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
38 #include "opt_atpic.h"
40 #include "opt_compat.h"
42 #include <machine/asmacros.h>
43 #include <machine/psl.h>
44 #include <machine/trap.h>
45 #include <machine/segments.h>
51 /*****************************************************************************/
53 /*****************************************************************************/
55 * Trap and fault vector routines.
57 * All traps are 'interrupt gates', SDT_SYSIGT. An interrupt gate pushes
58 * state on the stack but also disables interrupts. This is important for
59 * us for the use of the swapgs instruction. We cannot be interrupted
60 * until the GS.base value is correct. For most traps, we automatically
61 * then enable interrupts if the interrupted context had them enabled.
62 * This is equivalent to the i386 port's use of SDT_SYS386TGT.
64 * The cpu will push a certain amount of state onto the kernel stack for
65 * the current process. See x86_64/include/frame.h.
66 * This includes the current RFLAGS (status register, which includes
67 * the interrupt disable state prior to the trap), the code segment register,
68 * and the return instruction pointer are pushed by the cpu. The cpu
69 * will also push an 'error' code for certain traps. We push a dummy
70 * error code for those traps where the cpu doesn't in order to maintain
71 * a consistent frame. We also push a contrived 'trap number'.
73 * The cpu does not push the general registers, we must do that, and we
74 * must restore them prior to calling 'iret'. The cpu adjusts the %cs and
75 * %ss segment registers, but does not mess with %ds, %es, or %fs. Thus we
76 * must load them with appropriate values for supervisor mode operation.
82 /* Traps that we leave interrupts disabled for.. */
83 #define TRAP_NOEN(a) \
85 movq $0,TF_XFLAGS(%rsp) ; \
86 movq $(a),TF_TRAPNO(%rsp) ; \
87 movq $0,TF_ADDR(%rsp) ; \
88 movq $0,TF_ERR(%rsp) ; \
95 /* Regular traps; The cpu does not supply tf_err for these. */
98 movq $0,TF_XFLAGS(%rsp) ; \
99 movq $(a),TF_TRAPNO(%rsp) ; \
100 movq $0,TF_ADDR(%rsp) ; \
101 movq $0,TF_ERR(%rsp) ; \
124 /* This group of traps have tf_err already pushed by the cpu */
125 #define TRAP_ERR(a) \
127 movq $(a),TF_TRAPNO(%rsp) ; \
128 movq $0,TF_ADDR(%rsp) ; \
129 movq $0,TF_XFLAGS(%rsp) ; \
141 * alltraps entry point. Use swapgs if this is the first time in the
142 * kernel from userland. Reenable interrupts if they were enabled
143 * before the trap. This approximates SDT_SYS386TGT on the i386 port.
148 .type alltraps,@function
150 /* Fixup %gs if coming from userland */
151 testb $SEL_RPL_MASK,TF_CS(%rsp)
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 /* Fixup %gs if coming from userland */
194 testb $SEL_RPL_MASK,TF_CS(%rsp)
197 jmp alltraps_pushregs
201 movq $T_DOUBLEFLT,TF_TRAPNO(%rsp)
202 movq $0,TF_ADDR(%rsp)
204 movq $0,TF_XFLAGS(%rsp)
205 movq %rdi,TF_RDI(%rsp)
206 movq %rsi,TF_RSI(%rsp)
207 movq %rdx,TF_RDX(%rsp)
208 movq %rcx,TF_RCX(%rsp)
211 movq %rax,TF_RAX(%rsp)
212 movq %rbx,TF_RBX(%rsp)
213 movq %rbp,TF_RBP(%rsp)
214 movq %r10,TF_R10(%rsp)
215 movq %r11,TF_R11(%rsp)
216 movq %r12,TF_R12(%rsp)
217 movq %r13,TF_R13(%rsp)
218 movq %r14,TF_R14(%rsp)
219 movq %r15,TF_R15(%rsp)
220 testb $SEL_RPL_MASK,TF_CS(%rsp)
224 call dblfault_handler
230 movq $T_PAGEFLT,TF_TRAPNO(%rsp)
231 /* Fixup %gs if coming from userland */
232 testb $SEL_RPL_MASK,TF_CS(%rsp)
236 movq %rdi,TF_RDI(%rsp) /* free up a GP register */
237 movq %cr2,%rdi /* preserve %cr2 before .. */
238 movq %rdi,TF_ADDR(%rsp) /* enabling interrupts. */
239 movq $0,TF_XFLAGS(%rsp)
240 testq $PSL_I,TF_RFLAGS(%rsp)
241 jz alltraps_pushregs_no_rdi
243 jmp alltraps_pushregs_no_rdi
246 * We have to special-case this one. If we get a trap in doreti() at
247 * the iretq stage, we'll reenter with the wrong gs state. We'll have
248 * to do a special the swapgs in this case even coming from the kernel.
249 * XXX linux has a trap handler for their equivalent of load_gs().
253 movq $T_PROTFLT,TF_TRAPNO(%rsp)
254 movq $0,TF_ADDR(%rsp)
255 movq $0,TF_XFLAGS(%rsp)
256 movq %rdi,TF_RDI(%rsp) /* free up a GP register */
259 * Fixup %gs if coming from userland. Handle the special case where
260 * %fs faults in doreti at the iretq instruction itself.
262 leaq doreti_iret(%rip),%rdi
263 cmpq %rdi,TF_RIP(%rsp) /* special iretq fault case */
265 testb $SEL_RPL_MASK,TF_CS(%rsp) /* check if from userland */
270 testq $PSL_I,TF_RFLAGS(%rsp)
271 jz alltraps_pushregs_no_rdi
273 jmp alltraps_pushregs_no_rdi
276 * Fast syscall entry point. We enter here with just our new %cs/%ss set,
277 * and the new privilige level. We are still running on the old user stack
278 * pointer. We have to juggle a few things around to find our stack etc.
279 * swapgs gives us access to our PCPU space only.
283 movq %rsp,PCPU(scratch_rsp)
284 movq PCPU(common_tss) + TSS_RSP0, %rsp
285 /* Now emulate a trapframe. Make the 8 byte alignment odd for call. */
287 /* defer TF_RSP till we have a spare register */
288 movq %r11,TF_RFLAGS(%rsp)
289 movq %rcx,TF_RIP(%rsp) /* %rcx original value is in %r10 */
290 movq PCPU(scratch_rsp),%r11 /* %r11 already saved */
291 movq %r11,TF_RSP(%rsp) /* user stack pointer */
293 movq $KUDSEL,TF_SS(%rsp)
294 movq $KUCSEL,TF_CS(%rsp)
296 movq $T_FAST_SYSCALL,TF_TRAPNO(%rsp) /* for the vkernel */
297 movq $0,TF_XFLAGS(%rsp) /* note: used in signal frame */
298 movq %rdi,TF_RDI(%rsp) /* arg 1 */
299 movq %rsi,TF_RSI(%rsp) /* arg 2 */
300 movq %rdx,TF_RDX(%rsp) /* arg 3 */
301 movq %r10,TF_RCX(%rsp) /* arg 4 */
302 movq %r8,TF_R8(%rsp) /* arg 5 */
303 movq %r9,TF_R9(%rsp) /* arg 6 */
304 movq %rax,TF_RAX(%rsp) /* syscall number */
305 movq %rbx,TF_RBX(%rsp) /* C preserved */
306 movq %rbp,TF_RBP(%rsp) /* C preserved */
307 movq %r12,TF_R12(%rsp) /* C preserved */
308 movq %r13,TF_R13(%rsp) /* C preserved */
309 movq %r14,TF_R14(%rsp) /* C preserved */
310 movq %r15,TF_R15(%rsp) /* C preserved */
311 FAKE_MCOUNT(TF_RIP(%rsp))
318 * Here for CYA insurance, in case a "syscall" instruction gets
319 * issued from 32 bit compatability mode. MSR_CSTAR has to point
320 * to *something* if EFER_SCE is enabled.
322 IDTVEC(fast_syscall32)
326 * NMI handling is special.
328 * First, NMIs do not respect the state of the processor's RFLAGS.IF
329 * bit and the NMI handler may be invoked at any time, including when
330 * the processor is in a critical section with RFLAGS.IF == 0. In
331 * particular, this means that the processor's GS.base values could be
332 * inconsistent on entry to the handler, and so we need to read
333 * MSR_GSBASE to determine if a 'swapgs' is needed. We use '%ebx', a
334 * C-preserved register, to remember whether to swap GS back on the
337 * Second, the processor treats NMIs specially, blocking further NMIs
338 * until an 'iretq' instruction is executed. We therefore need to
339 * execute the NMI handler with interrupts disabled to prevent a
340 * nested interrupt from executing an 'iretq' instruction and
341 * inadvertently taking the processor out of NMI mode.
343 * Third, the NMI handler runs on its own stack (tss_ist1), shared
344 * with the double fault handler.
349 movq $(T_NMI),TF_TRAPNO(%rsp)
350 movq $0,TF_ADDR(%rsp)
352 movq $0,TF_XFLAGS(%rsp)
353 movq %rdi,TF_RDI(%rsp)
354 movq %rsi,TF_RSI(%rsp)
355 movq %rdx,TF_RDX(%rsp)
356 movq %rcx,TF_RCX(%rsp)
359 movq %rax,TF_RAX(%rsp)
360 movq %rbx,TF_RBX(%rsp)
361 movq %rbp,TF_RBP(%rsp)
362 movq %r10,TF_R10(%rsp)
363 movq %r11,TF_R11(%rsp)
364 movq %r12,TF_R12(%rsp)
365 movq %r13,TF_R13(%rsp)
366 movq %r14,TF_R14(%rsp)
367 movq %r15,TF_R15(%rsp)
369 testb $SEL_RPL_MASK,TF_CS(%rsp)
370 jnz nmi_needswapgs /* we came from userland */
371 movl $MSR_GSBASE,%ecx
373 cmpl $VM_MAX_USER_ADDRESS >> 32,%edx
374 jae nmi_calltrap /* GS.base holds a kernel VA */
378 /* Note: this label is also used by ddb and gdb: */
380 FAKE_MCOUNT(TF_RIP(%rsp))
388 movq TF_RDI(%rsp),%rdi
389 movq TF_RSI(%rsp),%rsi
390 movq TF_RDX(%rsp),%rdx
391 movq TF_RCX(%rsp),%rcx
394 movq TF_RAX(%rsp),%rax
395 movq TF_RBX(%rsp),%rbx
396 movq TF_RBP(%rsp),%rbp
397 movq TF_R10(%rsp),%r10
398 movq TF_R11(%rsp),%r11
399 movq TF_R12(%rsp),%r12
400 movq TF_R13(%rsp),%r13
401 movq TF_R14(%rsp),%r14
402 movq TF_R15(%rsp),%r15
407 * This function is what cpu_heavy_restore jumps to after a new process
408 * is created. The LWKT subsystem switches while holding a critical
409 * section and we maintain that abstraction here (e.g. because
410 * cpu_heavy_restore needs it due to PCB_*() manipulation), then get out of
411 * it before calling the initial function (typically fork_return()) and/or
412 * returning to user mode.
414 * The MP lock is not held at any point but the critcount is bumped
415 * on entry to prevent interruption of the trampoline at a bad point.
417 ENTRY(fork_trampoline)
418 movq PCPU(curthread),%rax
419 decl TD_CRITCOUNT(%rax)
422 * cpu_set_fork_handler intercepts this function call to
423 * have this call a non-return function to stay in kernel mode.
425 * initproc has its own fork handler, start_init(), which DOES
428 * %rbx - chaining function (typically fork_return)
429 * %r12 -> %rdi (argument)
430 * frame-> %rsi (trap frame)
432 * void (func:rbx)(arg:rdi, trapframe:rsi)
434 movq %rsp, %rsi /* pass trapframe by reference */
435 movq %r12, %rdi /* arg1 */
436 call *%rbx /* function */
438 /* cut from syscall */
444 * Return via doreti to handle ASTs.
446 * trapframe is at the top of the stack.
452 * To efficiently implement classification of trap and interrupt handlers
453 * for profiling, there must be only trap handlers between the labels btrap
454 * and bintr, and only interrupt handlers between the labels bintr and
455 * eintr. This is implemented (partly) by including files that contain
456 * some of the handlers. Before including the files, set up a normal asm
457 * environment so that the included files doen't need to know that they are
467 #include <x86_64/ia32/ia32_exception.S>
477 #include <x86_64/x86_64/apic_vector.S>
486 #include <x86_64/isa/atpic_vector.S>