */
560	movb	PT_OLDSS(%esp), %ah
561	movb	PT_CS(%esp), %al
562	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
563	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
564	je .Lldt_ss				# returning to user-space with LDT SS
565#endif
566.Lrestore_nocheck:
567	RESTORE_REGS 4				# skip orig_eax/error_code
568.Lirq_return:
569	INTERRUPT_RETURN
570
571.section .fixup, "ax"
572ENTRY(iret_exc	)
573	pushl	$0				# no error code
574	pushl	$do_iret_error
575	jmp	common_exception
576.previous
577	_ASM_EXTABLE(.Lirq_return, iret_exc)
578
579#ifdef CONFIG_X86_ESPFIX32
580.Lldt_ss:
581/*
582 * Setup and switch to ESPFIX stack
583 *
584 * We're returning to userspace with a 16 bit stack. The CPU will not
585 * restore the high word of ESP for us on executing iret... This is an
586 * "official" bug of all the x86-compatible CPUs, which we can work
587 * around to make dosemu and wine happy. We do this by preloading the
588 * high word of ESP with the high word of the userspace ESP while
589 * compensating for the offset by changing to the ESPFIX segment with
590 * a base address that matches for the difference.
591 */
592#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
593	mov	%esp, %edx			/* load kernel esp */
594	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
595	mov	%dx, %ax			/* eax: new kernel esp */
596	sub	%eax, %edx			/* offset (low word is 0) */
597	shr	$16, %edx
598	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
599	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
600	pushl	$__ESPFIX_SS
601	pushl	%eax				/* new kernel esp */
602	/*
603	 * Disable interrupts, but do not irqtrace this section: we
604	 * will soon execute iret and the tracer was already set to
605	 * the irqstate after the IRET:
606	 */
607	DISABLE_INTERRUPTS(CLBR_ANY)
608	lss	(%esp), %esp			/* switch to espfix segment */
609	jmp	.Lrestore_nocheck
610#endif
611ENDPROC(entry_INT80_32)
612
613.macro FIXUP_ESPFIX_STACK
614/*
615 * Switch back for ESPFIX stack to the normal zerobased stack
616 *
617 * We can't call C functions using the ESPFIX stack. This code reads
618 * the high word of the segment base from the GDT and swiches to the
619 * normal stack and adjusts ESP with the matching offset.
620 */
621#ifdef CONFIG_X86_ESPFIX32
622	/* fixup the stack */
623	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
624	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
625	shl	$16, %eax
626	addl	%esp, %eax			/* the adjusted stack pointer */
627	pushl	$__KERNEL_DS
628	pushl	%eax
629	lss	(%esp), %esp			/* switch to the normal stack segment */
630#endif
631.endm
632.macro UNWIND_ESPFIX_STACK
633#ifdef CONFIG_X86_ESPFIX32
634	movl	%ss, %eax
635	/* see if on espfix stack */
636	cmpw	$__ESPFIX_SS, %ax
637	jne	27f
638	movl	$__KERNEL_DS, %eax
639	movl	%eax, %ds
640	movl	%eax, %es
641	/* switch to normal stack */
642	FIXUP_ESPFIX_STACK
64327:
644#endif
645.endm
646
647/*
648 * Build the entry stubs with some assembler magic.
649 * We pack 1 stub into every 8-byte block.
650 */
651	.align 8
652ENTRY(irq_entries_start)
653    vector=FIRST_EXTERNAL_VECTOR
654    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
655	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
656    vector=vector+1
657	jmp	common_interrupt
658	.align	8
659    .endr
660END(irq_entries_start)
661
662/*
663 * the CPU automatically disables interrupts when executing an IRQ vector,
664 * so IRQ-flags tracing has to follow that:
665 */
666	.p2align CONFIG_X86_L1_CACHE_SHIFT
667common_interrupt:
668	ASM_CLAC
669	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
670	SAVE_ALL
671	ENCODE_FRAME_POINTER
672	TRACE_IRQS_OFF
673	movl	%esp, %eax
674	call	do_IRQ
675	jmp	ret_from_intr
676ENDPROC(common_interrupt)
677
678#define BUILD_INTERRUPT3(name, nr, fn)	\
679ENTRY(name)				\
680	ASM_CLAC;			\
681	pushl	$~(nr);			\
682	SAVE_ALL;			\
683	ENCODE_FRAME_POINTER;		\
684	TRACE_IRQS_OFF			\
685	movl	%esp, %eax;		\
686	call	fn;			\
687	jmp	ret_from_intr;		\
688ENDPROC(name)
689
690#define BUILD_INTERRUPT(name, nr)		\
691	BUILD_INTERRUPT3(name, nr, smp_##name);	\
692
693/* The include is where all of the SMP etc. interrupts come from */
694#include <asm/entry_arch.h>
695
696ENTRY(coprocessor_error)
697	ASM_CLAC
698	pushl	$0
699	pushl	$do_coprocessor_error
700	jmp	common_exception
701END(coprocessor_error)
702
703ENTRY(simd_coprocessor_error)
704	ASM_CLAC
705	pushl	$0
706#ifdef CONFIG_X86_INVD_BUG
707	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
708	ALTERNATIVE "pushl	$do_general_protection",	\
709		    "pushl	$do_simd_coprocessor_error",	\
710		    X86_FEATURE_XMM
711#else
712	pushl	$do_simd_coprocessor_error
713#endif
714	jmp	common_exception
715END(simd_coprocessor_error)
716
717ENTRY(device_not_available)
718	ASM_CLAC
719	pushl	$-1				# mark this as an int
720	pushl	$do_device_not_available
721	jmp	common_exception
722END(device_not_available)
723
724#ifdef CONFIG_PARAVIRT
725ENTRY(native_iret)
726	iret
727	_ASM_EXTABLE(native_iret, iret_exc)
728END(native_iret)
729#endif
730
731ENTRY(overflow)
732	ASM_CLAC
733	pushl	$0
734	pushl	$do_overflow
735	jmp	common_exception
736END(overflow)
737
738ENTRY(bounds)
739	ASM_CLAC
740	pushl	$0
741	pushl	$do_bounds
742	jmp	common_exception
743END(bounds)
744
745ENTRY(invalid_op)
746	ASM_CLAC
747	pushl	$0
748	pushl	$do_invalid_op
749	jmp	common_exception
750END(invalid_op)
751
752ENTRY(coprocessor_segment_overrun)
753	ASM_CLAC
754	pushl	$0
755	pushl	$do_coprocessor_segment_overrun
756	jmp	common_exception
757END(coprocessor_segment_overrun)
758
759ENTRY(invalid_TSS)
760	ASM_CLAC
761	pushl	$do_invalid_TSS
762	jmp	common_exception
763END(invalid_TSS)
764
765ENTRY(segment_not_present)
766	ASM_CLAC
767	pushl	$do_segment_not_present
768	jmp	common_exception
769END(segment_not_present)
770
771ENTRY(stack_segment)
772	ASM_CLAC
773	pushl	$do_stack_segment
774	jmp	common_exception
775END(stack_segment)
776
777ENTRY(alignment_check)
778	ASM_CLAC
779	pushl	$do_alignment_check
780	jmp	common_exception
781END(alignment_check)
782
783ENTRY(divide_error)
784	ASM_CLAC
785	pushl	$0				# no error code
786	pushl	$do_divide_error
787	jmp	common_exception
788END(divide_error)
789
790#ifdef CONFIG_X86_MCE
791ENTRY(machine_check)
792	ASM_CLAC
793	pushl	$0
794	pushl	machine_check_vector
795	jmp	common_exception
796END(machine_check)
797#endif
798
799ENTRY(spurious_interrupt_bug)
800	ASM_CLAC
801	pushl	$0
802	pushl	$do_spurious_interrupt_bug
803	jmp	common_exception
804END(spurious_interrupt_bug)
805
806#ifdef CONFIG_XEN
807ENTRY(xen_hypervisor_callback)
808	pushl	$-1				/* orig_ax = -1 => not a system call */
809	SAVE_ALL
810	ENCODE_FRAME_POINTER
811	TRACE_IRQS_OFF
812
813	/*
814	 * Check to see if we got the event in the critical
815	 * region in xen_iret_direct, after we've reenabled
816	 * events and checked for pending events.  This simulates
817	 * iret instruction's behaviour where it delivers a
818	 * pending interrupt when enabling interrupts:
819	 */
820	movl	PT_EIP(%esp), %eax
821	cmpl	$xen_iret_start_crit, %eax
822	jb	1f
823	cmpl	$xen_iret_end_crit, %eax
824	jae	1f
825
826	jmp	xen_iret_crit_fixup
827
828ENTRY(xen_do_upcall)
8291:	mov	%esp, %eax
830	call	xen_evtchn_do_upcall
831#ifndef CONFIG_PREEMPT
832	call	xen_maybe_preempt_hcall
833#endif
834	jmp	ret_from_intr
835ENDPROC(xen_hypervisor_callback)
836
837/*
838 * Hypervisor uses this for application faults while it executes.
839 * We get here for two reasons:
840 *  1. Fault while reloading DS, ES, FS or GS
841 *  2. Fault while executing IRET
842 * Category 1 we fix up by reattempting the load, and zeroing the segment
843 * register if the load fails.
844 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
845 * normal Linux return path in this case because if we use the IRET hypercall
846 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
847 * We distinguish between categories by maintaining a status value in EAX.
848 */
849ENTRY(xen_failsafe_callback)
850	pushl	%eax
851	movl	$1, %eax
8521:	mov	4(%esp), %ds
8532:	mov	8(%esp), %es
8543:	mov	12(%esp), %fs
8554:	mov	16(%esp), %gs
856	/* EAX == 0 => Category 1 (Bad segment)
857	   EAX != 0 => Category 2 (Bad IRET) */
858	testl	%eax, %eax
859	popl	%eax
860	lea	16(%esp), %esp
861	jz	5f
862	jmp	iret_exc
8635:	pushl	$-1				/* orig_ax = -1 => not a system call */
864	SAVE_ALL
865	ENCODE_FRAME_POINTER
866	jmp	ret_from_exception
867
868.section .fixup, "ax"
8696:	xorl	%eax, %eax
870	movl	%eax, 4(%esp)
871	jmp	1b
8727:	xorl	%eax, %eax
873	movl	%eax, 8(%esp)
874	jmp	2b
8758:	xorl	%eax, %eax
876	movl	%eax, 12(%esp)
877	jmp	3b
8789:	xorl	%eax, %eax
879	movl	%eax, 16(%esp)
880	jmp	4b
881.previous
882	_ASM_EXTABLE(1b, 6b)
883	_ASM_EXTABLE(2b, 7b)
884	_ASM_EXTABLE(3b, 8b)
885	_ASM_EXTABLE(4b, 9b)
886ENDPROC(xen_failsafe_callback)
887
888BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
889		 xen_evtchn_do_upcall)
890
891#endif /* CONFIG_XEN */
892
893#if IS_ENABLED(CONFIG_HYPERV)
894
895BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
896		 hyperv_vector_handler)
897
898#endif /* CONFIG_HYPERV */
899
900ENTRY(page_fault)
901	ASM_CLAC
902	pushl	$do_page_fault
903	ALIGN
904	jmp common_exception
905END(page_fault)
906
907common_exception:
908	/* the function address is in %gs's slot on the stack */
909	pushl	%fs
910	pushl	%es
911	pushl	%ds
912	pushl	%eax
913	pushl	%ebp
914	pushl	%edi
915	pushl	%esi
916	pushl	%edx
917	pushl	%ecx
918	pushl	%ebx
919	ENCODE_FRAME_POINTER
920	cld
921	movl	$(__KERNEL_PERCPU), %ecx
922	movl	%ecx, %fs
923	UNWIND_ESPFIX_STACK
924	GS_TO_REG %ecx
925	movl	PT_GS(%esp), %edi		# get the function address
926	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
927	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
928	REG_TO_PTGS %ecx
929	SET_KERNEL_GS %ecx
930	movl	$(__USER_DS), %ecx
931	movl	%ecx, %ds
932	movl	%ecx, %es
933	TRACE_IRQS_OFF
934	movl	%esp, %eax			# pt_regs pointer
935	CALL_NOSPEC %edi
936	jmp	ret_from_exception
937END(common_exception)
938
939ENTRY(debug)
940	/*
941	 * #DB can happen at the first instruction of
942	 * entry_SYSENTER_32 or in Xen's SYSENTER prologue.  If this
943	 * happens, then we will be running on a very small stack.  We
944	 * need to detect this condition and switch to the thread
945	 * stack before calling any C code at all.
946	 *
947	 * If you edit this code, keep in mind that NMIs can happen in here.
948	 */
949	ASM_CLAC
950	pushl	$-1				# mark this as an int
951	SAVE_ALL
952	ENCODE_FRAME_POINTER
953	xorl	%edx, %edx			# error code 0
954	movl	%esp, %eax			# pt_regs pointer
955
956	/* Are we currently on the SYSENTER stack? */
957	movl	PER_CPU_VAR(cpu_entry_area), %ecx
958	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
959	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
960	cmpl	$SIZEOF_entry_stack, %ecx
961	jb	.Ldebug_from_sysenter_stack
962
963	TRACE_IRQS_OFF
964	call	do_debug
965	jmp	ret_from_exception
966
967.Ldebug_from_sysenter_stack:
968	/* We're on the SYSENTER stack.  Switch off. */
969	movl	%esp, %ebx
970	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
971	TRACE_IRQS_OFF
972	call	do_debug
973	movl	%ebx, %esp
974	jmp	ret_from_exception
975END(debug)
976
977/*
978 * NMI is doubly nasty.  It can happen on the first instruction of
979 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
980 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
981 * switched stacks.  We handle both conditions by simply checking whether we
982 * interrupted kernel code running on the SYSENTER stack.
983 */
984ENTRY(nmi)
985	ASM_CLAC
986#ifdef CONFIG_X86_ESPFIX32
987	pushl	%eax
988	movl	%ss, %eax
989	cmpw	$__ESPFIX_SS, %ax
990	popl	%eax
991	je	.Lnmi_espfix_stack
992#endif
993
994	pushl	%eax				# pt_regs->orig_ax
995	SAVE_ALL
996	ENCODE_FRAME_POINTER
997	xorl	%edx, %edx			# zero error code
998	movl	%esp, %eax			# pt_regs pointer
999
1000	/* Are we currently on the SYSENTER stack? */
1001	movl	PER_CPU_VAR(cpu_entry_area), %ecx
1002	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1003	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
1004	cmpl	$SIZEOF_entry_stack, %ecx
1005	jb	.Lnmi_from_sysenter_stack
1006
1007	/* Not on SYSENTER stack. */
1008	call	do_nmi
1009	jmp	.Lrestore_all_notrace
1010
1011.Lnmi_from_sysenter_stack:
1012	/*
1013	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1014	 * is using the thread stack right now, so it's safe for us to use it.
1015	 */
1016	movl	%esp, %ebx
1017	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1018	call	do_nmi
1019	movl	%ebx, %esp
1020	jmp	.Lrestore_all_notrace
1021
1022#ifdef CONFIG_X86_ESPFIX32
1023.Lnmi_espfix_stack:
1024	/*
1025	 * create the pointer to lss back
1026	 */
1027	pushl	%ss
1028	pushl	%esp
1029	addl	$4, (%esp)
1030	/* copy the iret frame of 12 bytes */
1031	.rept 3
1032	pushl	16(%esp)
1033	.endr
1034	pushl	%eax
1035	SAVE_ALL
1036	ENCODE_FRAME_POINTER
1037	FIXUP_ESPFIX_STACK			# %eax == %esp
1038	xorl	%edx, %edx			# zero error code
1039	call	do_nmi
1040	RESTORE_REGS
1041	lss	12+4(%esp), %esp		# back to espfix stack
1042	jmp	.Lirq_return
1043#endif
1044END(nmi)
1045
1046ENTRY(int3)
1047	ASM_CLAC
1048	pushl	$-1				# mark this as an int
1049	SAVE_ALL
1050	ENCODE_FRAME_POINTER
1051	TRACE_IRQS_OFF
1052	xorl	%edx, %edx			# zero error code
1053	movl	%esp, %eax			# pt_regs pointer
1054	call	do_int3
1055	jmp	ret_from_exception
1056END(int3)
1057
1058ENTRY(general_protection)
1059	pushl	$do_general_protection
1060	jmp	common_exception
1061END(general_protection)
1062
1063#ifdef CONFIG_KVM_GUEST
1064ENTRY(async_page_fault)
1065	ASM_CLAC
1066	pushl	$do_async_page_fault
1067	jmp	common_exception
1068END(async_page_fault)
1069#endif
1070
1071ENTRY(rewind_stack_do_exit)
1072	/* Prevent any naive code from trying to unwind to our caller. */
1073	xorl	%ebp, %ebp
1074
1075	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1076	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1077
1078	call	do_exit
10791:	jmp 1b
1080END(rewind_stack_do_exit)
1081