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
55 dtrace_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
61 dtrace_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
97 MCOUNT_LABEL(user)
98 MCOUNT_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
107 IDTVEC(dbg)
108 TRAP_NOEN(T_TRCTRAP)
109 IDTVEC(bpt)
110 TRAP_NOEN(T_BPTFLT)
111 #ifdef KDTRACE_HOOKS
112 IDTVEC(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
123 IDTVEC(div)
124 TRAP(T_DIVIDE)
125 IDTVEC(ofl)
126 TRAP(T_OFLOW)
127 IDTVEC(bnd)
128 TRAP(T_BOUND)
129 IDTVEC(ill)
130 TRAP(T_PRIVINFLT)
131 IDTVEC(dna)
132 TRAP(T_DNA)
133 IDTVEC(fpusegm)
134 TRAP(T_FPOPFLT)
135 IDTVEC(mchk)
136 TRAP(T_MCHK)
137 IDTVEC(rsvd)
138 TRAP(T_RESERVED)
139 IDTVEC(fpu)
140 TRAP(T_ARITHTRAP)
141 IDTVEC(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
150 IDTVEC(tss)
151 TRAP_ERR(T_TSSFLT)
152 IDTVEC(missing)
153 subq $TF_ERR,%rsp
154 movl $T_SEGNPFLT,TF_TRAPNO(%rsp)
155 jmp prot_addrf
156 IDTVEC(stk)
157 subq $TF_ERR,%rsp
158 movl $T_STKFLT,TF_TRAPNO(%rsp)
159 jmp prot_addrf
160 IDTVEC(align)
161 TRAP_ERR(T_ALIGNFLT)
162
163 /*
164 * alltraps entry point. Use swapgs if this is the first time in the
165 * kernel from userland. Reenable interrupts if they were enabled
166 * before the trap. This approximates SDT_SYS386TGT on the i386 port.
167 */
168 SUPERALIGN_TEXT
169 .globl alltraps
170 .type alltraps,@function
171 alltraps:
172 movq %rdi,TF_RDI(%rsp)
173 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
174 jz alltraps_testi /* already running with kernel GS.base */
175 swapgs
176 movq PCPU(CURPCB),%rdi
177 andl $~PCB_FULL_IRET,PCB_FLAGS(%rdi)
178 movw %fs,TF_FS(%rsp)
179 movw %gs,TF_GS(%rsp)
180 movw %es,TF_ES(%rsp)
181 movw %ds,TF_DS(%rsp)
182 alltraps_testi:
183 testl $PSL_I,TF_RFLAGS(%rsp)
184 jz alltraps_pushregs_no_rdi
185 sti
186 alltraps_pushregs_no_rdi:
187 movq %rsi,TF_RSI(%rsp)
188 movq %rdx,TF_RDX(%rsp)
189 movq %rcx,TF_RCX(%rsp)
190 movq %r8,TF_R8(%rsp)
191 movq %r9,TF_R9(%rsp)
192 movq %rax,TF_RAX(%rsp)
193 movq %rbx,TF_RBX(%rsp)
194 movq %rbp,TF_RBP(%rsp)
195 movq %r10,TF_R10(%rsp)
196 movq %r11,TF_R11(%rsp)
197 movq %r12,TF_R12(%rsp)
198 movq %r13,TF_R13(%rsp)
199 movq %r14,TF_R14(%rsp)
200 movq %r15,TF_R15(%rsp)
201 movl $TF_HASSEGS,TF_FLAGS(%rsp)
202 cld
203 FAKE_MCOUNT(TF_RIP(%rsp))
204 #ifdef KDTRACE_HOOKS
205 /*
206 * DTrace Function Boundary Trace (fbt) probes are triggered
207 * by int3 (0xcc) which causes the #BP (T_BPTFLT) breakpoint
208 * interrupt. For all other trap types, just handle them in
209 * the usual way.
210 */
211 cmpl $T_BPTFLT,TF_TRAPNO(%rsp)
212 jne calltrap
213
214 /* Check if there is no DTrace hook registered. */
215 cmpq $0,dtrace_invop_jump_addr
216 je calltrap
217
218 /*
219 * Set our jump address for the jump back in the event that
220 * the breakpoint wasn't caused by DTrace at all.
221 */
222 movq $calltrap,dtrace_invop_calltrap_addr(%rip)
223
224 /* Jump to the code hooked in by DTrace. */
225 movq dtrace_invop_jump_addr,%rax
226 jmpq *dtrace_invop_jump_addr
227 #endif
228 .globl calltrap
229 .type calltrap,@function
230 calltrap:
231 movq %rsp,%rdi
232 call trap
233 MEXITCOUNT
234 jmp doreti /* Handle any pending ASTs */
235
236 /*
237 * alltraps_noen entry point. Unlike alltraps above, we want to
238 * leave the interrupts disabled. This corresponds to
239 * SDT_SYS386IGT on the i386 port.
240 */
241 SUPERALIGN_TEXT
242 .globl alltraps_noen
243 .type alltraps_noen,@function
244 alltraps_noen:
245 movq %rdi,TF_RDI(%rsp)
246 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
247 jz 1f /* already running with kernel GS.base */
248 swapgs
249 movq PCPU(CURPCB),%rdi
250 andl $~PCB_FULL_IRET,PCB_FLAGS(%rdi)
251 1: movw %fs,TF_FS(%rsp)
252 movw %gs,TF_GS(%rsp)
253 movw %es,TF_ES(%rsp)
254 movw %ds,TF_DS(%rsp)
255 jmp alltraps_pushregs_no_rdi
256
257 IDTVEC(dblfault)
258 subq $TF_ERR,%rsp
259 movl $T_DOUBLEFLT,TF_TRAPNO(%rsp)
260 movq $0,TF_ADDR(%rsp)
261 movq $0,TF_ERR(%rsp)
262 movq %rdi,TF_RDI(%rsp)
263 movq %rsi,TF_RSI(%rsp)
264 movq %rdx,TF_RDX(%rsp)
265 movq %rcx,TF_RCX(%rsp)
266 movq %r8,TF_R8(%rsp)
267 movq %r9,TF_R9(%rsp)
268 movq %rax,TF_RAX(%rsp)
269 movq %rbx,TF_RBX(%rsp)
270 movq %rbp,TF_RBP(%rsp)
271 movq %r10,TF_R10(%rsp)
272 movq %r11,TF_R11(%rsp)
273 movq %r12,TF_R12(%rsp)
274 movq %r13,TF_R13(%rsp)
275 movq %r14,TF_R14(%rsp)
276 movq %r15,TF_R15(%rsp)
277 movw %fs,TF_FS(%rsp)
278 movw %gs,TF_GS(%rsp)
279 movw %es,TF_ES(%rsp)
280 movw %ds,TF_DS(%rsp)
281 movl $TF_HASSEGS,TF_FLAGS(%rsp)
282 cld
283 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
284 jz 1f /* already running with kernel GS.base */
285 swapgs
286 1:
287 movq %rsp,%rdi
288 call dblfault_handler
289 2:
290 hlt
291 jmp 2b
292
293 IDTVEC(page)
294 subq $TF_ERR,%rsp
295 movl $T_PAGEFLT,TF_TRAPNO(%rsp)
296 movq %rdi,TF_RDI(%rsp) /* free up a GP register */
297 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
298 jz 1f /* already running with kernel GS.base */
299 swapgs
300 movq PCPU(CURPCB),%rdi
301 andl $~PCB_FULL_IRET,PCB_FLAGS(%rdi)
302 1: movq %cr2,%rdi /* preserve %cr2 before .. */
303 movq %rdi,TF_ADDR(%rsp) /* enabling interrupts. */
304 movw %fs,TF_FS(%rsp)
305 movw %gs,TF_GS(%rsp)
306 movw %es,TF_ES(%rsp)
307 movw %ds,TF_DS(%rsp)
308 testl $PSL_I,TF_RFLAGS(%rsp)
309 jz alltraps_pushregs_no_rdi
310 sti
311 jmp alltraps_pushregs_no_rdi
312
313 /*
314 * We have to special-case this one. If we get a trap in doreti() at
315 * the iretq stage, we'll reenter with the wrong gs state. We'll have
316 * to do a special the swapgs in this case even coming from the kernel.
317 * XXX linux has a trap handler for their equivalent of load_gs().
318 */
319 IDTVEC(prot)
320 subq $TF_ERR,%rsp
321 movl $T_PROTFLT,TF_TRAPNO(%rsp)
322 prot_addrf:
323 movq $0,TF_ADDR(%rsp)
324 movq %rdi,TF_RDI(%rsp) /* free up a GP register */
325 leaq doreti_iret(%rip),%rdi
326 cmpq %rdi,TF_RIP(%rsp)
327 je 1f /* kernel but with user gsbase!! */
328 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
329 jz 2f /* already running with kernel GS.base */
330 1: swapgs
331 2: movq PCPU(CURPCB),%rdi
332 orl $PCB_FULL_IRET,PCB_FLAGS(%rdi) /* always full iret from GPF */
333 movw %fs,TF_FS(%rsp)
334 movw %gs,TF_GS(%rsp)
335 movw %es,TF_ES(%rsp)
336 movw %ds,TF_DS(%rsp)
337 testl $PSL_I,TF_RFLAGS(%rsp)
338 jz alltraps_pushregs_no_rdi
339 sti
340 jmp alltraps_pushregs_no_rdi
341
342 /*
343 * Fast syscall entry point. We enter here with just our new %cs/%ss set,
344 * and the new privilige level. We are still running on the old user stack
345 * pointer. We have to juggle a few things around to find our stack etc.
346 * swapgs gives us access to our PCPU space only.
347 *
348 * We do not support invoking this from a custom %cs or %ss (e.g. using
349 * entries from an LDT).
350 */
351 IDTVEC(fast_syscall)
352 swapgs
353 movq %rsp,PCPU(SCRATCH_RSP)
354 movq PCPU(RSP0),%rsp
355 /* Now emulate a trapframe. Make the 8 byte alignment odd for call. */
356 subq $TF_SIZE,%rsp
357 /* defer TF_RSP till we have a spare register */
358 movq %r11,TF_RFLAGS(%rsp)
359 movq %rcx,TF_RIP(%rsp) /* %rcx original value is in %r10 */
360 movq PCPU(SCRATCH_RSP),%r11 /* %r11 already saved */
361 movq %r11,TF_RSP(%rsp) /* user stack pointer */
362 movw %fs,TF_FS(%rsp)
363 movw %gs,TF_GS(%rsp)
364 movw %es,TF_ES(%rsp)
365 movw %ds,TF_DS(%rsp)
366 movq PCPU(CURPCB),%r11
367 andl $~PCB_FULL_IRET,PCB_FLAGS(%r11)
368 sti
369 movq $KUDSEL,TF_SS(%rsp)
370 movq $KUCSEL,TF_CS(%rsp)
371 movq $2,TF_ERR(%rsp)
372 movq %rdi,TF_RDI(%rsp) /* arg 1 */
373 movq %rsi,TF_RSI(%rsp) /* arg 2 */
374 movq %rdx,TF_RDX(%rsp) /* arg 3 */
375 movq %r10,TF_RCX(%rsp) /* arg 4 */
376 movq %r8,TF_R8(%rsp) /* arg 5 */
377 movq %r9,TF_R9(%rsp) /* arg 6 */
378 movq %rax,TF_RAX(%rsp) /* syscall number */
379 movq %rbx,TF_RBX(%rsp) /* C preserved */
380 movq %rbp,TF_RBP(%rsp) /* C preserved */
381 movq %r12,TF_R12(%rsp) /* C preserved */
382 movq %r13,TF_R13(%rsp) /* C preserved */
383 movq %r14,TF_R14(%rsp) /* C preserved */
384 movq %r15,TF_R15(%rsp) /* C preserved */
385 movl $TF_HASSEGS,TF_FLAGS(%rsp)
386 cld
387 FAKE_MCOUNT(TF_RIP(%rsp))
388 movq PCPU(CURTHREAD),%rdi
389 movq %rsp,TD_FRAME(%rdi)
390 movl TF_RFLAGS(%rsp),%esi
391 andl $PSL_T,%esi
392 call amd64_syscall
393 1: movq PCPU(CURPCB),%rax
394 /* Disable interrupts before testing PCB_FULL_IRET. */
395 cli
396 testl $PCB_FULL_IRET,PCB_FLAGS(%rax)
397 jnz 3f
398 /* Check for and handle AST's on return to userland. */
399 movq PCPU(CURTHREAD),%rax
400 testl $TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax)
401 jne 2f
402 /* Restore preserved registers. */
403 MEXITCOUNT
404 movq TF_RDI(%rsp),%rdi /* bonus; preserve arg 1 */
405 movq TF_RSI(%rsp),%rsi /* bonus: preserve arg 2 */
406 movq TF_RDX(%rsp),%rdx /* return value 2 */
407 movq TF_RAX(%rsp),%rax /* return value 1 */
408 movq TF_RFLAGS(%rsp),%r11 /* original %rflags */
409 movq TF_RIP(%rsp),%rcx /* original %rip */
410 movq TF_RSP(%rsp),%rsp /* user stack pointer */
411 swapgs
412 sysretq
413
414 2: /* AST scheduled. */
415 sti
416 movq %rsp,%rdi
417 call ast
418 jmp 1b
419
420 3: /* Requested full context restore, use doreti for that. */
421 MEXITCOUNT
422 jmp doreti
423
424 /*
425 * Here for CYA insurance, in case a "syscall" instruction gets
426 * issued from 32 bit compatability mode. MSR_CSTAR has to point
427 * to *something* if EFER_SCE is enabled.
428 */
429 IDTVEC(fast_syscall32)
430 sysret
431
432 /*
433 * NMI handling is special.
434 *
435 * First, NMIs do not respect the state of the processor's RFLAGS.IF
436 * bit. The NMI handler may be entered at any time, including when
437 * the processor is in a critical section with RFLAGS.IF == 0.
438 * The processor's GS.base value could be invalid on entry to the
439 * handler.
440 *
441 * Second, the processor treats NMIs specially, blocking further NMIs
442 * until an 'iretq' instruction is executed. We thus need to execute
443 * the NMI handler with interrupts disabled, to prevent a nested interrupt
444 * from executing an 'iretq' instruction and inadvertently taking the
445 * processor out of NMI mode.
446 *
447 * Third, the NMI handler runs on its own stack (tss_ist2). The canonical
448 * GS.base value for the processor is stored just above the bottom of its
449 * NMI stack. For NMIs taken from kernel mode, the current value in
450 * the processor's GS.base is saved at entry to C-preserved register %r12,
451 * the canonical value for GS.base is then loaded into the processor, and
452 * the saved value is restored at exit time. For NMIs taken from user mode,
453 * the cheaper 'SWAPGS' instructions are used for swapping GS.base.
454 */
455
456 IDTVEC(nmi)
457 subq $TF_RIP,%rsp
458 movl $(T_NMI),TF_TRAPNO(%rsp)
459 movq $0,TF_ADDR(%rsp)
460 movq $0,TF_ERR(%rsp)
461 movq %rdi,TF_RDI(%rsp)
462 movq %rsi,TF_RSI(%rsp)
463 movq %rdx,TF_RDX(%rsp)
464 movq %rcx,TF_RCX(%rsp)
465 movq %r8,TF_R8(%rsp)
466 movq %r9,TF_R9(%rsp)
467 movq %rax,TF_RAX(%rsp)
468 movq %rbx,TF_RBX(%rsp)
469 movq %rbp,TF_RBP(%rsp)
470 movq %r10,TF_R10(%rsp)
471 movq %r11,TF_R11(%rsp)
472 movq %r12,TF_R12(%rsp)
473 movq %r13,TF_R13(%rsp)
474 movq %r14,TF_R14(%rsp)
475 movq %r15,TF_R15(%rsp)
476 movw %fs,TF_FS(%rsp)
477 movw %gs,TF_GS(%rsp)
478 movw %es,TF_ES(%rsp)
479 movw %ds,TF_DS(%rsp)
480 movl $TF_HASSEGS,TF_FLAGS(%rsp)
481 cld
482 xorl %ebx,%ebx
483 testb $SEL_RPL_MASK,TF_CS(%rsp)
484 jnz nmi_fromuserspace
485 /*
486 * We've interrupted the kernel. Preserve GS.base in %r12.
487 */
488 movl $MSR_GSBASE,%ecx
489 rdmsr
490 movq %rax,%r12
491 shlq $32,%rdx
492 orq %rdx,%r12
493 /* Retrieve and load the canonical value for GS.base. */
494 movq TF_SIZE(%rsp),%rdx
495 movl %edx,%eax
496 shrq $32,%rdx
497 wrmsr
498 jmp nmi_calltrap
499 nmi_fromuserspace:
500 incl %ebx
501 swapgs
502 /* Note: this label is also used by ddb and gdb: */
503 nmi_calltrap:
504 FAKE_MCOUNT(TF_RIP(%rsp))
505 movq %rsp,%rdi
506 call trap
507 MEXITCOUNT
508 #ifdef HWPMC_HOOKS
509 /*
510 * Capture a userspace callchain if needed.
511 *
512 * - Check if the current trap was from user mode.
513 * - Check if the current thread is valid.
514 * - Check if the thread requires a user call chain to be
515 * captured.
516 *
517 * We are still in NMI mode at this point.
518 */
519 testl %ebx,%ebx
520 jz nocallchain /* not from userspace */
521 movq PCPU(CURTHREAD),%rax
522 orq %rax,%rax /* curthread present? */
523 jz nocallchain
524 testl $TDP_CALLCHAIN,TD_PFLAGS(%rax) /* flagged for capture? */
525 jz nocallchain
526 /*
527 * A user callchain is to be captured, so:
528 * - Move execution to the regular kernel stack, to allow for
529 * nested NMI interrupts.
530 * - Take the processor out of "NMI" mode by faking an "iret".
531 * - Enable interrupts, so that copyin() can work.
532 */
533 movq %rsp,%rsi /* source stack pointer */
534 movq $TF_SIZE,%rcx
535 movq PCPU(RSP0),%rdx
536 subq %rcx,%rdx
537 movq %rdx,%rdi /* destination stack pointer */
538
539 shrq $3,%rcx /* trap frame size in long words */
540 cld
541 rep
542 movsq /* copy trapframe */
543
544 movl %ss,%eax
545 pushq %rax /* tf_ss */
546 pushq %rdx /* tf_rsp (on kernel stack) */
547 pushfq /* tf_rflags */
548 movl %cs,%eax
549 pushq %rax /* tf_cs */
550 pushq $outofnmi /* tf_rip */
551 iretq
552 outofnmi:
553 /*
554 * At this point the processor has exited NMI mode and is running
555 * with interrupts turned off on the normal kernel stack.
556 *
557 * If a pending NMI gets recognized at or after this point, it
558 * will cause a kernel callchain to be traced.
559 *
560 * We turn interrupts back on, and call the user callchain capture hook.
561 */
562 movq pmc_hook,%rax
563 orq %rax,%rax
564 jz nocallchain
565 movq PCPU(CURTHREAD),%rdi /* thread */
566 movq $PMC_FN_USER_CALLCHAIN,%rsi /* command */
567 movq %rsp,%rdx /* frame */
568 sti
569 call *%rax
570 cli
571 nocallchain:
572 #endif
573 testl %ebx,%ebx
574 jnz doreti_exit
575 nmi_kernelexit:
576 /*
577 * Put back the preserved MSR_GSBASE value.
578 */
579 movl $MSR_GSBASE,%ecx
580 movq %r12,%rdx
581 movl %edx,%eax
582 shrq $32,%rdx
583 wrmsr
584 nmi_restoreregs:
585 movq TF_RDI(%rsp),%rdi
586 movq TF_RSI(%rsp),%rsi
587 movq TF_RDX(%rsp),%rdx
588 movq TF_RCX(%rsp),%rcx
589 movq TF_R8(%rsp),%r8
590 movq TF_R9(%rsp),%r9
591 movq TF_RAX(%rsp),%rax
592 movq TF_RBX(%rsp),%rbx
593 movq TF_RBP(%rsp),%rbp
594 movq TF_R10(%rsp),%r10
595 movq TF_R11(%rsp),%r11
596 movq TF_R12(%rsp),%r12
597 movq TF_R13(%rsp),%r13
598 movq TF_R14(%rsp),%r14
599 movq TF_R15(%rsp),%r15
600 addq $TF_RIP,%rsp
601 jmp doreti_iret
602
603 ENTRY(fork_trampoline)
604 movq %r12,%rdi /* function */
605 movq %rbx,%rsi /* arg1 */
606 movq %rsp,%rdx /* trapframe pointer */
607 call fork_exit
608 MEXITCOUNT
609 jmp doreti /* Handle any ASTs */
610
611 /*
612 * To efficiently implement classification of trap and interrupt handlers
613 * for profiling, there must be only trap handlers between the labels btrap
614 * and bintr, and only interrupt handlers between the labels bintr and
615 * eintr. This is implemented (partly) by including files that contain
616 * some of the handlers. Before including the files, set up a normal asm
617 * environment so that the included files doen't need to know that they are
618 * included.
619 */
620
621 #ifdef COMPAT_FREEBSD32
622 .data
623 .p2align 4
624 .text
625 SUPERALIGN_TEXT
626
627 #include <amd64/ia32/ia32_exception.S>
628 #endif
629
630 .data
631 .p2align 4
632 .text
633 SUPERALIGN_TEXT
634 MCOUNT_LABEL(bintr)
635
636 #include <amd64/amd64/apic_vector.S>
637
638 #ifdef DEV_ATPIC
639 .data
640 .p2align 4
641 .text
642 SUPERALIGN_TEXT
643
644 #include <amd64/amd64/atpic_vector.S>
645 #endif
646
647 .text
648 MCOUNT_LABEL(eintr)
649
650 /*
651 * void doreti(struct trapframe)
652 *
653 * Handle return from interrupts, traps and syscalls.
654 */
655 .text
656 SUPERALIGN_TEXT
657 .type doreti,@function
658 doreti:
659 FAKE_MCOUNT($bintr) /* init "from" bintr -> doreti */
660 /*
661 * Check if ASTs can be handled now.
662 */
663 testb $SEL_RPL_MASK,TF_CS(%rsp) /* are we returning to user mode? */
664 jz doreti_exit /* can't handle ASTs now if not */
665
666 doreti_ast:
667 /*
668 * Check for ASTs atomically with returning. Disabling CPU
669 * interrupts provides sufficient locking even in the SMP case,
670 * since we will be informed of any new ASTs by an IPI.
671 */
672 cli
673 movq PCPU(CURTHREAD),%rax
674 testl $TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax)
675 je doreti_exit
676 sti
677 movq %rsp,%rdi /* pass a pointer to the trapframe */
678 call ast
679 jmp doreti_ast
680
681 /*
682 * doreti_exit: pop registers, iret.
683 *
684 * The segment register pop is a special case, since it may
685 * fault if (for example) a sigreturn specifies bad segment
686 * registers. The fault is handled in trap.c.
687 */
688 doreti_exit:
689 MEXITCOUNT
690 movq PCPU(CURPCB),%r8
691
692 /*
693 * Do not reload segment registers for kernel.
694 * Since we do not reload segments registers with sane
695 * values on kernel entry, descriptors referenced by
696 * segments registers might be not valid. This is fatal
697 * for user mode, but is not a problem for the kernel.
698 */
699 testb $SEL_RPL_MASK,TF_CS(%rsp)
700 jz ld_regs
701 testl $PCB_FULL_IRET,PCB_FLAGS(%r8)
702 jz ld_regs
703 testl $TF_HASSEGS,TF_FLAGS(%rsp)
704 je set_segs
705
706 do_segs:
707 /* Restore %fs and fsbase */
708 movw TF_FS(%rsp),%ax
709 .globl ld_fs
710 ld_fs:
711 movw %ax,%fs
712 cmpw $KUF32SEL,%ax
713 jne 1f
714 movl $MSR_FSBASE,%ecx
715 movl PCB_FSBASE(%r8),%eax
716 movl PCB_FSBASE+4(%r8),%edx
717 .globl ld_fsbase
718 ld_fsbase:
719 wrmsr
720 1:
721 /* Restore %gs and gsbase */
722 movw TF_GS(%rsp),%si
723 pushfq
724 cli
725 movl $MSR_GSBASE,%ecx
726 /* Save current kernel %gs base into %r12d:%r13d */
727 rdmsr
728 movl %eax,%r12d
729 movl %edx,%r13d
730 .globl ld_gs
731 ld_gs:
732 movw %si,%gs
733 /* Save user %gs base into %r14d:%r15d */
734 rdmsr
735 movl %eax,%r14d
736 movl %edx,%r15d
737 /* Restore kernel %gs base */
738 movl %r12d,%eax
739 movl %r13d,%edx
740 wrmsr
741 popfq
742 /*
743 * Restore user %gs base, either from PCB if used for TLS, or
744 * from the previously saved msr read.
745 */
746 movl $MSR_KGSBASE,%ecx
747 cmpw $KUG32SEL,%si
748 jne 1f
749 movl PCB_GSBASE(%r8),%eax
750 movl PCB_GSBASE+4(%r8),%edx
751 jmp ld_gsbase
752 1:
753 movl %r14d,%eax
754 movl %r15d,%edx
755 .globl ld_gsbase
756 ld_gsbase:
757 wrmsr /* May trap if non-canonical, but only for TLS. */
758 .globl ld_es
759 ld_es:
760 movw TF_ES(%rsp),%es
761 .globl ld_ds
762 ld_ds:
763 movw TF_DS(%rsp),%ds
764 ld_regs:
765 movq TF_RDI(%rsp),%rdi
766 movq TF_RSI(%rsp),%rsi
767 movq TF_RDX(%rsp),%rdx
768 movq TF_RCX(%rsp),%rcx
769 movq TF_R8(%rsp),%r8
770 movq TF_R9(%rsp),%r9
771 movq TF_RAX(%rsp),%rax
772 movq TF_RBX(%rsp),%rbx
773 movq TF_RBP(%rsp),%rbp
774 movq TF_R10(%rsp),%r10
775 movq TF_R11(%rsp),%r11
776 movq TF_R12(%rsp),%r12
777 movq TF_R13(%rsp),%r13
778 movq TF_R14(%rsp),%r14
779 movq TF_R15(%rsp),%r15
780 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
781 jz 1f /* keep running with kernel GS.base */
782 cli
783 swapgs
784 1:
785 addq $TF_RIP,%rsp /* skip over tf_err, tf_trapno */
786 .globl doreti_iret
787 doreti_iret:
788 iretq
789
790 set_segs:
791 movw $KUDSEL,%ax
792 movw %ax,TF_DS(%rsp)
793 movw %ax,TF_ES(%rsp)
794 movw $KUF32SEL,TF_FS(%rsp)
795 movw $KUG32SEL,TF_GS(%rsp)
796 jmp do_segs
797
798 /*
799 * doreti_iret_fault. Alternative return code for
800 * the case where we get a fault in the doreti_exit code
801 * above. trap() (amd64/amd64/trap.c) catches this specific
802 * case, sends the process a signal and continues in the
803 * corresponding place in the code below.
804 */
805 ALIGN_TEXT
806 .globl doreti_iret_fault
807 doreti_iret_fault:
808 subq $TF_RIP,%rsp /* space including tf_err, tf_trapno */
809 testl $PSL_I,TF_RFLAGS(%rsp)
810 jz 1f
811 sti
812 1:
813 movw %fs,TF_FS(%rsp)
814 movw %gs,TF_GS(%rsp)
815 movw %es,TF_ES(%rsp)
816 movw %ds,TF_DS(%rsp)
817 movl $TF_HASSEGS,TF_FLAGS(%rsp)
818 movq %rdi,TF_RDI(%rsp)
819 movq %rsi,TF_RSI(%rsp)
820 movq %rdx,TF_RDX(%rsp)
821 movq %rcx,TF_RCX(%rsp)
822 movq %r8,TF_R8(%rsp)
823 movq %r9,TF_R9(%rsp)
824 movq %rax,TF_RAX(%rsp)
825 movq %rbx,TF_RBX(%rsp)
826 movq %rbp,TF_RBP(%rsp)
827 movq %r10,TF_R10(%rsp)
828 movq %r11,TF_R11(%rsp)
829 movq %r12,TF_R12(%rsp)
830 movq %r13,TF_R13(%rsp)
831 movq %r14,TF_R14(%rsp)
832 movq %r15,TF_R15(%rsp)
833 movl $T_PROTFLT,TF_TRAPNO(%rsp)
834 movq $0,TF_ERR(%rsp) /* XXX should be the error code */
835 movq $0,TF_ADDR(%rsp)
836 FAKE_MCOUNT(TF_RIP(%rsp))
837 jmp calltrap
838
839 ALIGN_TEXT
840 .globl ds_load_fault
841 ds_load_fault:
842 movl $T_PROTFLT,TF_TRAPNO(%rsp)
843 testl $PSL_I,TF_RFLAGS(%rsp)
844 jz 1f
845 sti
846 1:
847 movq %rsp,%rdi
848 call trap
849 movw $KUDSEL,TF_DS(%rsp)
850 jmp doreti
851
852 ALIGN_TEXT
853 .globl es_load_fault
854 es_load_fault:
855 movl $T_PROTFLT,TF_TRAPNO(%rsp)
856 testl $PSL_I,TF_RFLAGS(%rsp)
857 jz 1f
858 sti
859 1:
860 movq %rsp,%rdi
861 call trap
862 movw $KUDSEL,TF_ES(%rsp)
863 jmp doreti
864
865 ALIGN_TEXT
866 .globl fs_load_fault
867 fs_load_fault:
868 testl $PSL_I,TF_RFLAGS(%rsp)
869 jz 1f
870 sti
871 1:
872 movl $T_PROTFLT,TF_TRAPNO(%rsp)
873 movq %rsp,%rdi
874 call trap
875 movw $KUF32SEL,TF_FS(%rsp)
876 jmp doreti
877
878 ALIGN_TEXT
879 .globl gs_load_fault
880 gs_load_fault:
881 popfq
882 movl $T_PROTFLT,TF_TRAPNO(%rsp)
883 testl $PSL_I,TF_RFLAGS(%rsp)
884 jz 1f
885 sti
886 1:
887 movq %rsp,%rdi
888 call trap
889 movw $KUG32SEL,TF_GS(%rsp)
890 jmp doreti
891
892 ALIGN_TEXT
893 .globl fsbase_load_fault
894 fsbase_load_fault:
895 movl $T_PROTFLT,TF_TRAPNO(%rsp)
896 testl $PSL_I,TF_RFLAGS(%rsp)
897 jz 1f
898 sti
899 1:
900 movq %rsp,%rdi
901 call trap
902 movq PCPU(CURTHREAD),%r8
903 movq TD_PCB(%r8),%r8
904 movq $0,PCB_FSBASE(%r8)
905 jmp doreti
906
907 ALIGN_TEXT
908 .globl gsbase_load_fault
909 gsbase_load_fault:
910 movl $T_PROTFLT,TF_TRAPNO(%rsp)
911 testl $PSL_I,TF_RFLAGS(%rsp)
912 jz 1f
913 sti
914 1:
915 movq %rsp,%rdi
916 call trap
917 movq PCPU(CURTHREAD),%r8
918 movq TD_PCB(%r8),%r8
919 movq $0,PCB_GSBASE(%r8)
920 jmp doreti
921
922 #ifdef HWPMC_HOOKS
923 ENTRY(end_exceptions)
924 #endif
Cache object: 3e3d9848530253d81a0cadd8ca70b483
|