xref: /freebsd/sys/amd64/amd64/exception.S (revision aa0a1e58)
1/*-
2 * Copyright (c) 1989, 1990 William F. Jolitz.
3 * Copyright (c) 1990 The Regents of the University of California.
4 * Copyright (c) 2007 The FreeBSD Foundation
5 * All rights reserved.
6 *
7 * Portions of this software were developed by A. Joseph Koshy under
8 * sponsorship from the FreeBSD Foundation and Google, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * $FreeBSD$
35 */
36
37#include "opt_atpic.h"
38#include "opt_compat.h"
39#include "opt_hwpmc_hooks.h"
40#include "opt_kdtrace.h"
41
42#include <machine/asmacros.h>
43#include <machine/psl.h>
44#include <machine/trap.h>
45#include <machine/specialreg.h>
46
47#include "assym.s"
48
49#ifdef KDTRACE_HOOKS
50	.bss
51	.globl	dtrace_invop_jump_addr
52	.align	8
53	.type	dtrace_invop_jump_addr,@object
54	.size	dtrace_invop_jump_addr,8
55dtrace_invop_jump_addr:
56	.zero	8
57	.globl	dtrace_invop_calltrap_addr
58	.align	8
59	.type	dtrace_invop_calltrap_addr,@object
60	.size	dtrace_invop_calltrap_addr,8
61dtrace_invop_calltrap_addr:
62	.zero	8
63#endif
64	.text
65#ifdef HWPMC_HOOKS
66	ENTRY(start_exceptions)
67#endif
68
69/*****************************************************************************/
70/* Trap handling                                                             */
71/*****************************************************************************/
72/*
73 * Trap and fault vector routines.
74 *
75 * All traps are 'interrupt gates', SDT_SYSIGT.  An interrupt gate pushes
76 * state on the stack but also disables interrupts.  This is important for
77 * us for the use of the swapgs instruction.  We cannot be interrupted
78 * until the GS.base value is correct.  For most traps, we automatically
79 * then enable interrupts if the interrupted context had them enabled.
80 * This is equivalent to the i386 port's use of SDT_SYS386TGT.
81 *
82 * The cpu will push a certain amount of state onto the kernel stack for
83 * the current process.  See amd64/include/frame.h.
84 * This includes the current RFLAGS (status register, which includes
85 * the interrupt disable state prior to the trap), the code segment register,
86 * and the return instruction pointer are pushed by the cpu.  The cpu
87 * will also push an 'error' code for certain traps.  We push a dummy
88 * error code for those traps where the cpu doesn't in order to maintain
89 * a consistent frame.  We also push a contrived 'trap number'.
90 *
91 * The cpu does not push the general registers, we must do that, and we
92 * must restore them prior to calling 'iret'.  The cpu adjusts the %cs and
93 * %ss segment registers, but does not mess with %ds, %es, or %fs.  Thus we
94 * must load them with appropriate values for supervisor mode operation.
95 */
96
97MCOUNT_LABEL(user)
98MCOUNT_LABEL(btrap)
99
100/* Traps that we leave interrupts disabled for.. */
101#define	TRAP_NOEN(a)	\
102	subq $TF_RIP,%rsp; \
103	movl $(a),TF_TRAPNO(%rsp) ; \
104	movq $0,TF_ADDR(%rsp) ; \
105	movq $0,TF_ERR(%rsp) ; \
106	jmp alltraps_noen
107IDTVEC(dbg)
108	TRAP_NOEN(T_TRCTRAP)
109IDTVEC(bpt)
110	TRAP_NOEN(T_BPTFLT)
111#ifdef KDTRACE_HOOKS
112IDTVEC(dtrace_ret)
113	TRAP_NOEN(T_DTRACE_RET)
114#endif
115
116/* Regular traps; The cpu does not supply tf_err for these. */
117#define	TRAP(a)	 \
118	subq $TF_RIP,%rsp; \
119	movl $(a),TF_TRAPNO(%rsp) ; \
120	movq $0,TF_ADDR(%rsp) ; \
121	movq $0,TF_ERR(%rsp) ; \
122	jmp alltraps
123IDTVEC(div)
124	TRAP(T_DIVIDE)
125IDTVEC(ofl)
126	TRAP(T_OFLOW)
127IDTVEC(bnd)
128	TRAP(T_BOUND)
129IDTVEC(ill)
130	TRAP(T_PRIVINFLT)
131IDTVEC(dna)
132	TRAP(T_DNA)
133IDTVEC(fpusegm)
134	TRAP(T_FPOPFLT)
135IDTVEC(mchk)
136	TRAP(T_MCHK)
137IDTVEC(rsvd)
138	TRAP(T_RESERVED)
139IDTVEC(fpu)
140	TRAP(T_ARITHTRAP)
141IDTVEC(xmm)
142	TRAP(T_XMMFLT)
143
144/* This group of traps have tf_err already pushed by the cpu */
145#define	TRAP_ERR(a)	\
146	subq $TF_ERR,%rsp; \
147	movl $(a),TF_TRAPNO(%rsp) ; \
148	movq $0,TF_ADDR(%rsp) ; \
149	jmp alltraps
150IDTVEC(tss)
151	TRAP_ERR(T_TSSFLT)
152IDTVEC(missing)
153	TRAP_ERR(T_SEGNPFLT)
154IDTVEC(stk)
155	TRAP_ERR(T_STKFLT)
156IDTVEC(align)
157	TRAP_ERR(T_ALIGNFLT)
158
159	/*
160	 * alltraps entry point.  Use swapgs if this is the first time in the
161	 * kernel from userland.  Reenable interrupts if they were enabled
162	 * before the trap.  This approximates SDT_SYS386TGT on the i386 port.
163	 */
164	SUPERALIGN_TEXT
165	.globl	alltraps
166	.type	alltraps,@function
167alltraps:
168	movq	%rdi,TF_RDI(%rsp)
169	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
170	jz	alltraps_testi		/* already running with kernel GS.base */
171	swapgs
172	movq	PCPU(CURPCB),%rdi
173	andl	$~PCB_FULL_IRET,PCB_FLAGS(%rdi)
174	movw	%fs,TF_FS(%rsp)
175	movw	%gs,TF_GS(%rsp)
176	movw	%es,TF_ES(%rsp)
177	movw	%ds,TF_DS(%rsp)
178alltraps_testi:
179	testl	$PSL_I,TF_RFLAGS(%rsp)
180	jz	alltraps_pushregs_no_rdi
181	sti
182alltraps_pushregs_no_rdi:
183	movq	%rsi,TF_RSI(%rsp)
184	movq	%rdx,TF_RDX(%rsp)
185	movq	%rcx,TF_RCX(%rsp)
186	movq	%r8,TF_R8(%rsp)
187	movq	%r9,TF_R9(%rsp)
188	movq	%rax,TF_RAX(%rsp)
189	movq	%rbx,TF_RBX(%rsp)
190	movq	%rbp,TF_RBP(%rsp)
191	movq	%r10,TF_R10(%rsp)
192	movq	%r11,TF_R11(%rsp)
193	movq	%r12,TF_R12(%rsp)
194	movq	%r13,TF_R13(%rsp)
195	movq	%r14,TF_R14(%rsp)
196	movq	%r15,TF_R15(%rsp)
197	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
198	cld
199	FAKE_MCOUNT(TF_RIP(%rsp))
200#ifdef KDTRACE_HOOKS
201	/*
202	 * DTrace Function Boundary Trace (fbt) probes are triggered
203	 * by int3 (0xcc) which causes the #BP (T_BPTFLT) breakpoint
204	 * interrupt. For all other trap types, just handle them in
205	 * the usual way.
206	 */
207	cmpl	$T_BPTFLT,TF_TRAPNO(%rsp)
208	jne	calltrap
209
210	/* Check if there is no DTrace hook registered. */
211	cmpq	$0,dtrace_invop_jump_addr
212	je	calltrap
213
214	/*
215	 * Set our jump address for the jump back in the event that
216	 * the breakpoint wasn't caused by DTrace at all.
217	 */
218	movq	$calltrap,dtrace_invop_calltrap_addr(%rip)
219
220	/* Jump to the code hooked in by DTrace. */
221	movq	dtrace_invop_jump_addr,%rax
222	jmpq	*dtrace_invop_jump_addr
223#endif
224	.globl	calltrap
225	.type	calltrap,@function
226calltrap:
227	movq	%rsp,%rdi
228	call	trap
229	MEXITCOUNT
230	jmp	doreti			/* Handle any pending ASTs */
231
232	/*
233	 * alltraps_noen entry point.  Unlike alltraps above, we want to
234	 * leave the interrupts disabled.  This corresponds to
235	 * SDT_SYS386IGT on the i386 port.
236	 */
237	SUPERALIGN_TEXT
238	.globl	alltraps_noen
239	.type	alltraps_noen,@function
240alltraps_noen:
241	movq	%rdi,TF_RDI(%rsp)
242	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
243	jz	1f	/* already running with kernel GS.base */
244	swapgs
245	movq	PCPU(CURPCB),%rdi
246	andl	$~PCB_FULL_IRET,PCB_FLAGS(%rdi)
2471:	movw	%fs,TF_FS(%rsp)
248	movw	%gs,TF_GS(%rsp)
249	movw	%es,TF_ES(%rsp)
250	movw	%ds,TF_DS(%rsp)
251	jmp	alltraps_pushregs_no_rdi
252
253IDTVEC(dblfault)
254	subq	$TF_ERR,%rsp
255	movl	$T_DOUBLEFLT,TF_TRAPNO(%rsp)
256	movq	$0,TF_ADDR(%rsp)
257	movq	$0,TF_ERR(%rsp)
258	movq	%rdi,TF_RDI(%rsp)
259	movq	%rsi,TF_RSI(%rsp)
260	movq	%rdx,TF_RDX(%rsp)
261	movq	%rcx,TF_RCX(%rsp)
262	movq	%r8,TF_R8(%rsp)
263	movq	%r9,TF_R9(%rsp)
264	movq	%rax,TF_RAX(%rsp)
265	movq	%rbx,TF_RBX(%rsp)
266	movq	%rbp,TF_RBP(%rsp)
267	movq	%r10,TF_R10(%rsp)
268	movq	%r11,TF_R11(%rsp)
269	movq	%r12,TF_R12(%rsp)
270	movq	%r13,TF_R13(%rsp)
271	movq	%r14,TF_R14(%rsp)
272	movq	%r15,TF_R15(%rsp)
273	movw	%fs,TF_FS(%rsp)
274	movw	%gs,TF_GS(%rsp)
275	movw	%es,TF_ES(%rsp)
276	movw	%ds,TF_DS(%rsp)
277	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
278	cld
279	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
280	jz	1f			/* already running with kernel GS.base */
281	swapgs
2821:
283	movq	%rsp,%rdi
284	call	dblfault_handler
2852:
286	hlt
287	jmp	2b
288
289IDTVEC(page)
290	subq	$TF_ERR,%rsp
291	movl	$T_PAGEFLT,TF_TRAPNO(%rsp)
292	movq	%rdi,TF_RDI(%rsp)	/* free up a GP register */
293	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
294	jz	1f			/* already running with kernel GS.base */
295	swapgs
296	movq	PCPU(CURPCB),%rdi
297	andl	$~PCB_FULL_IRET,PCB_FLAGS(%rdi)
2981:	movq	%cr2,%rdi		/* preserve %cr2 before ..  */
299	movq	%rdi,TF_ADDR(%rsp)	/* enabling interrupts. */
300	movw	%fs,TF_FS(%rsp)
301	movw	%gs,TF_GS(%rsp)
302	movw	%es,TF_ES(%rsp)
303	movw	%ds,TF_DS(%rsp)
304	testl	$PSL_I,TF_RFLAGS(%rsp)
305	jz	alltraps_pushregs_no_rdi
306	sti
307	jmp	alltraps_pushregs_no_rdi
308
309	/*
310	 * We have to special-case this one.  If we get a trap in doreti() at
311	 * the iretq stage, we'll reenter with the wrong gs state.  We'll have
312	 * to do a special the swapgs in this case even coming from the kernel.
313	 * XXX linux has a trap handler for their equivalent of load_gs().
314	 */
315IDTVEC(prot)
316	subq	$TF_ERR,%rsp
317	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
318	movq	$0,TF_ADDR(%rsp)
319	movq	%rdi,TF_RDI(%rsp)	/* free up a GP register */
320	leaq	doreti_iret(%rip),%rdi
321	cmpq	%rdi,TF_RIP(%rsp)
322	je	1f			/* kernel but with user gsbase!! */
323	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
324	jz	2f			/* already running with kernel GS.base */
3251:	swapgs
3262:	movq	PCPU(CURPCB),%rdi
327	orl	$PCB_FULL_IRET,PCB_FLAGS(%rdi)	/* always full iret from GPF */
328	movw	%fs,TF_FS(%rsp)
329	movw	%gs,TF_GS(%rsp)
330	movw	%es,TF_ES(%rsp)
331	movw	%ds,TF_DS(%rsp)
332	testl	$PSL_I,TF_RFLAGS(%rsp)
333	jz	alltraps_pushregs_no_rdi
334	sti
335	jmp	alltraps_pushregs_no_rdi
336
337/*
338 * Fast syscall entry point.  We enter here with just our new %cs/%ss set,
339 * and the new privilige level.  We are still running on the old user stack
340 * pointer.  We have to juggle a few things around to find our stack etc.
341 * swapgs gives us access to our PCPU space only.
342 */
343IDTVEC(fast_syscall)
344	swapgs
345	movq	%rsp,PCPU(SCRATCH_RSP)
346	movq	PCPU(RSP0),%rsp
347	/* Now emulate a trapframe. Make the 8 byte alignment odd for call. */
348	subq	$TF_SIZE,%rsp
349	/* defer TF_RSP till we have a spare register */
350	movq	%r11,TF_RFLAGS(%rsp)
351	movq	%rcx,TF_RIP(%rsp)	/* %rcx original value is in %r10 */
352	movq	PCPU(SCRATCH_RSP),%r11	/* %r11 already saved */
353	movq	%r11,TF_RSP(%rsp)	/* user stack pointer */
354	movw	%fs,TF_FS(%rsp)
355	movw	%gs,TF_GS(%rsp)
356	movw	%es,TF_ES(%rsp)
357	movw	%ds,TF_DS(%rsp)
358	movq	PCPU(CURPCB),%r11
359	andl	$~PCB_FULL_IRET,PCB_FLAGS(%r11)
360	sti
361	movq	$KUDSEL,TF_SS(%rsp)
362	movq	$KUCSEL,TF_CS(%rsp)
363	movq	$2,TF_ERR(%rsp)
364	movq	%rdi,TF_RDI(%rsp)	/* arg 1 */
365	movq	%rsi,TF_RSI(%rsp)	/* arg 2 */
366	movq	%rdx,TF_RDX(%rsp)	/* arg 3 */
367	movq	%r10,TF_RCX(%rsp)	/* arg 4 */
368	movq	%r8,TF_R8(%rsp)		/* arg 5 */
369	movq	%r9,TF_R9(%rsp)		/* arg 6 */
370	movq	%rax,TF_RAX(%rsp)	/* syscall number */
371	movq	%rbx,TF_RBX(%rsp)	/* C preserved */
372	movq	%rbp,TF_RBP(%rsp)	/* C preserved */
373	movq	%r12,TF_R12(%rsp)	/* C preserved */
374	movq	%r13,TF_R13(%rsp)	/* C preserved */
375	movq	%r14,TF_R14(%rsp)	/* C preserved */
376	movq	%r15,TF_R15(%rsp)	/* C preserved */
377	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
378	cld
379	FAKE_MCOUNT(TF_RIP(%rsp))
380	movq	%rsp,%rdi
381	call	syscall
382	movq	PCPU(CURPCB),%rax
383	MEXITCOUNT
384	jmp	doreti
385
386/*
387 * Here for CYA insurance, in case a "syscall" instruction gets
388 * issued from 32 bit compatability mode. MSR_CSTAR has to point
389 * to *something* if EFER_SCE is enabled.
390 */
391IDTVEC(fast_syscall32)
392	sysret
393
394/*
395 * NMI handling is special.
396 *
397 * First, NMIs do not respect the state of the processor's RFLAGS.IF
398 * bit.  The NMI handler may be entered at any time, including when
399 * the processor is in a critical section with RFLAGS.IF == 0.
400 * The processor's GS.base value could be invalid on entry to the
401 * handler.
402 *
403 * Second, the processor treats NMIs specially, blocking further NMIs
404 * until an 'iretq' instruction is executed.  We thus need to execute
405 * the NMI handler with interrupts disabled, to prevent a nested interrupt
406 * from executing an 'iretq' instruction and inadvertently taking the
407 * processor out of NMI mode.
408 *
409 * Third, the NMI handler runs on its own stack (tss_ist2). The canonical
410 * GS.base value for the processor is stored just above the bottom of its
411 * NMI stack.  For NMIs taken from kernel mode, the current value in
412 * the processor's GS.base is saved at entry to C-preserved register %r12,
413 * the canonical value for GS.base is then loaded into the processor, and
414 * the saved value is restored at exit time.  For NMIs taken from user mode,
415 * the cheaper 'SWAPGS' instructions are used for swapping GS.base.
416 */
417
418IDTVEC(nmi)
419	subq	$TF_RIP,%rsp
420	movl	$(T_NMI),TF_TRAPNO(%rsp)
421	movq	$0,TF_ADDR(%rsp)
422	movq	$0,TF_ERR(%rsp)
423	movq	%rdi,TF_RDI(%rsp)
424	movq	%rsi,TF_RSI(%rsp)
425	movq	%rdx,TF_RDX(%rsp)
426	movq	%rcx,TF_RCX(%rsp)
427	movq	%r8,TF_R8(%rsp)
428	movq	%r9,TF_R9(%rsp)
429	movq	%rax,TF_RAX(%rsp)
430	movq	%rbx,TF_RBX(%rsp)
431	movq	%rbp,TF_RBP(%rsp)
432	movq	%r10,TF_R10(%rsp)
433	movq	%r11,TF_R11(%rsp)
434	movq	%r12,TF_R12(%rsp)
435	movq	%r13,TF_R13(%rsp)
436	movq	%r14,TF_R14(%rsp)
437	movq	%r15,TF_R15(%rsp)
438	movw	%fs,TF_FS(%rsp)
439	movw	%gs,TF_GS(%rsp)
440	movw	%es,TF_ES(%rsp)
441	movw	%ds,TF_DS(%rsp)
442	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
443	cld
444	xorl	%ebx,%ebx
445	testb	$SEL_RPL_MASK,TF_CS(%rsp)
446	jnz	nmi_fromuserspace
447	/*
448	 * We've interrupted the kernel.  Preserve GS.base in %r12.
449	 */
450	movl	$MSR_GSBASE,%ecx
451	rdmsr
452	movq	%rax,%r12
453	shlq	$32,%rdx
454	orq	%rdx,%r12
455	/* Retrieve and load the canonical value for GS.base. */
456	movq	TF_SIZE(%rsp),%rdx
457	movl	%edx,%eax
458	shrq	$32,%rdx
459	wrmsr
460	jmp	nmi_calltrap
461nmi_fromuserspace:
462	incl	%ebx
463	swapgs
464/* Note: this label is also used by ddb and gdb: */
465nmi_calltrap:
466	FAKE_MCOUNT(TF_RIP(%rsp))
467	movq	%rsp,%rdi
468	call	trap
469	MEXITCOUNT
470#ifdef HWPMC_HOOKS
471	/*
472	 * Capture a userspace callchain if needed.
473	 *
474	 * - Check if the current trap was from user mode.
475	 * - Check if the current thread is valid.
476	 * - Check if the thread requires a user call chain to be
477	 *   captured.
478	 *
479	 * We are still in NMI mode at this point.
480	 */
481	testl	%ebx,%ebx
482	jz	nocallchain	/* not from userspace */
483	movq	PCPU(CURTHREAD),%rax
484	orq	%rax,%rax	/* curthread present? */
485	jz	nocallchain
486	testl	$TDP_CALLCHAIN,TD_PFLAGS(%rax) /* flagged for capture? */
487	jz	nocallchain
488	/*
489	 * A user callchain is to be captured, so:
490	 * - Move execution to the regular kernel stack, to allow for
491	 *   nested NMI interrupts.
492	 * - Take the processor out of "NMI" mode by faking an "iret".
493	 * - Enable interrupts, so that copyin() can work.
494	 */
495	movq	%rsp,%rsi	/* source stack pointer */
496	movq	$TF_SIZE,%rcx
497	movq	PCPU(RSP0),%rdx
498	subq	%rcx,%rdx
499	movq	%rdx,%rdi	/* destination stack pointer */
500
501	shrq	$3,%rcx		/* trap frame size in long words */
502	cld
503	rep
504	movsq			/* copy trapframe */
505
506	movl	%ss,%eax
507	pushq	%rax		/* tf_ss */
508	pushq	%rdx		/* tf_rsp (on kernel stack) */
509	pushfq			/* tf_rflags */
510	movl	%cs,%eax
511	pushq	%rax		/* tf_cs */
512	pushq	$outofnmi	/* tf_rip */
513	iretq
514outofnmi:
515	/*
516	 * At this point the processor has exited NMI mode and is running
517	 * with interrupts turned off on the normal kernel stack.
518	 *
519	 * If a pending NMI gets recognized at or after this point, it
520	 * will cause a kernel callchain to be traced.
521	 *
522	 * We turn interrupts back on, and call the user callchain capture hook.
523	 */
524	movq	pmc_hook,%rax
525	orq	%rax,%rax
526	jz	nocallchain
527	movq	PCPU(CURTHREAD),%rdi		/* thread */
528	movq	$PMC_FN_USER_CALLCHAIN,%rsi	/* command */
529	movq	%rsp,%rdx			/* frame */
530	sti
531	call	*%rax
532	cli
533nocallchain:
534#endif
535	testl	%ebx,%ebx
536	jnz	doreti_exit
537nmi_kernelexit:
538	/*
539	 * Put back the preserved MSR_GSBASE value.
540	 */
541	movl	$MSR_GSBASE,%ecx
542	movq	%r12,%rdx
543	movl	%edx,%eax
544	shrq	$32,%rdx
545	wrmsr
546nmi_restoreregs:
547	movq	TF_RDI(%rsp),%rdi
548	movq	TF_RSI(%rsp),%rsi
549	movq	TF_RDX(%rsp),%rdx
550	movq	TF_RCX(%rsp),%rcx
551	movq	TF_R8(%rsp),%r8
552	movq	TF_R9(%rsp),%r9
553	movq	TF_RAX(%rsp),%rax
554	movq	TF_RBX(%rsp),%rbx
555	movq	TF_RBP(%rsp),%rbp
556	movq	TF_R10(%rsp),%r10
557	movq	TF_R11(%rsp),%r11
558	movq	TF_R12(%rsp),%r12
559	movq	TF_R13(%rsp),%r13
560	movq	TF_R14(%rsp),%r14
561	movq	TF_R15(%rsp),%r15
562	addq	$TF_RIP,%rsp
563	jmp	doreti_iret
564
565ENTRY(fork_trampoline)
566	movq	%r12,%rdi		/* function */
567	movq	%rbx,%rsi		/* arg1 */
568	movq	%rsp,%rdx		/* trapframe pointer */
569	call	fork_exit
570	MEXITCOUNT
571	jmp	doreti			/* Handle any ASTs */
572
573/*
574 * To efficiently implement classification of trap and interrupt handlers
575 * for profiling, there must be only trap handlers between the labels btrap
576 * and bintr, and only interrupt handlers between the labels bintr and
577 * eintr.  This is implemented (partly) by including files that contain
578 * some of the handlers.  Before including the files, set up a normal asm
579 * environment so that the included files doen't need to know that they are
580 * included.
581 */
582
583#ifdef COMPAT_FREEBSD32
584	.data
585	.p2align 4
586	.text
587	SUPERALIGN_TEXT
588
589#include <amd64/ia32/ia32_exception.S>
590#endif
591
592	.data
593	.p2align 4
594	.text
595	SUPERALIGN_TEXT
596MCOUNT_LABEL(bintr)
597
598#include <amd64/amd64/apic_vector.S>
599
600#ifdef DEV_ATPIC
601	.data
602	.p2align 4
603	.text
604	SUPERALIGN_TEXT
605
606#include <amd64/amd64/atpic_vector.S>
607#endif
608
609	.text
610MCOUNT_LABEL(eintr)
611
612/*
613 * void doreti(struct trapframe)
614 *
615 * Handle return from interrupts, traps and syscalls.
616 */
617	.text
618	SUPERALIGN_TEXT
619	.type	doreti,@function
620doreti:
621	FAKE_MCOUNT($bintr)		/* init "from" bintr -> doreti */
622	/*
623	 * Check if ASTs can be handled now.
624	 */
625	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* are we returning to user mode? */
626	jz	doreti_exit		/* can't handle ASTs now if not */
627
628doreti_ast:
629	/*
630	 * Check for ASTs atomically with returning.  Disabling CPU
631	 * interrupts provides sufficient locking eve in the SMP case,
632	 * since we will be informed of any new ASTs by an IPI.
633	 */
634	cli
635	movq	PCPU(CURTHREAD),%rax
636	testl	$TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax)
637	je	doreti_exit
638	sti
639	movq	%rsp,%rdi	/* pass a pointer to the trapframe */
640	call	ast
641	jmp	doreti_ast
642
643	/*
644	 * doreti_exit:	pop registers, iret.
645	 *
646	 *	The segment register pop is a special case, since it may
647	 *	fault if (for example) a sigreturn specifies bad segment
648	 *	registers.  The fault is handled in trap.c.
649	 */
650doreti_exit:
651	MEXITCOUNT
652	movq	PCPU(CURTHREAD),%r8
653	movq	TD_PCB(%r8),%r8
654
655	/*
656	 * Do not reload segment registers for kernel.
657	 * Since we do not reload segments registers with sane
658	 * values on kernel entry, descriptors referenced by
659	 * segments registers might be not valid.  This is fatal
660	 * for user mode, but is not a problem for the kernel.
661	 */
662	testb	$SEL_RPL_MASK,TF_CS(%rsp)
663	jz	ld_regs
664	testl	$PCB_FULL_IRET,PCB_FLAGS(%r8)
665	jz	ld_regs
666	testl	$TF_HASSEGS,TF_FLAGS(%rsp)
667	je	set_segs
668
669do_segs:
670	/* Restore %fs and fsbase */
671	movw	TF_FS(%rsp),%ax
672	.globl	ld_fs
673ld_fs:
674	movw	%ax,%fs
675	cmpw	$KUF32SEL,%ax
676	jne	1f
677	movl	$MSR_FSBASE,%ecx
678	movl	PCB_FSBASE(%r8),%eax
679	movl	PCB_FSBASE+4(%r8),%edx
680	.globl	ld_fsbase
681ld_fsbase:
682	wrmsr
6831:
684	/* Restore %gs and gsbase */
685	movw	TF_GS(%rsp),%si
686	pushfq
687	cli
688	movl	$MSR_GSBASE,%ecx
689	rdmsr
690	.globl	ld_gs
691ld_gs:
692	movw	%si,%gs
693	wrmsr
694	popfq
695	cmpw	$KUG32SEL,%si
696	jne	1f
697	movl	$MSR_KGSBASE,%ecx
698	movl	PCB_GSBASE(%r8),%eax
699	movl	PCB_GSBASE+4(%r8),%edx
700	.globl	ld_gsbase
701ld_gsbase:
702	wrmsr
7031:
704	.globl	ld_es
705ld_es:
706	movw	TF_ES(%rsp),%es
707	.globl	ld_ds
708ld_ds:
709	movw	TF_DS(%rsp),%ds
710ld_regs:
711	movq	TF_RDI(%rsp),%rdi
712	movq	TF_RSI(%rsp),%rsi
713	movq	TF_RDX(%rsp),%rdx
714	movq	TF_RCX(%rsp),%rcx
715	movq	TF_R8(%rsp),%r8
716	movq	TF_R9(%rsp),%r9
717	movq	TF_RAX(%rsp),%rax
718	movq	TF_RBX(%rsp),%rbx
719	movq	TF_RBP(%rsp),%rbp
720	movq	TF_R10(%rsp),%r10
721	movq	TF_R11(%rsp),%r11
722	movq	TF_R12(%rsp),%r12
723	movq	TF_R13(%rsp),%r13
724	movq	TF_R14(%rsp),%r14
725	movq	TF_R15(%rsp),%r15
726	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
727	jz	1f			/* keep running with kernel GS.base */
728	cli
729	swapgs
7301:
731	addq	$TF_RIP,%rsp		/* skip over tf_err, tf_trapno */
732	.globl	doreti_iret
733doreti_iret:
734	iretq
735
736set_segs:
737	movw	$KUDSEL,%ax
738	movw	%ax,TF_DS(%rsp)
739	movw	%ax,TF_ES(%rsp)
740	movw	$KUF32SEL,TF_FS(%rsp)
741	movw	$KUG32SEL,TF_GS(%rsp)
742	jmp	do_segs
743
744	/*
745	 * doreti_iret_fault.  Alternative return code for
746	 * the case where we get a fault in the doreti_exit code
747	 * above.  trap() (amd64/amd64/trap.c) catches this specific
748	 * case, sends the process a signal and continues in the
749	 * corresponding place in the code below.
750	 */
751	ALIGN_TEXT
752	.globl	doreti_iret_fault
753doreti_iret_fault:
754	subq	$TF_RIP,%rsp		/* space including tf_err, tf_trapno */
755	testl	$PSL_I,TF_RFLAGS(%rsp)
756	jz	1f
757	sti
7581:
759	movw	%fs,TF_FS(%rsp)
760	movw	%gs,TF_GS(%rsp)
761	movw	%es,TF_ES(%rsp)
762	movw	%ds,TF_DS(%rsp)
763	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
764	movq	%rdi,TF_RDI(%rsp)
765	movq	%rsi,TF_RSI(%rsp)
766	movq	%rdx,TF_RDX(%rsp)
767	movq	%rcx,TF_RCX(%rsp)
768	movq	%r8,TF_R8(%rsp)
769	movq	%r9,TF_R9(%rsp)
770	movq	%rax,TF_RAX(%rsp)
771	movq	%rbx,TF_RBX(%rsp)
772	movq	%rbp,TF_RBP(%rsp)
773	movq	%r10,TF_R10(%rsp)
774	movq	%r11,TF_R11(%rsp)
775	movq	%r12,TF_R12(%rsp)
776	movq	%r13,TF_R13(%rsp)
777	movq	%r14,TF_R14(%rsp)
778	movq	%r15,TF_R15(%rsp)
779	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
780	movq	$0,TF_ERR(%rsp)	/* XXX should be the error code */
781	movq	$0,TF_ADDR(%rsp)
782	FAKE_MCOUNT(TF_RIP(%rsp))
783	jmp	calltrap
784
785	ALIGN_TEXT
786	.globl	ds_load_fault
787ds_load_fault:
788	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
789	movq	%rsp,%rdi
790	call	trap
791	movw	$KUDSEL,TF_DS(%rsp)
792	jmp	doreti
793
794	ALIGN_TEXT
795	.globl	es_load_fault
796es_load_fault:
797	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
798	movq	%rsp,%rdi
799	call	trap
800	movw	$KUDSEL,TF_ES(%rsp)
801	jmp	doreti
802
803	ALIGN_TEXT
804	.globl	fs_load_fault
805fs_load_fault:
806	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
807	movq	%rsp,%rdi
808	call	trap
809	movw	$KUF32SEL,TF_FS(%rsp)
810	jmp	doreti
811
812	ALIGN_TEXT
813	.globl	gs_load_fault
814gs_load_fault:
815	popfq
816	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
817	movq	%rsp,%rdi
818	call	trap
819	movw	$KUG32SEL,TF_GS(%rsp)
820	jmp	doreti
821
822	ALIGN_TEXT
823	.globl	fsbase_load_fault
824fsbase_load_fault:
825	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
826	movq	%rsp,%rdi
827	call	trap
828	movq	PCPU(CURTHREAD),%r8
829	movq	TD_PCB(%r8),%r8
830	movq	$0,PCB_FSBASE(%r8)
831	jmp	doreti
832
833	ALIGN_TEXT
834	.globl	gsbase_load_fault
835gsbase_load_fault:
836	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
837	movq	%rsp,%rdi
838	call	trap
839	movq	PCPU(CURTHREAD),%r8
840	movq	TD_PCB(%r8),%r8
841	movq	$0,PCB_GSBASE(%r8)
842	jmp	doreti
843
844#ifdef HWPMC_HOOKS
845	ENTRY(end_exceptions)
846#endif
847