xref: /linux/arch/x86/entry/entry_64.S (revision a2b0fe7435faee6f6fbb27409878013bc4727e98)
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 not 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	pushq	%rdi				/* pt_regs->di */
231	pushq	%rsi				/* pt_regs->si */
232	pushq	%rdx				/* pt_regs->dx */
233	pushq	%rcx				/* pt_regs->cx */
234	pushq	$-ENOSYS			/* pt_regs->ax */
235	pushq	%r8				/* pt_regs->r8 */
236	pushq	%r9				/* pt_regs->r9 */
237	pushq	%r10				/* pt_regs->r10 */
238	pushq	%r11				/* pt_regs->r11 */
239	sub	$(6*8), %rsp			/* pt_regs->bp, bx, r12-15 not saved */
240	UNWIND_HINT_REGS extra=0
241
242	TRACE_IRQS_OFF
243
244	/*
245	 * If we need to do entry work or if we guess we'll need to do
246	 * exit work, go straight to the slow path.
247	 */
248	movq	PER_CPU_VAR(current_task), %r11
249	testl	$_TIF_WORK_SYSCALL_ENTRY|_TIF_ALLWORK_MASK, TASK_TI_flags(%r11)
250	jnz	entry_SYSCALL64_slow_path
251
252entry_SYSCALL_64_fastpath:
253	/*
254	 * Easy case: enable interrupts and issue the syscall.  If the syscall
255	 * needs pt_regs, we'll call a stub that disables interrupts again
256	 * and jumps to the slow path.
257	 */
258	TRACE_IRQS_ON
259	ENABLE_INTERRUPTS(CLBR_NONE)
260#if __SYSCALL_MASK == ~0
261	cmpq	$__NR_syscall_max, %rax
262#else
263	andl	$__SYSCALL_MASK, %eax
264	cmpl	$__NR_syscall_max, %eax
265#endif
266	ja	1f				/* return -ENOSYS (already in pt_regs->ax) */
267	movq	%r10, %rcx
268
269	/*
270	 * This call instruction is handled specially in stub_ptregs_64.
271	 * It might end up jumping to the slow path.  If it jumps, RAX
272	 * and all argument registers are clobbered.
273	 */
274#ifdef CONFIG_RETPOLINE
275	movq	sys_call_table(, %rax, 8), %rax
276	call	__x86_indirect_thunk_rax
277#else
278	call	*sys_call_table(, %rax, 8)
279#endif
280.Lentry_SYSCALL_64_after_fastpath_call:
281
282	movq	%rax, RAX(%rsp)
2831:
284
285	/*
286	 * If we get here, then we know that pt_regs is clean for SYSRET64.
287	 * If we see that no exit work is required (which we are required
288	 * to check with IRQs off), then we can go straight to SYSRET64.
289	 */
290	DISABLE_INTERRUPTS(CLBR_ANY)
291	TRACE_IRQS_OFF
292	movq	PER_CPU_VAR(current_task), %r11
293	testl	$_TIF_ALLWORK_MASK, TASK_TI_flags(%r11)
294	jnz	1f
295
296	LOCKDEP_SYS_EXIT
297	TRACE_IRQS_ON		/* user mode is traced as IRQs on */
298	movq	RIP(%rsp), %rcx
299	movq	EFLAGS(%rsp), %r11
300	addq	$6*8, %rsp	/* skip extra regs -- they were preserved */
301	UNWIND_HINT_EMPTY
302	jmp	.Lpop_c_regs_except_rcx_r11_and_sysret
303
3041:
305	/*
306	 * The fast path looked good when we started, but something changed
307	 * along the way and we need to switch to the slow path.  Calling
308	 * raise(3) will trigger this, for example.  IRQs are off.
309	 */
310	TRACE_IRQS_ON
311	ENABLE_INTERRUPTS(CLBR_ANY)
312	SAVE_EXTRA_REGS
313	movq	%rsp, %rdi
314	call	syscall_return_slowpath	/* returns with IRQs disabled */
315	jmp	return_from_SYSCALL_64
316
317entry_SYSCALL64_slow_path:
318	/* IRQs are off. */
319	SAVE_EXTRA_REGS
320	movq	%rsp, %rdi
321	call	do_syscall_64		/* returns with IRQs disabled */
322
323return_from_SYSCALL_64:
324	TRACE_IRQS_IRETQ		/* we're about to change IF */
325
326	/*
327	 * Try to use SYSRET instead of IRET if we're returning to
328	 * a completely clean 64-bit userspace context.  If we're not,
329	 * go to the slow exit path.
330	 */
331	movq	RCX(%rsp), %rcx
332	movq	RIP(%rsp), %r11
333
334	cmpq	%rcx, %r11	/* SYSRET requires RCX == RIP */
335	jne	swapgs_restore_regs_and_return_to_usermode
336
337	/*
338	 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
339	 * in kernel space.  This essentially lets the user take over
340	 * the kernel, since userspace controls RSP.
341	 *
342	 * If width of "canonical tail" ever becomes variable, this will need
343	 * to be updated to remain correct on both old and new CPUs.
344	 *
345	 * Change top bits to match most significant bit (47th or 56th bit
346	 * depending on paging mode) in the address.
347	 */
348	shl	$(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
349	sar	$(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
350
351	/* If this changed %rcx, it was not canonical */
352	cmpq	%rcx, %r11
353	jne	swapgs_restore_regs_and_return_to_usermode
354
355	cmpq	$__USER_CS, CS(%rsp)		/* CS must match SYSRET */
356	jne	swapgs_restore_regs_and_return_to_usermode
357
358	movq	R11(%rsp), %r11
359	cmpq	%r11, EFLAGS(%rsp)		/* R11 == RFLAGS */
360	jne	swapgs_restore_regs_and_return_to_usermode
361
362	/*
363	 * SYSCALL clears RF when it saves RFLAGS in R11 and SYSRET cannot
364	 * restore RF properly. If the slowpath sets it for whatever reason, we
365	 * need to restore it correctly.
366	 *
367	 * SYSRET can restore TF, but unlike IRET, restoring TF results in a
368	 * trap from userspace immediately after SYSRET.  This would cause an
369	 * infinite loop whenever #DB happens with register state that satisfies
370	 * the opportunistic SYSRET conditions.  For example, single-stepping
371	 * this user code:
372	 *
373	 *           movq	$stuck_here, %rcx
374	 *           pushfq
375	 *           popq %r11
376	 *   stuck_here:
377	 *
378	 * would never get past 'stuck_here'.
379	 */
380	testq	$(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11
381	jnz	swapgs_restore_regs_and_return_to_usermode
382
383	/* nothing to check for RSP */
384
385	cmpq	$__USER_DS, SS(%rsp)		/* SS must match SYSRET */
386	jne	swapgs_restore_regs_and_return_to_usermode
387
388	/*
389	 * We win! This label is here just for ease of understanding
390	 * perf profiles. Nothing jumps here.
391	 */
392syscall_return_via_sysret:
393	/* rcx and r11 are already restored (see code above) */
394	UNWIND_HINT_EMPTY
395	POP_EXTRA_REGS
396.Lpop_c_regs_except_rcx_r11_and_sysret:
397	popq	%rsi	/* skip r11 */
398	popq	%r10
399	popq	%r9
400	popq	%r8
401	popq	%rax
402	popq	%rsi	/* skip rcx */
403	popq	%rdx
404	popq	%rsi
405
406	/*
407	 * Now all regs are restored except RSP and RDI.
408	 * Save old stack pointer and switch to trampoline stack.
409	 */
410	movq	%rsp, %rdi
411	movq	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
412
413	pushq	RSP-RDI(%rdi)	/* RSP */
414	pushq	(%rdi)		/* RDI */
415
416	/*
417	 * We are on the trampoline stack.  All regs except RDI are live.
418	 * We can do future final exit work right here.
419	 */
420	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
421
422	popq	%rdi
423	popq	%rsp
424	USERGS_SYSRET64
425END(entry_SYSCALL_64)
426
427ENTRY(stub_ptregs_64)
428	/*
429	 * Syscalls marked as needing ptregs land here.
430	 * If we are on the fast path, we need to save the extra regs,
431	 * which we achieve by trying again on the slow path.  If we are on
432	 * the slow path, the extra regs are already saved.
433	 *
434	 * RAX stores a pointer to the C function implementing the syscall.
435	 * IRQs are on.
436	 */
437	cmpq	$.Lentry_SYSCALL_64_after_fastpath_call, (%rsp)
438	jne	1f
439
440	/*
441	 * Called from fast path -- disable IRQs again, pop return address
442	 * and jump to slow path
443	 */
444	DISABLE_INTERRUPTS(CLBR_ANY)
445	TRACE_IRQS_OFF
446	popq	%rax
447	UNWIND_HINT_REGS extra=0
448	jmp	entry_SYSCALL64_slow_path
449
4501:
451	JMP_NOSPEC %rax				/* Called from C */
452END(stub_ptregs_64)
453
454.macro ptregs_stub func
455ENTRY(ptregs_\func)
456	UNWIND_HINT_FUNC
457	leaq	\func(%rip), %rax
458	jmp	stub_ptregs_64
459END(ptregs_\func)
460.endm
461
462/* Instantiate ptregs_stub for each ptregs-using syscall */
463#define __SYSCALL_64_QUAL_(sym)
464#define __SYSCALL_64_QUAL_ptregs(sym) ptregs_stub sym
465#define __SYSCALL_64(nr, sym, qual) __SYSCALL_64_QUAL_##qual(sym)
466#include <asm/syscalls_64.h>
467
468/*
469 * %rdi: prev task
470 * %rsi: next task
471 */
472ENTRY(__switch_to_asm)
473	UNWIND_HINT_FUNC
474	/*
475	 * Save callee-saved registers
476	 * This must match the order in inactive_task_frame
477	 */
478	pushq	%rbp
479	pushq	%rbx
480	pushq	%r12
481	pushq	%r13
482	pushq	%r14
483	pushq	%r15
484
485	/* switch stack */
486	movq	%rsp, TASK_threadsp(%rdi)
487	movq	TASK_threadsp(%rsi), %rsp
488
489#ifdef CONFIG_CC_STACKPROTECTOR
490	movq	TASK_stack_canary(%rsi), %rbx
491	movq	%rbx, PER_CPU_VAR(irq_stack_union)+stack_canary_offset
492#endif
493
494#ifdef CONFIG_RETPOLINE
495	/*
496	 * When switching from a shallower to a deeper call stack
497	 * the RSB may either underflow or use entries populated
498	 * with userspace addresses. On CPUs where those concerns
499	 * exist, overwrite the RSB with entries which capture
500	 * speculative execution to prevent attack.
501	 */
502	/* Clobbers %rbx */
503	FILL_RETURN_BUFFER RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
504#endif
505
506	/* restore callee-saved registers */
507	popq	%r15
508	popq	%r14
509	popq	%r13
510	popq	%r12
511	popq	%rbx
512	popq	%rbp
513
514	jmp	__switch_to
515END(__switch_to_asm)
516
517/*
518 * A newly forked process directly context switches into this address.
519 *
520 * rax: prev task we switched from
521 * rbx: kernel thread func (NULL for user thread)
522 * r12: kernel thread arg
523 */
524ENTRY(ret_from_fork)
525	UNWIND_HINT_EMPTY
526	movq	%rax, %rdi
527	call	schedule_tail			/* rdi: 'prev' task parameter */
528
529	testq	%rbx, %rbx			/* from kernel_thread? */
530	jnz	1f				/* kernel threads are uncommon */
531
5322:
533	UNWIND_HINT_REGS
534	movq	%rsp, %rdi
535	call	syscall_return_slowpath	/* returns with IRQs disabled */
536	TRACE_IRQS_ON			/* user mode is traced as IRQS on */
537	jmp	swapgs_restore_regs_and_return_to_usermode
538
5391:
540	/* kernel thread */
541	movq	%r12, %rdi
542	CALL_NOSPEC %rbx
543	/*
544	 * A kernel thread is allowed to return here after successfully
545	 * calling do_execve().  Exit to userspace to complete the execve()
546	 * syscall.
547	 */
548	movq	$0, RAX(%rsp)
549	jmp	2b
550END(ret_from_fork)
551
552/*
553 * Build the entry stubs with some assembler magic.
554 * We pack 1 stub into every 8-byte block.
555 */
556	.align 8
557ENTRY(irq_entries_start)
558    vector=FIRST_EXTERNAL_VECTOR
559    .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
560	UNWIND_HINT_IRET_REGS
561	pushq	$(~vector+0x80)			/* Note: always in signed byte range */
562	jmp	common_interrupt
563	.align	8
564	vector=vector+1
565    .endr
566END(irq_entries_start)
567
568.macro DEBUG_ENTRY_ASSERT_IRQS_OFF
569#ifdef CONFIG_DEBUG_ENTRY
570	pushq %rax
571	SAVE_FLAGS(CLBR_RAX)
572	testl $X86_EFLAGS_IF, %eax
573	jz .Lokay_\@
574	ud2
575.Lokay_\@:
576	popq %rax
577#endif
578.endm
579
580/*
581 * Enters the IRQ stack if we're not already using it.  NMI-safe.  Clobbers
582 * flags and puts old RSP into old_rsp, and leaves all other GPRs alone.
583 * Requires kernel GSBASE.
584 *
585 * The invariant is that, if irq_count != -1, then the IRQ stack is in use.
586 */
587.macro ENTER_IRQ_STACK regs=1 old_rsp
588	DEBUG_ENTRY_ASSERT_IRQS_OFF
589	movq	%rsp, \old_rsp
590
591	.if \regs
592	UNWIND_HINT_REGS base=\old_rsp
593	.endif
594
595	incl	PER_CPU_VAR(irq_count)
596	jnz	.Lirq_stack_push_old_rsp_\@
597
598	/*
599	 * Right now, if we just incremented irq_count to zero, we've
600	 * claimed the IRQ stack but we haven't switched to it yet.
601	 *
602	 * If anything is added that can interrupt us here without using IST,
603	 * it must be *extremely* careful to limit its stack usage.  This
604	 * could include kprobes and a hypothetical future IST-less #DB
605	 * handler.
606	 *
607	 * The OOPS unwinder relies on the word at the top of the IRQ
608	 * stack linking back to the previous RSP for the entire time we're
609	 * on the IRQ stack.  For this to work reliably, we need to write
610	 * it before we actually move ourselves to the IRQ stack.
611	 */
612
613	movq	\old_rsp, PER_CPU_VAR(irq_stack_union + IRQ_STACK_SIZE - 8)
614	movq	PER_CPU_VAR(irq_stack_ptr), %rsp
615
616#ifdef CONFIG_DEBUG_ENTRY
617	/*
618	 * If the first movq above becomes wrong due to IRQ stack layout
619	 * changes, the only way we'll notice is if we try to unwind right
620	 * here.  Assert that we set up the stack right to catch this type
621	 * of bug quickly.
622	 */
623	cmpq	-8(%rsp), \old_rsp
624	je	.Lirq_stack_okay\@
625	ud2
626	.Lirq_stack_okay\@:
627#endif
628
629.Lirq_stack_push_old_rsp_\@:
630	pushq	\old_rsp
631
632	.if \regs
633	UNWIND_HINT_REGS indirect=1
634	.endif
635.endm
636
637/*
638 * Undoes ENTER_IRQ_STACK.
639 */
640.macro LEAVE_IRQ_STACK regs=1
641	DEBUG_ENTRY_ASSERT_IRQS_OFF
642	/* We need to be off the IRQ stack before decrementing irq_count. */
643	popq	%rsp
644
645	.if \regs
646	UNWIND_HINT_REGS
647	.endif
648
649	/*
650	 * As in ENTER_IRQ_STACK, irq_count == 0, we are still claiming
651	 * the irq stack but we're not on it.
652	 */
653
654	decl	PER_CPU_VAR(irq_count)
655.endm
656
657/*
658 * Interrupt entry/exit.
659 *
660 * Interrupt entry points save only callee clobbered registers in fast path.
661 *
662 * Entry runs with interrupts off.
663 */
664
665/* 0(%rsp): ~(interrupt number) */
666	.macro interrupt func
667	cld
668
669	testb	$3, CS-ORIG_RAX(%rsp)
670	jz	1f
671	SWAPGS
672	call	switch_to_thread_stack
6731:
674
675	ALLOC_PT_GPREGS_ON_STACK
676	SAVE_C_REGS
677	SAVE_EXTRA_REGS
678	ENCODE_FRAME_POINTER
679
680	testb	$3, CS(%rsp)
681	jz	1f
682
683	/*
684	 * IRQ from user mode.
685	 *
686	 * We need to tell lockdep that IRQs are off.  We can't do this until
687	 * we fix gsbase, and we should do it before enter_from_user_mode
688	 * (which can take locks).  Since TRACE_IRQS_OFF idempotent,
689	 * the simplest way to handle it is to just call it twice if
690	 * we enter from user mode.  There's no reason to optimize this since
691	 * TRACE_IRQS_OFF is a no-op if lockdep is off.
692	 */
693	TRACE_IRQS_OFF
694
695	CALL_enter_from_user_mode
696
6971:
698	ENTER_IRQ_STACK old_rsp=%rdi
699	/* We entered an interrupt context - irqs are off: */
700	TRACE_IRQS_OFF
701
702	call	\func	/* rdi points to pt_regs */
703	.endm
704
705	/*
706	 * The interrupt stubs push (~vector+0x80) onto the stack and
707	 * then jump to common_interrupt.
708	 */
709	.p2align CONFIG_X86_L1_CACHE_SHIFT
710common_interrupt:
711	ASM_CLAC
712	addq	$-0x80, (%rsp)			/* Adjust vector to [-256, -1] range */
713	interrupt do_IRQ
714	/* 0(%rsp): old RSP */
715ret_from_intr:
716	DISABLE_INTERRUPTS(CLBR_ANY)
717	TRACE_IRQS_OFF
718
719	LEAVE_IRQ_STACK
720
721	testb	$3, CS(%rsp)
722	jz	retint_kernel
723
724	/* Interrupt came from user space */
725GLOBAL(retint_user)
726	mov	%rsp,%rdi
727	call	prepare_exit_to_usermode
728	TRACE_IRQS_IRETQ
729
730GLOBAL(swapgs_restore_regs_and_return_to_usermode)
731#ifdef CONFIG_DEBUG_ENTRY
732	/* Assert that pt_regs indicates user mode. */
733	testb	$3, CS(%rsp)
734	jnz	1f
735	ud2
7361:
737#endif
738	POP_EXTRA_REGS
739	popq	%r11
740	popq	%r10
741	popq	%r9
742	popq	%r8
743	popq	%rax
744	popq	%rcx
745	popq	%rdx
746	popq	%rsi
747
748	/*
749	 * The stack is now user RDI, orig_ax, RIP, CS, EFLAGS, RSP, SS.
750	 * Save old stack pointer and switch to trampoline stack.
751	 */
752	movq	%rsp, %rdi
753	movq	PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp
754
755	/* Copy the IRET frame to the trampoline stack. */
756	pushq	6*8(%rdi)	/* SS */
757	pushq	5*8(%rdi)	/* RSP */
758	pushq	4*8(%rdi)	/* EFLAGS */
759	pushq	3*8(%rdi)	/* CS */
760	pushq	2*8(%rdi)	/* RIP */
761
762	/* Push user RDI on the trampoline stack. */
763	pushq	(%rdi)
764
765	/*
766	 * We are on the trampoline stack.  All regs except RDI are live.
767	 * We can do future final exit work right here.
768	 */
769
770	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
771
772	/* Restore RDI. */
773	popq	%rdi
774	SWAPGS
775	INTERRUPT_RETURN
776
777
778/* Returning to kernel space */
779retint_kernel:
780#ifdef CONFIG_PREEMPT
781	/* Interrupts are off */
782	/* Check if we need preemption */
783	bt	$9, EFLAGS(%rsp)		/* were interrupts off? */
784	jnc	1f
7850:	cmpl	$0, PER_CPU_VAR(__preempt_count)
786	jnz	1f
787	call	preempt_schedule_irq
788	jmp	0b
7891:
790#endif
791	/*
792	 * The iretq could re-enable interrupts:
793	 */
794	TRACE_IRQS_IRETQ
795
796GLOBAL(restore_regs_and_return_to_kernel)
797#ifdef CONFIG_DEBUG_ENTRY
798	/* Assert that pt_regs indicates kernel mode. */
799	testb	$3, CS(%rsp)
800	jz	1f
801	ud2
8021:
803#endif
804	POP_EXTRA_REGS
805	POP_C_REGS
806	addq	$8, %rsp	/* skip regs->orig_ax */
807	INTERRUPT_RETURN
808
809ENTRY(native_iret)
810	UNWIND_HINT_IRET_REGS
811	/*
812	 * Are we returning to a stack segment from the LDT?  Note: in
813	 * 64-bit mode SS:RSP on the exception stack is always valid.
814	 */
815#ifdef CONFIG_X86_ESPFIX64
816	testb	$4, (SS-RIP)(%rsp)
817	jnz	native_irq_return_ldt
818#endif
819
820.global native_irq_return_iret
821native_irq_return_iret:
822	/*
823	 * This may fault.  Non-paranoid faults on return to userspace are
824	 * handled by fixup_bad_iret.  These include #SS, #GP, and #NP.
825	 * Double-faults due to espfix64 are handled in do_double_fault.
826	 * Other faults here are fatal.
827	 */
828	iretq
829
830#ifdef CONFIG_X86_ESPFIX64
831native_irq_return_ldt:
832	/*
833	 * We are running with user GSBASE.  All GPRs contain their user
834	 * values.  We have a percpu ESPFIX stack that is eight slots
835	 * long (see ESPFIX_STACK_SIZE).  espfix_waddr points to the bottom
836	 * of the ESPFIX stack.
837	 *
838	 * We clobber RAX and RDI in this code.  We stash RDI on the
839	 * normal stack and RAX on the ESPFIX stack.
840	 *
841	 * The ESPFIX stack layout we set up looks like this:
842	 *
843	 * --- top of ESPFIX stack ---
844	 * SS
845	 * RSP
846	 * RFLAGS
847	 * CS
848	 * RIP  <-- RSP points here when we're done
849	 * RAX  <-- espfix_waddr points here
850	 * --- bottom of ESPFIX stack ---
851	 */
852
853	pushq	%rdi				/* Stash user RDI */
854	SWAPGS					/* to kernel GS */
855	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi	/* to kernel CR3 */
856
857	movq	PER_CPU_VAR(espfix_waddr), %rdi
858	movq	%rax, (0*8)(%rdi)		/* user RAX */
859	movq	(1*8)(%rsp), %rax		/* user RIP */
860	movq	%rax, (1*8)(%rdi)
861	movq	(2*8)(%rsp), %rax		/* user CS */
862	movq	%rax, (2*8)(%rdi)
863	movq	(3*8)(%rsp), %rax		/* user RFLAGS */
864	movq	%rax, (3*8)(%rdi)
865	movq	(5*8)(%rsp), %rax		/* user SS */
866	movq	%rax, (5*8)(%rdi)
867	movq	(4*8)(%rsp), %rax		/* user RSP */
868	movq	%rax, (4*8)(%rdi)
869	/* Now RAX == RSP. */
870
871	andl	$0xffff0000, %eax		/* RAX = (RSP & 0xffff0000) */
872
873	/*
874	 * espfix_stack[31:16] == 0.  The page tables are set up such that
875	 * (espfix_stack | (X & 0xffff0000)) points to a read-only alias of
876	 * espfix_waddr for any X.  That is, there are 65536 RO aliases of
877	 * the same page.  Set up RSP so that RSP[31:16] contains the
878	 * respective 16 bits of the /userspace/ RSP and RSP nonetheless
879	 * still points to an RO alias of the ESPFIX stack.
880	 */
881	orq	PER_CPU_VAR(espfix_stack), %rax
882
883	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
884	SWAPGS					/* to user GS */
885	popq	%rdi				/* Restore user RDI */
886
887	movq	%rax, %rsp
888	UNWIND_HINT_IRET_REGS offset=8
889
890	/*
891	 * At this point, we cannot write to the stack any more, but we can
892	 * still read.
893	 */
894	popq	%rax				/* Restore user RAX */
895
896	/*
897	 * RSP now points to an ordinary IRET frame, except that the page
898	 * is read-only and RSP[31:16] are preloaded with the userspace
899	 * values.  We can now IRET back to userspace.
900	 */
901	jmp	native_irq_return_iret
902#endif
903END(common_interrupt)
904
905/*
906 * APIC interrupts.
907 */
908.macro apicinterrupt3 num sym do_sym
909ENTRY(\sym)
910	UNWIND_HINT_IRET_REGS
911	ASM_CLAC
912	pushq	$~(\num)
913.Lcommon_\sym:
914	interrupt \do_sym
915	jmp	ret_from_intr
916END(\sym)
917.endm
918
919/* Make sure APIC interrupt handlers end up in the irqentry section: */
920#define PUSH_SECTION_IRQENTRY	.pushsection .irqentry.text, "ax"
921#define POP_SECTION_IRQENTRY	.popsection
922
923.macro apicinterrupt num sym do_sym
924PUSH_SECTION_IRQENTRY
925apicinterrupt3 \num \sym \do_sym
926POP_SECTION_IRQENTRY
927.endm
928
929#ifdef CONFIG_SMP
930apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR		irq_move_cleanup_interrupt	smp_irq_move_cleanup_interrupt
931apicinterrupt3 REBOOT_VECTOR			reboot_interrupt		smp_reboot_interrupt
932#endif
933
934#ifdef CONFIG_X86_UV
935apicinterrupt3 UV_BAU_MESSAGE			uv_bau_message_intr1		uv_bau_message_interrupt
936#endif
937
938apicinterrupt LOCAL_TIMER_VECTOR		apic_timer_interrupt		smp_apic_timer_interrupt
939apicinterrupt X86_PLATFORM_IPI_VECTOR		x86_platform_ipi		smp_x86_platform_ipi
940
941#ifdef CONFIG_HAVE_KVM
942apicinterrupt3 POSTED_INTR_VECTOR		kvm_posted_intr_ipi		smp_kvm_posted_intr_ipi
943apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR	kvm_posted_intr_wakeup_ipi	smp_kvm_posted_intr_wakeup_ipi
944apicinterrupt3 POSTED_INTR_NESTED_VECTOR	kvm_posted_intr_nested_ipi	smp_kvm_posted_intr_nested_ipi
945#endif
946
947#ifdef CONFIG_X86_MCE_THRESHOLD
948apicinterrupt THRESHOLD_APIC_VECTOR		threshold_interrupt		smp_threshold_interrupt
949#endif
950
951#ifdef CONFIG_X86_MCE_AMD
952apicinterrupt DEFERRED_ERROR_VECTOR		deferred_error_interrupt	smp_deferred_error_interrupt
953#endif
954
955#ifdef CONFIG_X86_THERMAL_VECTOR
956apicinterrupt THERMAL_APIC_VECTOR		thermal_interrupt		smp_thermal_interrupt
957#endif
958
959#ifdef CONFIG_SMP
960apicinterrupt CALL_FUNCTION_SINGLE_VECTOR	call_function_single_interrupt	smp_call_function_single_interrupt
961apicinterrupt CALL_FUNCTION_VECTOR		call_function_interrupt		smp_call_function_interrupt
962apicinterrupt RESCHEDULE_VECTOR			reschedule_interrupt		smp_reschedule_interrupt
963#endif
964
965apicinterrupt ERROR_APIC_VECTOR			error_interrupt			smp_error_interrupt
966apicinterrupt SPURIOUS_APIC_VECTOR		spurious_interrupt		smp_spurious_interrupt
967
968#ifdef CONFIG_IRQ_WORK
969apicinterrupt IRQ_WORK_VECTOR			irq_work_interrupt		smp_irq_work_interrupt
970#endif
971
972/*
973 * Exception entry points.
974 */
975#define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss_rw) + (TSS_ist + ((x) - 1) * 8)
976
977/*
978 * Switch to the thread stack.  This is called with the IRET frame and
979 * orig_ax on the stack.  (That is, RDI..R12 are not on the stack and
980 * space has not been allocated for them.)
981 */
982ENTRY(switch_to_thread_stack)
983	UNWIND_HINT_FUNC
984
985	pushq	%rdi
986	/* Need to switch before accessing the thread stack. */
987	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi
988	movq	%rsp, %rdi
989	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
990	UNWIND_HINT sp_offset=16 sp_reg=ORC_REG_DI
991
992	pushq	7*8(%rdi)		/* regs->ss */
993	pushq	6*8(%rdi)		/* regs->rsp */
994	pushq	5*8(%rdi)		/* regs->eflags */
995	pushq	4*8(%rdi)		/* regs->cs */
996	pushq	3*8(%rdi)		/* regs->ip */
997	pushq	2*8(%rdi)		/* regs->orig_ax */
998	pushq	8(%rdi)			/* return address */
999	UNWIND_HINT_FUNC
1000
1001	movq	(%rdi), %rdi
1002	ret
1003END(switch_to_thread_stack)
1004
1005.macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
1006ENTRY(\sym)
1007	UNWIND_HINT_IRET_REGS offset=\has_error_code*8
1008
1009	/* Sanity check */
1010	.if \shift_ist != -1 && \paranoid == 0
1011	.error "using shift_ist requires paranoid=1"
1012	.endif
1013
1014	ASM_CLAC
1015
1016	.if \has_error_code == 0
1017	pushq	$-1				/* ORIG_RAX: no syscall to restart */
1018	.endif
1019
1020	ALLOC_PT_GPREGS_ON_STACK
1021
1022	.if \paranoid < 2
1023	testb	$3, CS(%rsp)			/* If coming from userspace, switch stacks */
1024	jnz	.Lfrom_usermode_switch_stack_\@
1025	.endif
1026
1027	.if \paranoid
1028	call	paranoid_entry
1029	.else
1030	call	error_entry
1031	.endif
1032	UNWIND_HINT_REGS
1033	/* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */
1034
1035	.if \paranoid
1036	.if \shift_ist != -1
1037	TRACE_IRQS_OFF_DEBUG			/* reload IDT in case of recursion */
1038	.else
1039	TRACE_IRQS_OFF
1040	.endif
1041	.endif
1042
1043	movq	%rsp, %rdi			/* pt_regs pointer */
1044
1045	.if \has_error_code
1046	movq	ORIG_RAX(%rsp), %rsi		/* get error code */
1047	movq	$-1, ORIG_RAX(%rsp)		/* no syscall to restart */
1048	.else
1049	xorl	%esi, %esi			/* no error code */
1050	.endif
1051
1052	.if \shift_ist != -1
1053	subq	$EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
1054	.endif
1055
1056	call	\do_sym
1057
1058	.if \shift_ist != -1
1059	addq	$EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
1060	.endif
1061
1062	/* these procedures expect "no swapgs" flag in ebx */
1063	.if \paranoid
1064	jmp	paranoid_exit
1065	.else
1066	jmp	error_exit
1067	.endif
1068
1069	.if \paranoid < 2
1070	/*
1071	 * Entry from userspace.  Switch stacks and treat it
1072	 * as a normal entry.  This means that paranoid handlers
1073	 * run in real process context if user_mode(regs).
1074	 */
1075.Lfrom_usermode_switch_stack_\@:
1076	call	error_entry
1077
1078	movq	%rsp, %rdi			/* pt_regs pointer */
1079
1080	.if \has_error_code
1081	movq	ORIG_RAX(%rsp), %rsi		/* get error code */
1082	movq	$-1, ORIG_RAX(%rsp)		/* no syscall to restart */
1083	.else
1084	xorl	%esi, %esi			/* no error code */
1085	.endif
1086
1087	call	\do_sym
1088
1089	jmp	error_exit			/* %ebx: no swapgs flag */
1090	.endif
1091END(\sym)
1092.endm
1093
1094idtentry divide_error			do_divide_error			has_error_code=0
1095idtentry overflow			do_overflow			has_error_code=0
1096idtentry bounds				do_bounds			has_error_code=0
1097idtentry invalid_op			do_invalid_op			has_error_code=0
1098idtentry device_not_available		do_device_not_available		has_error_code=0
1099idtentry double_fault			do_double_fault			has_error_code=1 paranoid=2
1100idtentry coprocessor_segment_overrun	do_coprocessor_segment_overrun	has_error_code=0
1101idtentry invalid_TSS			do_invalid_TSS			has_error_code=1
1102idtentry segment_not_present		do_segment_not_present		has_error_code=1
1103idtentry spurious_interrupt_bug		do_spurious_interrupt_bug	has_error_code=0
1104idtentry coprocessor_error		do_coprocessor_error		has_error_code=0
1105idtentry alignment_check		do_alignment_check		has_error_code=1
1106idtentry simd_coprocessor_error		do_simd_coprocessor_error	has_error_code=0
1107
1108
1109	/*
1110	 * Reload gs selector with exception handling
1111	 * edi:  new selector
1112	 */
1113ENTRY(native_load_gs_index)
1114	FRAME_BEGIN
1115	pushfq
1116	DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1117	TRACE_IRQS_OFF
1118	SWAPGS
1119.Lgs_change:
1120	movl	%edi, %gs
11212:	ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE
1122	SWAPGS
1123	TRACE_IRQS_FLAGS (%rsp)
1124	popfq
1125	FRAME_END
1126	ret
1127ENDPROC(native_load_gs_index)
1128EXPORT_SYMBOL(native_load_gs_index)
1129
1130	_ASM_EXTABLE(.Lgs_change, bad_gs)
1131	.section .fixup, "ax"
1132	/* running with kernelgs */
1133bad_gs:
1134	SWAPGS					/* switch back to user gs */
1135.macro ZAP_GS
1136	/* This can't be a string because the preprocessor needs to see it. */
1137	movl $__USER_DS, %eax
1138	movl %eax, %gs
1139.endm
1140	ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG
1141	xorl	%eax, %eax
1142	movl	%eax, %gs
1143	jmp	2b
1144	.previous
1145
1146/* Call softirq on interrupt stack. Interrupts are off. */
1147ENTRY(do_softirq_own_stack)
1148	pushq	%rbp
1149	mov	%rsp, %rbp
1150	ENTER_IRQ_STACK regs=0 old_rsp=%r11
1151	call	__do_softirq
1152	LEAVE_IRQ_STACK regs=0
1153	leaveq
1154	ret
1155ENDPROC(do_softirq_own_stack)
1156
1157#ifdef CONFIG_XEN
1158idtentry hypervisor_callback xen_do_hypervisor_callback has_error_code=0
1159
1160/*
1161 * A note on the "critical region" in our callback handler.
1162 * We want to avoid stacking callback handlers due to events occurring
1163 * during handling of the last event. To do this, we keep events disabled
1164 * until we've done all processing. HOWEVER, we must enable events before
1165 * popping the stack frame (can't be done atomically) and so it would still
1166 * be possible to get enough handler activations to overflow the stack.
1167 * Although unlikely, bugs of that kind are hard to track down, so we'd
1168 * like to avoid the possibility.
1169 * So, on entry to the handler we detect whether we interrupted an
1170 * existing activation in its critical region -- if so, we pop the current
1171 * activation and restart the handler using the previous one.
1172 */
1173ENTRY(xen_do_hypervisor_callback)		/* do_hypervisor_callback(struct *pt_regs) */
1174
1175/*
1176 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1177 * see the correct pointer to the pt_regs
1178 */
1179	UNWIND_HINT_FUNC
1180	movq	%rdi, %rsp			/* we don't return, adjust the stack frame */
1181	UNWIND_HINT_REGS
1182
1183	ENTER_IRQ_STACK old_rsp=%r10
1184	call	xen_evtchn_do_upcall
1185	LEAVE_IRQ_STACK
1186
1187#ifndef CONFIG_PREEMPT
1188	call	xen_maybe_preempt_hcall
1189#endif
1190	jmp	error_exit
1191END(xen_do_hypervisor_callback)
1192
1193/*
1194 * Hypervisor uses this for application faults while it executes.
1195 * We get here for two reasons:
1196 *  1. Fault while reloading DS, ES, FS or GS
1197 *  2. Fault while executing IRET
1198 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1199 * registers that could be reloaded and zeroed the others.
1200 * Category 2 we fix up by killing the current process. We cannot use the
1201 * normal Linux return path in this case because if we use the IRET hypercall
1202 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1203 * We distinguish between categories by comparing each saved segment register
1204 * with its current contents: any discrepancy means we in category 1.
1205 */
1206ENTRY(xen_failsafe_callback)
1207	UNWIND_HINT_EMPTY
1208	movl	%ds, %ecx
1209	cmpw	%cx, 0x10(%rsp)
1210	jne	1f
1211	movl	%es, %ecx
1212	cmpw	%cx, 0x18(%rsp)
1213	jne	1f
1214	movl	%fs, %ecx
1215	cmpw	%cx, 0x20(%rsp)
1216	jne	1f
1217	movl	%gs, %ecx
1218	cmpw	%cx, 0x28(%rsp)
1219	jne	1f
1220	/* All segments match their saved values => Category 2 (Bad IRET). */
1221	movq	(%rsp), %rcx
1222	movq	8(%rsp), %r11
1223	addq	$0x30, %rsp
1224	pushq	$0				/* RIP */
1225	UNWIND_HINT_IRET_REGS offset=8
1226	jmp	general_protection
12271:	/* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1228	movq	(%rsp), %rcx
1229	movq	8(%rsp), %r11
1230	addq	$0x30, %rsp
1231	UNWIND_HINT_IRET_REGS
1232	pushq	$-1 /* orig_ax = -1 => not a system call */
1233	ALLOC_PT_GPREGS_ON_STACK
1234	SAVE_C_REGS
1235	SAVE_EXTRA_REGS
1236	ENCODE_FRAME_POINTER
1237	jmp	error_exit
1238END(xen_failsafe_callback)
1239
1240apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1241	xen_hvm_callback_vector xen_evtchn_do_upcall
1242
1243#endif /* CONFIG_XEN */
1244
1245#if IS_ENABLED(CONFIG_HYPERV)
1246apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
1247	hyperv_callback_vector hyperv_vector_handler
1248#endif /* CONFIG_HYPERV */
1249
1250idtentry debug			do_debug		has_error_code=0	paranoid=1 shift_ist=DEBUG_STACK
1251idtentry int3			do_int3			has_error_code=0	paranoid=1 shift_ist=DEBUG_STACK
1252idtentry stack_segment		do_stack_segment	has_error_code=1
1253
1254#ifdef CONFIG_XEN
1255idtentry xennmi			do_nmi			has_error_code=0
1256idtentry xendebug		do_debug		has_error_code=0
1257idtentry xenint3		do_int3			has_error_code=0
1258#endif
1259
1260idtentry general_protection	do_general_protection	has_error_code=1
1261idtentry page_fault		do_page_fault		has_error_code=1
1262
1263#ifdef CONFIG_KVM_GUEST
1264idtentry async_page_fault	do_async_page_fault	has_error_code=1
1265#endif
1266
1267#ifdef CONFIG_X86_MCE
1268idtentry machine_check		do_mce			has_error_code=0	paranoid=1
1269#endif
1270
1271/*
1272 * Save all registers in pt_regs, and switch gs if needed.
1273 * Use slow, but surefire "are we in kernel?" check.
1274 * Return: ebx=0: need swapgs on exit, ebx=1: otherwise
1275 */
1276ENTRY(paranoid_entry)
1277	UNWIND_HINT_FUNC
1278	cld
1279	SAVE_C_REGS 8
1280	SAVE_EXTRA_REGS 8
1281	ENCODE_FRAME_POINTER 8
1282	movl	$1, %ebx
1283	movl	$MSR_GS_BASE, %ecx
1284	rdmsr
1285	testl	%edx, %edx
1286	js	1f				/* negative -> in kernel */
1287	SWAPGS
1288	xorl	%ebx, %ebx
1289
12901:
1291	SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14
1292
1293	ret
1294END(paranoid_entry)
1295
1296/*
1297 * "Paranoid" exit path from exception stack.  This is invoked
1298 * only on return from non-NMI IST interrupts that came
1299 * from kernel space.
1300 *
1301 * We may be returning to very strange contexts (e.g. very early
1302 * in syscall entry), so checking for preemption here would
1303 * be complicated.  Fortunately, we there's no good reason
1304 * to try to handle preemption here.
1305 *
1306 * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it)
1307 */
1308ENTRY(paranoid_exit)
1309	UNWIND_HINT_REGS
1310	DISABLE_INTERRUPTS(CLBR_ANY)
1311	TRACE_IRQS_OFF_DEBUG
1312	testl	%ebx, %ebx			/* swapgs needed? */
1313	jnz	.Lparanoid_exit_no_swapgs
1314	TRACE_IRQS_IRETQ
1315	RESTORE_CR3	scratch_reg=%rbx save_reg=%r14
1316	SWAPGS_UNSAFE_STACK
1317	jmp	.Lparanoid_exit_restore
1318.Lparanoid_exit_no_swapgs:
1319	TRACE_IRQS_IRETQ_DEBUG
1320.Lparanoid_exit_restore:
1321	jmp restore_regs_and_return_to_kernel
1322END(paranoid_exit)
1323
1324/*
1325 * Save all registers in pt_regs, and switch gs if needed.
1326 * Return: EBX=0: came from user mode; EBX=1: otherwise
1327 */
1328ENTRY(error_entry)
1329	UNWIND_HINT_FUNC
1330	cld
1331	SAVE_C_REGS 8
1332	SAVE_EXTRA_REGS 8
1333	ENCODE_FRAME_POINTER 8
1334	xorl	%ebx, %ebx
1335	testb	$3, CS+8(%rsp)
1336	jz	.Lerror_kernelspace
1337
1338	/*
1339	 * We entered from user mode or we're pretending to have entered
1340	 * from user mode due to an IRET fault.
1341	 */
1342	SWAPGS
1343	/* We have user CR3.  Change to kernel CR3. */
1344	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1345
1346.Lerror_entry_from_usermode_after_swapgs:
1347	/* Put us onto the real thread stack. */
1348	popq	%r12				/* save return addr in %12 */
1349	movq	%rsp, %rdi			/* arg0 = pt_regs pointer */
1350	call	sync_regs
1351	movq	%rax, %rsp			/* switch stack */
1352	ENCODE_FRAME_POINTER
1353	pushq	%r12
1354
1355	/*
1356	 * We need to tell lockdep that IRQs are off.  We can't do this until
1357	 * we fix gsbase, and we should do it before enter_from_user_mode
1358	 * (which can take locks).
1359	 */
1360	TRACE_IRQS_OFF
1361	CALL_enter_from_user_mode
1362	ret
1363
1364.Lerror_entry_done:
1365	TRACE_IRQS_OFF
1366	ret
1367
1368	/*
1369	 * There are two places in the kernel that can potentially fault with
1370	 * usergs. Handle them here.  B stepping K8s sometimes report a
1371	 * truncated RIP for IRET exceptions returning to compat mode. Check
1372	 * for these here too.
1373	 */
1374.Lerror_kernelspace:
1375	incl	%ebx
1376	leaq	native_irq_return_iret(%rip), %rcx
1377	cmpq	%rcx, RIP+8(%rsp)
1378	je	.Lerror_bad_iret
1379	movl	%ecx, %eax			/* zero extend */
1380	cmpq	%rax, RIP+8(%rsp)
1381	je	.Lbstep_iret
1382	cmpq	$.Lgs_change, RIP+8(%rsp)
1383	jne	.Lerror_entry_done
1384
1385	/*
1386	 * hack: .Lgs_change can fail with user gsbase.  If this happens, fix up
1387	 * gsbase and proceed.  We'll fix up the exception and land in
1388	 * .Lgs_change's error handler with kernel gsbase.
1389	 */
1390	SWAPGS
1391	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1392	jmp .Lerror_entry_done
1393
1394.Lbstep_iret:
1395	/* Fix truncated RIP */
1396	movq	%rcx, RIP+8(%rsp)
1397	/* fall through */
1398
1399.Lerror_bad_iret:
1400	/*
1401	 * We came from an IRET to user mode, so we have user
1402	 * gsbase and CR3.  Switch to kernel gsbase and CR3:
1403	 */
1404	SWAPGS
1405	SWITCH_TO_KERNEL_CR3 scratch_reg=%rax
1406
1407	/*
1408	 * Pretend that the exception came from user mode: set up pt_regs
1409	 * as if we faulted immediately after IRET and clear EBX so that
1410	 * error_exit knows that we will be returning to user mode.
1411	 */
1412	mov	%rsp, %rdi
1413	call	fixup_bad_iret
1414	mov	%rax, %rsp
1415	decl	%ebx
1416	jmp	.Lerror_entry_from_usermode_after_swapgs
1417END(error_entry)
1418
1419
1420/*
1421 * On entry, EBX is a "return to kernel mode" flag:
1422 *   1: already in kernel mode, don't need SWAPGS
1423 *   0: user gsbase is loaded, we need SWAPGS and standard preparation for return to usermode
1424 */
1425ENTRY(error_exit)
1426	UNWIND_HINT_REGS
1427	DISABLE_INTERRUPTS(CLBR_ANY)
1428	TRACE_IRQS_OFF
1429	testl	%ebx, %ebx
1430	jnz	retint_kernel
1431	jmp	retint_user
1432END(error_exit)
1433
1434/*
1435 * Runs on exception stack.  Xen PV does not go through this path at all,
1436 * so we can use real assembly here.
1437 *
1438 * Registers:
1439 *	%r14: Used to save/restore the CR3 of the interrupted context
1440 *	      when PAGE_TABLE_ISOLATION is in use.  Do not clobber.
1441 */
1442ENTRY(nmi)
1443	UNWIND_HINT_IRET_REGS
1444
1445	/*
1446	 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1447	 * the iretq it performs will take us out of NMI context.
1448	 * This means that we can have nested NMIs where the next
1449	 * NMI is using the top of the stack of the previous NMI. We
1450	 * can't let it execute because the nested NMI will corrupt the
1451	 * stack of the previous NMI. NMI handlers are not re-entrant
1452	 * anyway.
1453	 *
1454	 * To handle this case we do the following:
1455	 *  Check the a special location on the stack that contains
1456	 *  a variable that is set when NMIs are executing.
1457	 *  The interrupted task's stack is also checked to see if it
1458	 *  is an NMI stack.
1459	 *  If the variable is not set and the stack is not the NMI
1460	 *  stack then:
1461	 *    o Set the special variable on the stack
1462	 *    o Copy the interrupt frame into an "outermost" location on the
1463	 *      stack
1464	 *    o Copy the interrupt frame into an "iret" location on the stack
1465	 *    o Continue processing the NMI
1466	 *  If the variable is set or the previous stack is the NMI stack:
1467	 *    o Modify the "iret" location to jump to the repeat_nmi
1468	 *    o return back to the first NMI
1469	 *
1470	 * Now on exit of the first NMI, we first clear the stack variable
1471	 * The NMI stack will tell any nested NMIs at that point that it is
1472	 * nested. Then we pop the stack normally with iret, and if there was
1473	 * a nested NMI that updated the copy interrupt stack frame, a
1474	 * jump will be made to the repeat_nmi code that will handle the second
1475	 * NMI.
1476	 *
1477	 * However, espfix prevents us from directly returning to userspace
1478	 * with a single IRET instruction.  Similarly, IRET to user mode
1479	 * can fault.  We therefore handle NMIs from user space like
1480	 * other IST entries.
1481	 */
1482
1483	ASM_CLAC
1484
1485	/* Use %rdx as our temp variable throughout */
1486	pushq	%rdx
1487
1488	testb	$3, CS-RIP+8(%rsp)
1489	jz	.Lnmi_from_kernel
1490
1491	/*
1492	 * NMI from user mode.  We need to run on the thread stack, but we
1493	 * can't go through the normal entry paths: NMIs are masked, and
1494	 * we don't want to enable interrupts, because then we'll end
1495	 * up in an awkward situation in which IRQs are on but NMIs
1496	 * are off.
1497	 *
1498	 * We also must not push anything to the stack before switching
1499	 * stacks lest we corrupt the "NMI executing" variable.
1500	 */
1501
1502	swapgs
1503	cld
1504	SWITCH_TO_KERNEL_CR3 scratch_reg=%rdx
1505	movq	%rsp, %rdx
1506	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rsp
1507	UNWIND_HINT_IRET_REGS base=%rdx offset=8
1508	pushq	5*8(%rdx)	/* pt_regs->ss */
1509	pushq	4*8(%rdx)	/* pt_regs->rsp */
1510	pushq	3*8(%rdx)	/* pt_regs->flags */
1511	pushq	2*8(%rdx)	/* pt_regs->cs */
1512	pushq	1*8(%rdx)	/* pt_regs->rip */
1513	UNWIND_HINT_IRET_REGS
1514	pushq   $-1		/* pt_regs->orig_ax */
1515	pushq   %rdi		/* pt_regs->di */
1516	pushq   %rsi		/* pt_regs->si */
1517	pushq   (%rdx)		/* pt_regs->dx */
1518	pushq   %rcx		/* pt_regs->cx */
1519	pushq   %rax		/* pt_regs->ax */
1520	pushq   %r8		/* pt_regs->r8 */
1521	pushq   %r9		/* pt_regs->r9 */
1522	pushq   %r10		/* pt_regs->r10 */
1523	pushq   %r11		/* pt_regs->r11 */
1524	pushq	%rbx		/* pt_regs->rbx */
1525	pushq	%rbp		/* pt_regs->rbp */
1526	pushq	%r12		/* pt_regs->r12 */
1527	pushq	%r13		/* pt_regs->r13 */
1528	pushq	%r14		/* pt_regs->r14 */
1529	pushq	%r15		/* pt_regs->r15 */
1530	UNWIND_HINT_REGS
1531	ENCODE_FRAME_POINTER
1532
1533	/*
1534	 * At this point we no longer need to worry about stack damage
1535	 * due to nesting -- we're on the normal thread stack and we're
1536	 * done with the NMI stack.
1537	 */
1538
1539	movq	%rsp, %rdi
1540	movq	$-1, %rsi
1541	call	do_nmi
1542
1543	/*
1544	 * Return back to user mode.  We must *not* do the normal exit
1545	 * work, because we don't want to enable interrupts.
1546	 */
1547	jmp	swapgs_restore_regs_and_return_to_usermode
1548
1549.Lnmi_from_kernel:
1550	/*
1551	 * Here's what our stack frame will look like:
1552	 * +---------------------------------------------------------+
1553	 * | original SS                                             |
1554	 * | original Return RSP                                     |
1555	 * | original RFLAGS                                         |
1556	 * | original CS                                             |
1557	 * | original RIP                                            |
1558	 * +---------------------------------------------------------+
1559	 * | temp storage for rdx                                    |
1560	 * +---------------------------------------------------------+
1561	 * | "NMI executing" variable                                |
1562	 * +---------------------------------------------------------+
1563	 * | iret SS          } Copied from "outermost" frame        |
1564	 * | iret Return RSP  } on each loop iteration; overwritten  |
1565	 * | iret RFLAGS      } by a nested NMI to force another     |
1566	 * | iret CS          } iteration if needed.                 |
1567	 * | iret RIP         }                                      |
1568	 * +---------------------------------------------------------+
1569	 * | outermost SS          } initialized in first_nmi;       |
1570	 * | outermost Return RSP  } will not be changed before      |
1571	 * | outermost RFLAGS      } NMI processing is done.         |
1572	 * | outermost CS          } Copied to "iret" frame on each  |
1573	 * | outermost RIP         } iteration.                      |
1574	 * +---------------------------------------------------------+
1575	 * | pt_regs                                                 |
1576	 * +---------------------------------------------------------+
1577	 *
1578	 * The "original" frame is used by hardware.  Before re-enabling
1579	 * NMIs, we need to be done with it, and we need to leave enough
1580	 * space for the asm code here.
1581	 *
1582	 * We return by executing IRET while RSP points to the "iret" frame.
1583	 * That will either return for real or it will loop back into NMI
1584	 * processing.
1585	 *
1586	 * The "outermost" frame is copied to the "iret" frame on each
1587	 * iteration of the loop, so each iteration starts with the "iret"
1588	 * frame pointing to the final return target.
1589	 */
1590
1591	/*
1592	 * Determine whether we're a nested NMI.
1593	 *
1594	 * If we interrupted kernel code between repeat_nmi and
1595	 * end_repeat_nmi, then we are a nested NMI.  We must not
1596	 * modify the "iret" frame because it's being written by
1597	 * the outer NMI.  That's okay; the outer NMI handler is
1598	 * about to about to call do_nmi anyway, so we can just
1599	 * resume the outer NMI.
1600	 */
1601
1602	movq	$repeat_nmi, %rdx
1603	cmpq	8(%rsp), %rdx
1604	ja	1f
1605	movq	$end_repeat_nmi, %rdx
1606	cmpq	8(%rsp), %rdx
1607	ja	nested_nmi_out
16081:
1609
1610	/*
1611	 * Now check "NMI executing".  If it's set, then we're nested.
1612	 * This will not detect if we interrupted an outer NMI just
1613	 * before IRET.
1614	 */
1615	cmpl	$1, -8(%rsp)
1616	je	nested_nmi
1617
1618	/*
1619	 * Now test if the previous stack was an NMI stack.  This covers
1620	 * the case where we interrupt an outer NMI after it clears
1621	 * "NMI executing" but before IRET.  We need to be careful, though:
1622	 * there is one case in which RSP could point to the NMI stack
1623	 * despite there being no NMI active: naughty userspace controls
1624	 * RSP at the very beginning of the SYSCALL targets.  We can
1625	 * pull a fast one on naughty userspace, though: we program
1626	 * SYSCALL to mask DF, so userspace cannot cause DF to be set
1627	 * if it controls the kernel's RSP.  We set DF before we clear
1628	 * "NMI executing".
1629	 */
1630	lea	6*8(%rsp), %rdx
1631	/* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */
1632	cmpq	%rdx, 4*8(%rsp)
1633	/* If the stack pointer is above the NMI stack, this is a normal NMI */
1634	ja	first_nmi
1635
1636	subq	$EXCEPTION_STKSZ, %rdx
1637	cmpq	%rdx, 4*8(%rsp)
1638	/* If it is below the NMI stack, it is a normal NMI */
1639	jb	first_nmi
1640
1641	/* Ah, it is within the NMI stack. */
1642
1643	testb	$(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp)
1644	jz	first_nmi	/* RSP was user controlled. */
1645
1646	/* This is a nested NMI. */
1647
1648nested_nmi:
1649	/*
1650	 * Modify the "iret" frame to point to repeat_nmi, forcing another
1651	 * iteration of NMI handling.
1652	 */
1653	subq	$8, %rsp
1654	leaq	-10*8(%rsp), %rdx
1655	pushq	$__KERNEL_DS
1656	pushq	%rdx
1657	pushfq
1658	pushq	$__KERNEL_CS
1659	pushq	$repeat_nmi
1660
1661	/* Put stack back */
1662	addq	$(6*8), %rsp
1663
1664nested_nmi_out:
1665	popq	%rdx
1666
1667	/* We are returning to kernel mode, so this cannot result in a fault. */
1668	iretq
1669
1670first_nmi:
1671	/* Restore rdx. */
1672	movq	(%rsp), %rdx
1673
1674	/* Make room for "NMI executing". */
1675	pushq	$0
1676
1677	/* Leave room for the "iret" frame */
1678	subq	$(5*8), %rsp
1679
1680	/* Copy the "original" frame to the "outermost" frame */
1681	.rept 5
1682	pushq	11*8(%rsp)
1683	.endr
1684	UNWIND_HINT_IRET_REGS
1685
1686	/* Everything up to here is safe from nested NMIs */
1687
1688#ifdef CONFIG_DEBUG_ENTRY
1689	/*
1690	 * For ease of testing, unmask NMIs right away.  Disabled by
1691	 * default because IRET is very expensive.
1692	 */
1693	pushq	$0		/* SS */
1694	pushq	%rsp		/* RSP (minus 8 because of the previous push) */
1695	addq	$8, (%rsp)	/* Fix up RSP */
1696	pushfq			/* RFLAGS */
1697	pushq	$__KERNEL_CS	/* CS */
1698	pushq	$1f		/* RIP */
1699	iretq			/* continues at repeat_nmi below */
1700	UNWIND_HINT_IRET_REGS
17011:
1702#endif
1703
1704repeat_nmi:
1705	/*
1706	 * If there was a nested NMI, the first NMI's iret will return
1707	 * here. But NMIs are still enabled and we can take another
1708	 * nested NMI. The nested NMI checks the interrupted RIP to see
1709	 * if it is between repeat_nmi and end_repeat_nmi, and if so
1710	 * it will just return, as we are about to repeat an NMI anyway.
1711	 * This makes it safe to copy to the stack frame that a nested
1712	 * NMI will update.
1713	 *
1714	 * RSP is pointing to "outermost RIP".  gsbase is unknown, but, if
1715	 * we're repeating an NMI, gsbase has the same value that it had on
1716	 * the first iteration.  paranoid_entry will load the kernel
1717	 * gsbase if needed before we call do_nmi.  "NMI executing"
1718	 * is zero.
1719	 */
1720	movq	$1, 10*8(%rsp)		/* Set "NMI executing". */
1721
1722	/*
1723	 * Copy the "outermost" frame to the "iret" frame.  NMIs that nest
1724	 * here must not modify the "iret" frame while we're writing to
1725	 * it or it will end up containing garbage.
1726	 */
1727	addq	$(10*8), %rsp
1728	.rept 5
1729	pushq	-6*8(%rsp)
1730	.endr
1731	subq	$(5*8), %rsp
1732end_repeat_nmi:
1733
1734	/*
1735	 * Everything below this point can be preempted by a nested NMI.
1736	 * If this happens, then the inner NMI will change the "iret"
1737	 * frame to point back to repeat_nmi.
1738	 */
1739	pushq	$-1				/* ORIG_RAX: no syscall to restart */
1740	ALLOC_PT_GPREGS_ON_STACK
1741
1742	/*
1743	 * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit
1744	 * as we should not be calling schedule in NMI context.
1745	 * Even with normal interrupts enabled. An NMI should not be
1746	 * setting NEED_RESCHED or anything that normal interrupts and
1747	 * exceptions might do.
1748	 */
1749	call	paranoid_entry
1750	UNWIND_HINT_REGS
1751
1752	/* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1753	movq	%rsp, %rdi
1754	movq	$-1, %rsi
1755	call	do_nmi
1756
1757	RESTORE_CR3 scratch_reg=%r15 save_reg=%r14
1758
1759	testl	%ebx, %ebx			/* swapgs needed? */
1760	jnz	nmi_restore
1761nmi_swapgs:
1762	SWAPGS_UNSAFE_STACK
1763nmi_restore:
1764	POP_EXTRA_REGS
1765	POP_C_REGS
1766
1767	/*
1768	 * Skip orig_ax and the "outermost" frame to point RSP at the "iret"
1769	 * at the "iret" frame.
1770	 */
1771	addq	$6*8, %rsp
1772
1773	/*
1774	 * Clear "NMI executing".  Set DF first so that we can easily
1775	 * distinguish the remaining code between here and IRET from
1776	 * the SYSCALL entry and exit paths.
1777	 *
1778	 * We arguably should just inspect RIP instead, but I (Andy) wrote
1779	 * this code when I had the misapprehension that Xen PV supported
1780	 * NMIs, and Xen PV would break that approach.
1781	 */
1782	std
1783	movq	$0, 5*8(%rsp)		/* clear "NMI executing" */
1784
1785	/*
1786	 * iretq reads the "iret" frame and exits the NMI stack in a
1787	 * single instruction.  We are returning to kernel mode, so this
1788	 * cannot result in a fault.  Similarly, we don't need to worry
1789	 * about espfix64 on the way back to kernel mode.
1790	 */
1791	iretq
1792END(nmi)
1793
1794ENTRY(ignore_sysret)
1795	UNWIND_HINT_EMPTY
1796	mov	$-ENOSYS, %eax
1797	sysret
1798END(ignore_sysret)
1799
1800ENTRY(rewind_stack_do_exit)
1801	UNWIND_HINT_FUNC
1802	/* Prevent any naive code from trying to unwind to our caller. */
1803	xorl	%ebp, %ebp
1804
1805	movq	PER_CPU_VAR(cpu_current_top_of_stack), %rax
1806	leaq	-PTREGS_SIZE(%rax), %rsp
1807	UNWIND_HINT_FUNC sp_offset=PTREGS_SIZE
1808
1809	call	do_exit
1810END(rewind_stack_do_exit)
1811