exception.S revision 287147
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: releng/9.3/sys/amd64/amd64/exception.S 287147 2015-08-25 20:49:05Z delphij $
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, so we must do that, and we
92 * must restore them prior to calling 'iret'.  The CPU adjusts %cs and %ss
93 * but does not mess with %ds, %es, %gs or %fs.  We swap the %gs base for
94 * for the kernel mode operation shortly, without changes to the selector
95 * loaded.  Since superuser long mode works with any selectors loaded into
96 * segment registers other then %cs, which makes them mostly unused in long
97 * mode, and kernel does not reference %fs, leave them alone.  The segment
98 * registers are reloaded on return to the usermode.
99 */
100
101MCOUNT_LABEL(user)
102MCOUNT_LABEL(btrap)
103
104/* Traps that we leave interrupts disabled for.. */
105#define	TRAP_NOEN(a)	\
106	subq $TF_RIP,%rsp; \
107	movl $(a),TF_TRAPNO(%rsp) ; \
108	movq $0,TF_ADDR(%rsp) ; \
109	movq $0,TF_ERR(%rsp) ; \
110	jmp alltraps_noen
111IDTVEC(dbg)
112	TRAP_NOEN(T_TRCTRAP)
113IDTVEC(bpt)
114	TRAP_NOEN(T_BPTFLT)
115#ifdef KDTRACE_HOOKS
116IDTVEC(dtrace_ret)
117	TRAP_NOEN(T_DTRACE_RET)
118#endif
119
120/* Regular traps; The cpu does not supply tf_err for these. */
121#define	TRAP(a)	 \
122	subq $TF_RIP,%rsp; \
123	movl $(a),TF_TRAPNO(%rsp) ; \
124	movq $0,TF_ADDR(%rsp) ; \
125	movq $0,TF_ERR(%rsp) ; \
126	jmp alltraps
127IDTVEC(div)
128	TRAP(T_DIVIDE)
129IDTVEC(ofl)
130	TRAP(T_OFLOW)
131IDTVEC(bnd)
132	TRAP(T_BOUND)
133IDTVEC(ill)
134	TRAP(T_PRIVINFLT)
135IDTVEC(dna)
136	TRAP(T_DNA)
137IDTVEC(fpusegm)
138	TRAP(T_FPOPFLT)
139IDTVEC(mchk)
140	TRAP(T_MCHK)
141IDTVEC(rsvd)
142	TRAP(T_RESERVED)
143IDTVEC(fpu)
144	TRAP(T_ARITHTRAP)
145IDTVEC(xmm)
146	TRAP(T_XMMFLT)
147
148/* This group of traps have tf_err already pushed by the cpu */
149#define	TRAP_ERR(a)	\
150	subq $TF_ERR,%rsp; \
151	movl $(a),TF_TRAPNO(%rsp) ; \
152	movq $0,TF_ADDR(%rsp) ; \
153	jmp alltraps
154IDTVEC(tss)
155	TRAP_ERR(T_TSSFLT)
156IDTVEC(missing)
157	subq	$TF_ERR,%rsp
158	movl	$T_SEGNPFLT,TF_TRAPNO(%rsp)
159	jmp	prot_addrf
160IDTVEC(stk)
161	subq	$TF_ERR,%rsp
162	movl	$T_STKFLT,TF_TRAPNO(%rsp)
163	jmp	prot_addrf
164IDTVEC(align)
165	TRAP_ERR(T_ALIGNFLT)
166
167	/*
168	 * alltraps entry point.  Use swapgs if this is the first time in the
169	 * kernel from userland.  Reenable interrupts if they were enabled
170	 * before the trap.  This approximates SDT_SYS386TGT on the i386 port.
171	 */
172	SUPERALIGN_TEXT
173	.globl	alltraps
174	.type	alltraps,@function
175alltraps:
176	movq	%rdi,TF_RDI(%rsp)
177	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
178	jz	alltraps_testi		/* already running with kernel GS.base */
179	swapgs
180	movq	PCPU(CURPCB),%rdi
181	andl	$~PCB_FULL_IRET,PCB_FLAGS(%rdi)
182	movw	%fs,TF_FS(%rsp)
183	movw	%gs,TF_GS(%rsp)
184	movw	%es,TF_ES(%rsp)
185	movw	%ds,TF_DS(%rsp)
186alltraps_testi:
187	testl	$PSL_I,TF_RFLAGS(%rsp)
188	jz	alltraps_pushregs_no_rdi
189	sti
190alltraps_pushregs_no_rdi:
191	movq	%rsi,TF_RSI(%rsp)
192	movq	%rdx,TF_RDX(%rsp)
193	movq	%rcx,TF_RCX(%rsp)
194	movq	%r8,TF_R8(%rsp)
195	movq	%r9,TF_R9(%rsp)
196	movq	%rax,TF_RAX(%rsp)
197	movq	%rbx,TF_RBX(%rsp)
198	movq	%rbp,TF_RBP(%rsp)
199	movq	%r10,TF_R10(%rsp)
200	movq	%r11,TF_R11(%rsp)
201	movq	%r12,TF_R12(%rsp)
202	movq	%r13,TF_R13(%rsp)
203	movq	%r14,TF_R14(%rsp)
204	movq	%r15,TF_R15(%rsp)
205	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
206	cld
207	FAKE_MCOUNT(TF_RIP(%rsp))
208#ifdef KDTRACE_HOOKS
209	/*
210	 * DTrace Function Boundary Trace (fbt) probes are triggered
211	 * by int3 (0xcc) which causes the #BP (T_BPTFLT) breakpoint
212	 * interrupt. For all other trap types, just handle them in
213	 * the usual way.
214	 */
215	cmpl	$T_BPTFLT,TF_TRAPNO(%rsp)
216	jne	calltrap
217
218	/* Check if there is no DTrace hook registered. */
219	cmpq	$0,dtrace_invop_jump_addr
220	je	calltrap
221
222	/*
223	 * Set our jump address for the jump back in the event that
224	 * the breakpoint wasn't caused by DTrace at all.
225	 */
226	movq	$calltrap,dtrace_invop_calltrap_addr(%rip)
227
228	/* Jump to the code hooked in by DTrace. */
229	movq	dtrace_invop_jump_addr,%rax
230	jmpq	*dtrace_invop_jump_addr
231#endif
232	.globl	calltrap
233	.type	calltrap,@function
234calltrap:
235	movq	%rsp,%rdi
236	call	trap
237	MEXITCOUNT
238	jmp	doreti			/* Handle any pending ASTs */
239
240	/*
241	 * alltraps_noen entry point.  Unlike alltraps above, we want to
242	 * leave the interrupts disabled.  This corresponds to
243	 * SDT_SYS386IGT on the i386 port.
244	 */
245	SUPERALIGN_TEXT
246	.globl	alltraps_noen
247	.type	alltraps_noen,@function
248alltraps_noen:
249	movq	%rdi,TF_RDI(%rsp)
250	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
251	jz	1f	/* already running with kernel GS.base */
252	swapgs
253	movq	PCPU(CURPCB),%rdi
254	andl	$~PCB_FULL_IRET,PCB_FLAGS(%rdi)
2551:	movw	%fs,TF_FS(%rsp)
256	movw	%gs,TF_GS(%rsp)
257	movw	%es,TF_ES(%rsp)
258	movw	%ds,TF_DS(%rsp)
259	jmp	alltraps_pushregs_no_rdi
260
261IDTVEC(dblfault)
262	subq	$TF_ERR,%rsp
263	movl	$T_DOUBLEFLT,TF_TRAPNO(%rsp)
264	movq	$0,TF_ADDR(%rsp)
265	movq	$0,TF_ERR(%rsp)
266	movq	%rdi,TF_RDI(%rsp)
267	movq	%rsi,TF_RSI(%rsp)
268	movq	%rdx,TF_RDX(%rsp)
269	movq	%rcx,TF_RCX(%rsp)
270	movq	%r8,TF_R8(%rsp)
271	movq	%r9,TF_R9(%rsp)
272	movq	%rax,TF_RAX(%rsp)
273	movq	%rbx,TF_RBX(%rsp)
274	movq	%rbp,TF_RBP(%rsp)
275	movq	%r10,TF_R10(%rsp)
276	movq	%r11,TF_R11(%rsp)
277	movq	%r12,TF_R12(%rsp)
278	movq	%r13,TF_R13(%rsp)
279	movq	%r14,TF_R14(%rsp)
280	movq	%r15,TF_R15(%rsp)
281	movw	%fs,TF_FS(%rsp)
282	movw	%gs,TF_GS(%rsp)
283	movw	%es,TF_ES(%rsp)
284	movw	%ds,TF_DS(%rsp)
285	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
286	cld
287	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
288	jz	1f			/* already running with kernel GS.base */
289	swapgs
2901:
291	movq	%rsp,%rdi
292	call	dblfault_handler
2932:
294	hlt
295	jmp	2b
296
297IDTVEC(page)
298	subq	$TF_ERR,%rsp
299	movl	$T_PAGEFLT,TF_TRAPNO(%rsp)
300	movq	%rdi,TF_RDI(%rsp)	/* free up a GP register */
301	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
302	jz	1f			/* already running with kernel GS.base */
303	swapgs
304	movq	PCPU(CURPCB),%rdi
305	andl	$~PCB_FULL_IRET,PCB_FLAGS(%rdi)
3061:	movq	%cr2,%rdi		/* preserve %cr2 before ..  */
307	movq	%rdi,TF_ADDR(%rsp)	/* enabling interrupts. */
308	movw	%fs,TF_FS(%rsp)
309	movw	%gs,TF_GS(%rsp)
310	movw	%es,TF_ES(%rsp)
311	movw	%ds,TF_DS(%rsp)
312	testl	$PSL_I,TF_RFLAGS(%rsp)
313	jz	alltraps_pushregs_no_rdi
314	sti
315	jmp	alltraps_pushregs_no_rdi
316
317	/*
318	 * We have to special-case this one.  If we get a trap in doreti() at
319	 * the iretq stage, we'll reenter with the wrong gs state.  We'll have
320	 * to do a special the swapgs in this case even coming from the kernel.
321	 * XXX linux has a trap handler for their equivalent of load_gs().
322	 */
323IDTVEC(prot)
324	subq	$TF_ERR,%rsp
325	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
326prot_addrf:
327	movq	$0,TF_ADDR(%rsp)
328	movq	%rdi,TF_RDI(%rsp)	/* free up a GP register */
329	leaq	doreti_iret(%rip),%rdi
330	cmpq	%rdi,TF_RIP(%rsp)
331	je	1f			/* kernel but with user gsbase!! */
332	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
333	jz	2f			/* already running with kernel GS.base */
3341:	swapgs
3352:	movq	PCPU(CURPCB),%rdi
336	orl	$PCB_FULL_IRET,PCB_FLAGS(%rdi)	/* always full iret from GPF */
337	movw	%fs,TF_FS(%rsp)
338	movw	%gs,TF_GS(%rsp)
339	movw	%es,TF_ES(%rsp)
340	movw	%ds,TF_DS(%rsp)
341	testl	$PSL_I,TF_RFLAGS(%rsp)
342	jz	alltraps_pushregs_no_rdi
343	sti
344	jmp	alltraps_pushregs_no_rdi
345
346/*
347 * Fast syscall entry point.  We enter here with just our new %cs/%ss set,
348 * and the new privilige level.  We are still running on the old user stack
349 * pointer.  We have to juggle a few things around to find our stack etc.
350 * swapgs gives us access to our PCPU space only.
351 *
352 * We do not support invoking this from a custom %cs or %ss (e.g. using
353 * entries from an LDT).
354 */
355IDTVEC(fast_syscall)
356	swapgs
357	movq	%rsp,PCPU(SCRATCH_RSP)
358	movq	PCPU(RSP0),%rsp
359	/* Now emulate a trapframe. Make the 8 byte alignment odd for call. */
360	subq	$TF_SIZE,%rsp
361	/* defer TF_RSP till we have a spare register */
362	movq	%r11,TF_RFLAGS(%rsp)
363	movq	%rcx,TF_RIP(%rsp)	/* %rcx original value is in %r10 */
364	movq	PCPU(SCRATCH_RSP),%r11	/* %r11 already saved */
365	movq	%r11,TF_RSP(%rsp)	/* user stack pointer */
366	movw	%fs,TF_FS(%rsp)
367	movw	%gs,TF_GS(%rsp)
368	movw	%es,TF_ES(%rsp)
369	movw	%ds,TF_DS(%rsp)
370	movq	PCPU(CURPCB),%r11
371	andl	$~PCB_FULL_IRET,PCB_FLAGS(%r11)
372	sti
373	movq	$KUDSEL,TF_SS(%rsp)
374	movq	$KUCSEL,TF_CS(%rsp)
375	movq	$2,TF_ERR(%rsp)
376	movq	%rdi,TF_RDI(%rsp)	/* arg 1 */
377	movq	%rsi,TF_RSI(%rsp)	/* arg 2 */
378	movq	%rdx,TF_RDX(%rsp)	/* arg 3 */
379	movq	%r10,TF_RCX(%rsp)	/* arg 4 */
380	movq	%r8,TF_R8(%rsp)		/* arg 5 */
381	movq	%r9,TF_R9(%rsp)		/* arg 6 */
382	movq	%rax,TF_RAX(%rsp)	/* syscall number */
383	movq	%rbx,TF_RBX(%rsp)	/* C preserved */
384	movq	%rbp,TF_RBP(%rsp)	/* C preserved */
385	movq	%r12,TF_R12(%rsp)	/* C preserved */
386	movq	%r13,TF_R13(%rsp)	/* C preserved */
387	movq	%r14,TF_R14(%rsp)	/* C preserved */
388	movq	%r15,TF_R15(%rsp)	/* C preserved */
389	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
390	cld
391	FAKE_MCOUNT(TF_RIP(%rsp))
392	movq	PCPU(CURTHREAD),%rdi
393	movq	%rsp,TD_FRAME(%rdi)
394	movl	TF_RFLAGS(%rsp),%esi
395	andl	$PSL_T,%esi
396	call	amd64_syscall
3971:	movq	PCPU(CURPCB),%rax
398	/* Disable interrupts before testing PCB_FULL_IRET. */
399	cli
400	testl	$PCB_FULL_IRET,PCB_FLAGS(%rax)
401	jnz	3f
402	/* Check for and handle AST's on return to userland. */
403	movq	PCPU(CURTHREAD),%rax
404	testl	$TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax)
405	jne	2f
406	/* Restore preserved registers. */
407	MEXITCOUNT
408	movq	TF_RDI(%rsp),%rdi	/* bonus; preserve arg 1 */
409	movq	TF_RSI(%rsp),%rsi	/* bonus: preserve arg 2 */
410	movq	TF_RDX(%rsp),%rdx	/* return value 2 */
411	movq	TF_RAX(%rsp),%rax	/* return value 1 */
412	movq	TF_RFLAGS(%rsp),%r11	/* original %rflags */
413	movq	TF_RIP(%rsp),%rcx	/* original %rip */
414	movq	TF_RSP(%rsp),%rsp	/* user stack pointer */
415	swapgs
416	sysretq
417
4182:	/* AST scheduled. */
419	sti
420	movq	%rsp,%rdi
421	call	ast
422	jmp	1b
423
4243:	/* Requested full context restore, use doreti for that. */
425	MEXITCOUNT
426	jmp	doreti
427
428/*
429 * Here for CYA insurance, in case a "syscall" instruction gets
430 * issued from 32 bit compatability mode. MSR_CSTAR has to point
431 * to *something* if EFER_SCE is enabled.
432 */
433IDTVEC(fast_syscall32)
434	sysret
435
436/*
437 * NMI handling is special.
438 *
439 * First, NMIs do not respect the state of the processor's RFLAGS.IF
440 * bit.  The NMI handler may be entered at any time, including when
441 * the processor is in a critical section with RFLAGS.IF == 0.
442 * The processor's GS.base value could be invalid on entry to the
443 * handler.
444 *
445 * Second, the processor treats NMIs specially, blocking further NMIs
446 * until an 'iretq' instruction is executed.  We thus need to execute
447 * the NMI handler with interrupts disabled, to prevent a nested interrupt
448 * from executing an 'iretq' instruction and inadvertently taking the
449 * processor out of NMI mode.
450 *
451 * Third, the NMI handler runs on its own stack (tss_ist2). The canonical
452 * GS.base value for the processor is stored just above the bottom of its
453 * NMI stack.  For NMIs taken from kernel mode, the current value in
454 * the processor's GS.base is saved at entry to C-preserved register %r12,
455 * the canonical value for GS.base is then loaded into the processor, and
456 * the saved value is restored at exit time.  For NMIs taken from user mode,
457 * the cheaper 'SWAPGS' instructions are used for swapping GS.base.
458 */
459
460IDTVEC(nmi)
461	subq	$TF_RIP,%rsp
462	movl	$(T_NMI),TF_TRAPNO(%rsp)
463	movq	$0,TF_ADDR(%rsp)
464	movq	$0,TF_ERR(%rsp)
465	movq	%rdi,TF_RDI(%rsp)
466	movq	%rsi,TF_RSI(%rsp)
467	movq	%rdx,TF_RDX(%rsp)
468	movq	%rcx,TF_RCX(%rsp)
469	movq	%r8,TF_R8(%rsp)
470	movq	%r9,TF_R9(%rsp)
471	movq	%rax,TF_RAX(%rsp)
472	movq	%rbx,TF_RBX(%rsp)
473	movq	%rbp,TF_RBP(%rsp)
474	movq	%r10,TF_R10(%rsp)
475	movq	%r11,TF_R11(%rsp)
476	movq	%r12,TF_R12(%rsp)
477	movq	%r13,TF_R13(%rsp)
478	movq	%r14,TF_R14(%rsp)
479	movq	%r15,TF_R15(%rsp)
480	movw	%fs,TF_FS(%rsp)
481	movw	%gs,TF_GS(%rsp)
482	movw	%es,TF_ES(%rsp)
483	movw	%ds,TF_DS(%rsp)
484	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
485	cld
486	xorl	%ebx,%ebx
487	testb	$SEL_RPL_MASK,TF_CS(%rsp)
488	jnz	nmi_fromuserspace
489	/*
490	 * We've interrupted the kernel.  Preserve GS.base in %r12.
491	 */
492	movl	$MSR_GSBASE,%ecx
493	rdmsr
494	movq	%rax,%r12
495	shlq	$32,%rdx
496	orq	%rdx,%r12
497	/* Retrieve and load the canonical value for GS.base. */
498	movq	TF_SIZE(%rsp),%rdx
499	movl	%edx,%eax
500	shrq	$32,%rdx
501	wrmsr
502	jmp	nmi_calltrap
503nmi_fromuserspace:
504	incl	%ebx
505	swapgs
506/* Note: this label is also used by ddb and gdb: */
507nmi_calltrap:
508	FAKE_MCOUNT(TF_RIP(%rsp))
509	movq	%rsp,%rdi
510	call	trap
511	MEXITCOUNT
512#ifdef HWPMC_HOOKS
513	/*
514	 * Capture a userspace callchain if needed.
515	 *
516	 * - Check if the current trap was from user mode.
517	 * - Check if the current thread is valid.
518	 * - Check if the thread requires a user call chain to be
519	 *   captured.
520	 *
521	 * We are still in NMI mode at this point.
522	 */
523	testl	%ebx,%ebx
524	jz	nocallchain	/* not from userspace */
525	movq	PCPU(CURTHREAD),%rax
526	orq	%rax,%rax	/* curthread present? */
527	jz	nocallchain
528	testl	$TDP_CALLCHAIN,TD_PFLAGS(%rax) /* flagged for capture? */
529	jz	nocallchain
530	/*
531	 * A user callchain is to be captured, so:
532	 * - Move execution to the regular kernel stack, to allow for
533	 *   nested NMI interrupts.
534	 * - Take the processor out of "NMI" mode by faking an "iret".
535	 * - Enable interrupts, so that copyin() can work.
536	 */
537	movq	%rsp,%rsi	/* source stack pointer */
538	movq	$TF_SIZE,%rcx
539	movq	PCPU(RSP0),%rdx
540	subq	%rcx,%rdx
541	movq	%rdx,%rdi	/* destination stack pointer */
542
543	shrq	$3,%rcx		/* trap frame size in long words */
544	cld
545	rep
546	movsq			/* copy trapframe */
547
548	movl	%ss,%eax
549	pushq	%rax		/* tf_ss */
550	pushq	%rdx		/* tf_rsp (on kernel stack) */
551	pushfq			/* tf_rflags */
552	movl	%cs,%eax
553	pushq	%rax		/* tf_cs */
554	pushq	$outofnmi	/* tf_rip */
555	iretq
556outofnmi:
557	/*
558	 * At this point the processor has exited NMI mode and is running
559	 * with interrupts turned off on the normal kernel stack.
560	 *
561	 * If a pending NMI gets recognized at or after this point, it
562	 * will cause a kernel callchain to be traced.
563	 *
564	 * We turn interrupts back on, and call the user callchain capture hook.
565	 */
566	movq	pmc_hook,%rax
567	orq	%rax,%rax
568	jz	nocallchain
569	movq	PCPU(CURTHREAD),%rdi		/* thread */
570	movq	$PMC_FN_USER_CALLCHAIN,%rsi	/* command */
571	movq	%rsp,%rdx			/* frame */
572	sti
573	call	*%rax
574	cli
575nocallchain:
576#endif
577	testl	%ebx,%ebx
578	jnz	doreti_exit
579nmi_kernelexit:
580	/*
581	 * Put back the preserved MSR_GSBASE value.
582	 */
583	movl	$MSR_GSBASE,%ecx
584	movq	%r12,%rdx
585	movl	%edx,%eax
586	shrq	$32,%rdx
587	wrmsr
588nmi_restoreregs:
589	movq	TF_RDI(%rsp),%rdi
590	movq	TF_RSI(%rsp),%rsi
591	movq	TF_RDX(%rsp),%rdx
592	movq	TF_RCX(%rsp),%rcx
593	movq	TF_R8(%rsp),%r8
594	movq	TF_R9(%rsp),%r9
595	movq	TF_RAX(%rsp),%rax
596	movq	TF_RBX(%rsp),%rbx
597	movq	TF_RBP(%rsp),%rbp
598	movq	TF_R10(%rsp),%r10
599	movq	TF_R11(%rsp),%r11
600	movq	TF_R12(%rsp),%r12
601	movq	TF_R13(%rsp),%r13
602	movq	TF_R14(%rsp),%r14
603	movq	TF_R15(%rsp),%r15
604	addq	$TF_RIP,%rsp
605	jmp	doreti_iret
606
607ENTRY(fork_trampoline)
608	movq	%r12,%rdi		/* function */
609	movq	%rbx,%rsi		/* arg1 */
610	movq	%rsp,%rdx		/* trapframe pointer */
611	call	fork_exit
612	MEXITCOUNT
613	jmp	doreti			/* Handle any ASTs */
614
615/*
616 * To efficiently implement classification of trap and interrupt handlers
617 * for profiling, there must be only trap handlers between the labels btrap
618 * and bintr, and only interrupt handlers between the labels bintr and
619 * eintr.  This is implemented (partly) by including files that contain
620 * some of the handlers.  Before including the files, set up a normal asm
621 * environment so that the included files doen't need to know that they are
622 * included.
623 */
624
625#ifdef COMPAT_FREEBSD32
626	.data
627	.p2align 4
628	.text
629	SUPERALIGN_TEXT
630
631#include <amd64/ia32/ia32_exception.S>
632#endif
633
634	.data
635	.p2align 4
636	.text
637	SUPERALIGN_TEXT
638MCOUNT_LABEL(bintr)
639
640#include <amd64/amd64/apic_vector.S>
641
642#ifdef DEV_ATPIC
643	.data
644	.p2align 4
645	.text
646	SUPERALIGN_TEXT
647
648#include <amd64/amd64/atpic_vector.S>
649#endif
650
651	.text
652MCOUNT_LABEL(eintr)
653
654/*
655 * void doreti(struct trapframe)
656 *
657 * Handle return from interrupts, traps and syscalls.
658 */
659	.text
660	SUPERALIGN_TEXT
661	.type	doreti,@function
662doreti:
663	FAKE_MCOUNT($bintr)		/* init "from" bintr -> doreti */
664	/*
665	 * Check if ASTs can be handled now.
666	 */
667	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* are we returning to user mode? */
668	jz	doreti_exit		/* can't handle ASTs now if not */
669
670doreti_ast:
671	/*
672	 * Check for ASTs atomically with returning.  Disabling CPU
673	 * interrupts provides sufficient locking even in the SMP case,
674	 * since we will be informed of any new ASTs by an IPI.
675	 */
676	cli
677	movq	PCPU(CURTHREAD),%rax
678	testl	$TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax)
679	je	doreti_exit
680	sti
681	movq	%rsp,%rdi	/* pass a pointer to the trapframe */
682	call	ast
683	jmp	doreti_ast
684
685	/*
686	 * doreti_exit:	pop registers, iret.
687	 *
688	 *	The segment register pop is a special case, since it may
689	 *	fault if (for example) a sigreturn specifies bad segment
690	 *	registers.  The fault is handled in trap.c.
691	 */
692doreti_exit:
693	MEXITCOUNT
694	movq	PCPU(CURPCB),%r8
695
696	/*
697	 * Do not reload segment registers for kernel.
698	 * Since we do not reload segments registers with sane
699	 * values on kernel entry, descriptors referenced by
700	 * segments registers might be not valid.  This is fatal
701	 * for user mode, but is not a problem for the kernel.
702	 */
703	testb	$SEL_RPL_MASK,TF_CS(%rsp)
704	jz	ld_regs
705	testl	$PCB_FULL_IRET,PCB_FLAGS(%r8)
706	jz	ld_regs
707	testl	$TF_HASSEGS,TF_FLAGS(%rsp)
708	je	set_segs
709
710do_segs:
711	/* Restore %fs and fsbase */
712	movw	TF_FS(%rsp),%ax
713	.globl	ld_fs
714ld_fs:
715	movw	%ax,%fs
716	cmpw	$KUF32SEL,%ax
717	jne	1f
718	movl	$MSR_FSBASE,%ecx
719	movl	PCB_FSBASE(%r8),%eax
720	movl	PCB_FSBASE+4(%r8),%edx
721	.globl	ld_fsbase
722ld_fsbase:
723	wrmsr
7241:
725	/* Restore %gs and gsbase */
726	movw	TF_GS(%rsp),%si
727	pushfq
728	cli
729	movl	$MSR_GSBASE,%ecx
730	/* Save current kernel %gs base into %r12d:%r13d */
731	rdmsr
732	movl	%eax,%r12d
733	movl	%edx,%r13d
734	.globl	ld_gs
735ld_gs:
736	movw	%si,%gs
737	/* Save user %gs base into %r14d:%r15d */
738	rdmsr
739	movl	%eax,%r14d
740	movl	%edx,%r15d
741	/* Restore kernel %gs base */
742	movl	%r12d,%eax
743	movl	%r13d,%edx
744	wrmsr
745	popfq
746	/*
747	 * Restore user %gs base, either from PCB if used for TLS, or
748	 * from the previously saved msr read.
749	 */
750	movl	$MSR_KGSBASE,%ecx
751	cmpw	$KUG32SEL,%si
752	jne	1f
753	movl	PCB_GSBASE(%r8),%eax
754	movl	PCB_GSBASE+4(%r8),%edx
755	jmp	ld_gsbase
7561:
757	movl	%r14d,%eax
758	movl	%r15d,%edx
759	.globl	ld_gsbase
760ld_gsbase:
761	wrmsr	/* May trap if non-canonical, but only for TLS. */
762	.globl	ld_es
763ld_es:
764	movw	TF_ES(%rsp),%es
765	.globl	ld_ds
766ld_ds:
767	movw	TF_DS(%rsp),%ds
768ld_regs:
769	movq	TF_RDI(%rsp),%rdi
770	movq	TF_RSI(%rsp),%rsi
771	movq	TF_RDX(%rsp),%rdx
772	movq	TF_RCX(%rsp),%rcx
773	movq	TF_R8(%rsp),%r8
774	movq	TF_R9(%rsp),%r9
775	movq	TF_RAX(%rsp),%rax
776	movq	TF_RBX(%rsp),%rbx
777	movq	TF_RBP(%rsp),%rbp
778	movq	TF_R10(%rsp),%r10
779	movq	TF_R11(%rsp),%r11
780	movq	TF_R12(%rsp),%r12
781	movq	TF_R13(%rsp),%r13
782	movq	TF_R14(%rsp),%r14
783	movq	TF_R15(%rsp),%r15
784	testb	$SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
785	jz	1f			/* keep running with kernel GS.base */
786	cli
787	swapgs
7881:
789	addq	$TF_RIP,%rsp		/* skip over tf_err, tf_trapno */
790	.globl	doreti_iret
791doreti_iret:
792	iretq
793
794set_segs:
795	movw	$KUDSEL,%ax
796	movw	%ax,TF_DS(%rsp)
797	movw	%ax,TF_ES(%rsp)
798	movw	$KUF32SEL,TF_FS(%rsp)
799	movw	$KUG32SEL,TF_GS(%rsp)
800	jmp	do_segs
801
802	/*
803	 * doreti_iret_fault.  Alternative return code for
804	 * the case where we get a fault in the doreti_exit code
805	 * above.  trap() (amd64/amd64/trap.c) catches this specific
806	 * case, sends the process a signal and continues in the
807	 * corresponding place in the code below.
808	 */
809	ALIGN_TEXT
810	.globl	doreti_iret_fault
811doreti_iret_fault:
812	subq	$TF_RIP,%rsp		/* space including tf_err, tf_trapno */
813	testl	$PSL_I,TF_RFLAGS(%rsp)
814	jz	1f
815	sti
8161:
817	movw	%fs,TF_FS(%rsp)
818	movw	%gs,TF_GS(%rsp)
819	movw	%es,TF_ES(%rsp)
820	movw	%ds,TF_DS(%rsp)
821	movl	$TF_HASSEGS,TF_FLAGS(%rsp)
822	movq	%rdi,TF_RDI(%rsp)
823	movq	%rsi,TF_RSI(%rsp)
824	movq	%rdx,TF_RDX(%rsp)
825	movq	%rcx,TF_RCX(%rsp)
826	movq	%r8,TF_R8(%rsp)
827	movq	%r9,TF_R9(%rsp)
828	movq	%rax,TF_RAX(%rsp)
829	movq	%rbx,TF_RBX(%rsp)
830	movq	%rbp,TF_RBP(%rsp)
831	movq	%r10,TF_R10(%rsp)
832	movq	%r11,TF_R11(%rsp)
833	movq	%r12,TF_R12(%rsp)
834	movq	%r13,TF_R13(%rsp)
835	movq	%r14,TF_R14(%rsp)
836	movq	%r15,TF_R15(%rsp)
837	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
838	movq	$0,TF_ERR(%rsp)	/* XXX should be the error code */
839	movq	$0,TF_ADDR(%rsp)
840	FAKE_MCOUNT(TF_RIP(%rsp))
841	jmp	calltrap
842
843	ALIGN_TEXT
844	.globl	ds_load_fault
845ds_load_fault:
846	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
847	testl	$PSL_I,TF_RFLAGS(%rsp)
848	jz	1f
849	sti
8501:
851	movq	%rsp,%rdi
852	call	trap
853	movw	$KUDSEL,TF_DS(%rsp)
854	jmp	doreti
855
856	ALIGN_TEXT
857	.globl	es_load_fault
858es_load_fault:
859	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
860	testl	$PSL_I,TF_RFLAGS(%rsp)
861	jz	1f
862	sti
8631:
864	movq	%rsp,%rdi
865	call	trap
866	movw	$KUDSEL,TF_ES(%rsp)
867	jmp	doreti
868
869	ALIGN_TEXT
870	.globl	fs_load_fault
871fs_load_fault:
872	testl	$PSL_I,TF_RFLAGS(%rsp)
873	jz	1f
874	sti
8751:
876	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
877	movq	%rsp,%rdi
878	call	trap
879	movw	$KUF32SEL,TF_FS(%rsp)
880	jmp	doreti
881
882	ALIGN_TEXT
883	.globl	gs_load_fault
884gs_load_fault:
885	popfq
886	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
887	testl	$PSL_I,TF_RFLAGS(%rsp)
888	jz	1f
889	sti
8901:
891	movq	%rsp,%rdi
892	call	trap
893	movw	$KUG32SEL,TF_GS(%rsp)
894	jmp	doreti
895
896	ALIGN_TEXT
897	.globl	fsbase_load_fault
898fsbase_load_fault:
899	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
900	testl	$PSL_I,TF_RFLAGS(%rsp)
901	jz	1f
902	sti
9031:
904	movq	%rsp,%rdi
905	call	trap
906	movq	PCPU(CURTHREAD),%r8
907	movq	TD_PCB(%r8),%r8
908	movq	$0,PCB_FSBASE(%r8)
909	jmp	doreti
910
911	ALIGN_TEXT
912	.globl	gsbase_load_fault
913gsbase_load_fault:
914	movl	$T_PROTFLT,TF_TRAPNO(%rsp)
915	testl	$PSL_I,TF_RFLAGS(%rsp)
916	jz	1f
917	sti
9181:
919	movq	%rsp,%rdi
920	call	trap
921	movq	PCPU(CURTHREAD),%r8
922	movq	TD_PCB(%r8),%r8
923	movq	$0,PCB_GSBASE(%r8)
924	jmp	doreti
925
926#ifdef HWPMC_HOOKS
927	ENTRY(end_exceptions)
928#endif
929