xref: /openbmc/linux/arch/x86/entry/entry_32.S (revision 67bf4745)
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#include "calling.h"
50
51	.section .entry.text, "ax"
52
53/*
54 * We use macros for low-level operations which need to be overridden
55 * for paravirtualization.  The following will never clobber any registers:
56 *   INTERRUPT_RETURN (aka. "iret")
57 *   GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
58 *   ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
59 *
60 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
61 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
62 * Allowing a register to be clobbered can shrink the paravirt replacement
63 * enough to patch inline, increasing performance.
64 */
65
66#ifdef CONFIG_PREEMPT
67# define preempt_stop(clobbers)	DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
68#else
69# define preempt_stop(clobbers)
70#endif
71
72.macro TRACE_IRQS_IRET
73#ifdef CONFIG_TRACE_IRQFLAGS
74	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)     # interrupts off?
75	jz	1f
76	TRACE_IRQS_ON
771:
78#endif
79.endm
80
81#define PTI_SWITCH_MASK         (1 << PAGE_SHIFT)
82
83/*
84 * User gs save/restore
85 *
86 * %gs is used for userland TLS and kernel only uses it for stack
87 * canary which is required to be at %gs:20 by gcc.  Read the comment
88 * at the top of stackprotector.h for more info.
89 *
90 * Local labels 98 and 99 are used.
91 */
92#ifdef CONFIG_X86_32_LAZY_GS
93
94 /* unfortunately push/pop can't be no-op */
95.macro PUSH_GS
96	pushl	$0
97.endm
98.macro POP_GS pop=0
99	addl	$(4 + \pop), %esp
100.endm
101.macro POP_GS_EX
102.endm
103
104 /* all the rest are no-op */
105.macro PTGS_TO_GS
106.endm
107.macro PTGS_TO_GS_EX
108.endm
109.macro GS_TO_REG reg
110.endm
111.macro REG_TO_PTGS reg
112.endm
113.macro SET_KERNEL_GS reg
114.endm
115
116#else	/* CONFIG_X86_32_LAZY_GS */
117
118.macro PUSH_GS
119	pushl	%gs
120.endm
121
122.macro POP_GS pop=0
12398:	popl	%gs
124  .if \pop <> 0
125	add	$\pop, %esp
126  .endif
127.endm
128.macro POP_GS_EX
129.pushsection .fixup, "ax"
13099:	movl	$0, (%esp)
131	jmp	98b
132.popsection
133	_ASM_EXTABLE(98b, 99b)
134.endm
135
136.macro PTGS_TO_GS
13798:	mov	PT_GS(%esp), %gs
138.endm
139.macro PTGS_TO_GS_EX
140.pushsection .fixup, "ax"
14199:	movl	$0, PT_GS(%esp)
142	jmp	98b
143.popsection
144	_ASM_EXTABLE(98b, 99b)
145.endm
146
147.macro GS_TO_REG reg
148	movl	%gs, \reg
149.endm
150.macro REG_TO_PTGS reg
151	movl	\reg, PT_GS(%esp)
152.endm
153.macro SET_KERNEL_GS reg
154	movl	$(__KERNEL_STACK_CANARY), \reg
155	movl	\reg, %gs
156.endm
157
158#endif /* CONFIG_X86_32_LAZY_GS */
159
160/* Unconditionally switch to user cr3 */
161.macro SWITCH_TO_USER_CR3 scratch_reg:req
162	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
163
164	movl	%cr3, \scratch_reg
165	orl	$PTI_SWITCH_MASK, \scratch_reg
166	movl	\scratch_reg, %cr3
167.Lend_\@:
168.endm
169
170.macro BUG_IF_WRONG_CR3 no_user_check=0
171#ifdef CONFIG_DEBUG_ENTRY
172	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
173	.if \no_user_check == 0
174	/* coming from usermode? */
175	testl	$SEGMENT_RPL_MASK, PT_CS(%esp)
176	jz	.Lend_\@
177	.endif
178	/* On user-cr3? */
179	movl	%cr3, %eax
180	testl	$PTI_SWITCH_MASK, %eax
181	jnz	.Lend_\@
182	/* From userspace with kernel cr3 - BUG */
183	ud2
184.Lend_\@:
185#endif
186.endm
187
188/*
189 * Switch to kernel cr3 if not already loaded and return current cr3 in
190 * \scratch_reg
191 */
192.macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
193	ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
194	movl	%cr3, \scratch_reg
195	/* Test if we are already on kernel CR3 */
196	testl	$PTI_SWITCH_MASK, \scratch_reg
197	jz	.Lend_\@
198	andl	$(~PTI_SWITCH_MASK), \scratch_reg
199	movl	\scratch_reg, %cr3
200	/* Return original CR3 in \scratch_reg */
201	orl	$PTI_SWITCH_MASK, \scratch_reg
202.Lend_\@:
203.endm
204
205#define CS_FROM_ENTRY_STACK	(1 << 31)
206#define CS_FROM_USER_CR3	(1 << 30)
207#define CS_FROM_KERNEL		(1 << 29)
208
209.macro FIXUP_FRAME
210	/*
211	 * The high bits of the CS dword (__csh) are used for CS_FROM_*.
212	 * Clear them in case hardware didn't do this for us.
213	 */
214	andl	$0x0000ffff, 3*4(%esp)
215
216#ifdef CONFIG_VM86
217	testl	$X86_EFLAGS_VM, 4*4(%esp)
218	jnz	.Lfrom_usermode_no_fixup_\@
219#endif
220	testl	$SEGMENT_RPL_MASK, 3*4(%esp)
221	jnz	.Lfrom_usermode_no_fixup_\@
222
223	orl	$CS_FROM_KERNEL, 3*4(%esp)
224
225	/*
226	 * When we're here from kernel mode; the (exception) stack looks like:
227	 *
228	 *  5*4(%esp) - <previous context>
229	 *  4*4(%esp) - flags
230	 *  3*4(%esp) - cs
231	 *  2*4(%esp) - ip
232	 *  1*4(%esp) - orig_eax
233	 *  0*4(%esp) - gs / function
234	 *
235	 * Lets build a 5 entry IRET frame after that, such that struct pt_regs
236	 * is complete and in particular regs->sp is correct. This gives us
237	 * the original 5 enties as gap:
238	 *
239	 * 12*4(%esp) - <previous context>
240	 * 11*4(%esp) - gap / flags
241	 * 10*4(%esp) - gap / cs
242	 *  9*4(%esp) - gap / ip
243	 *  8*4(%esp) - gap / orig_eax
244	 *  7*4(%esp) - gap / gs / function
245	 *  6*4(%esp) - ss
246	 *  5*4(%esp) - sp
247	 *  4*4(%esp) - flags
248	 *  3*4(%esp) - cs
249	 *  2*4(%esp) - ip
250	 *  1*4(%esp) - orig_eax
251	 *  0*4(%esp) - gs / function
252	 */
253
254	pushl	%ss		# ss
255	pushl	%esp		# sp (points at ss)
256	addl	$6*4, (%esp)	# point sp back at the previous context
257	pushl	6*4(%esp)	# flags
258	pushl	6*4(%esp)	# cs
259	pushl	6*4(%esp)	# ip
260	pushl	6*4(%esp)	# orig_eax
261	pushl	6*4(%esp)	# gs / function
262.Lfrom_usermode_no_fixup_\@:
263.endm
264
265.macro IRET_FRAME
266	testl $CS_FROM_KERNEL, 1*4(%esp)
267	jz .Lfinished_frame_\@
268
269	/*
270	 * Reconstruct the 3 entry IRET frame right after the (modified)
271	 * regs->sp without lowering %esp in between, such that an NMI in the
272	 * middle doesn't scribble our stack.
273	 */
274	pushl	%eax
275	pushl	%ecx
276	movl	5*4(%esp), %eax		# (modified) regs->sp
277
278	movl	4*4(%esp), %ecx		# flags
279	movl	%ecx, -4(%eax)
280
281	movl	3*4(%esp), %ecx		# cs
282	andl	$0x0000ffff, %ecx
283	movl	%ecx, -8(%eax)
284
285	movl	2*4(%esp), %ecx		# ip
286	movl	%ecx, -12(%eax)
287
288	movl	1*4(%esp), %ecx		# eax
289	movl	%ecx, -16(%eax)
290
291	popl	%ecx
292	lea	-16(%eax), %esp
293	popl	%eax
294.Lfinished_frame_\@:
295.endm
296
297.macro SAVE_ALL pt_regs_ax=%eax switch_stacks=0
298	cld
299	PUSH_GS
300	FIXUP_FRAME
301	pushl	%fs
302	pushl	%es
303	pushl	%ds
304	pushl	\pt_regs_ax
305	pushl	%ebp
306	pushl	%edi
307	pushl	%esi
308	pushl	%edx
309	pushl	%ecx
310	pushl	%ebx
311	movl	$(__USER_DS), %edx
312	movl	%edx, %ds
313	movl	%edx, %es
314	movl	$(__KERNEL_PERCPU), %edx
315	movl	%edx, %fs
316	SET_KERNEL_GS %edx
317
318	/* Switch to kernel stack if necessary */
319.if \switch_stacks > 0
320	SWITCH_TO_KERNEL_STACK
321.endif
322
323.endm
324
325.macro SAVE_ALL_NMI cr3_reg:req
326	SAVE_ALL
327
328	BUG_IF_WRONG_CR3
329
330	/*
331	 * Now switch the CR3 when PTI is enabled.
332	 *
333	 * We can enter with either user or kernel cr3, the code will
334	 * store the old cr3 in \cr3_reg and switches to the kernel cr3
335	 * if necessary.
336	 */
337	SWITCH_TO_KERNEL_CR3 scratch_reg=\cr3_reg
338
339.Lend_\@:
340.endm
341
342.macro RESTORE_INT_REGS
343	popl	%ebx
344	popl	%ecx
345	popl	%edx
346	popl	%esi
347	popl	%edi
348	popl	%ebp
349	popl	%eax
350.endm
351
352.macro RESTORE_REGS pop=0
353	RESTORE_INT_REGS
3541:	popl	%ds
3552:	popl	%es
3563:	popl	%fs
357	POP_GS \pop
358.pushsection .fixup, "ax"
3594:	movl	$0, (%esp)
360	jmp	1b
3615:	movl	$0, (%esp)
362	jmp	2b
3636:	movl	$0, (%esp)
364	jmp	3b
365.popsection
366	_ASM_EXTABLE(1b, 4b)
367	_ASM_EXTABLE(2b, 5b)
368	_ASM_EXTABLE(3b, 6b)
369	POP_GS_EX
370.endm
371
372.macro RESTORE_ALL_NMI cr3_reg:req pop=0
373	/*
374	 * Now switch the CR3 when PTI is enabled.
375	 *
376	 * We enter with kernel cr3 and switch the cr3 to the value
377	 * stored on \cr3_reg, which is either a user or a kernel cr3.
378	 */
379	ALTERNATIVE "jmp .Lswitched_\@", "", X86_FEATURE_PTI
380
381	testl	$PTI_SWITCH_MASK, \cr3_reg
382	jz	.Lswitched_\@
383
384	/* User cr3 in \cr3_reg - write it to hardware cr3 */
385	movl	\cr3_reg, %cr3
386
387.Lswitched_\@:
388
389	BUG_IF_WRONG_CR3
390
391	RESTORE_REGS pop=\pop
392.endm
393
394.macro CHECK_AND_APPLY_ESPFIX
395#ifdef CONFIG_X86_ESPFIX32
396#define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
397
398	ALTERNATIVE	"jmp .Lend_\@", "", X86_BUG_ESPFIX
399
400	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS, SS and CS
401	/*
402	 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
403	 * are returning to the kernel.
404	 * See comments in process.c:copy_thread() for details.
405	 */
406	movb	PT_OLDSS(%esp), %ah
407	movb	PT_CS(%esp), %al
408	andl	$(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
409	cmpl	$((SEGMENT_LDT << 8) | USER_RPL), %eax
410	jne	.Lend_\@	# returning to user-space with LDT SS
411
412	/*
413	 * Setup and switch to ESPFIX stack
414	 *
415	 * We're returning to userspace with a 16 bit stack. The CPU will not
416	 * restore the high word of ESP for us on executing iret... This is an
417	 * "official" bug of all the x86-compatible CPUs, which we can work
418	 * around to make dosemu and wine happy. We do this by preloading the
419	 * high word of ESP with the high word of the userspace ESP while
420	 * compensating for the offset by changing to the ESPFIX segment with
421	 * a base address that matches for the difference.
422	 */
423	mov	%esp, %edx			/* load kernel esp */
424	mov	PT_OLDESP(%esp), %eax		/* load userspace esp */
425	mov	%dx, %ax			/* eax: new kernel esp */
426	sub	%eax, %edx			/* offset (low word is 0) */
427	shr	$16, %edx
428	mov	%dl, GDT_ESPFIX_SS + 4		/* bits 16..23 */
429	mov	%dh, GDT_ESPFIX_SS + 7		/* bits 24..31 */
430	pushl	$__ESPFIX_SS
431	pushl	%eax				/* new kernel esp */
432	/*
433	 * Disable interrupts, but do not irqtrace this section: we
434	 * will soon execute iret and the tracer was already set to
435	 * the irqstate after the IRET:
436	 */
437	DISABLE_INTERRUPTS(CLBR_ANY)
438	lss	(%esp), %esp			/* switch to espfix segment */
439.Lend_\@:
440#endif /* CONFIG_X86_ESPFIX32 */
441.endm
442
443/*
444 * Called with pt_regs fully populated and kernel segments loaded,
445 * so we can access PER_CPU and use the integer registers.
446 *
447 * We need to be very careful here with the %esp switch, because an NMI
448 * can happen everywhere. If the NMI handler finds itself on the
449 * entry-stack, it will overwrite the task-stack and everything we
450 * copied there. So allocate the stack-frame on the task-stack and
451 * switch to it before we do any copying.
452 */
453
454.macro SWITCH_TO_KERNEL_STACK
455
456	ALTERNATIVE     "", "jmp .Lend_\@", X86_FEATURE_XENPV
457
458	BUG_IF_WRONG_CR3
459
460	SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
461
462	/*
463	 * %eax now contains the entry cr3 and we carry it forward in
464	 * that register for the time this macro runs
465	 */
466
467	/* Are we on the entry stack? Bail out if not! */
468	movl	PER_CPU_VAR(cpu_entry_area), %ecx
469	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
470	subl	%esp, %ecx	/* ecx = (end of entry_stack) - esp */
471	cmpl	$SIZEOF_entry_stack, %ecx
472	jae	.Lend_\@
473
474	/* Load stack pointer into %esi and %edi */
475	movl	%esp, %esi
476	movl	%esi, %edi
477
478	/* Move %edi to the top of the entry stack */
479	andl	$(MASK_entry_stack), %edi
480	addl	$(SIZEOF_entry_stack), %edi
481
482	/* Load top of task-stack into %edi */
483	movl	TSS_entry2task_stack(%edi), %edi
484
485	/* Special case - entry from kernel mode via entry stack */
486#ifdef CONFIG_VM86
487	movl	PT_EFLAGS(%esp), %ecx		# mix EFLAGS and CS
488	movb	PT_CS(%esp), %cl
489	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %ecx
490#else
491	movl	PT_CS(%esp), %ecx
492	andl	$SEGMENT_RPL_MASK, %ecx
493#endif
494	cmpl	$USER_RPL, %ecx
495	jb	.Lentry_from_kernel_\@
496
497	/* Bytes to copy */
498	movl	$PTREGS_SIZE, %ecx
499
500#ifdef CONFIG_VM86
501	testl	$X86_EFLAGS_VM, PT_EFLAGS(%esi)
502	jz	.Lcopy_pt_regs_\@
503
504	/*
505	 * Stack-frame contains 4 additional segment registers when
506	 * coming from VM86 mode
507	 */
508	addl	$(4 * 4), %ecx
509
510#endif
511.Lcopy_pt_regs_\@:
512
513	/* Allocate frame on task-stack */
514	subl	%ecx, %edi
515
516	/* Switch to task-stack */
517	movl	%edi, %esp
518
519	/*
520	 * We are now on the task-stack and can safely copy over the
521	 * stack-frame
522	 */
523	shrl	$2, %ecx
524	cld
525	rep movsl
526
527	jmp .Lend_\@
528
529.Lentry_from_kernel_\@:
530
531	/*
532	 * This handles the case when we enter the kernel from
533	 * kernel-mode and %esp points to the entry-stack. When this
534	 * happens we need to switch to the task-stack to run C code,
535	 * but switch back to the entry-stack again when we approach
536	 * iret and return to the interrupted code-path. This usually
537	 * happens when we hit an exception while restoring user-space
538	 * segment registers on the way back to user-space or when the
539	 * sysenter handler runs with eflags.tf set.
540	 *
541	 * When we switch to the task-stack here, we can't trust the
542	 * contents of the entry-stack anymore, as the exception handler
543	 * might be scheduled out or moved to another CPU. Therefore we
544	 * copy the complete entry-stack to the task-stack and set a
545	 * marker in the iret-frame (bit 31 of the CS dword) to detect
546	 * what we've done on the iret path.
547	 *
548	 * On the iret path we copy everything back and switch to the
549	 * entry-stack, so that the interrupted kernel code-path
550	 * continues on the same stack it was interrupted with.
551	 *
552	 * Be aware that an NMI can happen anytime in this code.
553	 *
554	 * %esi: Entry-Stack pointer (same as %esp)
555	 * %edi: Top of the task stack
556	 * %eax: CR3 on kernel entry
557	 */
558
559	/* Calculate number of bytes on the entry stack in %ecx */
560	movl	%esi, %ecx
561
562	/* %ecx to the top of entry-stack */
563	andl	$(MASK_entry_stack), %ecx
564	addl	$(SIZEOF_entry_stack), %ecx
565
566	/* Number of bytes on the entry stack to %ecx */
567	sub	%esi, %ecx
568
569	/* Mark stackframe as coming from entry stack */
570	orl	$CS_FROM_ENTRY_STACK, PT_CS(%esp)
571
572	/*
573	 * Test the cr3 used to enter the kernel and add a marker
574	 * so that we can switch back to it before iret.
575	 */
576	testl	$PTI_SWITCH_MASK, %eax
577	jz	.Lcopy_pt_regs_\@
578	orl	$CS_FROM_USER_CR3, PT_CS(%esp)
579
580	/*
581	 * %esi and %edi are unchanged, %ecx contains the number of
582	 * bytes to copy. The code at .Lcopy_pt_regs_\@ will allocate
583	 * the stack-frame on task-stack and copy everything over
584	 */
585	jmp .Lcopy_pt_regs_\@
586
587.Lend_\@:
588.endm
589
590/*
591 * Switch back from the kernel stack to the entry stack.
592 *
593 * The %esp register must point to pt_regs on the task stack. It will
594 * first calculate the size of the stack-frame to copy, depending on
595 * whether we return to VM86 mode or not. With that it uses 'rep movsl'
596 * to copy the contents of the stack over to the entry stack.
597 *
598 * We must be very careful here, as we can't trust the contents of the
599 * task-stack once we switched to the entry-stack. When an NMI happens
600 * while on the entry-stack, the NMI handler will switch back to the top
601 * of the task stack, overwriting our stack-frame we are about to copy.
602 * Therefore we switch the stack only after everything is copied over.
603 */
604.macro SWITCH_TO_ENTRY_STACK
605
606	ALTERNATIVE     "", "jmp .Lend_\@", X86_FEATURE_XENPV
607
608	/* Bytes to copy */
609	movl	$PTREGS_SIZE, %ecx
610
611#ifdef CONFIG_VM86
612	testl	$(X86_EFLAGS_VM), PT_EFLAGS(%esp)
613	jz	.Lcopy_pt_regs_\@
614
615	/* Additional 4 registers to copy when returning to VM86 mode */
616	addl    $(4 * 4), %ecx
617
618.Lcopy_pt_regs_\@:
619#endif
620
621	/* Initialize source and destination for movsl */
622	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
623	subl	%ecx, %edi
624	movl	%esp, %esi
625
626	/* Save future stack pointer in %ebx */
627	movl	%edi, %ebx
628
629	/* Copy over the stack-frame */
630	shrl	$2, %ecx
631	cld
632	rep movsl
633
634	/*
635	 * Switch to entry-stack - needs to happen after everything is
636	 * copied because the NMI handler will overwrite the task-stack
637	 * when on entry-stack
638	 */
639	movl	%ebx, %esp
640
641.Lend_\@:
642.endm
643
644/*
645 * This macro handles the case when we return to kernel-mode on the iret
646 * path and have to switch back to the entry stack and/or user-cr3
647 *
648 * See the comments below the .Lentry_from_kernel_\@ label in the
649 * SWITCH_TO_KERNEL_STACK macro for more details.
650 */
651.macro PARANOID_EXIT_TO_KERNEL_MODE
652
653	/*
654	 * Test if we entered the kernel with the entry-stack. Most
655	 * likely we did not, because this code only runs on the
656	 * return-to-kernel path.
657	 */
658	testl	$CS_FROM_ENTRY_STACK, PT_CS(%esp)
659	jz	.Lend_\@
660
661	/* Unlikely slow-path */
662
663	/* Clear marker from stack-frame */
664	andl	$(~CS_FROM_ENTRY_STACK), PT_CS(%esp)
665
666	/* Copy the remaining task-stack contents to entry-stack */
667	movl	%esp, %esi
668	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
669
670	/* Bytes on the task-stack to ecx */
671	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp1), %ecx
672	subl	%esi, %ecx
673
674	/* Allocate stack-frame on entry-stack */
675	subl	%ecx, %edi
676
677	/*
678	 * Save future stack-pointer, we must not switch until the
679	 * copy is done, otherwise the NMI handler could destroy the
680	 * contents of the task-stack we are about to copy.
681	 */
682	movl	%edi, %ebx
683
684	/* Do the copy */
685	shrl	$2, %ecx
686	cld
687	rep movsl
688
689	/* Safe to switch to entry-stack now */
690	movl	%ebx, %esp
691
692	/*
693	 * We came from entry-stack and need to check if we also need to
694	 * switch back to user cr3.
695	 */
696	testl	$CS_FROM_USER_CR3, PT_CS(%esp)
697	jz	.Lend_\@
698
699	/* Clear marker from stack-frame */
700	andl	$(~CS_FROM_USER_CR3), PT_CS(%esp)
701
702	SWITCH_TO_USER_CR3 scratch_reg=%eax
703
704.Lend_\@:
705.endm
706/*
707 * %eax: prev task
708 * %edx: next task
709 */
710ENTRY(__switch_to_asm)
711	/*
712	 * Save callee-saved registers
713	 * This must match the order in struct inactive_task_frame
714	 */
715	pushl	%ebp
716	pushl	%ebx
717	pushl	%edi
718	pushl	%esi
719	pushfl
720
721	/* switch stack */
722	movl	%esp, TASK_threadsp(%eax)
723	movl	TASK_threadsp(%edx), %esp
724
725#ifdef CONFIG_STACKPROTECTOR
726	movl	TASK_stack_canary(%edx), %ebx
727	movl	%ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
728#endif
729
730#ifdef CONFIG_RETPOLINE
731	/*
732	 * When switching from a shallower to a deeper call stack
733	 * the RSB may either underflow or use entries populated
734	 * with userspace addresses. On CPUs where those concerns
735	 * exist, overwrite the RSB with entries which capture
736	 * speculative execution to prevent attack.
737	 */
738	FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
739#endif
740
741	/* restore callee-saved registers */
742	popfl
743	popl	%esi
744	popl	%edi
745	popl	%ebx
746	popl	%ebp
747
748	jmp	__switch_to
749END(__switch_to_asm)
750
751/*
752 * The unwinder expects the last frame on the stack to always be at the same
753 * offset from the end of the page, which allows it to validate the stack.
754 * Calling schedule_tail() directly would break that convention because its an
755 * asmlinkage function so its argument has to be pushed on the stack.  This
756 * wrapper creates a proper "end of stack" frame header before the call.
757 */
758ENTRY(schedule_tail_wrapper)
759	FRAME_BEGIN
760
761	pushl	%eax
762	call	schedule_tail
763	popl	%eax
764
765	FRAME_END
766	ret
767ENDPROC(schedule_tail_wrapper)
768/*
769 * A newly forked process directly context switches into this address.
770 *
771 * eax: prev task we switched from
772 * ebx: kernel thread func (NULL for user thread)
773 * edi: kernel thread arg
774 */
775ENTRY(ret_from_fork)
776	call	schedule_tail_wrapper
777
778	testl	%ebx, %ebx
779	jnz	1f		/* kernel threads are uncommon */
780
7812:
782	/* When we fork, we trace the syscall return in the child, too. */
783	movl    %esp, %eax
784	call    syscall_return_slowpath
785	STACKLEAK_ERASE
786	jmp     restore_all
787
788	/* kernel thread */
7891:	movl	%edi, %eax
790	CALL_NOSPEC %ebx
791	/*
792	 * A kernel thread is allowed to return here after successfully
793	 * calling do_execve().  Exit to userspace to complete the execve()
794	 * syscall.
795	 */
796	movl	$0, PT_EAX(%esp)
797	jmp	2b
798END(ret_from_fork)
799
800/*
801 * Return to user mode is not as complex as all this looks,
802 * but we want the default path for a system call return to
803 * go as quickly as possible which is why some of this is
804 * less clear than it otherwise should be.
805 */
806
807	# userspace resumption stub bypassing syscall exit tracing
808	ALIGN
809ret_from_exception:
810	preempt_stop(CLBR_ANY)
811ret_from_intr:
812#ifdef CONFIG_VM86
813	movl	PT_EFLAGS(%esp), %eax		# mix EFLAGS and CS
814	movb	PT_CS(%esp), %al
815	andl	$(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
816#else
817	/*
818	 * We can be coming here from child spawned by kernel_thread().
819	 */
820	movl	PT_CS(%esp), %eax
821	andl	$SEGMENT_RPL_MASK, %eax
822#endif
823	cmpl	$USER_RPL, %eax
824	jb	restore_all_kernel		# not returning to v8086 or userspace
825
826ENTRY(resume_userspace)
827	DISABLE_INTERRUPTS(CLBR_ANY)
828	TRACE_IRQS_OFF
829	movl	%esp, %eax
830	call	prepare_exit_to_usermode
831	jmp	restore_all
832END(ret_from_exception)
833
834GLOBAL(__begin_SYSENTER_singlestep_region)
835/*
836 * All code from here through __end_SYSENTER_singlestep_region is subject
837 * to being single-stepped if a user program sets TF and executes SYSENTER.
838 * There is absolutely nothing that we can do to prevent this from happening
839 * (thanks Intel!).  To keep our handling of this situation as simple as
840 * possible, we handle TF just like AC and NT, except that our #DB handler
841 * will ignore all of the single-step traps generated in this range.
842 */
843
844#ifdef CONFIG_XEN_PV
845/*
846 * Xen doesn't set %esp to be precisely what the normal SYSENTER
847 * entry point expects, so fix it up before using the normal path.
848 */
849ENTRY(xen_sysenter_target)
850	addl	$5*4, %esp			/* remove xen-provided frame */
851	jmp	.Lsysenter_past_esp
852#endif
853
854/*
855 * 32-bit SYSENTER entry.
856 *
857 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
858 * if X86_FEATURE_SEP is available.  This is the preferred system call
859 * entry on 32-bit systems.
860 *
861 * The SYSENTER instruction, in principle, should *only* occur in the
862 * vDSO.  In practice, a small number of Android devices were shipped
863 * with a copy of Bionic that inlined a SYSENTER instruction.  This
864 * never happened in any of Google's Bionic versions -- it only happened
865 * in a narrow range of Intel-provided versions.
866 *
867 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
868 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
869 * SYSENTER does not save anything on the stack,
870 * and does not save old EIP (!!!), ESP, or EFLAGS.
871 *
872 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
873 * user and/or vm86 state), we explicitly disable the SYSENTER
874 * instruction in vm86 mode by reprogramming the MSRs.
875 *
876 * Arguments:
877 * eax  system call number
878 * ebx  arg1
879 * ecx  arg2
880 * edx  arg3
881 * esi  arg4
882 * edi  arg5
883 * ebp  user stack
884 * 0(%ebp) arg6
885 */
886ENTRY(entry_SYSENTER_32)
887	/*
888	 * On entry-stack with all userspace-regs live - save and
889	 * restore eflags and %eax to use it as scratch-reg for the cr3
890	 * switch.
891	 */
892	pushfl
893	pushl	%eax
894	BUG_IF_WRONG_CR3 no_user_check=1
895	SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
896	popl	%eax
897	popfl
898
899	/* Stack empty again, switch to task stack */
900	movl	TSS_entry2task_stack(%esp), %esp
901
902.Lsysenter_past_esp:
903	pushl	$__USER_DS		/* pt_regs->ss */
904	pushl	%ebp			/* pt_regs->sp (stashed in bp) */
905	pushfl				/* pt_regs->flags (except IF = 0) */
906	orl	$X86_EFLAGS_IF, (%esp)	/* Fix IF */
907	pushl	$__USER_CS		/* pt_regs->cs */
908	pushl	$0			/* pt_regs->ip = 0 (placeholder) */
909	pushl	%eax			/* pt_regs->orig_ax */
910	SAVE_ALL pt_regs_ax=$-ENOSYS	/* save rest, stack already switched */
911
912	/*
913	 * SYSENTER doesn't filter flags, so we need to clear NT, AC
914	 * and TF ourselves.  To save a few cycles, we can check whether
915	 * either was set instead of doing an unconditional popfq.
916	 * This needs to happen before enabling interrupts so that
917	 * we don't get preempted with NT set.
918	 *
919	 * If TF is set, we will single-step all the way to here -- do_debug
920	 * will ignore all the traps.  (Yes, this is slow, but so is
921	 * single-stepping in general.  This allows us to avoid having
922	 * a more complicated code to handle the case where a user program
923	 * forces us to single-step through the SYSENTER entry code.)
924	 *
925	 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
926	 * out-of-line as an optimization: NT is unlikely to be set in the
927	 * majority of the cases and instead of polluting the I$ unnecessarily,
928	 * we're keeping that code behind a branch which will predict as
929	 * not-taken and therefore its instructions won't be fetched.
930	 */
931	testl	$X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
932	jnz	.Lsysenter_fix_flags
933.Lsysenter_flags_fixed:
934
935	/*
936	 * User mode is traced as though IRQs are on, and SYSENTER
937	 * turned them off.
938	 */
939	TRACE_IRQS_OFF
940
941	movl	%esp, %eax
942	call	do_fast_syscall_32
943	/* XEN PV guests always use IRET path */
944	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
945		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
946
947	STACKLEAK_ERASE
948
949/* Opportunistic SYSEXIT */
950	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
951
952	/*
953	 * Setup entry stack - we keep the pointer in %eax and do the
954	 * switch after almost all user-state is restored.
955	 */
956
957	/* Load entry stack pointer and allocate frame for eflags/eax */
958	movl	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %eax
959	subl	$(2*4), %eax
960
961	/* Copy eflags and eax to entry stack */
962	movl	PT_EFLAGS(%esp), %edi
963	movl	PT_EAX(%esp), %esi
964	movl	%edi, (%eax)
965	movl	%esi, 4(%eax)
966
967	/* Restore user registers and segments */
968	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
969	movl	PT_OLDESP(%esp), %ecx	/* pt_regs->sp */
9701:	mov	PT_FS(%esp), %fs
971	PTGS_TO_GS
972
973	popl	%ebx			/* pt_regs->bx */
974	addl	$2*4, %esp		/* skip pt_regs->cx and pt_regs->dx */
975	popl	%esi			/* pt_regs->si */
976	popl	%edi			/* pt_regs->di */
977	popl	%ebp			/* pt_regs->bp */
978
979	/* Switch to entry stack */
980	movl	%eax, %esp
981
982	/* Now ready to switch the cr3 */
983	SWITCH_TO_USER_CR3 scratch_reg=%eax
984
985	/*
986	 * Restore all flags except IF. (We restore IF separately because
987	 * STI gives a one-instruction window in which we won't be interrupted,
988	 * whereas POPF does not.)
989	 */
990	btrl	$X86_EFLAGS_IF_BIT, (%esp)
991	BUG_IF_WRONG_CR3 no_user_check=1
992	popfl
993	popl	%eax
994
995	/*
996	 * Return back to the vDSO, which will pop ecx and edx.
997	 * Don't bother with DS and ES (they already contain __USER_DS).
998	 */
999	sti
1000	sysexit
1001
1002.pushsection .fixup, "ax"
10032:	movl	$0, PT_FS(%esp)
1004	jmp	1b
1005.popsection
1006	_ASM_EXTABLE(1b, 2b)
1007	PTGS_TO_GS_EX
1008
1009.Lsysenter_fix_flags:
1010	pushl	$X86_EFLAGS_FIXED
1011	popfl
1012	jmp	.Lsysenter_flags_fixed
1013GLOBAL(__end_SYSENTER_singlestep_region)
1014ENDPROC(entry_SYSENTER_32)
1015
1016/*
1017 * 32-bit legacy system call entry.
1018 *
1019 * 32-bit x86 Linux system calls traditionally used the INT $0x80
1020 * instruction.  INT $0x80 lands here.
1021 *
1022 * This entry point can be used by any 32-bit perform system calls.
1023 * Instances of INT $0x80 can be found inline in various programs and
1024 * libraries.  It is also used by the vDSO's __kernel_vsyscall
1025 * fallback for hardware that doesn't support a faster entry method.
1026 * Restarted 32-bit system calls also fall back to INT $0x80
1027 * regardless of what instruction was originally used to do the system
1028 * call.  (64-bit programs can use INT $0x80 as well, but they can
1029 * only run on 64-bit kernels and therefore land in
1030 * entry_INT80_compat.)
1031 *
1032 * This is considered a slow path.  It is not used by most libc
1033 * implementations on modern hardware except during process startup.
1034 *
1035 * Arguments:
1036 * eax  system call number
1037 * ebx  arg1
1038 * ecx  arg2
1039 * edx  arg3
1040 * esi  arg4
1041 * edi  arg5
1042 * ebp  arg6
1043 */
1044ENTRY(entry_INT80_32)
1045	ASM_CLAC
1046	pushl	%eax			/* pt_regs->orig_ax */
1047
1048	SAVE_ALL pt_regs_ax=$-ENOSYS switch_stacks=1	/* save rest */
1049
1050	/*
1051	 * User mode is traced as though IRQs are on, and the interrupt gate
1052	 * turned them off.
1053	 */
1054	TRACE_IRQS_OFF
1055
1056	movl	%esp, %eax
1057	call	do_int80_syscall_32
1058.Lsyscall_32_done:
1059
1060	STACKLEAK_ERASE
1061
1062restore_all:
1063	TRACE_IRQS_IRET
1064	SWITCH_TO_ENTRY_STACK
1065.Lrestore_all_notrace:
1066	CHECK_AND_APPLY_ESPFIX
1067.Lrestore_nocheck:
1068	/* Switch back to user CR3 */
1069	SWITCH_TO_USER_CR3 scratch_reg=%eax
1070
1071	BUG_IF_WRONG_CR3
1072
1073	/* Restore user state */
1074	RESTORE_REGS pop=4			# skip orig_eax/error_code
1075.Lirq_return:
1076	IRET_FRAME
1077	/*
1078	 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
1079	 * when returning from IPI handler and when returning from
1080	 * scheduler to user-space.
1081	 */
1082	INTERRUPT_RETURN
1083
1084restore_all_kernel:
1085#ifdef CONFIG_PREEMPT
1086	DISABLE_INTERRUPTS(CLBR_ANY)
1087	cmpl	$0, PER_CPU_VAR(__preempt_count)
1088	jnz	.Lno_preempt
1089	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
1090	jz	.Lno_preempt
1091	call	preempt_schedule_irq
1092.Lno_preempt:
1093#endif
1094	TRACE_IRQS_IRET
1095	PARANOID_EXIT_TO_KERNEL_MODE
1096	BUG_IF_WRONG_CR3
1097	RESTORE_REGS 4
1098	jmp	.Lirq_return
1099
1100.section .fixup, "ax"
1101ENTRY(iret_exc	)
1102	pushl	$0				# no error code
1103	pushl	$do_iret_error
1104
1105#ifdef CONFIG_DEBUG_ENTRY
1106	/*
1107	 * The stack-frame here is the one that iret faulted on, so its a
1108	 * return-to-user frame. We are on kernel-cr3 because we come here from
1109	 * the fixup code. This confuses the CR3 checker, so switch to user-cr3
1110	 * as the checker expects it.
1111	 */
1112	pushl	%eax
1113	SWITCH_TO_USER_CR3 scratch_reg=%eax
1114	popl	%eax
1115#endif
1116
1117	jmp	common_exception
1118.previous
1119	_ASM_EXTABLE(.Lirq_return, iret_exc)
1120ENDPROC(entry_INT80_32)
1121
1122.macro FIXUP_ESPFIX_STACK
1123/*
1124 * Switch back for ESPFIX stack to the normal zerobased stack
1125 *
1126 * We can't call C functions using the ESPFIX stack. This code reads
1127 * the high word of the segment base from the GDT and swiches to the
1128 * normal stack and adjusts ESP with the matching offset.
1129 */
1130#ifdef CONFIG_X86_ESPFIX32
1131	/* fixup the stack */
1132	mov	GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
1133	mov	GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
1134	shl	$16, %eax
1135	addl	%esp, %eax			/* the adjusted stack pointer */
1136	pushl	$__KERNEL_DS
1137	pushl	%eax
1138	lss	(%esp), %esp			/* switch to the normal stack segment */
1139#endif
1140.endm
1141.macro UNWIND_ESPFIX_STACK
1142#ifdef CONFIG_X86_ESPFIX32
1143	movl	%ss, %eax
1144	/* see if on espfix stack */
1145	cmpw	$__ESPFIX_SS, %ax
1146	jne	27f
1147	movl	$__KERNEL_DS, %eax
1148	movl	%eax, %ds
1149	movl	%eax, %es
1150	/* switch to normal stack */
1151	FIXUP_ESPFIX_STACK
115227:
1153#endif
1154.endm
1155
1156/*
1157 * Build the entry stubs with some assembler magic.
1158 * We pack 1 stub into every 8-byte block.
1159 */
1160	.align 8
1161ENTRY(irq_entries_start)
1162    vector=FIRST_EXTERNAL_VECTOR
1163    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
1164	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
1165    vector=vector+1
1166	jmp	common_interrupt
1167	.align	8
1168    .endr
1169END(irq_entries_start)
1170
1171#ifdef CONFIG_X86_LOCAL_APIC
1172	.align 8
1173ENTRY(spurious_entries_start)
1174    vector=FIRST_SYSTEM_VECTOR
1175    .rept (NR_VECTORS - FIRST_SYSTEM_VECTOR)
1176	pushl	$(~vector+0x80)			/* Note: always in signed byte range */
1177    vector=vector+1
1178	jmp	common_spurious
1179	.align	8
1180    .endr
1181END(spurious_entries_start)
1182
1183common_spurious:
1184	ASM_CLAC
1185	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
1186	SAVE_ALL switch_stacks=1
1187	ENCODE_FRAME_POINTER
1188	TRACE_IRQS_OFF
1189	movl	%esp, %eax
1190	call	smp_spurious_interrupt
1191	jmp	ret_from_intr
1192ENDPROC(common_spurious)
1193#endif
1194
1195/*
1196 * the CPU automatically disables interrupts when executing an IRQ vector,
1197 * so IRQ-flags tracing has to follow that:
1198 */
1199	.p2align CONFIG_X86_L1_CACHE_SHIFT
1200common_interrupt:
1201	ASM_CLAC
1202	addl	$-0x80, (%esp)			/* Adjust vector into the [-256, -1] range */
1203
1204	SAVE_ALL switch_stacks=1
1205	ENCODE_FRAME_POINTER
1206	TRACE_IRQS_OFF
1207	movl	%esp, %eax
1208	call	do_IRQ
1209	jmp	ret_from_intr
1210ENDPROC(common_interrupt)
1211
1212#define BUILD_INTERRUPT3(name, nr, fn)			\
1213ENTRY(name)						\
1214	ASM_CLAC;					\
1215	pushl	$~(nr);					\
1216	SAVE_ALL switch_stacks=1;			\
1217	ENCODE_FRAME_POINTER;				\
1218	TRACE_IRQS_OFF					\
1219	movl	%esp, %eax;				\
1220	call	fn;					\
1221	jmp	ret_from_intr;				\
1222ENDPROC(name)
1223
1224#define BUILD_INTERRUPT(name, nr)		\
1225	BUILD_INTERRUPT3(name, nr, smp_##name);	\
1226
1227/* The include is where all of the SMP etc. interrupts come from */
1228#include <asm/entry_arch.h>
1229
1230ENTRY(coprocessor_error)
1231	ASM_CLAC
1232	pushl	$0
1233	pushl	$do_coprocessor_error
1234	jmp	common_exception
1235END(coprocessor_error)
1236
1237ENTRY(simd_coprocessor_error)
1238	ASM_CLAC
1239	pushl	$0
1240#ifdef CONFIG_X86_INVD_BUG
1241	/* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
1242	ALTERNATIVE "pushl	$do_general_protection",	\
1243		    "pushl	$do_simd_coprocessor_error",	\
1244		    X86_FEATURE_XMM
1245#else
1246	pushl	$do_simd_coprocessor_error
1247#endif
1248	jmp	common_exception
1249END(simd_coprocessor_error)
1250
1251ENTRY(device_not_available)
1252	ASM_CLAC
1253	pushl	$-1				# mark this as an int
1254	pushl	$do_device_not_available
1255	jmp	common_exception
1256END(device_not_available)
1257
1258#ifdef CONFIG_PARAVIRT
1259ENTRY(native_iret)
1260	iret
1261	_ASM_EXTABLE(native_iret, iret_exc)
1262END(native_iret)
1263#endif
1264
1265ENTRY(overflow)
1266	ASM_CLAC
1267	pushl	$0
1268	pushl	$do_overflow
1269	jmp	common_exception
1270END(overflow)
1271
1272ENTRY(bounds)
1273	ASM_CLAC
1274	pushl	$0
1275	pushl	$do_bounds
1276	jmp	common_exception
1277END(bounds)
1278
1279ENTRY(invalid_op)
1280	ASM_CLAC
1281	pushl	$0
1282	pushl	$do_invalid_op
1283	jmp	common_exception
1284END(invalid_op)
1285
1286ENTRY(coprocessor_segment_overrun)
1287	ASM_CLAC
1288	pushl	$0
1289	pushl	$do_coprocessor_segment_overrun
1290	jmp	common_exception
1291END(coprocessor_segment_overrun)
1292
1293ENTRY(invalid_TSS)
1294	ASM_CLAC
1295	pushl	$do_invalid_TSS
1296	jmp	common_exception
1297END(invalid_TSS)
1298
1299ENTRY(segment_not_present)
1300	ASM_CLAC
1301	pushl	$do_segment_not_present
1302	jmp	common_exception
1303END(segment_not_present)
1304
1305ENTRY(stack_segment)
1306	ASM_CLAC
1307	pushl	$do_stack_segment
1308	jmp	common_exception
1309END(stack_segment)
1310
1311ENTRY(alignment_check)
1312	ASM_CLAC
1313	pushl	$do_alignment_check
1314	jmp	common_exception
1315END(alignment_check)
1316
1317ENTRY(divide_error)
1318	ASM_CLAC
1319	pushl	$0				# no error code
1320	pushl	$do_divide_error
1321	jmp	common_exception
1322END(divide_error)
1323
1324#ifdef CONFIG_X86_MCE
1325ENTRY(machine_check)
1326	ASM_CLAC
1327	pushl	$0
1328	pushl	machine_check_vector
1329	jmp	common_exception
1330END(machine_check)
1331#endif
1332
1333ENTRY(spurious_interrupt_bug)
1334	ASM_CLAC
1335	pushl	$0
1336	pushl	$do_spurious_interrupt_bug
1337	jmp	common_exception
1338END(spurious_interrupt_bug)
1339
1340#ifdef CONFIG_XEN_PV
1341ENTRY(xen_hypervisor_callback)
1342	pushl	$-1				/* orig_ax = -1 => not a system call */
1343	SAVE_ALL
1344	ENCODE_FRAME_POINTER
1345	TRACE_IRQS_OFF
1346
1347	/*
1348	 * Check to see if we got the event in the critical
1349	 * region in xen_iret_direct, after we've reenabled
1350	 * events and checked for pending events.  This simulates
1351	 * iret instruction's behaviour where it delivers a
1352	 * pending interrupt when enabling interrupts:
1353	 */
1354	movl	PT_EIP(%esp), %eax
1355	cmpl	$xen_iret_start_crit, %eax
1356	jb	1f
1357	cmpl	$xen_iret_end_crit, %eax
1358	jae	1f
1359
1360	jmp	xen_iret_crit_fixup
1361
1362ENTRY(xen_do_upcall)
13631:	mov	%esp, %eax
1364	call	xen_evtchn_do_upcall
1365#ifndef CONFIG_PREEMPT
1366	call	xen_maybe_preempt_hcall
1367#endif
1368	jmp	ret_from_intr
1369ENDPROC(xen_hypervisor_callback)
1370
1371/*
1372 * Hypervisor uses this for application faults while it executes.
1373 * We get here for two reasons:
1374 *  1. Fault while reloading DS, ES, FS or GS
1375 *  2. Fault while executing IRET
1376 * Category 1 we fix up by reattempting the load, and zeroing the segment
1377 * register if the load fails.
1378 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
1379 * normal Linux return path in this case because if we use the IRET hypercall
1380 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1381 * We distinguish between categories by maintaining a status value in EAX.
1382 */
1383ENTRY(xen_failsafe_callback)
1384	pushl	%eax
1385	movl	$1, %eax
13861:	mov	4(%esp), %ds
13872:	mov	8(%esp), %es
13883:	mov	12(%esp), %fs
13894:	mov	16(%esp), %gs
1390	/* EAX == 0 => Category 1 (Bad segment)
1391	   EAX != 0 => Category 2 (Bad IRET) */
1392	testl	%eax, %eax
1393	popl	%eax
1394	lea	16(%esp), %esp
1395	jz	5f
1396	jmp	iret_exc
13975:	pushl	$-1				/* orig_ax = -1 => not a system call */
1398	SAVE_ALL
1399	ENCODE_FRAME_POINTER
1400	jmp	ret_from_exception
1401
1402.section .fixup, "ax"
14036:	xorl	%eax, %eax
1404	movl	%eax, 4(%esp)
1405	jmp	1b
14067:	xorl	%eax, %eax
1407	movl	%eax, 8(%esp)
1408	jmp	2b
14098:	xorl	%eax, %eax
1410	movl	%eax, 12(%esp)
1411	jmp	3b
14129:	xorl	%eax, %eax
1413	movl	%eax, 16(%esp)
1414	jmp	4b
1415.previous
1416	_ASM_EXTABLE(1b, 6b)
1417	_ASM_EXTABLE(2b, 7b)
1418	_ASM_EXTABLE(3b, 8b)
1419	_ASM_EXTABLE(4b, 9b)
1420ENDPROC(xen_failsafe_callback)
1421#endif /* CONFIG_XEN_PV */
1422
1423#ifdef CONFIG_XEN_PVHVM
1424BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1425		 xen_evtchn_do_upcall)
1426#endif
1427
1428
1429#if IS_ENABLED(CONFIG_HYPERV)
1430
1431BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1432		 hyperv_vector_handler)
1433
1434BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
1435		 hyperv_reenlightenment_intr)
1436
1437BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
1438		 hv_stimer0_vector_handler)
1439
1440#endif /* CONFIG_HYPERV */
1441
1442ENTRY(page_fault)
1443	ASM_CLAC
1444	pushl	$do_page_fault
1445	ALIGN
1446	jmp common_exception
1447END(page_fault)
1448
1449common_exception:
1450	/* the function address is in %gs's slot on the stack */
1451	FIXUP_FRAME
1452	pushl	%fs
1453	pushl	%es
1454	pushl	%ds
1455	pushl	%eax
1456	movl	$(__USER_DS), %eax
1457	movl	%eax, %ds
1458	movl	%eax, %es
1459	movl	$(__KERNEL_PERCPU), %eax
1460	movl	%eax, %fs
1461	pushl	%ebp
1462	pushl	%edi
1463	pushl	%esi
1464	pushl	%edx
1465	pushl	%ecx
1466	pushl	%ebx
1467	SWITCH_TO_KERNEL_STACK
1468	ENCODE_FRAME_POINTER
1469	cld
1470	UNWIND_ESPFIX_STACK
1471	GS_TO_REG %ecx
1472	movl	PT_GS(%esp), %edi		# get the function address
1473	movl	PT_ORIG_EAX(%esp), %edx		# get the error code
1474	movl	$-1, PT_ORIG_EAX(%esp)		# no syscall to restart
1475	REG_TO_PTGS %ecx
1476	SET_KERNEL_GS %ecx
1477	TRACE_IRQS_OFF
1478	movl	%esp, %eax			# pt_regs pointer
1479	CALL_NOSPEC %edi
1480	jmp	ret_from_exception
1481END(common_exception)
1482
1483ENTRY(debug)
1484	/*
1485	 * Entry from sysenter is now handled in common_exception
1486	 */
1487	ASM_CLAC
1488	pushl	$-1				# mark this as an int
1489	pushl	$do_debug
1490	jmp	common_exception
1491END(debug)
1492
1493/*
1494 * NMI is doubly nasty.  It can happen on the first instruction of
1495 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1496 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1497 * switched stacks.  We handle both conditions by simply checking whether we
1498 * interrupted kernel code running on the SYSENTER stack.
1499 */
1500ENTRY(nmi)
1501	ASM_CLAC
1502
1503#ifdef CONFIG_X86_ESPFIX32
1504	pushl	%eax
1505	movl	%ss, %eax
1506	cmpw	$__ESPFIX_SS, %ax
1507	popl	%eax
1508	je	.Lnmi_espfix_stack
1509#endif
1510
1511	pushl	%eax				# pt_regs->orig_ax
1512	SAVE_ALL_NMI cr3_reg=%edi
1513	ENCODE_FRAME_POINTER
1514	xorl	%edx, %edx			# zero error code
1515	movl	%esp, %eax			# pt_regs pointer
1516
1517	/* Are we currently on the SYSENTER stack? */
1518	movl	PER_CPU_VAR(cpu_entry_area), %ecx
1519	addl	$CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1520	subl	%eax, %ecx	/* ecx = (end of entry_stack) - esp */
1521	cmpl	$SIZEOF_entry_stack, %ecx
1522	jb	.Lnmi_from_sysenter_stack
1523
1524	/* Not on SYSENTER stack. */
1525	call	do_nmi
1526	jmp	.Lnmi_return
1527
1528.Lnmi_from_sysenter_stack:
1529	/*
1530	 * We're on the SYSENTER stack.  Switch off.  No one (not even debug)
1531	 * is using the thread stack right now, so it's safe for us to use it.
1532	 */
1533	movl	%esp, %ebx
1534	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esp
1535	call	do_nmi
1536	movl	%ebx, %esp
1537
1538.Lnmi_return:
1539	CHECK_AND_APPLY_ESPFIX
1540	RESTORE_ALL_NMI cr3_reg=%edi pop=4
1541	jmp	.Lirq_return
1542
1543#ifdef CONFIG_X86_ESPFIX32
1544.Lnmi_espfix_stack:
1545	/*
1546	 * create the pointer to lss back
1547	 */
1548	pushl	%ss
1549	pushl	%esp
1550	addl	$4, (%esp)
1551	/* copy the iret frame of 12 bytes */
1552	.rept 3
1553	pushl	16(%esp)
1554	.endr
1555	pushl	%eax
1556	SAVE_ALL_NMI cr3_reg=%edi
1557	ENCODE_FRAME_POINTER
1558	FIXUP_ESPFIX_STACK			# %eax == %esp
1559	xorl	%edx, %edx			# zero error code
1560	call	do_nmi
1561	RESTORE_ALL_NMI cr3_reg=%edi
1562	lss	12+4(%esp), %esp		# back to espfix stack
1563	jmp	.Lirq_return
1564#endif
1565END(nmi)
1566
1567ENTRY(int3)
1568	ASM_CLAC
1569	pushl	$-1				# mark this as an int
1570
1571	SAVE_ALL switch_stacks=1
1572	ENCODE_FRAME_POINTER
1573	TRACE_IRQS_OFF
1574	xorl	%edx, %edx			# zero error code
1575	movl	%esp, %eax			# pt_regs pointer
1576	call	do_int3
1577	jmp	ret_from_exception
1578END(int3)
1579
1580ENTRY(general_protection)
1581	pushl	$do_general_protection
1582	jmp	common_exception
1583END(general_protection)
1584
1585#ifdef CONFIG_KVM_GUEST
1586ENTRY(async_page_fault)
1587	ASM_CLAC
1588	pushl	$do_async_page_fault
1589	jmp	common_exception
1590END(async_page_fault)
1591#endif
1592
1593ENTRY(rewind_stack_do_exit)
1594	/* Prevent any naive code from trying to unwind to our caller. */
1595	xorl	%ebp, %ebp
1596
1597	movl	PER_CPU_VAR(cpu_current_top_of_stack), %esi
1598	leal	-TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1599
1600	call	do_exit
16011:	jmp 1b
1602END(rewind_stack_do_exit)
1603