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 * Copyright (c) 2008 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
9 * Portions of this software were developed by A. Joseph Koshy under
10 * sponsorship from the FreeBSD Foundation and Google, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * $DragonFly: src/sys/platform/pc64/amd64/exception.S,v 1.1 2008/08/29 17:07:10 dillon Exp $
41 #include "opt_atpic.h"
43 #include "opt_compat.h"
45 #include <machine/asmacros.h>
46 #include <machine/psl.h>
47 #include <machine/trap.h>
53 /*****************************************************************************/
55 /*****************************************************************************/
57 * Trap and fault vector routines.
59 * All traps are 'interrupt gates', SDT_SYSIGT. An interrupt gate pushes
60 * state on the stack but also disables interrupts. This is important for
61 * us for the use of the swapgs instruction. We cannot be interrupted
62 * until the GS.base value is correct. For most traps, we automatically
63 * then enable interrupts if the interrupted context had them enabled.
64 * This is equivalent to the i386 port's use of SDT_SYS386TGT.
66 * The cpu will push a certain amount of state onto the kernel stack for
67 * the current process. See amd64/include/frame.h.
68 * This includes the current RFLAGS (status register, which includes
69 * the interrupt disable state prior to the trap), the code segment register,
70 * and the return instruction pointer are pushed by the cpu. The cpu
71 * will also push an 'error' code for certain traps. We push a dummy
72 * error code for those traps where the cpu doesn't in order to maintain
73 * a consistent frame. We also push a contrived 'trap number'.
75 * The cpu does not push the general registers, we must do that, and we
76 * must restore them prior to calling 'iret'. The cpu adjusts the %cs and
77 * %ss segment registers, but does not mess with %ds, %es, or %fs. Thus we
78 * must load them with appropriate values for supervisor mode operation.
84 /* Traps that we leave interrupts disabled for.. */
85 #define TRAP_NOEN(a) \
87 movq $0,TF_XFLAGS(%rsp) ; \
88 movq $(a),TF_TRAPNO(%rsp) ; \
89 movq $0,TF_ADDR(%rsp) ; \
90 movq $0,TF_ERR(%rsp) ; \
97 /* Regular traps; The cpu does not supply tf_err for these. */
100 movq $0,TF_XFLAGS(%rsp) ; \
101 movq $(a),TF_TRAPNO(%rsp) ; \
102 movq $0,TF_ADDR(%rsp) ; \
103 movq $0,TF_ERR(%rsp) ; \
126 /* This group of traps have tf_err already pushed by the cpu */
127 #define TRAP_ERR(a) \
129 movq $(a),TF_TRAPNO(%rsp) ; \
130 movq $0,TF_ADDR(%rsp) ; \
142 * alltraps entry point. Use swapgs if this is the first time in the
143 * kernel from userland. Reenable interrupts if they were enabled
144 * before the trap. This approximates SDT_SYS386TGT on the i386 port.
149 .type alltraps,@function
151 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
152 jz alltraps_testi /* already running with kernel GS.base */
155 testq $PSL_I,TF_RFLAGS(%rsp)
159 movq %rdi,TF_RDI(%rsp)
160 alltraps_pushregs_no_rdi:
161 movq %rsi,TF_RSI(%rsp)
162 movq %rdx,TF_RDX(%rsp)
163 movq %rcx,TF_RCX(%rsp)
166 movq %rax,TF_RAX(%rsp)
167 movq %rbx,TF_RBX(%rsp)
168 movq %rbp,TF_RBP(%rsp)
169 movq %r10,TF_R10(%rsp)
170 movq %r11,TF_R11(%rsp)
171 movq %r12,TF_R12(%rsp)
172 movq %r13,TF_R13(%rsp)
173 movq %r14,TF_R14(%rsp)
174 movq %r15,TF_R15(%rsp)
175 FAKE_MCOUNT(TF_RIP(%rsp))
177 .type calltrap,@function
182 jmp doreti /* Handle any pending ASTs */
185 * alltraps_noen entry point. Unlike alltraps above, we want to
186 * leave the interrupts disabled. This corresponds to
187 * SDT_SYS386IGT on the i386 port.
191 .type alltraps_noen,@function
193 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
194 jz alltraps_pushregs /* already running with kernel GS.base */
196 jmp alltraps_pushregs
200 movq $T_DOUBLEFLT,TF_TRAPNO(%rsp)
201 movq $0,TF_ADDR(%rsp)
203 movq %rdi,TF_RDI(%rsp)
204 movq %rsi,TF_RSI(%rsp)
205 movq %rdx,TF_RDX(%rsp)
206 movq %rcx,TF_RCX(%rsp)
209 movq %rax,TF_RAX(%rsp)
210 movq %rbx,TF_RBX(%rsp)
211 movq %rbp,TF_RBP(%rsp)
212 movq %r10,TF_R10(%rsp)
213 movq %r11,TF_R11(%rsp)
214 movq %r12,TF_R12(%rsp)
215 movq %r13,TF_R13(%rsp)
216 movq %r14,TF_R14(%rsp)
217 movq %r15,TF_R15(%rsp)
218 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
219 jz 1f /* already running with kernel GS.base */
222 call dblfault_handler
228 movq $T_PAGEFLT,TF_TRAPNO(%rsp)
229 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
230 jz 1f /* already running with kernel GS.base */
233 movq %rdi,TF_RDI(%rsp) /* free up a GP register */
234 movq %cr2,%rdi /* preserve %cr2 before .. */
235 movq %rdi,TF_ADDR(%rsp) /* enabling interrupts. */
236 testq $PSL_I,TF_RFLAGS(%rsp)
237 jz alltraps_pushregs_no_rdi
239 jmp alltraps_pushregs_no_rdi
242 * We have to special-case this one. If we get a trap in doreti() at
243 * the iretq stage, we'll reenter with the wrong gs state. We'll have
244 * to do a special the swapgs in this case even coming from the kernel.
245 * XXX linux has a trap handler for their equivalent of load_gs().
249 movq $T_PROTFLT,TF_TRAPNO(%rsp)
250 movq $0,TF_ADDR(%rsp)
251 movq %rdi,TF_RDI(%rsp) /* free up a GP register */
252 leaq doreti_iret(%rip),%rdi
253 cmpq %rdi,TF_RIP(%rsp)
254 je 2f /* kernel but with user gsbase!! */
255 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */
256 jz 1f /* already running with kernel GS.base */
260 testq $PSL_I,TF_RFLAGS(%rsp)
261 jz alltraps_pushregs_no_rdi
263 jmp alltraps_pushregs_no_rdi
266 * Fast syscall entry point. We enter here with just our new %cs/%ss set,
267 * and the new privilige level. We are still running on the old user stack
268 * pointer. We have to juggle a few things around to find our stack etc.
269 * swapgs gives us access to our PCPU space only.
273 movq %rsp,PCPU(scratch_rsp)
275 /* Now emulate a trapframe. Make the 8 byte alignment odd for call. */
277 /* defer TF_RSP till we have a spare register */
278 movq %r11,TF_RFLAGS(%rsp)
279 movq %rcx,TF_RIP(%rsp) /* %rcx original value is in %r10 */
280 movq PCPU(scratch_rsp),%r11 /* %r11 already saved */
281 movq %r11,TF_RSP(%rsp) /* user stack pointer */
283 movq $KUDSEL,TF_SS(%rsp)
284 movq $KUCSEL,TF_CS(%rsp)
286 movq %rdi,TF_RDI(%rsp) /* arg 1 */
287 movq %rsi,TF_RSI(%rsp) /* arg 2 */
288 movq %rdx,TF_RDX(%rsp) /* arg 3 */
289 movq %r10,TF_RCX(%rsp) /* arg 4 */
290 movq %r8,TF_R8(%rsp) /* arg 5 */
291 movq %r9,TF_R9(%rsp) /* arg 6 */
292 movq %rax,TF_RAX(%rsp) /* syscall number */
293 movq %rbx,TF_RBX(%rsp) /* C preserved */
294 movq %rbp,TF_RBP(%rsp) /* C preserved */
295 movq %r12,TF_R12(%rsp) /* C preserved */
296 movq %r13,TF_R13(%rsp) /* C preserved */
297 movq %r14,TF_R14(%rsp) /* C preserved */
298 movq %r15,TF_R15(%rsp) /* C preserved */
299 FAKE_MCOUNT(TF_RIP(%rsp))
302 /* JGXXX handle AST's? */
303 /* restore preserved registers */
305 movq TF_RDI(%rsp),%rdi /* bonus; preserve arg 1 */
306 movq TF_RSI(%rsp),%rsi /* bonus: preserve arg 2 */
307 movq TF_RDX(%rsp),%rdx /* return value 2 */
308 movq TF_RAX(%rsp),%rax /* return value 1 */
309 movq TF_RBX(%rsp),%rbx /* C preserved */
310 movq TF_RBP(%rsp),%rbp /* C preserved */
311 movq TF_R12(%rsp),%r12 /* C preserved */
312 movq TF_R13(%rsp),%r13 /* C preserved */
313 movq TF_R14(%rsp),%r14 /* C preserved */
314 movq TF_R15(%rsp),%r15 /* C preserved */
315 movq TF_RFLAGS(%rsp),%r11 /* original %rflags */
316 movq TF_RIP(%rsp),%rcx /* original %rip */
317 movq TF_RSP(%rsp),%r9 /* user stack pointer */
318 movq %r9,%rsp /* original %rsp */
325 * Here for CYA insurance, in case a "syscall" instruction gets
326 * issued from 32 bit compatability mode. MSR_CSTAR has to point
327 * to *something* if EFER_SCE is enabled.
329 IDTVEC(fast_syscall32)
333 * NMI handling is special.
335 * First, NMIs do not respect the state of the processor's RFLAGS.IF
336 * bit and the NMI handler may be invoked at any time, including when
337 * the processor is in a critical section with RFLAGS.IF == 0. In
338 * particular, this means that the processor's GS.base values could be
339 * inconsistent on entry to the handler, and so we need to read
340 * MSR_GSBASE to determine if a 'swapgs' is needed. We use '%ebx', a
341 * C-preserved register, to remember whether to swap GS back on the
344 * Second, the processor treats NMIs specially, blocking further NMIs
345 * until an 'iretq' instruction is executed. We therefore need to
346 * execute the NMI handler with interrupts disabled to prevent a
347 * nested interrupt from executing an 'iretq' instruction and
348 * inadvertently taking the processor out of NMI mode.
350 * Third, the NMI handler runs on its own stack (tss_ist1), shared
351 * with the double fault handler.
356 movq $(T_NMI),TF_TRAPNO(%rsp)
357 movq $0,TF_ADDR(%rsp)
359 movq %rdi,TF_RDI(%rsp)
360 movq %rsi,TF_RSI(%rsp)
361 movq %rdx,TF_RDX(%rsp)
362 movq %rcx,TF_RCX(%rsp)
365 movq %rax,TF_RAX(%rsp)
366 movq %rbx,TF_RBX(%rsp)
367 movq %rbp,TF_RBP(%rsp)
368 movq %r10,TF_R10(%rsp)
369 movq %r11,TF_R11(%rsp)
370 movq %r12,TF_R12(%rsp)
371 movq %r13,TF_R13(%rsp)
372 movq %r14,TF_R14(%rsp)
373 movq %r15,TF_R15(%rsp)
375 testb $SEL_RPL_MASK,TF_CS(%rsp)
376 jnz nmi_needswapgs /* we came from userland */
377 movl $MSR_GSBASE,%ecx
379 cmpl $VM_MAXUSER_ADDRESS >> 32,%edx
380 jae nmi_calltrap /* GS.base holds a kernel VA */
384 /* Note: this label is also used by ddb and gdb: */
386 FAKE_MCOUNT(TF_RIP(%rsp))
394 movq TF_RDI(%rsp),%rdi
395 movq TF_RSI(%rsp),%rsi
396 movq TF_RDX(%rsp),%rdx
397 movq TF_RCX(%rsp),%rcx
400 movq TF_RAX(%rsp),%rax
401 movq TF_RBX(%rsp),%rbx
402 movq TF_RBP(%rsp),%rbp
403 movq TF_R10(%rsp),%r10
404 movq TF_R11(%rsp),%r11
405 movq TF_R12(%rsp),%r12
406 movq TF_R13(%rsp),%r13
407 movq TF_R14(%rsp),%r14
408 movq TF_R15(%rsp),%r15
413 * This function is what cpu_heavy_restore jumps to after a new process
414 * is created. The LWKT subsystem switches while holding a critical
415 * section and we maintain that abstraction here (e.g. because
416 * cpu_heavy_restore needs it due to PCB_*() manipulation), then get out of
417 * it before calling the initial function (typically fork_return()) and/or
418 * returning to user mode.
420 * The MP lock is held on entry, but for processes fork_return(esi)
421 * releases it. 'doreti' always runs without the MP lock.
423 ENTRY(fork_trampoline)
424 movq PCPU(curthread),%rax
425 subl $TDPRI_CRIT,TD_PRI(%rax)
428 * cpu_set_fork_handler intercepts this function call to
429 * have this call a non-return function to stay in kernel mode.
431 * initproc has its own fork handler, start_init(), which DOES
434 * %rbx - chaining function (typically fork_return)
435 * %r12 -> %rdi (argument)
436 * frame-> %rsi (trap frame)
438 * void (func:rbx)(arg:rdi, trapframe:rsi)
440 movq %rsp, %rsi /* pass trapframe by reference */
441 movq %r12, %rdi /* arg1 */
442 call *%rbx /* function */
444 /* cut from syscall */
449 #if defined(INVARIANTS) && defined(SMP)
450 movq PCPU(curthread),%rax
451 cmpl $0,TD_MPCOUNT(%rax)
454 movl TD_MPCOUNT(%rax), %rsi
458 pmsg4: .asciz "fork_trampoline mpcount %d after calling %p"
459 /* JG what's the purpose of this alignment and is it enough on amd64? */
464 * Return via doreti to handle ASTs.
466 * trapframe is at the top of the stack.
472 * To efficiently implement classification of trap and interrupt handlers
473 * for profiling, there must be only trap handlers between the labels btrap
474 * and bintr, and only interrupt handlers between the labels bintr and
475 * eintr. This is implemented (partly) by including files that contain
476 * some of the handlers. Before including the files, set up a normal asm
477 * environment so that the included files doen't need to know that they are
487 #include <amd64/ia32/ia32_exception.S>
497 #include <amd64/amd64/apic_vector.S>
506 #include <amd64/isa/atpic_vector.S>