xref: /openbmc/linux/arch/x86/entry/entry_32.S (revision 4b0aaacee51eb6592a03fdefd5ce97558518e291)
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_kernel
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#define PTI_SWITCH_MASK         (1 << PAGE_SHIFT)
81
82/*
83 * User gs save/restore
84 *
85 * %gs is used for userland TLS and kernel only uses it for stack
86 * canary which is required to be at %gs:20 by gcc.  Read the comment
87 * at the top of stackprotector.h for more info.
88 *
89 * Local labels 98 and 99 are used.
90 */
91#ifdef CONFIG_X86_32_LAZY_GS
92
93 /* unfortunately push/pop can't be no-op */
94.macro PUSH_GS
95	pushl	$0
96.endm
97.macro POP_GS pop=0
98	addl	$(4 + \pop), %esp
99.endm
100.macro POP_GS_EX
101.endm
102
103 /* all the rest are no-op */
104.macro PTGS_TO_GS
105.endm
106.macro PTGS_TO_GS_EX
107.endm
108.macro GS_TO_REG reg
109.endm
110.macro REG_TO_PTGS reg
111.endm
112.macro SET_KERNEL_GS reg
113.endm
114
115#else	/* CONFIG_X86_32_LAZY_GS */
116
117.macro PUSH_GS
118	pushl	%gs
119.endm
120
121.macro POP_GS pop=0
12298:	popl	%gs
123  .if \pop <> 0
124	add	$\pop, %esp
125  .endif
126.endm
127.macro POP_GS_EX
128.pushsection .fixup, "ax"
12999:	movl	$0, (%esp)
130	jmp	98b
131.popsection
132	_ASM_EXTABLE(98b, 99b)
133.endm
134
135.macro PTGS_TO_GS
13698:	mov	PT_GS(%esp), %gs
137.endm
138.macro PTGS_TO_GS_EX
139.pushsection .fixup, "ax"
14099:	movl	$0, PT_GS(%esp)
141	jmp	98b
142.popsection
143	_ASM_EXTABLE(98b, 99b)
144.endm
145
146.macro GS_TO_REG reg
147	movl	%gs, \reg
148.endm
149.macro REG_TO_PTGS reg
150	movl	\reg, PT_GS(%esp)
151.endm
152.macro SET_KERNEL_GS reg
153	movl	$(__KERNEL_STACK_CANARY), \reg
154	movl	\reg, %gs
155.endm
156
157#endif /* CONFIG_X86_32_LAZY_GS */
158
159/* Unconditionally switch to user cr3 */
160.macro SWITCH_TO_USER_CR3 scratch_reg:req
161	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
162
163	movl	%cr3, \scratch_reg
164	orl	$PTI_SWITCH_MASK, \scratch_reg
165	movl	\scratch_reg, %cr3
166.Lend_\@:
167.endm
168
169.macro BUG_IF_WRONG_CR3 no_user_check=0
170#ifdef CONFIG_DEBUG_ENTRY
171	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
172	.if \no_user_check == 0
173	/* coming from usermode? */
174	testl	$SEGMENT_RPL_MASK, PT_CS(%esp)
175	jz	.Lend_\@
176	.endif
177	/* On user-cr3? */
178	movl	%cr3, %eax
179	testl	$PTI_SWITCH_MASK, %eax
180	jnz	.Lend_\@
181	/* From userspace with kernel cr3 - BUG */
182	ud2
183.Lend_\@:
184#endif
185.endm
186
187/*
188 * Switch to kernel cr3 if not already loaded and return current cr3 in
189 * \scratch_reg
190 */
191.macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
192	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
193	movl	%cr3, \scratch_reg
194	/* Test if we are already on kernel CR3 */
195	testl	$PTI_SWITCH_MASK, \scratch_reg
196	jz	.Lend_\@
197	andl	$(~PTI_SWITCH_MASK), \scratch_reg
198	movl	\scratch_reg, %cr3
199	/* Return original CR3 in \scratch_reg */
200	orl	$PTI_SWITCH_MASK, \scratch_reg
201.Lend_\@:
202.endm
203
204.macro SAVE_ALL pt_regs_ax=%eax switch_stacks=0
205	cld
206	PUSH_GS
207	pushl	%fs
208	pushl	%es
209	pushl	%ds
210	pushl	\pt_regs_ax
211	pushl	%ebp
212	pushl	%edi
213	pushl	%esi
214	pushl	%edx
215	pushl	%ecx
216	pushl	%ebx
217	movl	$(__USER_DS), %edx
218	movl	%edx, %ds
219	movl	%edx, %es
220	movl	$(__KERNEL_PERCPU), %edx
221	movl	%edx, %fs
222	SET_KERNEL_GS %edx
223
224	/* Switch to kernel stack if necessary */
225.if \switch_stacks > 0
226	SWITCH_TO_KERNEL_STACK
227.endif
228
229.endm
230
231.macro SAVE_ALL_NMI cr3_reg:req
232	SAVE_ALL
233
234	BUG_IF_WRONG_CR3
235
236	/*
237	 * Now switch the CR3 when PTI is enabled.
238	 *
239	 * We can enter with either user or kernel cr3, the code will
240	 * store the old cr3 in \cr3_reg and switches to the kernel cr3
241	 * if necessary.
242	 */
243	SWITCH_TO_KERNEL_CR3 scratch_reg=\cr3_reg
244
245.Lend_\@:
246.endm
247
248/*
249 * This is a sneaky trick to help the unwinder find pt_regs on the stack.  The
250 * frame pointer is replaced with an encoded pointer to pt_regs.  The encoding
251 * is just clearing the MSB, which makes it an invalid stack address and is also
252 * a signal to the unwinder that it's a pt_regs pointer in disguise.
253 *
254 * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the
255 * original rbp.
256 */
257.macro ENCODE_FRAME_POINTER
258#ifdef CONFIG_FRAME_POINTER
259	mov %esp, %ebp
260	andl $0x7fffffff, %ebp
261#endif
262.endm
263
264.macro RESTORE_INT_REGS
265	popl	%ebx
266	popl	%ecx
267	popl	%edx
268	popl	%esi
269	popl	%edi
270	popl	%ebp
271	popl	%eax
272.endm
273
274.macro RESTORE_REGS pop=0
275	RESTORE_INT_REGS
2761:	popl	%ds
2772:	popl	%es
2783:	popl	%fs
279	POP_GS \pop
280.pushsection .fixup, "ax"
2814:	movl	$0, (%esp)
282	jmp	1b
2835:	movl	$0, (%esp)
284	jmp	2b
2856:	movl	$0, (%esp)
286	jmp	3b
287.popsection
288	_ASM_EXTABLE(1b, 4b)
289	_ASM_EXTABLE(2b, 5b)
290	_ASM_EXTABLE(3b, 6b)
291	POP_GS_EX
292.endm
293
294.macro RESTORE_ALL_NMI cr3_reg:req pop=0
295	/*
296	 * Now switch the CR3 when PTI is enabled.
297	 *
298	 * We enter with kernel cr3 and switch the cr3 to the value
299	 * stored on \cr3_reg, which is either a user or a kernel cr3.
300	 */
301	ALTERNATIVE "jmp .Lswitched_\@", "", X86_FEATURE_PTI
302
303	testl	$PTI_SWITCH_MASK, \cr3_reg
304	jz	.Lswitched_\@
305
306	/* User cr3 in \cr3_reg - write it to hardware cr3 */
307	movl	\cr3_reg, %cr3
308
309.Lswitched_\@:
310
311	BUG_IF_WRONG_CR3
312
313	RESTORE_REGS pop=\pop
314.endm
315
316.macro CHECK_AND_APPLY_ESPFIX
317#ifdef CONFIG_X86_ESPFIX32
318#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
319
320	ALTERNATIVE	"jmp .Lend_\@", "", X86_BUG_ESPFIX
321
322	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
323	/*
324	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
325	 * are returning to the kernel.
326	 * See comments in process.c:copy_thread() for details.
327	 */
328	movb	PT_OLDSS(%esp), %ah
329	movb	PT_CS(%esp), %al
330	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
331	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
332	jne	.Lend_\@	# returning to user-space with LDT SS
333
334	/*
335	 * Setup and switch to ESPFIX stack
336	 *
337	 * We're returning to userspace with a 16 bit stack. The CPU will not
338	 * restore the high word of ESP for us on executing iret... This is an
339	 * "official" bug of all the x86-compatible CPUs, which we can work
340	 * around to make dosemu and wine happy. We do this by preloading the
341	 * high word of ESP with the high word of the userspace ESP while
342	 * compensating for the offset by changing to the ESPFIX segment with
343	 * a base address that matches for the difference.
344	 */
345	mov	%esp, %edx			/* load kernel esp */
346	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
347	mov	%dx, %ax			/* eax: new kernel esp */
348	sub	%eax, %edx			/* offset (low word is 0) */
349	shr	$16, %edx
350	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
351	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
352	pushl	$__ESPFIX_SS
353	pushl	%eax				/* new kernel esp */
354	/*
355	 * Disable interrupts, but do not irqtrace this section: we
356	 * will soon execute iret and the tracer was already set to
357	 * the irqstate after the IRET:
358	 */
359	DISABLE_INTERRUPTS(CLBR_ANY)
360	lss	(%esp), %esp			/* switch to espfix segment */
361.Lend_\@:
362#endif /* CONFIG_X86_ESPFIX32 */
363.endm
364
365/*
366 * Called with pt_regs fully populated and kernel segments loaded,
367 * so we can access PER_CPU and use the integer registers.
368 *
369 * We need to be very careful here with the %esp switch, because an NMI
370 * can happen everywhere. If the NMI handler finds itself on the
371 * entry-stack, it will overwrite the task-stack and everything we
372 * copied there. So allocate the stack-frame on the task-stack and
373 * switch to it before we do any copying.
374 */
375
376#define CS_FROM_ENTRY_STACK	(1 << 31)
377#define CS_FROM_USER_CR3	(1 << 30)
378
379.macro SWITCH_TO_KERNEL_STACK
380
381	ALTERNATIVE     "", "jmp .Lend_\@", X86_FEATURE_XENPV
382
383	BUG_IF_WRONG_CR3
384
385	SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
386
387	/*
388	 * %eax now contains the entry cr3 and we carry it forward in
389	 * that register for the time this macro runs
390	 */
391
392	/* Are we on the entry stack? Bail out if not! */
393	movl	PER_CPU_VAR(cpu_entry_area), %ecx
394	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
395	subl	%esp, %ecx	/* ecx = (end of entry_stack) - esp */
396	cmpl	$SIZEOF_entry_stack, %ecx
397	jae	.Lend_\@
398
399	/* Load stack pointer into %esi and %edi */
400	movl	%esp, %esi
401	movl	%esi, %edi
402
403	/* Move %edi to the top of the entry stack */
404	andl	$(MASK_entry_stack), %edi
405	addl	$(SIZEOF_entry_stack), %edi
406
407	/* Load top of task-stack into %edi */
408	movl	TSS_entry2task_stack(%edi), %edi
409
410	/*
411	 * Clear unused upper bits of the dword containing the word-sized CS
412	 * slot in pt_regs in case hardware didn't clear it for us.
413	 */
414	andl	$(0x0000ffff), PT_CS(%esp)
415
416	/* Special case - entry from kernel mode via entry stack */
417#ifdef CONFIG_VM86
418	movl	PT_EFLAGS(%esp), %ecx		# mix EFLAGS and CS
419	movb	PT_CS(%esp), %cl
420	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %ecx
421#else
422	movl	PT_CS(%esp), %ecx
423	andl	$SEGMENT_RPL_MASK, %ecx
424#endif
425	cmpl	$USER_RPL, %ecx
426	jb	.Lentry_from_kernel_\@
427
428	/* Bytes to copy */
429	movl	$PTREGS_SIZE, %ecx
430
431#ifdef CONFIG_VM86
432	testl	$X86_EFLAGS_VM, PT_EFLAGS(%esi)
433	jz	.Lcopy_pt_regs_\@
434
435	/*
436	 * Stack-frame contains 4 additional segment registers when
437	 * coming from VM86 mode
438	 */
439	addl	$(4 * 4), %ecx
440
441#endif
442.Lcopy_pt_regs_\@:
443
444	/* Allocate frame on task-stack */
445	subl	%ecx, %edi
446
447	/* Switch to task-stack */
448	movl	%edi, %esp
449
450	/*
451	 * We are now on the task-stack and can safely copy over the
452	 * stack-frame
453	 */
454	shrl	$2, %ecx
455	cld
456	rep movsl
457
458	jmp .Lend_\@
459
460.Lentry_from_kernel_\@:
461
462	/*
463	 * This handles the case when we enter the kernel from
464	 * kernel-mode and %esp points to the entry-stack. When this
465	 * happens we need to switch to the task-stack to run C code,
466	 * but switch back to the entry-stack again when we approach
467	 * iret and return to the interrupted code-path. This usually
468	 * happens when we hit an exception while restoring user-space
469	 * segment registers on the way back to user-space or when the
470	 * sysenter handler runs with eflags.tf set.
471	 *
472	 * When we switch to the task-stack here, we can't trust the
473	 * contents of the entry-stack anymore, as the exception handler
474	 * might be scheduled out or moved to another CPU. Therefore we
475	 * copy the complete entry-stack to the task-stack and set a
476	 * marker in the iret-frame (bit 31 of the CS dword) to detect
477	 * what we've done on the iret path.
478	 *
479	 * On the iret path we copy everything back and switch to the
480	 * entry-stack, so that the interrupted kernel code-path
481	 * continues on the same stack it was interrupted with.
482	 *
483	 * Be aware that an NMI can happen anytime in this code.
484	 *
485	 * %esi: Entry-Stack pointer (same as %esp)
486	 * %edi: Top of the task stack
487	 * %eax: CR3 on kernel entry
488	 */
489
490	/* Calculate number of bytes on the entry stack in %ecx */
491	movl	%esi, %ecx
492
493	/* %ecx to the top of entry-stack */
494	andl	$(MASK_entry_stack), %ecx
495	addl	$(SIZEOF_entry_stack), %ecx
496
497	/* Number of bytes on the entry stack to %ecx */
498	sub	%esi, %ecx
499
500	/* Mark stackframe as coming from entry stack */
501	orl	$CS_FROM_ENTRY_STACK, PT_CS(%esp)
502
503	/*
504	 * Test the cr3 used to enter the kernel and add a marker
505	 * so that we can switch back to it before iret.
506	 */
507	testl	$PTI_SWITCH_MASK, %eax
508	jz	.Lcopy_pt_regs_\@
509	orl	$CS_FROM_USER_CR3, PT_CS(%esp)
510
511	/*
512	 * %esi and %edi are unchanged, %ecx contains the number of
513	 * bytes to copy. The code at .Lcopy_pt_regs_\@ will allocate
514	 * the stack-frame on task-stack and copy everything over
515	 */
516	jmp .Lcopy_pt_regs_\@
517
518.Lend_\@:
519.endm
520
521/*
522 * Switch back from the kernel stack to the entry stack.
523 *
524 * The %esp register must point to pt_regs on the task stack. It will
525 * first calculate the size of the stack-frame to copy, depending on
526 * whether we return to VM86 mode or not. With that it uses 'rep movsl'
527 * to copy the contents of the stack over to the entry stack.
528 *
529 * We must be very careful here, as we can't trust the contents of the
530 * task-stack once we switched to the entry-stack. When an NMI happens
531 * while on the entry-stack, the NMI handler will switch back to the top
532 * of the task stack, overwriting our stack-frame we are about to copy.
533 * Therefore we switch the stack only after everything is copied over.
534 */
535.macro SWITCH_TO_ENTRY_STACK
536
537	ALTERNATIVE     "", "jmp .Lend_\@", X86_FEATURE_XENPV
538
539	/* Bytes to copy */
540	movl	$PTREGS_SIZE, %ecx
541
542#ifdef CONFIG_VM86
543	testl	$(X86_EFLAGS_VM), PT_EFLAGS(%esp)
544	jz	.Lcopy_pt_regs_\@
545
546	/* Additional 4 registers to copy when returning to VM86 mode */
547	addl    $(4 * 4), %ecx
548
549.Lcopy_pt_regs_\@:
550#endif
551
552	/* Initialize source and destination for movsl */
553	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
554	subl	%ecx, %edi
555	movl	%esp, %esi
556
557	/* Save future stack pointer in %ebx */
558	movl	%edi, %ebx
559
560	/* Copy over the stack-frame */
561	shrl	$2, %ecx
562	cld
563	rep movsl
564
565	/*
566	 * Switch to entry-stack - needs to happen after everything is
567	 * copied because the NMI handler will overwrite the task-stack
568	 * when on entry-stack
569	 */
570	movl	%ebx, %esp
571
572.Lend_\@:
573.endm
574
575/*
576 * This macro handles the case when we return to kernel-mode on the iret
577 * path and have to switch back to the entry stack and/or user-cr3
578 *
579 * See the comments below the .Lentry_from_kernel_\@ label in the
580 * SWITCH_TO_KERNEL_STACK macro for more details.
581 */
582.macro PARANOID_EXIT_TO_KERNEL_MODE
583
584	/*
585	 * Test if we entered the kernel with the entry-stack. Most
586	 * likely we did not, because this code only runs on the
587	 * return-to-kernel path.
588	 */
589	testl	$CS_FROM_ENTRY_STACK, PT_CS(%esp)
590	jz	.Lend_\@
591
592	/* Unlikely slow-path */
593
594	/* Clear marker from stack-frame */
595	andl	$(~CS_FROM_ENTRY_STACK), PT_CS(%esp)
596
597	/* Copy the remaining task-stack contents to entry-stack */
598	movl	%esp, %esi
599	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
600
601	/* Bytes on the task-stack to ecx */
602	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp1), %ecx
603	subl	%esi, %ecx
604
605	/* Allocate stack-frame on entry-stack */
606	subl	%ecx, %edi
607
608	/*
609	 * Save future stack-pointer, we must not switch until the
610	 * copy is done, otherwise the NMI handler could destroy the
611	 * contents of the task-stack we are about to copy.
612	 */
613	movl	%edi, %ebx
614
615	/* Do the copy */
616	shrl	$2, %ecx
617	cld
618	rep movsl
619
620	/* Safe to switch to entry-stack now */
621	movl	%ebx, %esp
622
623	/*
624	 * We came from entry-stack and need to check if we also need to
625	 * switch back to user cr3.
626	 */
627	testl	$CS_FROM_USER_CR3, PT_CS(%esp)
628	jz	.Lend_\@
629
630	/* Clear marker from stack-frame */
631	andl	$(~CS_FROM_USER_CR3), PT_CS(%esp)
632
633	SWITCH_TO_USER_CR3 scratch_reg=%eax
634
635.Lend_\@:
636.endm
637/*
638 * %eax: prev task
639 * %edx: next task
640 */
641ENTRY(__switch_to_asm)
642	/*
643	 * Save callee-saved registers
644	 * This must match the order in struct inactive_task_frame
645	 */
646	pushl	%ebp
647	pushl	%ebx
648	pushl	%edi
649	pushl	%esi
650
651	/* switch stack */
652	movl	%esp, TASK_threadsp(%eax)
653	movl	TASK_threadsp(%edx), %esp
654
655#ifdef CONFIG_STACKPROTECTOR
656	movl	TASK_stack_canary(%edx), %ebx
657	movl	%ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
658#endif
659
660#ifdef CONFIG_RETPOLINE
661	/*
662	 * When switching from a shallower to a deeper call stack
663	 * the RSB may either underflow or use entries populated
664	 * with userspace addresses. On CPUs where those concerns
665	 * exist, overwrite the RSB with entries which capture
666	 * speculative execution to prevent attack.
667	 */
668	FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
669#endif
670
671	/* restore callee-saved registers */
672	popl	%esi
673	popl	%edi
674	popl	%ebx
675	popl	%ebp
676
677	jmp	__switch_to
678END(__switch_to_asm)
679
680/*
681 * The unwinder expects the last frame on the stack to always be at the same
682 * offset from the end of the page, which allows it to validate the stack.
683 * Calling schedule_tail() directly would break that convention because its an
684 * asmlinkage function so its argument has to be pushed on the stack.  This
685 * wrapper creates a proper "end of stack" frame header before the call.
686 */
687ENTRY(schedule_tail_wrapper)
688	FRAME_BEGIN
689
690	pushl	%eax
691	call	schedule_tail
692	popl	%eax
693
694	FRAME_END
695	ret
696ENDPROC(schedule_tail_wrapper)
697/*
698 * A newly forked process directly context switches into this address.
699 *
700 * eax: prev task we switched from
701 * ebx: kernel thread func (NULL for user thread)
702 * edi: kernel thread arg
703 */
704ENTRY(ret_from_fork)
705	call	schedule_tail_wrapper
706
707	testl	%ebx, %ebx
708	jnz	1f		/* kernel threads are uncommon */
709
7102:
711	/* When we fork, we trace the syscall return in the child, too. */
712	movl    %esp, %eax
713	call    syscall_return_slowpath
714	jmp     restore_all
715
716	/* kernel thread */
7171:	movl	%edi, %eax
718	CALL_NOSPEC %ebx
719	/*
720	 * A kernel thread is allowed to return here after successfully
721	 * calling do_execve().  Exit to userspace to complete the execve()
722	 * syscall.
723	 */
724	movl	$0, PT_EAX(%esp)
725	jmp	2b
726END(ret_from_fork)
727
728/*
729 * Return to user mode is not as complex as all this looks,
730 * but we want the default path for a system call return to
731 * go as quickly as possible which is why some of this is
732 * less clear than it otherwise should be.
733 */
734
735	# userspace resumption stub bypassing syscall exit tracing
736	ALIGN
737ret_from_exception:
738	preempt_stop(CLBR_ANY)
739ret_from_intr:
740#ifdef CONFIG_VM86
741	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
742	movb	PT_CS(%esp), %al
743	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
744#else
745	/*
746	 * We can be coming here from child spawned by kernel_thread().
747	 */
748	movl	PT_CS(%esp), %eax
749	andl	$SEGMENT_RPL_MASK, %eax
750#endif
751	cmpl	$USER_RPL, %eax
752	jb	resume_kernel			# not returning to v8086 or userspace
753
754ENTRY(resume_userspace)
755	DISABLE_INTERRUPTS(CLBR_ANY)
756	TRACE_IRQS_OFF
757	movl	%esp, %eax
758	call	prepare_exit_to_usermode
759	jmp	restore_all
760END(ret_from_exception)
761
762#ifdef CONFIG_PREEMPT
763ENTRY(resume_kernel)
764	DISABLE_INTERRUPTS(CLBR_ANY)
765.Lneed_resched:
766	cmpl	$0, PER_CPU_VAR(__preempt_count)
767	jnz	restore_all_kernel
768	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
769	jz	restore_all_kernel
770	call	preempt_schedule_irq
771	jmp	.Lneed_resched
772END(resume_kernel)
773#endif
774
775GLOBAL(__begin_SYSENTER_singlestep_region)
776/*
777 * All code from here through __end_SYSENTER_singlestep_region is subject
778 * to being single-stepped if a user program sets TF and executes SYSENTER.
779 * There is absolutely nothing that we can do to prevent this from happening
780 * (thanks Intel!).  To keep our handling of this situation as simple as
781 * possible, we handle TF just like AC and NT, except that our #DB handler
782 * will ignore all of the single-step traps generated in this range.
783 */
784
785#ifdef CONFIG_XEN
786/*
787 * Xen doesn't set %esp to be precisely what the normal SYSENTER
788 * entry point expects, so fix it up before using the normal path.
789 */
790ENTRY(xen_sysenter_target)
791	addl	$5*4, %esp			/* remove xen-provided frame */
792	jmp	.Lsysenter_past_esp
793#endif
794
795/*
796 * 32-bit SYSENTER entry.
797 *
798 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
799 * if X86_FEATURE_SEP is available.  This is the preferred system call
800 * entry on 32-bit systems.
801 *
802 * The SYSENTER instruction, in principle, should *only* occur in the
803 * vDSO.  In practice, a small number of Android devices were shipped
804 * with a copy of Bionic that inlined a SYSENTER instruction.  This
805 * never happened in any of Google's Bionic versions -- it only happened
806 * in a narrow range of Intel-provided versions.
807 *
808 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
809 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
810 * SYSENTER does not save anything on the stack,
811 * and does not save old EIP (!!!), ESP, or EFLAGS.
812 *
813 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
814 * user and/or vm86 state), we explicitly disable the SYSENTER
815 * instruction in vm86 mode by reprogramming the MSRs.
816 *
817 * Arguments:
818 * eax  system call number
819 * ebx  arg1
820 * ecx  arg2
821 * edx  arg3
822 * esi  arg4
823 * edi  arg5
824 * ebp  user stack
825 * 0(%ebp) arg6
826 */
827ENTRY(entry_SYSENTER_32)
828	/*
829	 * On entry-stack with all userspace-regs live - save and
830	 * restore eflags and %eax to use it as scratch-reg for the cr3
831	 * switch.
832	 */
833	pushfl
834	pushl	%eax
835	BUG_IF_WRONG_CR3 no_user_check=1
836	SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
837	popl	%eax
838	popfl
839
840	/* Stack empty again, switch to task stack */
841	movl	TSS_entry2task_stack(%esp), %esp
842
843.Lsysenter_past_esp:
844	pushl	$__USER_DS		/* pt_regs->ss */
845	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
846	pushfl				/* pt_regs->flags (except IF = 0) */
847	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
848	pushl	$__USER_CS		/* pt_regs->cs */
849	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
850	pushl	%eax			/* pt_regs->orig_ax */
851	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest, stack already switched */
852
853	/*
854	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
855	 * and TF ourselves.  To save a few cycles, we can check whether
856	 * either was set instead of doing an unconditional popfq.
857	 * This needs to happen before enabling interrupts so that
858	 * we don't get preempted with NT set.
859	 *
860	 * If TF is set, we will single-step all the way to here -- do_debug
861	 * will ignore all the traps.  (Yes, this is slow, but so is
862	 * single-stepping in general.  This allows us to avoid having
863	 * a more complicated code to handle the case where a user program
864	 * forces us to single-step through the SYSENTER entry code.)
865	 *
866	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
867	 * out-of-line as an optimization: NT is unlikely to be set in the
868	 * majority of the cases and instead of polluting the I$ unnecessarily,
869	 * we're keeping that code behind a branch which will predict as
870	 * not-taken and therefore its instructions won't be fetched.
871	 */
872	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
873	jnz	.Lsysenter_fix_flags
874.Lsysenter_flags_fixed:
875
876	/*
877	 * User mode is traced as though IRQs are on, and SYSENTER
878	 * turned them off.
879	 */
880	TRACE_IRQS_OFF
881
882	movl	%esp, %eax
883	call	do_fast_syscall_32
884	/* XEN PV guests always use IRET path */
885	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
886		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
887
888/* Opportunistic SYSEXIT */
889	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
890
891	/*
892	 * Setup entry stack - we keep the pointer in %eax and do the
893	 * switch after almost all user-state is restored.
894	 */
895
896	/* Load entry stack pointer and allocate frame for eflags/eax */
897	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %eax
898	subl	$(2*4), %eax
899
900	/* Copy eflags and eax to entry stack */
901	movl	PT_EFLAGS(%esp), %edi
902	movl	PT_EAX(%esp), %esi
903	movl	%edi, (%eax)
904	movl	%esi, 4(%eax)
905
906	/* Restore user registers and segments */
907	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
908	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
9091:	mov	PT_FS(%esp), %fs
910	PTGS_TO_GS
911
912	popl	%ebx			/* pt_regs->bx */
913	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
914	popl	%esi			/* pt_regs->si */
915	popl	%edi			/* pt_regs->di */
916	popl	%ebp			/* pt_regs->bp */
917
918	/* Switch to entry stack */
919	movl	%eax, %esp
920
921	/* Now ready to switch the cr3 */
922	SWITCH_TO_USER_CR3 scratch_reg=%eax
923
924	/*
925	 * Restore all flags except IF. (We restore IF separately because
926	 * STI gives a one-instruction window in which we won't be interrupted,
927	 * whereas POPF does not.)
928	 */
929	btrl	$X86_EFLAGS_IF_BIT, (%esp)
930	BUG_IF_WRONG_CR3 no_user_check=1
931	popfl
932	popl	%eax
933
934	/*
935	 * Return back to the vDSO, which will pop ecx and edx.
936	 * Don't bother with DS and ES (they already contain __USER_DS).
937	 */
938	sti
939	sysexit
940
941.pushsection .fixup, "ax"
9422:	movl	$0, PT_FS(%esp)
943	jmp	1b
944.popsection
945	_ASM_EXTABLE(1b, 2b)
946	PTGS_TO_GS_EX
947
948.Lsysenter_fix_flags:
949	pushl	$X86_EFLAGS_FIXED
950	popfl
951	jmp	.Lsysenter_flags_fixed
952GLOBAL(__end_SYSENTER_singlestep_region)
953ENDPROC(entry_SYSENTER_32)
954
955/*
956 * 32-bit legacy system call entry.
957 *
958 * 32-bit x86 Linux system calls traditionally used the INT $0x80
959 * instruction.  INT $0x80 lands here.
960 *
961 * This entry point can be used by any 32-bit perform system calls.
962 * Instances of INT $0x80 can be found inline in various programs and
963 * libraries.  It is also used by the vDSO's __kernel_vsyscall
964 * fallback for hardware that doesn't support a faster entry method.
965 * Restarted 32-bit system calls also fall back to INT $0x80
966 * regardless of what instruction was originally used to do the system
967 * call.  (64-bit programs can use INT $0x80 as well, but they can
968 * only run on 64-bit kernels and therefore land in
969 * entry_INT80_compat.)
970 *
971 * This is considered a slow path.  It is not used by most libc
972 * implementations on modern hardware except during process startup.
973 *
974 * Arguments:
975 * eax  system call number
976 * ebx  arg1
977 * ecx  arg2
978 * edx  arg3
979 * esi  arg4
980 * edi  arg5
981 * ebp  arg6
982 */
983ENTRY(entry_INT80_32)
984	ASM_CLAC
985	pushl	%eax			/* pt_regs->orig_ax */
986
987	SAVE_ALL pt_regs_ax=$-ENOSYS switch_stacks=1	/* save rest */
988
989	/*
990	 * User mode is traced as though IRQs are on, and the interrupt gate
991	 * turned them off.
992	 */
993	TRACE_IRQS_OFF
994
995	movl	%esp, %eax
996	call	do_int80_syscall_32
997.Lsyscall_32_done:
998
999restore_all:
1000	TRACE_IRQS_IRET
1001	SWITCH_TO_ENTRY_STACK
1002.Lrestore_all_notrace:
1003	CHECK_AND_APPLY_ESPFIX
1004.Lrestore_nocheck:
1005	/* Switch back to user CR3 */
1006	SWITCH_TO_USER_CR3 scratch_reg=%eax
1007
1008	BUG_IF_WRONG_CR3
1009
1010	/* Restore user state */
1011	RESTORE_REGS pop=4			# skip orig_eax/error_code
1012.Lirq_return:
1013	/*
1014	 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
1015	 * when returning from IPI handler and when returning from
1016	 * scheduler to user-space.
1017	 */
1018	INTERRUPT_RETURN
1019
1020restore_all_kernel:
1021	TRACE_IRQS_IRET
1022	PARANOID_EXIT_TO_KERNEL_MODE
1023	BUG_IF_WRONG_CR3
1024	RESTORE_REGS 4
1025	jmp	.Lirq_return
1026
1027.section .fixup, "ax"
1028ENTRY(iret_exc	)
1029	pushl	$0				# no error code
1030	pushl	$do_iret_error
1031
1032#ifdef CONFIG_DEBUG_ENTRY
1033	/*
1034	 * The stack-frame here is the one that iret faulted on, so its a
1035	 * return-to-user frame. We are on kernel-cr3 because we come here from
1036	 * the fixup code. This confuses the CR3 checker, so switch to user-cr3
1037	 * as the checker expects it.
1038	 */
1039	pushl	%eax
1040	SWITCH_TO_USER_CR3 scratch_reg=%eax
1041	popl	%eax
1042#endif
1043
1044	jmp	common_exception
1045.previous
1046	_ASM_EXTABLE(.Lirq_return, iret_exc)
1047ENDPROC(entry_INT80_32)
1048
1049.macro FIXUP_ESPFIX_STACK
1050/*
1051 * Switch back for ESPFIX stack to the normal zerobased stack
1052 *
1053 * We can't call C functions using the ESPFIX stack. This code reads
1054 * the high word of the segment base from the GDT and swiches to the
1055 * normal stack and adjusts ESP with the matching offset.
1056 */
1057#ifdef CONFIG_X86_ESPFIX32
1058	/* fixup the stack */
1059	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
1060	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
1061	shl	$16, %eax
1062	addl	%esp, %eax			/* the adjusted stack pointer */
1063	pushl	$__KERNEL_DS
1064	pushl	%eax
1065	lss	(%esp), %esp			/* switch to the normal stack segment */
1066#endif
1067.endm
1068.macro UNWIND_ESPFIX_STACK
1069#ifdef CONFIG_X86_ESPFIX32
1070	movl	%ss, %eax
1071	/* see if on espfix stack */
1072	cmpw	$__ESPFIX_SS, %ax
1073	jne	27f
1074	movl	$__KERNEL_DS, %eax
1075	movl	%eax, %ds
1076	movl	%eax, %es
1077	/* switch to normal stack */
1078	FIXUP_ESPFIX_STACK
107927:
1080#endif
1081.endm
1082
1083/*
1084 * Build the entry stubs with some assembler magic.
1085 * We pack 1 stub into every 8-byte block.
1086 */
1087	.align 8
1088ENTRY(irq_entries_start)
1089    vector=FIRST_EXTERNAL_VECTOR
1090    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
1091	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
1092    vector=vector+1
1093	jmp	common_interrupt
1094	.align	8
1095    .endr
1096END(irq_entries_start)
1097
1098/*
1099 * the CPU automatically disables interrupts when executing an IRQ vector,
1100 * so IRQ-flags tracing has to follow that:
1101 */
1102	.p2align CONFIG_X86_L1_CACHE_SHIFT
1103common_interrupt:
1104	ASM_CLAC
1105	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
1106
1107	SAVE_ALL switch_stacks=1
1108	ENCODE_FRAME_POINTER
1109	TRACE_IRQS_OFF
1110	movl	%esp, %eax
1111	call	do_IRQ
1112	jmp	ret_from_intr
1113ENDPROC(common_interrupt)
1114
1115#define BUILD_INTERRUPT3(name, nr, fn)			\
1116ENTRY(name)						\
1117	ASM_CLAC;					\
1118	pushl	$~(nr);					\
1119	SAVE_ALL switch_stacks=1;			\
1120	ENCODE_FRAME_POINTER;				\
1121	TRACE_IRQS_OFF					\
1122	movl	%esp, %eax;				\
1123	call	fn;					\
1124	jmp	ret_from_intr;				\
1125ENDPROC(name)
1126
1127#define BUILD_INTERRUPT(name, nr)		\
1128	BUILD_INTERRUPT3(name, nr, smp_##name);	\
1129
1130/* The include is where all of the SMP etc. interrupts come from */
1131#include <asm/entry_arch.h>
1132
1133ENTRY(coprocessor_error)
1134	ASM_CLAC
1135	pushl	$0
1136	pushl	$do_coprocessor_error
1137	jmp	common_exception
1138END(coprocessor_error)
1139
1140ENTRY(simd_coprocessor_error)
1141	ASM_CLAC
1142	pushl	$0
1143#ifdef CONFIG_X86_INVD_BUG
1144	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
1145	ALTERNATIVE "pushl	$do_general_protection",	\
1146		    "pushl	$do_simd_coprocessor_error",	\
1147		    X86_FEATURE_XMM
1148#else
1149	pushl	$do_simd_coprocessor_error
1150#endif
1151	jmp	common_exception
1152END(simd_coprocessor_error)
1153
1154ENTRY(device_not_available)
1155	ASM_CLAC
1156	pushl	$-1				# mark this as an int
1157	pushl	$do_device_not_available
1158	jmp	common_exception
1159END(device_not_available)
1160
1161#ifdef CONFIG_PARAVIRT
1162ENTRY(native_iret)
1163	iret
1164	_ASM_EXTABLE(native_iret, iret_exc)
1165END(native_iret)
1166#endif
1167
1168ENTRY(overflow)
1169	ASM_CLAC
1170	pushl	$0
1171	pushl	$do_overflow
1172	jmp	common_exception
1173END(overflow)
1174
1175ENTRY(bounds)
1176	ASM_CLAC
1177	pushl	$0
1178	pushl	$do_bounds
1179	jmp	common_exception
1180END(bounds)
1181
1182ENTRY(invalid_op)
1183	ASM_CLAC
1184	pushl	$0
1185	pushl	$do_invalid_op
1186	jmp	common_exception
1187END(invalid_op)
1188
1189ENTRY(coprocessor_segment_overrun)
1190	ASM_CLAC
1191	pushl	$0
1192	pushl	$do_coprocessor_segment_overrun
1193	jmp	common_exception
1194END(coprocessor_segment_overrun)
1195
1196ENTRY(invalid_TSS)
1197	ASM_CLAC
1198	pushl	$do_invalid_TSS
1199	jmp	common_exception
1200END(invalid_TSS)
1201
1202ENTRY(segment_not_present)
1203	ASM_CLAC
1204	pushl	$do_segment_not_present
1205	jmp	common_exception
1206END(segment_not_present)
1207
1208ENTRY(stack_segment)
1209	ASM_CLAC
1210	pushl	$do_stack_segment
1211	jmp	common_exception
1212END(stack_segment)
1213
1214ENTRY(alignment_check)
1215	ASM_CLAC
1216	pushl	$do_alignment_check
1217	jmp	common_exception
1218END(alignment_check)
1219
1220ENTRY(divide_error)
1221	ASM_CLAC
1222	pushl	$0				# no error code
1223	pushl	$do_divide_error
1224	jmp	common_exception
1225END(divide_error)
1226
1227#ifdef CONFIG_X86_MCE
1228ENTRY(machine_check)
1229	ASM_CLAC
1230	pushl	$0
1231	pushl	machine_check_vector
1232	jmp	common_exception
1233END(machine_check)
1234#endif
1235
1236ENTRY(spurious_interrupt_bug)
1237	ASM_CLAC
1238	pushl	$0
1239	pushl	$do_spurious_interrupt_bug
1240	jmp	common_exception
1241END(spurious_interrupt_bug)
1242
1243#ifdef CONFIG_XEN
1244ENTRY(xen_hypervisor_callback)
1245	pushl	$-1				/* orig_ax = -1 => not a system call */
1246	SAVE_ALL
1247	ENCODE_FRAME_POINTER
1248	TRACE_IRQS_OFF
1249
1250	/*
1251	 * Check to see if we got the event in the critical
1252	 * region in xen_iret_direct, after we've reenabled
1253	 * events and checked for pending events.  This simulates
1254	 * iret instruction's behaviour where it delivers a
1255	 * pending interrupt when enabling interrupts:
1256	 */
1257	movl	PT_EIP(%esp), %eax
1258	cmpl	$xen_iret_start_crit, %eax
1259	jb	1f
1260	cmpl	$xen_iret_end_crit, %eax
1261	jae	1f
1262
1263	jmp	xen_iret_crit_fixup
1264
1265ENTRY(xen_do_upcall)
12661:	mov	%esp, %eax
1267	call	xen_evtchn_do_upcall
1268#ifndef CONFIG_PREEMPT
1269	call	xen_maybe_preempt_hcall
1270#endif
1271	jmp	ret_from_intr
1272ENDPROC(xen_hypervisor_callback)
1273
1274/*
1275 * Hypervisor uses this for application faults while it executes.
1276 * We get here for two reasons:
1277 *  1. Fault while reloading DS, ES, FS or GS
1278 *  2. Fault while executing IRET
1279 * Category 1 we fix up by reattempting the load, and zeroing the segment
1280 * register if the load fails.
1281 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
1282 * normal Linux return path in this case because if we use the IRET hypercall
1283 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1284 * We distinguish between categories by maintaining a status value in EAX.
1285 */
1286ENTRY(xen_failsafe_callback)
1287	pushl	%eax
1288	movl	$1, %eax
12891:	mov	4(%esp), %ds
12902:	mov	8(%esp), %es
12913:	mov	12(%esp), %fs
12924:	mov	16(%esp), %gs
1293	/* EAX == 0 => Category 1 (Bad segment)
1294	   EAX != 0 => Category 2 (Bad IRET) */
1295	testl	%eax, %eax
1296	popl	%eax
1297	lea	16(%esp), %esp
1298	jz	5f
1299	jmp	iret_exc
13005:	pushl	$-1				/* orig_ax = -1 => not a system call */
1301	SAVE_ALL
1302	ENCODE_FRAME_POINTER
1303	jmp	ret_from_exception
1304
1305.section .fixup, "ax"
13066:	xorl	%eax, %eax
1307	movl	%eax, 4(%esp)
1308	jmp	1b
13097:	xorl	%eax, %eax
1310	movl	%eax, 8(%esp)
1311	jmp	2b
13128:	xorl	%eax, %eax
1313	movl	%eax, 12(%esp)
1314	jmp	3b
13159:	xorl	%eax, %eax
1316	movl	%eax, 16(%esp)
1317	jmp	4b
1318.previous
1319	_ASM_EXTABLE(1b, 6b)
1320	_ASM_EXTABLE(2b, 7b)
1321	_ASM_EXTABLE(3b, 8b)
1322	_ASM_EXTABLE(4b, 9b)
1323ENDPROC(xen_failsafe_callback)
1324
1325BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1326		 xen_evtchn_do_upcall)
1327
1328#endif /* CONFIG_XEN */
1329
1330#if IS_ENABLED(CONFIG_HYPERV)
1331
1332BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1333		 hyperv_vector_handler)
1334
1335BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
1336		 hyperv_reenlightenment_intr)
1337
1338BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
1339		 hv_stimer0_vector_handler)
1340
1341#endif /* CONFIG_HYPERV */
1342
1343ENTRY(page_fault)
1344	ASM_CLAC
1345	pushl	$do_page_fault
1346	ALIGN
1347	jmp common_exception
1348END(page_fault)
1349
1350common_exception:
1351	/* the function address is in %gs's slot on the stack */
1352	pushl	%fs
1353	pushl	%es
1354	pushl	%ds
1355	pushl	%eax
1356	movl	$(__USER_DS), %eax
1357	movl	%eax, %ds
1358	movl	%eax, %es
1359	movl	$(__KERNEL_PERCPU), %eax
1360	movl	%eax, %fs
1361	pushl	%ebp
1362	pushl	%edi
1363	pushl	%esi
1364	pushl	%edx
1365	pushl	%ecx
1366	pushl	%ebx
1367	SWITCH_TO_KERNEL_STACK
1368	ENCODE_FRAME_POINTER
1369	cld
1370	UNWIND_ESPFIX_STACK
1371	GS_TO_REG %ecx
1372	movl	PT_GS(%esp), %edi		# get the function address
1373	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
1374	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
1375	REG_TO_PTGS %ecx
1376	SET_KERNEL_GS %ecx
1377	TRACE_IRQS_OFF
1378	movl	%esp, %eax			# pt_regs pointer
1379	CALL_NOSPEC %edi
1380	jmp	ret_from_exception
1381END(common_exception)
1382
1383ENTRY(debug)
1384	/*
1385	 * Entry from sysenter is now handled in common_exception
1386	 */
1387	ASM_CLAC
1388	pushl	$-1				# mark this as an int
1389	pushl	$do_debug
1390	jmp	common_exception
1391END(debug)
1392
1393/*
1394 * NMI is doubly nasty.  It can happen on the first instruction of
1395 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1396 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1397 * switched stacks.  We handle both conditions by simply checking whether we
1398 * interrupted kernel code running on the SYSENTER stack.
1399 */
1400ENTRY(nmi)
1401	ASM_CLAC
1402
1403#ifdef CONFIG_X86_ESPFIX32
1404	pushl	%eax
1405	movl	%ss, %eax
1406	cmpw	$__ESPFIX_SS, %ax
1407	popl	%eax
1408	je	.Lnmi_espfix_stack
1409#endif
1410
1411	pushl	%eax				# pt_regs->orig_ax
1412	SAVE_ALL_NMI cr3_reg=%edi
1413	ENCODE_FRAME_POINTER
1414	xorl	%edx, %edx			# zero error code
1415	movl	%esp, %eax			# pt_regs pointer
1416
1417	/* Are we currently on the SYSENTER stack? */
1418	movl	PER_CPU_VAR(cpu_entry_area), %ecx
1419	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1420	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
1421	cmpl	$SIZEOF_entry_stack, %ecx
1422	jb	.Lnmi_from_sysenter_stack
1423
1424	/* Not on SYSENTER stack. */
1425	call	do_nmi
1426	jmp	.Lnmi_return
1427
1428.Lnmi_from_sysenter_stack:
1429	/*
1430	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1431	 * is using the thread stack right now, so it's safe for us to use it.
1432	 */
1433	movl	%esp, %ebx
1434	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1435	call	do_nmi
1436	movl	%ebx, %esp
1437
1438.Lnmi_return:
1439	CHECK_AND_APPLY_ESPFIX
1440	RESTORE_ALL_NMI cr3_reg=%edi pop=4
1441	jmp	.Lirq_return
1442
1443#ifdef CONFIG_X86_ESPFIX32
1444.Lnmi_espfix_stack:
1445	/*
1446	 * create the pointer to lss back
1447	 */
1448	pushl	%ss
1449	pushl	%esp
1450	addl	$4, (%esp)
1451	/* copy the iret frame of 12 bytes */
1452	.rept 3
1453	pushl	16(%esp)
1454	.endr
1455	pushl	%eax
1456	SAVE_ALL_NMI cr3_reg=%edi
1457	ENCODE_FRAME_POINTER
1458	FIXUP_ESPFIX_STACK			# %eax == %esp
1459	xorl	%edx, %edx			# zero error code
1460	call	do_nmi
1461	RESTORE_ALL_NMI cr3_reg=%edi
1462	lss	12+4(%esp), %esp		# back to espfix stack
1463	jmp	.Lirq_return
1464#endif
1465END(nmi)
1466
1467ENTRY(int3)
1468	ASM_CLAC
1469	pushl	$-1				# mark this as an int
1470
1471	SAVE_ALL switch_stacks=1
1472	ENCODE_FRAME_POINTER
1473	TRACE_IRQS_OFF
1474	xorl	%edx, %edx			# zero error code
1475	movl	%esp, %eax			# pt_regs pointer
1476	call	do_int3
1477	jmp	ret_from_exception
1478END(int3)
1479
1480ENTRY(general_protection)
1481	pushl	$do_general_protection
1482	jmp	common_exception
1483END(general_protection)
1484
1485#ifdef CONFIG_KVM_GUEST
1486ENTRY(async_page_fault)
1487	ASM_CLAC
1488	pushl	$do_async_page_fault
1489	jmp	common_exception
1490END(async_page_fault)
1491#endif
1492
1493ENTRY(rewind_stack_do_exit)
1494	/* Prevent any naive code from trying to unwind to our caller. */
1495	xorl	%ebp, %ebp
1496
1497	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1498	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1499
1500	call	do_exit
15011:	jmp 1b
1502END(rewind_stack_do_exit)
1503