xref: /openbmc/linux/arch/x86/entry/entry_64.S (revision 9977a8c3497a8f7f7f951994f298a8e4d961234f)
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 *  linux/arch/x86_64/entry.S
4 *
5 *  Copyright (C) 1991, 1992  Linus Torvalds
6 *  Copyright (C) 2000, 2001, 2002  Andi Kleen SuSE Labs
7 *  Copyright (C) 2000  Pavel Machek <pavel@suse.cz>
8 *
9 * entry.S contains the system-call and fault low-level handling routines.
10 *
11 * Some of this is documented in Documentation/x86/entry_64.txt
12 *
13 * A note on terminology:
14 * - iret frame:	Architecture defined interrupt frame from SS to RIP
15 *			at the top of the kernel process stack.
16 *
17 * Some macro usage:
18 * - ENTRY/END:		Define functions in the symbol table.
19 * - TRACE_IRQ_*:	Trace hardirq state for lock debugging.
20 * - idtentry:		Define exception entry points.
21 */
22#include <linux/linkage.h>
23#include <asm/segment.h>
24#include <asm/cache.h>
25#include <asm/errno.h>
26#include <asm/asm-offsets.h>
27#include <asm/msr.h>
28#include <asm/unistd.h>
29#include <asm/thread_info.h>
30#include <asm/hw_irq.h>
31#include <asm/page_types.h>
32#include <asm/irqflags.h>
33#include <asm/paravirt.h>
34#include <asm/percpu.h>
35#include <asm/asm.h>
36#include <asm/smap.h>
37#include <asm/pgtable_types.h>
38#include <asm/export.h>
39#include <asm/frame.h>
40#include <asm/nospec-branch.h>
41#include <linux/err.h>
42
43#include "calling.h"
44
45.code64
46.section .entry.text, "ax"
47
48#ifdef CONFIG_PARAVIRT
49ENTRY(native_usergs_sysret64)
50	UNWIND_HINT_EMPTY
51	swapgs
52	sysretq
53END(native_usergs_sysret64)
54#endif /* CONFIG_PARAVIRT */
55
56.macro TRACE_IRQS_FLAGS flags:req
57#ifdef CONFIG_TRACE_IRQFLAGS
58	bt	$9, \flags		/* interrupts off? */
59	jnc	1f
60	TRACE_IRQS_ON
611:
62#endif
63.endm
64
65.macro TRACE_IRQS_IRETQ
66	TRACE_IRQS_FLAGS EFLAGS(%rsp)
67.endm
68
69/*
70 * When dynamic function tracer is enabled it will add a breakpoint
71 * to all locations that it is about to modify, sync CPUs, update
72 * all the code, sync CPUs, then remove the breakpoints. In this time
73 * if lockdep is enabled, it might jump back into the debug handler
74 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
75 *
76 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
77 * make sure the stack pointer does not get reset back to the top
78 * of the debug stack, and instead just reuses the current stack.
79 */
80#if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
81
82.macro TRACE_IRQS_OFF_DEBUG
83	call	debug_stack_set_zero
84	TRACE_IRQS_OFF
85	call	debug_stack_reset
86.endm
87
88.macro TRACE_IRQS_ON_DEBUG
89	call	debug_stack_set_zero
90	TRACE_IRQS_ON
91	call	debug_stack_reset
92.endm
93
94.macro TRACE_IRQS_IRETQ_DEBUG
95	bt	$9, EFLAGS(%rsp)		/* interrupts off? */
96	jnc	1f
97	TRACE_IRQS_ON_DEBUG
981:
99.endm
100
101#else
102# define TRACE_IRQS_OFF_DEBUG			TRACE_IRQS_OFF
103# define TRACE_IRQS_ON_DEBUG			TRACE_IRQS_ON
104# define TRACE_IRQS_IRETQ_DEBUG			TRACE_IRQS_IRETQ
105#endif
106
107/*
108 * 64-bit SYSCALL instruction entry. Up to 6 arguments in registers.
109 *
110 * This is the only entry point used for 64-bit system calls.  The
111 * hardware interface is reasonably well designed and the register to
112 * argument mapping Linux uses fits well with the registers that are
113 * available when SYSCALL is used.
114 *
115 * SYSCALL instructions can be found inlined in libc implementations as
116 * well as some other programs and libraries.  There are also a handful
117 * of SYSCALL instructions in the vDSO used, for example, as a
118 * clock_gettimeofday fallback.
119 *
120 * 64-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
121 * then loads new ss, cs, and rip from previously programmed MSRs.
122 * rflags gets masked by a value from another MSR (so CLD and CLAC
123 * are not needed). SYSCALL does not save anything on the stack
124 * and does not change rsp.
125 *
126 * Registers on entry:
127 * rax  system call number
128 * rcx  return address
129 * r11  saved rflags (note: r11 is callee-clobbered register in C ABI)
130 * rdi  arg0
131 * rsi  arg1
132 * rdx  arg2
133 * r10  arg3 (needs to be moved to rcx to conform to C ABI)
134 * r8   arg4
135 * r9   arg5
136 * (note: r12-r15, rbp, rbx are callee-preserved in C ABI)
137 *
138 * Only called from user space.
139 *
140 * When user can change pt_regs->foo always force IRET. That is because
141 * it deals with uncanonical addresses better. SYSRET has trouble
142 * with them due to bugs in both AMD and Intel CPUs.
143 */
144
145	.pushsection .entry_trampoline, "ax"
146
147/*
148 * The code in here gets remapped into cpu_entry_area's trampoline.  This means
149 * that the assembler and linker have the wrong idea as to where this code
150 * lives (and, in fact, it's mapped more than once, so it's not even at a
151 * fixed address).  So we can't reference any symbols outside the entry
152 * trampoline and expect it to work.
153 *
154 * Instead, we carefully abuse %rip-relative addressing.
155 * _entry_trampoline(%rip) refers to the start of the remapped) entry
156 * trampoline.  We can thus find cpu_entry_area with this macro:
157 */
158
159#define CPU_ENTRY_AREA \
160	_entry_trampoline - CPU_ENTRY_AREA_entry_trampoline(%rip)
161
162/* The top word of the SYSENTER stack is hot and is usable as scratch space. */
163#define RSP_SCRATCH	CPU_ENTRY_AREA_entry_stack + \
164			SIZEOF_entry_stack - 8 + CPU_ENTRY_AREA
165
166ENTRY(entry_SYSCALL_64_trampoline)
167	UNWIND_HINT_EMPTY
168	swapgs
169
170	/* Stash the user RSP. */
171	movq	%rsp, RSP_SCRATCH
172
173	/* Note: using %rsp as a scratch reg. */
174	SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp
175
176	/* Load the top of the task stack into RSP */
177	movq	CPU_ENTRY_AREA_tss + TSS_sp1 + CPU_ENTRY_AREA, %rsp
178
179	/* Start building the simulated IRET frame. */
180	pushq	$__USER_DS			/* pt_regs->ss */
181	pushq	RSP_SCRATCH			/* pt_regs->sp */
182	pushq	%r11				/* pt_regs->flags */
183	pushq	$__USER_CS			/* pt_regs->cs */
184	pushq	%rcx				/* pt_regs->ip */
185
186	/*
187	 * x86 lacks a near absolute jump, and we can't jump to the real
188	 * entry text with a relative jump.  We could push the target
189	 * address and then use retq, but this destroys the pipeline on
190	 * many CPUs (wasting over 20 cycles on Sandy Bridge).  Instead,
191	 * spill RDI and restore it in a second-stage trampoline.
192	 */
193	pushq	%rdi
194	movq	$entry_SYSCALL_64_stage2, %rdi
195	JMP_NOSPEC %rdi
196END(entry_SYSCALL_64_trampoline)
197
198	.popsection
199
200ENTRY(entry_SYSCALL_64_stage2)
201	UNWIND_HINT_EMPTY
202	popq	%rdi
203	jmp	entry_SYSCALL_64_after_hwframe
204END(entry_SYSCALL_64_stage2)
205
206ENTRY(entry_SYSCALL_64)
207	UNWIND_HINT_EMPTY
208	/*
209	 * Interrupts are off on entry.
210	 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
211	 * it is too small to ever cause noticeable irq latency.
212	 */
213
214	swapgs
215	/*
216	 * This path is only taken when PAGE_TABLE_ISOLATION is disabled so it
217	 * is not required to switch CR3.
218	 */
219	movq	%rsp, PER_CPU_VAR(rsp_scratch)
220	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
221
222	/* Construct struct pt_regs on stack */
223	pushq	$__USER_DS			/* pt_regs->ss */
224	pushq	PER_CPU_VAR(rsp_scratch)	/* pt_regs->sp */
225	pushq	%r11				/* pt_regs->flags */
226	pushq	$__USER_CS			/* pt_regs->cs */
227	pushq	%rcx				/* pt_regs->ip */
228GLOBAL(entry_SYSCALL_64_after_hwframe)
229	pushq	%rax				/* pt_regs->orig_ax */
230
231	PUSH_AND_CLEAR_REGS rax=$-ENOSYS
232
233	TRACE_IRQS_OFF
234
235	/* IRQs are off. */
236	movq	%rsp, %rdi
237	call	do_syscall_64		/* returns with IRQs disabled */
238
239	TRACE_IRQS_IRETQ		/* we're about to change IF */
240
241	/*
242	 * Try to use SYSRET instead of IRET if we're returning to
243	 * a completely clean 64-bit userspace context.  If we're not,
244	 * go to the slow exit path.
245	 */
246	movq	RCX(%rsp), %rcx
247	movq	RIP(%rsp), %r11
248
249	cmpq	%rcx, %r11	/* SYSRET requires RCX == RIP */
250	jne	swapgs_restore_regs_and_return_to_usermode
251
252	/*
253	 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
254	 * in kernel space.  This essentially lets the user take over
255	 * the kernel, since userspace controls RSP.
256	 *
257	 * If width of "canonical tail" ever becomes variable, this will need
258	 * to be updated to remain correct on both old and new CPUs.
259	 *
260	 * Change top bits to match most significant bit (47th or 56th bit
261	 * depending on paging mode) in the address.
262	 */
263	shl	$(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
264	sar	$(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
265
266	/* If this changed %rcx, it was not canonical */
267	cmpq	%rcx, %r11
268	jne	swapgs_restore_regs_and_return_to_usermode
269
270	cmpq	$__USER_CS, CS(%rsp)		/* CS must match SYSRET */
271	jne	swapgs_restore_regs_and_return_to_usermode
272
273	movq	R11(%rsp), %r11
274	cmpq	%r11, EFLAGS(%rsp)		/* R11 == RFLAGS */
275	jne	swapgs_restore_regs_and_return_to_usermode
276
277	/*
278	 * SYSCALL clears RF when it saves RFLAGS in R11 and SYSRET cannot
279	 * restore RF properly. If the slowpath sets it for whatever reason, we
280	 * need to restore it correctly.
281	 *
282	 * SYSRET can restore TF, but unlike IRET, restoring TF results in a
283	 * trap from userspace immediately after SYSRET.  This would cause an
284	 * infinite loop whenever #DB happens with register state that satisfies
285	 * the opportunistic SYSRET conditions.  For example, single-stepping
286	 * this user code:
287	 *
288	 *           movq	$stuck_here, %rcx
289	 *           pushfq
290	 *           popq %r11
291	 *   stuck_here:
292	 *
293	 * would never get past 'stuck_here'.
294	 */
295	testq	$(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11
296	jnz	swapgs_restore_regs_and_return_to_usermode
297
298	/* nothing to check for RSP */
299
300	cmpq	$__USER_DS, SS(%rsp)		/* SS must match SYSRET */
301	jne	swapgs_restore_regs_and_return_to_usermode
302
303	/*
304	 * We win! This label is here just for ease of understanding
305	 * perf profiles. Nothing jumps here.
306	 */
307syscall_return_via_sysret:
308	/* rcx and r11 are already restored (see code above) */
309	UNWIND_HINT_EMPTY
310	POP_REGS pop_rdi=0 skip_r11rcx=1
311
312	/*
313	 * Now all regs are restored except RSP and RDI.
314	 * Save old stack pointer and switch to trampoline stack.
315	 */
316	movq	%rsp, %rdi
317	movq	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
318
319	pushq	RSP-RDI(%rdi)	/* RSP */
320	pushq	(%rdi)		/* RDI */
321
322	/*
323	 * We are on the trampoline stack.  All regs except RDI are live.
324	 * We can do future final exit work right here.
325	 */
326	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
327
328	popq	%rdi
329	popq	%rsp
330	USERGS_SYSRET64
331END(entry_SYSCALL_64)
332
333/*
334 * %rdi: prev task
335 * %rsi: next task
336 */
337ENTRY(__switch_to_asm)
338	UNWIND_HINT_FUNC
339	/*
340	 * Save callee-saved registers
341	 * This must match the order in inactive_task_frame
342	 */
343	pushq	%rbp
344	pushq	%rbx
345	pushq	%r12
346	pushq	%r13
347	pushq	%r14
348	pushq	%r15
349
350	/* switch stack */
351	movq	%rsp, TASK_threadsp(%rdi)
352	movq	TASK_threadsp(%rsi), %rsp
353
354#ifdef CONFIG_CC_STACKPROTECTOR
355	movq	TASK_stack_canary(%rsi), %rbx
356	movq	%rbx, PER_CPU_VAR(irq_stack_union)+stack_canary_offset
357#endif
358
359#ifdef CONFIG_RETPOLINE
360	/*
361	 * When switching from a shallower to a deeper call stack
362	 * the RSB may either underflow or use entries populated
363	 * with userspace addresses. On CPUs where those concerns
364	 * exist, overwrite the RSB with entries which capture
365	 * speculative execution to prevent attack.
366	 */
367	/* Clobbers %rbx */
368	FILL_RETURN_BUFFER RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
369#endif
370
371	/* restore callee-saved registers */
372	popq	%r15
373	popq	%r14
374	popq	%r13
375	popq	%r12
376	popq	%rbx
377	popq	%rbp
378
379	jmp	__switch_to
380END(__switch_to_asm)
381
382/*
383 * A newly forked process directly context switches into this address.
384 *
385 * rax: prev task we switched from
386 * rbx: kernel thread func (NULL for user thread)
387 * r12: kernel thread arg
388 */
389ENTRY(ret_from_fork)
390	UNWIND_HINT_EMPTY
391	movq	%rax, %rdi
392	call	schedule_tail			/* rdi: 'prev' task parameter */
393
394	testq	%rbx, %rbx			/* from kernel_thread? */
395	jnz	1f				/* kernel threads are uncommon */
396
3972:
398	UNWIND_HINT_REGS
399	movq	%rsp, %rdi
400	call	syscall_return_slowpath	/* returns with IRQs disabled */
401	TRACE_IRQS_ON			/* user mode is traced as IRQS on */
402	jmp	swapgs_restore_regs_and_return_to_usermode
403
4041:
405	/* kernel thread */
406	movq	%r12, %rdi
407	CALL_NOSPEC %rbx
408	/*
409	 * A kernel thread is allowed to return here after successfully
410	 * calling do_execve().  Exit to userspace to complete the execve()
411	 * syscall.
412	 */
413	movq	$0, RAX(%rsp)
414	jmp	2b
415END(ret_from_fork)
416
417/*
418 * Build the entry stubs with some assembler magic.
419 * We pack 1 stub into every 8-byte block.
420 */
421	.align 8
422ENTRY(irq_entries_start)
423    vector=FIRST_EXTERNAL_VECTOR
424    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
425	UNWIND_HINT_IRET_REGS
426	pushq	$(~vector+0x80)			/* Note: always in signed byte range */
427	jmp	common_interrupt
428	.align	8
429	vector=vector+1
430    .endr
431END(irq_entries_start)
432
433.macro DEBUG_ENTRY_ASSERT_IRQS_OFF
434#ifdef CONFIG_DEBUG_ENTRY
435	pushq %rax
436	SAVE_FLAGS(CLBR_RAX)
437	testl $X86_EFLAGS_IF, %eax
438	jz .Lokay_\@
439	ud2
440.Lokay_\@:
441	popq %rax
442#endif
443.endm
444
445/*
446 * Enters the IRQ stack if we're not already using it.  NMI-safe.  Clobbers
447 * flags and puts old RSP into old_rsp, and leaves all other GPRs alone.
448 * Requires kernel GSBASE.
449 *
450 * The invariant is that, if irq_count != -1, then the IRQ stack is in use.
451 */
452.macro ENTER_IRQ_STACK regs=1 old_rsp
453	DEBUG_ENTRY_ASSERT_IRQS_OFF
454	movq	%rsp, \old_rsp
455
456	.if \regs
457	UNWIND_HINT_REGS base=\old_rsp
458	.endif
459
460	incl	PER_CPU_VAR(irq_count)
461	jnz	.Lirq_stack_push_old_rsp_\@
462
463	/*
464	 * Right now, if we just incremented irq_count to zero, we've
465	 * claimed the IRQ stack but we haven't switched to it yet.
466	 *
467	 * If anything is added that can interrupt us here without using IST,
468	 * it must be *extremely* careful to limit its stack usage.  This
469	 * could include kprobes and a hypothetical future IST-less #DB
470	 * handler.
471	 *
472	 * The OOPS unwinder relies on the word at the top of the IRQ
473	 * stack linking back to the previous RSP for the entire time we're
474	 * on the IRQ stack.  For this to work reliably, we need to write
475	 * it before we actually move ourselves to the IRQ stack.
476	 */
477
478	movq	\old_rsp, PER_CPU_VAR(irq_stack_union + IRQ_STACK_SIZE - 8)
479	movq	PER_CPU_VAR(irq_stack_ptr), %rsp
480
481#ifdef CONFIG_DEBUG_ENTRY
482	/*
483	 * If the first movq above becomes wrong due to IRQ stack layout
484	 * changes, the only way we'll notice is if we try to unwind right
485	 * here.  Assert that we set up the stack right to catch this type
486	 * of bug quickly.
487	 */
488	cmpq	-8(%rsp), \old_rsp
489	je	.Lirq_stack_okay\@
490	ud2
491	.Lirq_stack_okay\@:
492#endif
493
494.Lirq_stack_push_old_rsp_\@:
495	pushq	\old_rsp
496
497	.if \regs
498	UNWIND_HINT_REGS indirect=1
499	.endif
500.endm
501
502/*
503 * Undoes ENTER_IRQ_STACK.
504 */
505.macro LEAVE_IRQ_STACK regs=1
506	DEBUG_ENTRY_ASSERT_IRQS_OFF
507	/* We need to be off the IRQ stack before decrementing irq_count. */
508	popq	%rsp
509
510	.if \regs
511	UNWIND_HINT_REGS
512	.endif
513
514	/*
515	 * As in ENTER_IRQ_STACK, irq_count == 0, we are still claiming
516	 * the irq stack but we're not on it.
517	 */
518
519	decl	PER_CPU_VAR(irq_count)
520.endm
521
522/*
523 * Interrupt entry/exit.
524 *
525 * Interrupt entry points save only callee clobbered registers in fast path.
526 *
527 * Entry runs with interrupts off.
528 */
529
530/* 0(%rsp): ~(interrupt number) */
531	.macro interrupt func
532	cld
533
534	testb	$3, CS-ORIG_RAX(%rsp)
535	jz	1f
536	SWAPGS
537	call	switch_to_thread_stack
5381:
539
540	PUSH_AND_CLEAR_REGS
541	ENCODE_FRAME_POINTER
542
543	testb	$3, CS(%rsp)
544	jz	1f
545
546	/*
547	 * IRQ from user mode.
548	 *
549	 * We need to tell lockdep that IRQs are off.  We can't do this until
550	 * we fix gsbase, and we should do it before enter_from_user_mode
551	 * (which can take locks).  Since TRACE_IRQS_OFF idempotent,
552	 * the simplest way to handle it is to just call it twice if
553	 * we enter from user mode.  There's no reason to optimize this since
554	 * TRACE_IRQS_OFF is a no-op if lockdep is off.
555	 */
556	TRACE_IRQS_OFF
557
558	CALL_enter_from_user_mode
559
5601:
561	ENTER_IRQ_STACK old_rsp=%rdi
562	/* We entered an interrupt context - irqs are off: */
563	TRACE_IRQS_OFF
564
565	call	\func	/* rdi points to pt_regs */
566	.endm
567
568	/*
569	 * The interrupt stubs push (~vector+0x80) onto the stack and
570	 * then jump to common_interrupt.
571	 */
572	.p2align CONFIG_X86_L1_CACHE_SHIFT
573common_interrupt:
574	ASM_CLAC
575	addq	$-0x80, (%rsp)			/* Adjust vector to [-256, -1] range */
576	interrupt do_IRQ
577	/* 0(%rsp): old RSP */
578ret_from_intr:
579	DISABLE_INTERRUPTS(CLBR_ANY)
580	TRACE_IRQS_OFF
581
582	LEAVE_IRQ_STACK
583
584	testb	$3, CS(%rsp)
585	jz	retint_kernel
586
587	/* Interrupt came from user space */
588GLOBAL(retint_user)
589	mov	%rsp,%rdi
590	call	prepare_exit_to_usermode
591	TRACE_IRQS_IRETQ
592
593GLOBAL(swapgs_restore_regs_and_return_to_usermode)
594#ifdef CONFIG_DEBUG_ENTRY
595	/* Assert that pt_regs indicates user mode. */
596	testb	$3, CS(%rsp)
597	jnz	1f
598	ud2
5991:
600#endif
601	POP_REGS pop_rdi=0
602
603	/*
604	 * The stack is now user RDI, orig_ax, RIP, CS, EFLAGS, RSP, SS.
605	 * Save old stack pointer and switch to trampoline stack.
606	 */
607	movq	%rsp, %rdi
608	movq	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
609
610	/* Copy the IRET frame to the trampoline stack. */
611	pushq	6*8(%rdi)	/* SS */
612	pushq	5*8(%rdi)	/* RSP */
613	pushq	4*8(%rdi)	/* EFLAGS */
614	pushq	3*8(%rdi)	/* CS */
615	pushq	2*8(%rdi)	/* RIP */
616
617	/* Push user RDI on the trampoline stack. */
618	pushq	(%rdi)
619
620	/*
621	 * We are on the trampoline stack.  All regs except RDI are live.
622	 * We can do future final exit work right here.
623	 */
624
625	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
626
627	/* Restore RDI. */
628	popq	%rdi
629	SWAPGS
630	INTERRUPT_RETURN
631
632
633/* Returning to kernel space */
634retint_kernel:
635#ifdef CONFIG_PREEMPT
636	/* Interrupts are off */
637	/* Check if we need preemption */
638	bt	$9, EFLAGS(%rsp)		/* were interrupts off? */
639	jnc	1f
6400:	cmpl	$0, PER_CPU_VAR(__preempt_count)
641	jnz	1f
642	call	preempt_schedule_irq
643	jmp	0b
6441:
645#endif
646	/*
647	 * The iretq could re-enable interrupts:
648	 */
649	TRACE_IRQS_IRETQ
650
651GLOBAL(restore_regs_and_return_to_kernel)
652#ifdef CONFIG_DEBUG_ENTRY
653	/* Assert that pt_regs indicates kernel mode. */
654	testb	$3, CS(%rsp)
655	jz	1f
656	ud2
6571:
658#endif
659	POP_REGS
660	addq	$8, %rsp	/* skip regs->orig_ax */
661	/*
662	 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
663	 * when returning from IPI handler.
664	 */
665	INTERRUPT_RETURN
666
667ENTRY(native_iret)
668	UNWIND_HINT_IRET_REGS
669	/*
670	 * Are we returning to a stack segment from the LDT?  Note: in
671	 * 64-bit mode SS:RSP on the exception stack is always valid.
672	 */
673#ifdef CONFIG_X86_ESPFIX64
674	testb	$4, (SS-RIP)(%rsp)
675	jnz	native_irq_return_ldt
676#endif
677
678.global native_irq_return_iret
679native_irq_return_iret:
680	/*
681	 * This may fault.  Non-paranoid faults on return to userspace are
682	 * handled by fixup_bad_iret.  These include #SS, #GP, and #NP.
683	 * Double-faults due to espfix64 are handled in do_double_fault.
684	 * Other faults here are fatal.
685	 */
686	iretq
687
688#ifdef CONFIG_X86_ESPFIX64
689native_irq_return_ldt:
690	/*
691	 * We are running with user GSBASE.  All GPRs contain their user
692	 * values.  We have a percpu ESPFIX stack that is eight slots
693	 * long (see ESPFIX_STACK_SIZE).  espfix_waddr points to the bottom
694	 * of the ESPFIX stack.
695	 *
696	 * We clobber RAX and RDI in this code.  We stash RDI on the
697	 * normal stack and RAX on the ESPFIX stack.
698	 *
699	 * The ESPFIX stack layout we set up looks like this:
700	 *
701	 * --- top of ESPFIX stack ---
702	 * SS
703	 * RSP
704	 * RFLAGS
705	 * CS
706	 * RIP  <-- RSP points here when we're done
707	 * RAX  <-- espfix_waddr points here
708	 * --- bottom of ESPFIX stack ---
709	 */
710
711	pushq	%rdi				/* Stash user RDI */
712	SWAPGS					/* to kernel GS */
713	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi	/* to kernel CR3 */
714
715	movq	PER_CPU_VAR(espfix_waddr), %rdi
716	movq	%rax, (0*8)(%rdi)		/* user RAX */
717	movq	(1*8)(%rsp), %rax		/* user RIP */
718	movq	%rax, (1*8)(%rdi)
719	movq	(2*8)(%rsp), %rax		/* user CS */
720	movq	%rax, (2*8)(%rdi)
721	movq	(3*8)(%rsp), %rax		/* user RFLAGS */
722	movq	%rax, (3*8)(%rdi)
723	movq	(5*8)(%rsp), %rax		/* user SS */
724	movq	%rax, (5*8)(%rdi)
725	movq	(4*8)(%rsp), %rax		/* user RSP */
726	movq	%rax, (4*8)(%rdi)
727	/* Now RAX == RSP. */
728
729	andl	$0xffff0000, %eax		/* RAX = (RSP & 0xffff0000) */
730
731	/*
732	 * espfix_stack[31:16] == 0.  The page tables are set up such that
733	 * (espfix_stack | (X & 0xffff0000)) points to a read-only alias of
734	 * espfix_waddr for any X.  That is, there are 65536 RO aliases of
735	 * the same page.  Set up RSP so that RSP[31:16] contains the
736	 * respective 16 bits of the /userspace/ RSP and RSP nonetheless
737	 * still points to an RO alias of the ESPFIX stack.
738	 */
739	orq	PER_CPU_VAR(espfix_stack), %rax
740
741	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
742	SWAPGS					/* to user GS */
743	popq	%rdi				/* Restore user RDI */
744
745	movq	%rax, %rsp
746	UNWIND_HINT_IRET_REGS offset=8
747
748	/*
749	 * At this point, we cannot write to the stack any more, but we can
750	 * still read.
751	 */
752	popq	%rax				/* Restore user RAX */
753
754	/*
755	 * RSP now points to an ordinary IRET frame, except that the page
756	 * is read-only and RSP[31:16] are preloaded with the userspace
757	 * values.  We can now IRET back to userspace.
758	 */
759	jmp	native_irq_return_iret
760#endif
761END(common_interrupt)
762
763/*
764 * APIC interrupts.
765 */
766.macro apicinterrupt3 num sym do_sym
767ENTRY(\sym)
768	UNWIND_HINT_IRET_REGS
769	ASM_CLAC
770	pushq	$~(\num)
771.Lcommon_\sym:
772	interrupt \do_sym
773	jmp	ret_from_intr
774END(\sym)
775.endm
776
777/* Make sure APIC interrupt handlers end up in the irqentry section: */
778#define PUSH_SECTION_IRQENTRY	.pushsection .irqentry.text, "ax"
779#define POP_SECTION_IRQENTRY	.popsection
780
781.macro apicinterrupt num sym do_sym
782PUSH_SECTION_IRQENTRY
783apicinterrupt3 \num \sym \do_sym
784POP_SECTION_IRQENTRY
785.endm
786
787#ifdef CONFIG_SMP
788apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR		irq_move_cleanup_interrupt	smp_irq_move_cleanup_interrupt
789apicinterrupt3 REBOOT_VECTOR			reboot_interrupt		smp_reboot_interrupt
790#endif
791
792#ifdef CONFIG_X86_UV
793apicinterrupt3 UV_BAU_MESSAGE			uv_bau_message_intr1		uv_bau_message_interrupt
794#endif
795
796apicinterrupt LOCAL_TIMER_VECTOR		apic_timer_interrupt		smp_apic_timer_interrupt
797apicinterrupt X86_PLATFORM_IPI_VECTOR		x86_platform_ipi		smp_x86_platform_ipi
798
799#ifdef CONFIG_HAVE_KVM
800apicinterrupt3 POSTED_INTR_VECTOR		kvm_posted_intr_ipi		smp_kvm_posted_intr_ipi
801apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR	kvm_posted_intr_wakeup_ipi	smp_kvm_posted_intr_wakeup_ipi
802apicinterrupt3 POSTED_INTR_NESTED_VECTOR	kvm_posted_intr_nested_ipi	smp_kvm_posted_intr_nested_ipi
803#endif
804
805#ifdef CONFIG_X86_MCE_THRESHOLD
806apicinterrupt THRESHOLD_APIC_VECTOR		threshold_interrupt		smp_threshold_interrupt
807#endif
808
809#ifdef CONFIG_X86_MCE_AMD
810apicinterrupt DEFERRED_ERROR_VECTOR		deferred_error_interrupt	smp_deferred_error_interrupt
811#endif
812
813#ifdef CONFIG_X86_THERMAL_VECTOR
814apicinterrupt THERMAL_APIC_VECTOR		thermal_interrupt		smp_thermal_interrupt
815#endif
816
817#ifdef CONFIG_SMP
818apicinterrupt CALL_FUNCTION_SINGLE_VECTOR	call_function_single_interrupt	smp_call_function_single_interrupt
819apicinterrupt CALL_FUNCTION_VECTOR		call_function_interrupt		smp_call_function_interrupt
820apicinterrupt RESCHEDULE_VECTOR			reschedule_interrupt		smp_reschedule_interrupt
821#endif
822
823apicinterrupt ERROR_APIC_VECTOR			error_interrupt			smp_error_interrupt
824apicinterrupt SPURIOUS_APIC_VECTOR		spurious_interrupt		smp_spurious_interrupt
825
826#ifdef CONFIG_IRQ_WORK
827apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
828#endif
829
830/*
831 * Exception entry points.
832 */
833#define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss_rw) + (TSS_ist + ((x) - 1) * 8)
834
835/*
836 * Switch to the thread stack.  This is called with the IRET frame and
837 * orig_ax on the stack.  (That is, RDI..R12 are not on the stack and
838 * space has not been allocated for them.)
839 */
840ENTRY(switch_to_thread_stack)
841	UNWIND_HINT_FUNC
842
843	pushq	%rdi
844	/* Need to switch before accessing the thread stack. */
845	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi
846	movq	%rsp, %rdi
847	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
848	UNWIND_HINT sp_offset=16 sp_reg=ORC_REG_DI
849
850	pushq	7*8(%rdi)		/* regs->ss */
851	pushq	6*8(%rdi)		/* regs->rsp */
852	pushq	5*8(%rdi)		/* regs->eflags */
853	pushq	4*8(%rdi)		/* regs->cs */
854	pushq	3*8(%rdi)		/* regs->ip */
855	pushq	2*8(%rdi)		/* regs->orig_ax */
856	pushq	8(%rdi)			/* return address */
857	UNWIND_HINT_FUNC
858
859	movq	(%rdi), %rdi
860	ret
861END(switch_to_thread_stack)
862
863.macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
864ENTRY(\sym)
865	UNWIND_HINT_IRET_REGS offset=\has_error_code*8
866
867	/* Sanity check */
868	.if \shift_ist != -1 && \paranoid == 0
869	.error "using shift_ist requires paranoid=1"
870	.endif
871
872	ASM_CLAC
873
874	.if \has_error_code == 0
875	pushq	$-1				/* ORIG_RAX: no syscall to restart */
876	.endif
877
878	/* Save all registers in pt_regs */
879	PUSH_AND_CLEAR_REGS
880	ENCODE_FRAME_POINTER
881
882	.if \paranoid < 2
883	testb	$3, CS(%rsp)			/* If coming from userspace, switch stacks */
884	jnz	.Lfrom_usermode_switch_stack_\@
885	.endif
886
887	.if \paranoid
888	call	paranoid_entry
889	.else
890	call	error_entry
891	.endif
892	UNWIND_HINT_REGS
893	/* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */
894
895	.if \paranoid
896	.if \shift_ist != -1
897	TRACE_IRQS_OFF_DEBUG			/* reload IDT in case of recursion */
898	.else
899	TRACE_IRQS_OFF
900	.endif
901	.endif
902
903	movq	%rsp, %rdi			/* pt_regs pointer */
904
905	.if \has_error_code
906	movq	ORIG_RAX(%rsp), %rsi		/* get error code */
907	movq	$-1, ORIG_RAX(%rsp)		/* no syscall to restart */
908	.else
909	xorl	%esi, %esi			/* no error code */
910	.endif
911
912	.if \shift_ist != -1
913	subq	$EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
914	.endif
915
916	call	\do_sym
917
918	.if \shift_ist != -1
919	addq	$EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
920	.endif
921
922	/* these procedures expect "no swapgs" flag in ebx */
923	.if \paranoid
924	jmp	paranoid_exit
925	.else
926	jmp	error_exit
927	.endif
928
929	.if \paranoid < 2
930	/*
931	 * Entry from userspace.  Switch stacks and treat it
932	 * as a normal entry.  This means that paranoid handlers
933	 * run in real process context if user_mode(regs).
934	 */
935.Lfrom_usermode_switch_stack_\@:
936	call	error_entry
937
938	movq	%rsp, %rdi			/* pt_regs pointer */
939
940	.if \has_error_code
941	movq	ORIG_RAX(%rsp), %rsi		/* get error code */
942	movq	$-1, ORIG_RAX(%rsp)		/* no syscall to restart */
943	.else
944	xorl	%esi, %esi			/* no error code */
945	.endif
946
947	call	\do_sym
948
949	jmp	error_exit			/* %ebx: no swapgs flag */
950	.endif
951END(\sym)
952.endm
953
954idtentry divide_error			do_divide_error			has_error_code=0
955idtentry overflow			do_overflow			has_error_code=0
956idtentry bounds				do_bounds			has_error_code=0
957idtentry invalid_op			do_invalid_op			has_error_code=0
958idtentry device_not_available		do_device_not_available		has_error_code=0
959idtentry double_fault			do_double_fault			has_error_code=1 paranoid=2
960idtentry coprocessor_segment_overrun	do_coprocessor_segment_overrun	has_error_code=0
961idtentry invalid_TSS			do_invalid_TSS			has_error_code=1
962idtentry segment_not_present		do_segment_not_present		has_error_code=1
963idtentry spurious_interrupt_bug		do_spurious_interrupt_bug	has_error_code=0
964idtentry coprocessor_error		do_coprocessor_error		has_error_code=0
965idtentry alignment_check		do_alignment_check		has_error_code=1
966idtentry simd_coprocessor_error		do_simd_coprocessor_error	has_error_code=0
967
968
969	/*
970	 * Reload gs selector with exception handling
971	 * edi:  new selector
972	 */
973ENTRY(native_load_gs_index)
974	FRAME_BEGIN
975	pushfq
976	DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
977	TRACE_IRQS_OFF
978	SWAPGS
979.Lgs_change:
980	movl	%edi, %gs
9812:	ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE
982	SWAPGS
983	TRACE_IRQS_FLAGS (%rsp)
984	popfq
985	FRAME_END
986	ret
987ENDPROC(native_load_gs_index)
988EXPORT_SYMBOL(native_load_gs_index)
989
990	_ASM_EXTABLE(.Lgs_change, bad_gs)
991	.section .fixup, "ax"
992	/* running with kernelgs */
993bad_gs:
994	SWAPGS					/* switch back to user gs */
995.macro ZAP_GS
996	/* This can't be a string because the preprocessor needs to see it. */
997	movl $__USER_DS, %eax
998	movl %eax, %gs
999.endm
1000	ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG
1001	xorl	%eax, %eax
1002	movl	%eax, %gs
1003	jmp	2b
1004	.previous
1005
1006/* Call softirq on interrupt stack. Interrupts are off. */
1007ENTRY(do_softirq_own_stack)
1008	pushq	%rbp
1009	mov	%rsp, %rbp
1010	ENTER_IRQ_STACK regs=0 old_rsp=%r11
1011	call	__do_softirq
1012	LEAVE_IRQ_STACK regs=0
1013	leaveq
1014	ret
1015ENDPROC(do_softirq_own_stack)
1016
1017#ifdef CONFIG_XEN
1018idtentry hypervisor_callback xen_do_hypervisor_callback has_error_code=0
1019
1020/*
1021 * A note on the "critical region" in our callback handler.
1022 * We want to avoid stacking callback handlers due to events occurring
1023 * during handling of the last event. To do this, we keep events disabled
1024 * until we've done all processing. HOWEVER, we must enable events before
1025 * popping the stack frame (can't be done atomically) and so it would still
1026 * be possible to get enough handler activations to overflow the stack.
1027 * Although unlikely, bugs of that kind are hard to track down, so we'd
1028 * like to avoid the possibility.
1029 * So, on entry to the handler we detect whether we interrupted an
1030 * existing activation in its critical region -- if so, we pop the current
1031 * activation and restart the handler using the previous one.
1032 */
1033ENTRY(xen_do_hypervisor_callback)		/* do_hypervisor_callback(struct *pt_regs) */
1034
1035/*
1036 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1037 * see the correct pointer to the pt_regs
1038 */
1039	UNWIND_HINT_FUNC
1040	movq	%rdi, %rsp			/* we don't return, adjust the stack frame */
1041	UNWIND_HINT_REGS
1042
1043	ENTER_IRQ_STACK old_rsp=%r10
1044	call	xen_evtchn_do_upcall
1045	LEAVE_IRQ_STACK
1046
1047#ifndef CONFIG_PREEMPT
1048	call	xen_maybe_preempt_hcall
1049#endif
1050	jmp	error_exit
1051END(xen_do_hypervisor_callback)
1052
1053/*
1054 * Hypervisor uses this for application faults while it executes.
1055 * We get here for two reasons:
1056 *  1. Fault while reloading DS, ES, FS or GS
1057 *  2. Fault while executing IRET
1058 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1059 * registers that could be reloaded and zeroed the others.
1060 * Category 2 we fix up by killing the current process. We cannot use the
1061 * normal Linux return path in this case because if we use the IRET hypercall
1062 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1063 * We distinguish between categories by comparing each saved segment register
1064 * with its current contents: any discrepancy means we in category 1.
1065 */
1066ENTRY(xen_failsafe_callback)
1067	UNWIND_HINT_EMPTY
1068	movl	%ds, %ecx
1069	cmpw	%cx, 0x10(%rsp)
1070	jne	1f
1071	movl	%es, %ecx
1072	cmpw	%cx, 0x18(%rsp)
1073	jne	1f
1074	movl	%fs, %ecx
1075	cmpw	%cx, 0x20(%rsp)
1076	jne	1f
1077	movl	%gs, %ecx
1078	cmpw	%cx, 0x28(%rsp)
1079	jne	1f
1080	/* All segments match their saved values => Category 2 (Bad IRET). */
1081	movq	(%rsp), %rcx
1082	movq	8(%rsp), %r11
1083	addq	$0x30, %rsp
1084	pushq	$0				/* RIP */
1085	UNWIND_HINT_IRET_REGS offset=8
1086	jmp	general_protection
10871:	/* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1088	movq	(%rsp), %rcx
1089	movq	8(%rsp), %r11
1090	addq	$0x30, %rsp
1091	UNWIND_HINT_IRET_REGS
1092	pushq	$-1 /* orig_ax = -1 => not a system call */
1093	PUSH_AND_CLEAR_REGS
1094	ENCODE_FRAME_POINTER
1095	jmp	error_exit
1096END(xen_failsafe_callback)
1097
1098apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1099	xen_hvm_callback_vector xen_evtchn_do_upcall
1100
1101#endif /* CONFIG_XEN */
1102
1103#if IS_ENABLED(CONFIG_HYPERV)
1104apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1105	hyperv_callback_vector hyperv_vector_handler
1106
1107apicinterrupt3 HYPERV_REENLIGHTENMENT_VECTOR \
1108	hyperv_reenlightenment_vector hyperv_reenlightenment_intr
1109#endif /* CONFIG_HYPERV */
1110
1111idtentry debug			do_debug		has_error_code=0	paranoid=1 shift_ist=DEBUG_STACK
1112idtentry int3			do_int3			has_error_code=0	paranoid=1 shift_ist=DEBUG_STACK
1113idtentry stack_segment		do_stack_segment	has_error_code=1
1114
1115#ifdef CONFIG_XEN
1116idtentry xennmi			do_nmi			has_error_code=0
1117idtentry xendebug		do_debug		has_error_code=0
1118idtentry xenint3		do_int3			has_error_code=0
1119#endif
1120
1121idtentry general_protection	do_general_protection	has_error_code=1
1122idtentry page_fault		do_page_fault		has_error_code=1
1123
1124#ifdef CONFIG_KVM_GUEST
1125idtentry async_page_fault	do_async_page_fault	has_error_code=1
1126#endif
1127
1128#ifdef CONFIG_X86_MCE
1129idtentry machine_check		do_mce			has_error_code=0	paranoid=1
1130#endif
1131
1132/*
1133 * Switch gs if needed.
1134 * Use slow, but surefire "are we in kernel?" check.
1135 * Return: ebx=0: need swapgs on exit, ebx=1: otherwise
1136 */
1137ENTRY(paranoid_entry)
1138	UNWIND_HINT_FUNC
1139	cld
1140	movl	$1, %ebx
1141	movl	$MSR_GS_BASE, %ecx
1142	rdmsr
1143	testl	%edx, %edx
1144	js	1f				/* negative -> in kernel */
1145	SWAPGS
1146	xorl	%ebx, %ebx
1147
11481:
1149	SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14
1150
1151	ret
1152END(paranoid_entry)
1153
1154/*
1155 * "Paranoid" exit path from exception stack.  This is invoked
1156 * only on return from non-NMI IST interrupts that came
1157 * from kernel space.
1158 *
1159 * We may be returning to very strange contexts (e.g. very early
1160 * in syscall entry), so checking for preemption here would
1161 * be complicated.  Fortunately, we there's no good reason
1162 * to try to handle preemption here.
1163 *
1164 * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it)
1165 */
1166ENTRY(paranoid_exit)
1167	UNWIND_HINT_REGS
1168	DISABLE_INTERRUPTS(CLBR_ANY)
1169	TRACE_IRQS_OFF_DEBUG
1170	testl	%ebx, %ebx			/* swapgs needed? */
1171	jnz	.Lparanoid_exit_no_swapgs
1172	TRACE_IRQS_IRETQ
1173	RESTORE_CR3	scratch_reg=%rbx save_reg=%r14
1174	SWAPGS_UNSAFE_STACK
1175	jmp	.Lparanoid_exit_restore
1176.Lparanoid_exit_no_swapgs:
1177	TRACE_IRQS_IRETQ_DEBUG
1178	RESTORE_CR3	scratch_reg=%rbx save_reg=%r14
1179.Lparanoid_exit_restore:
1180	jmp restore_regs_and_return_to_kernel
1181END(paranoid_exit)
1182
1183/*
1184 * Switch gs if needed.
1185 * Return: EBX=0: came from user mode; EBX=1: otherwise
1186 */
1187ENTRY(error_entry)
1188	UNWIND_HINT_REGS offset=8
1189	cld
1190	testb	$3, CS+8(%rsp)
1191	jz	.Lerror_kernelspace
1192
1193	/*
1194	 * We entered from user mode or we're pretending to have entered
1195	 * from user mode due to an IRET fault.
1196	 */
1197	SWAPGS
1198	/* We have user CR3.  Change to kernel CR3. */
1199	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1200
1201.Lerror_entry_from_usermode_after_swapgs:
1202	/* Put us onto the real thread stack. */
1203	popq	%r12				/* save return addr in %12 */
1204	movq	%rsp, %rdi			/* arg0 = pt_regs pointer */
1205	call	sync_regs
1206	movq	%rax, %rsp			/* switch stack */
1207	ENCODE_FRAME_POINTER
1208	pushq	%r12
1209
1210	/*
1211	 * We need to tell lockdep that IRQs are off.  We can't do this until
1212	 * we fix gsbase, and we should do it before enter_from_user_mode
1213	 * (which can take locks).
1214	 */
1215	TRACE_IRQS_OFF
1216	CALL_enter_from_user_mode
1217	ret
1218
1219.Lerror_entry_done:
1220	TRACE_IRQS_OFF
1221	ret
1222
1223	/*
1224	 * There are two places in the kernel that can potentially fault with
1225	 * usergs. Handle them here.  B stepping K8s sometimes report a
1226	 * truncated RIP for IRET exceptions returning to compat mode. Check
1227	 * for these here too.
1228	 */
1229.Lerror_kernelspace:
1230	incl	%ebx
1231	leaq	native_irq_return_iret(%rip), %rcx
1232	cmpq	%rcx, RIP+8(%rsp)
1233	je	.Lerror_bad_iret
1234	movl	%ecx, %eax			/* zero extend */
1235	cmpq	%rax, RIP+8(%rsp)
1236	je	.Lbstep_iret
1237	cmpq	$.Lgs_change, RIP+8(%rsp)
1238	jne	.Lerror_entry_done
1239
1240	/*
1241	 * hack: .Lgs_change can fail with user gsbase.  If this happens, fix up
1242	 * gsbase and proceed.  We'll fix up the exception and land in
1243	 * .Lgs_change's error handler with kernel gsbase.
1244	 */
1245	SWAPGS
1246	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1247	jmp .Lerror_entry_done
1248
1249.Lbstep_iret:
1250	/* Fix truncated RIP */
1251	movq	%rcx, RIP+8(%rsp)
1252	/* fall through */
1253
1254.Lerror_bad_iret:
1255	/*
1256	 * We came from an IRET to user mode, so we have user
1257	 * gsbase and CR3.  Switch to kernel gsbase and CR3:
1258	 */
1259	SWAPGS
1260	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1261
1262	/*
1263	 * Pretend that the exception came from user mode: set up pt_regs
1264	 * as if we faulted immediately after IRET and clear EBX so that
1265	 * error_exit knows that we will be returning to user mode.
1266	 */
1267	mov	%rsp, %rdi
1268	call	fixup_bad_iret
1269	mov	%rax, %rsp
1270	decl	%ebx
1271	jmp	.Lerror_entry_from_usermode_after_swapgs
1272END(error_entry)
1273
1274
1275/*
1276 * On entry, EBX is a "return to kernel mode" flag:
1277 *   1: already in kernel mode, don't need SWAPGS
1278 *   0: user gsbase is loaded, we need SWAPGS and standard preparation for return to usermode
1279 */
1280ENTRY(error_exit)
1281	UNWIND_HINT_REGS
1282	DISABLE_INTERRUPTS(CLBR_ANY)
1283	TRACE_IRQS_OFF
1284	testl	%ebx, %ebx
1285	jnz	retint_kernel
1286	jmp	retint_user
1287END(error_exit)
1288
1289/*
1290 * Runs on exception stack.  Xen PV does not go through this path at all,
1291 * so we can use real assembly here.
1292 *
1293 * Registers:
1294 *	%r14: Used to save/restore the CR3 of the interrupted context
1295 *	      when PAGE_TABLE_ISOLATION is in use.  Do not clobber.
1296 */
1297ENTRY(nmi)
1298	UNWIND_HINT_IRET_REGS
1299
1300	/*
1301	 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1302	 * the iretq it performs will take us out of NMI context.
1303	 * This means that we can have nested NMIs where the next
1304	 * NMI is using the top of the stack of the previous NMI. We
1305	 * can't let it execute because the nested NMI will corrupt the
1306	 * stack of the previous NMI. NMI handlers are not re-entrant
1307	 * anyway.
1308	 *
1309	 * To handle this case we do the following:
1310	 *  Check the a special location on the stack that contains
1311	 *  a variable that is set when NMIs are executing.
1312	 *  The interrupted task's stack is also checked to see if it
1313	 *  is an NMI stack.
1314	 *  If the variable is not set and the stack is not the NMI
1315	 *  stack then:
1316	 *    o Set the special variable on the stack
1317	 *    o Copy the interrupt frame into an "outermost" location on the
1318	 *      stack
1319	 *    o Copy the interrupt frame into an "iret" location on the stack
1320	 *    o Continue processing the NMI
1321	 *  If the variable is set or the previous stack is the NMI stack:
1322	 *    o Modify the "iret" location to jump to the repeat_nmi
1323	 *    o return back to the first NMI
1324	 *
1325	 * Now on exit of the first NMI, we first clear the stack variable
1326	 * The NMI stack will tell any nested NMIs at that point that it is
1327	 * nested. Then we pop the stack normally with iret, and if there was
1328	 * a nested NMI that updated the copy interrupt stack frame, a
1329	 * jump will be made to the repeat_nmi code that will handle the second
1330	 * NMI.
1331	 *
1332	 * However, espfix prevents us from directly returning to userspace
1333	 * with a single IRET instruction.  Similarly, IRET to user mode
1334	 * can fault.  We therefore handle NMIs from user space like
1335	 * other IST entries.
1336	 */
1337
1338	ASM_CLAC
1339
1340	/* Use %rdx as our temp variable throughout */
1341	pushq	%rdx
1342
1343	testb	$3, CS-RIP+8(%rsp)
1344	jz	.Lnmi_from_kernel
1345
1346	/*
1347	 * NMI from user mode.  We need to run on the thread stack, but we
1348	 * can't go through the normal entry paths: NMIs are masked, and
1349	 * we don't want to enable interrupts, because then we'll end
1350	 * up in an awkward situation in which IRQs are on but NMIs
1351	 * are off.
1352	 *
1353	 * We also must not push anything to the stack before switching
1354	 * stacks lest we corrupt the "NMI executing" variable.
1355	 */
1356
1357	swapgs
1358	cld
1359	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdx
1360	movq	%rsp, %rdx
1361	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
1362	UNWIND_HINT_IRET_REGS base=%rdx offset=8
1363	pushq	5*8(%rdx)	/* pt_regs->ss */
1364	pushq	4*8(%rdx)	/* pt_regs->rsp */
1365	pushq	3*8(%rdx)	/* pt_regs->flags */
1366	pushq	2*8(%rdx)	/* pt_regs->cs */
1367	pushq	1*8(%rdx)	/* pt_regs->rip */
1368	UNWIND_HINT_IRET_REGS
1369	pushq   $-1		/* pt_regs->orig_ax */
1370	PUSH_AND_CLEAR_REGS rdx=(%rdx)
1371	ENCODE_FRAME_POINTER
1372
1373	/*
1374	 * At this point we no longer need to worry about stack damage
1375	 * due to nesting -- we're on the normal thread stack and we're
1376	 * done with the NMI stack.
1377	 */
1378
1379	movq	%rsp, %rdi
1380	movq	$-1, %rsi
1381	call	do_nmi
1382
1383	/*
1384	 * Return back to user mode.  We must *not* do the normal exit
1385	 * work, because we don't want to enable interrupts.
1386	 */
1387	jmp	swapgs_restore_regs_and_return_to_usermode
1388
1389.Lnmi_from_kernel:
1390	/*
1391	 * Here's what our stack frame will look like:
1392	 * +---------------------------------------------------------+
1393	 * | original SS                                             |
1394	 * | original Return RSP                                     |
1395	 * | original RFLAGS                                         |
1396	 * | original CS                                             |
1397	 * | original RIP                                            |
1398	 * +---------------------------------------------------------+
1399	 * | temp storage for rdx                                    |
1400	 * +---------------------------------------------------------+
1401	 * | "NMI executing" variable                                |
1402	 * +---------------------------------------------------------+
1403	 * | iret SS          } Copied from "outermost" frame        |
1404	 * | iret Return RSP  } on each loop iteration; overwritten  |
1405	 * | iret RFLAGS      } by a nested NMI to force another     |
1406	 * | iret CS          } iteration if needed.                 |
1407	 * | iret RIP         }                                      |
1408	 * +---------------------------------------------------------+
1409	 * | outermost SS          } initialized in first_nmi;       |
1410	 * | outermost Return RSP  } will not be changed before      |
1411	 * | outermost RFLAGS      } NMI processing is done.         |
1412	 * | outermost CS          } Copied to "iret" frame on each  |
1413	 * | outermost RIP         } iteration.                      |
1414	 * +---------------------------------------------------------+
1415	 * | pt_regs                                                 |
1416	 * +---------------------------------------------------------+
1417	 *
1418	 * The "original" frame is used by hardware.  Before re-enabling
1419	 * NMIs, we need to be done with it, and we need to leave enough
1420	 * space for the asm code here.
1421	 *
1422	 * We return by executing IRET while RSP points to the "iret" frame.
1423	 * That will either return for real or it will loop back into NMI
1424	 * processing.
1425	 *
1426	 * The "outermost" frame is copied to the "iret" frame on each
1427	 * iteration of the loop, so each iteration starts with the "iret"
1428	 * frame pointing to the final return target.
1429	 */
1430
1431	/*
1432	 * Determine whether we're a nested NMI.
1433	 *
1434	 * If we interrupted kernel code between repeat_nmi and
1435	 * end_repeat_nmi, then we are a nested NMI.  We must not
1436	 * modify the "iret" frame because it's being written by
1437	 * the outer NMI.  That's okay; the outer NMI handler is
1438	 * about to about to call do_nmi anyway, so we can just
1439	 * resume the outer NMI.
1440	 */
1441
1442	movq	$repeat_nmi, %rdx
1443	cmpq	8(%rsp), %rdx
1444	ja	1f
1445	movq	$end_repeat_nmi, %rdx
1446	cmpq	8(%rsp), %rdx
1447	ja	nested_nmi_out
14481:
1449
1450	/*
1451	 * Now check "NMI executing".  If it's set, then we're nested.
1452	 * This will not detect if we interrupted an outer NMI just
1453	 * before IRET.
1454	 */
1455	cmpl	$1, -8(%rsp)
1456	je	nested_nmi
1457
1458	/*
1459	 * Now test if the previous stack was an NMI stack.  This covers
1460	 * the case where we interrupt an outer NMI after it clears
1461	 * "NMI executing" but before IRET.  We need to be careful, though:
1462	 * there is one case in which RSP could point to the NMI stack
1463	 * despite there being no NMI active: naughty userspace controls
1464	 * RSP at the very beginning of the SYSCALL targets.  We can
1465	 * pull a fast one on naughty userspace, though: we program
1466	 * SYSCALL to mask DF, so userspace cannot cause DF to be set
1467	 * if it controls the kernel's RSP.  We set DF before we clear
1468	 * "NMI executing".
1469	 */
1470	lea	6*8(%rsp), %rdx
1471	/* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */
1472	cmpq	%rdx, 4*8(%rsp)
1473	/* If the stack pointer is above the NMI stack, this is a normal NMI */
1474	ja	first_nmi
1475
1476	subq	$EXCEPTION_STKSZ, %rdx
1477	cmpq	%rdx, 4*8(%rsp)
1478	/* If it is below the NMI stack, it is a normal NMI */
1479	jb	first_nmi
1480
1481	/* Ah, it is within the NMI stack. */
1482
1483	testb	$(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp)
1484	jz	first_nmi	/* RSP was user controlled. */
1485
1486	/* This is a nested NMI. */
1487
1488nested_nmi:
1489	/*
1490	 * Modify the "iret" frame to point to repeat_nmi, forcing another
1491	 * iteration of NMI handling.
1492	 */
1493	subq	$8, %rsp
1494	leaq	-10*8(%rsp), %rdx
1495	pushq	$__KERNEL_DS
1496	pushq	%rdx
1497	pushfq
1498	pushq	$__KERNEL_CS
1499	pushq	$repeat_nmi
1500
1501	/* Put stack back */
1502	addq	$(6*8), %rsp
1503
1504nested_nmi_out:
1505	popq	%rdx
1506
1507	/* We are returning to kernel mode, so this cannot result in a fault. */
1508	iretq
1509
1510first_nmi:
1511	/* Restore rdx. */
1512	movq	(%rsp), %rdx
1513
1514	/* Make room for "NMI executing". */
1515	pushq	$0
1516
1517	/* Leave room for the "iret" frame */
1518	subq	$(5*8), %rsp
1519
1520	/* Copy the "original" frame to the "outermost" frame */
1521	.rept 5
1522	pushq	11*8(%rsp)
1523	.endr
1524	UNWIND_HINT_IRET_REGS
1525
1526	/* Everything up to here is safe from nested NMIs */
1527
1528#ifdef CONFIG_DEBUG_ENTRY
1529	/*
1530	 * For ease of testing, unmask NMIs right away.  Disabled by
1531	 * default because IRET is very expensive.
1532	 */
1533	pushq	$0		/* SS */
1534	pushq	%rsp		/* RSP (minus 8 because of the previous push) */
1535	addq	$8, (%rsp)	/* Fix up RSP */
1536	pushfq			/* RFLAGS */
1537	pushq	$__KERNEL_CS	/* CS */
1538	pushq	$1f		/* RIP */
1539	iretq			/* continues at repeat_nmi below */
1540	UNWIND_HINT_IRET_REGS
15411:
1542#endif
1543
1544repeat_nmi:
1545	/*
1546	 * If there was a nested NMI, the first NMI's iret will return
1547	 * here. But NMIs are still enabled and we can take another
1548	 * nested NMI. The nested NMI checks the interrupted RIP to see
1549	 * if it is between repeat_nmi and end_repeat_nmi, and if so
1550	 * it will just return, as we are about to repeat an NMI anyway.
1551	 * This makes it safe to copy to the stack frame that a nested
1552	 * NMI will update.
1553	 *
1554	 * RSP is pointing to "outermost RIP".  gsbase is unknown, but, if
1555	 * we're repeating an NMI, gsbase has the same value that it had on
1556	 * the first iteration.  paranoid_entry will load the kernel
1557	 * gsbase if needed before we call do_nmi.  "NMI executing"
1558	 * is zero.
1559	 */
1560	movq	$1, 10*8(%rsp)		/* Set "NMI executing". */
1561
1562	/*
1563	 * Copy the "outermost" frame to the "iret" frame.  NMIs that nest
1564	 * here must not modify the "iret" frame while we're writing to
1565	 * it or it will end up containing garbage.
1566	 */
1567	addq	$(10*8), %rsp
1568	.rept 5
1569	pushq	-6*8(%rsp)
1570	.endr
1571	subq	$(5*8), %rsp
1572end_repeat_nmi:
1573
1574	/*
1575	 * Everything below this point can be preempted by a nested NMI.
1576	 * If this happens, then the inner NMI will change the "iret"
1577	 * frame to point back to repeat_nmi.
1578	 */
1579	pushq	$-1				/* ORIG_RAX: no syscall to restart */
1580	PUSH_AND_CLEAR_REGS
1581	ENCODE_FRAME_POINTER
1582
1583	/*
1584	 * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit
1585	 * as we should not be calling schedule in NMI context.
1586	 * Even with normal interrupts enabled. An NMI should not be
1587	 * setting NEED_RESCHED or anything that normal interrupts and
1588	 * exceptions might do.
1589	 */
1590	call	paranoid_entry
1591	UNWIND_HINT_REGS
1592
1593	/* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1594	movq	%rsp, %rdi
1595	movq	$-1, %rsi
1596	call	do_nmi
1597
1598	RESTORE_CR3 scratch_reg=%r15 save_reg=%r14
1599
1600	testl	%ebx, %ebx			/* swapgs needed? */
1601	jnz	nmi_restore
1602nmi_swapgs:
1603	SWAPGS_UNSAFE_STACK
1604nmi_restore:
1605	POP_REGS
1606
1607	/*
1608	 * Skip orig_ax and the "outermost" frame to point RSP at the "iret"
1609	 * at the "iret" frame.
1610	 */
1611	addq	$6*8, %rsp
1612
1613	/*
1614	 * Clear "NMI executing".  Set DF first so that we can easily
1615	 * distinguish the remaining code between here and IRET from
1616	 * the SYSCALL entry and exit paths.
1617	 *
1618	 * We arguably should just inspect RIP instead, but I (Andy) wrote
1619	 * this code when I had the misapprehension that Xen PV supported
1620	 * NMIs, and Xen PV would break that approach.
1621	 */
1622	std
1623	movq	$0, 5*8(%rsp)		/* clear "NMI executing" */
1624
1625	/*
1626	 * iretq reads the "iret" frame and exits the NMI stack in a
1627	 * single instruction.  We are returning to kernel mode, so this
1628	 * cannot result in a fault.  Similarly, we don't need to worry
1629	 * about espfix64 on the way back to kernel mode.
1630	 */
1631	iretq
1632END(nmi)
1633
1634ENTRY(ignore_sysret)
1635	UNWIND_HINT_EMPTY
1636	mov	$-ENOSYS, %eax
1637	sysret
1638END(ignore_sysret)
1639
1640ENTRY(rewind_stack_do_exit)
1641	UNWIND_HINT_FUNC
1642	/* Prevent any naive code from trying to unwind to our caller. */
1643	xorl	%ebp, %ebp
1644
1645	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rax
1646	leaq	-PTREGS_SIZE(%rax), %rsp
1647	UNWIND_HINT_FUNC sp_offset=PTREGS_SIZE
1648
1649	call	do_exit
1650END(rewind_stack_do_exit)
1651