xref: /freebsd-src/sys/amd64/amd64/exec_machdep.c (revision 32a95656b51ebefcdf3e0b02c110825f59abd26f)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 2003 Peter Wemm.
5  * Copyright (c) 1992 Terrence R. Lambert.
6  * Copyright (c) 1982, 1987, 1990 The Regents of the University of California.
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * William Jolitz.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
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  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	from: @(#)machdep.c	7.4 (Berkeley) 6/3/91
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_cpu.h"
47 #include "opt_ddb.h"
48 #include "opt_kstack_pages.h"
49 
50 #include <sys/param.h>
51 #include <sys/proc.h>
52 #include <sys/systm.h>
53 #include <sys/exec.h>
54 #include <sys/imgact.h>
55 #include <sys/kdb.h>
56 #include <sys/kernel.h>
57 #include <sys/ktr.h>
58 #include <sys/linker.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/pcpu.h>
63 #include <sys/reg.h>
64 #include <sys/rwlock.h>
65 #include <sys/signalvar.h>
66 #ifdef SMP
67 #include <sys/smp.h>
68 #endif
69 #include <sys/syscallsubr.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/sysproto.h>
73 #include <sys/ucontext.h>
74 #include <sys/vmmeter.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/vm_extern.h>
79 #include <vm/pmap.h>
80 
81 #ifdef DDB
82 #ifndef KDB
83 #error KDB must be enabled in order for DDB to work!
84 #endif
85 #include <ddb/ddb.h>
86 #include <ddb/db_sym.h>
87 #endif
88 
89 #include <machine/vmparam.h>
90 #include <machine/frame.h>
91 #include <machine/md_var.h>
92 #include <machine/pcb.h>
93 #include <machine/proc.h>
94 #include <machine/sigframe.h>
95 #include <machine/specialreg.h>
96 #include <machine/trap.h>
97 
98 /*
99  * Send an interrupt to process.
100  *
101  * Stack is set up to allow sigcode stored at top to call routine,
102  * followed by call to sigreturn routine below.  After sigreturn
103  * resets the signal mask, the stack, and the frame pointer, it
104  * returns to the user specified pc, psl.
105  */
106 void
107 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
108 {
109 	struct sigframe sf, *sfp;
110 	struct pcb *pcb;
111 	struct proc *p;
112 	struct thread *td;
113 	struct sigacts *psp;
114 	char *sp;
115 	struct trapframe *regs;
116 	char *xfpusave;
117 	size_t xfpusave_len;
118 	int sig;
119 	int oonstack;
120 
121 	td = curthread;
122 	pcb = td->td_pcb;
123 	p = td->td_proc;
124 	PROC_LOCK_ASSERT(p, MA_OWNED);
125 	sig = ksi->ksi_signo;
126 	psp = p->p_sigacts;
127 	mtx_assert(&psp->ps_mtx, MA_OWNED);
128 	regs = td->td_frame;
129 	oonstack = sigonstack(regs->tf_rsp);
130 
131 	/* Save user context. */
132 	bzero(&sf, sizeof(sf));
133 	sf.sf_uc.uc_sigmask = *mask;
134 	sf.sf_uc.uc_stack = td->td_sigstk;
135 	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
136 	    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
137 	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
138 	bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(*regs));
139 	sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */
140 	get_fpcontext(td, &sf.sf_uc.uc_mcontext, &xfpusave, &xfpusave_len);
141 	update_pcb_bases(pcb);
142 	sf.sf_uc.uc_mcontext.mc_fsbase = pcb->pcb_fsbase;
143 	sf.sf_uc.uc_mcontext.mc_gsbase = pcb->pcb_gsbase;
144 	bzero(sf.sf_uc.uc_mcontext.mc_spare,
145 	    sizeof(sf.sf_uc.uc_mcontext.mc_spare));
146 
147 	/* Allocate space for the signal handler context. */
148 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
149 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
150 		sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
151 #if defined(COMPAT_43)
152 		td->td_sigstk.ss_flags |= SS_ONSTACK;
153 #endif
154 	} else
155 		sp = (char *)regs->tf_rsp - 128;
156 	if (xfpusave != NULL) {
157 		sp -= xfpusave_len;
158 		sp = (char *)((unsigned long)sp & ~0x3Ful);
159 		sf.sf_uc.uc_mcontext.mc_xfpustate = (register_t)sp;
160 	}
161 	sp -= sizeof(struct sigframe);
162 	/* Align to 16 bytes. */
163 	sfp = (struct sigframe *)((unsigned long)sp & ~0xFul);
164 
165 	/* Build the argument list for the signal handler. */
166 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
167 	regs->tf_rdx = (register_t)&sfp->sf_uc;	/* arg 3 in %rdx */
168 	bzero(&sf.sf_si, sizeof(sf.sf_si));
169 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
170 		/* Signal handler installed with SA_SIGINFO. */
171 		regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
172 		sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
173 
174 		/* Fill in POSIX parts */
175 		sf.sf_si = ksi->ksi_info;
176 		sf.sf_si.si_signo = sig; /* maybe a translated signal */
177 		regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */
178 	} else {
179 		/* Old FreeBSD-style arguments. */
180 		regs->tf_rsi = ksi->ksi_code;	/* arg 2 in %rsi */
181 		regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */
182 		sf.sf_ahu.sf_handler = catcher;
183 	}
184 	mtx_unlock(&psp->ps_mtx);
185 	PROC_UNLOCK(p);
186 
187 	/*
188 	 * Copy the sigframe out to the user's stack.
189 	 */
190 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0 ||
191 	    (xfpusave != NULL && copyout(xfpusave,
192 	    (void *)sf.sf_uc.uc_mcontext.mc_xfpustate, xfpusave_len)
193 	    != 0)) {
194 		uprintf("pid %d comm %s has trashed its stack, killing\n",
195 		    p->p_pid, p->p_comm);
196 		PROC_LOCK(p);
197 		sigexit(td, SIGILL);
198 	}
199 
200 	fpstate_drop(td);
201 	regs->tf_rsp = (long)sfp;
202 	regs->tf_rip = p->p_sysent->sv_sigcode_base;
203 	regs->tf_rflags &= ~(PSL_T | PSL_D);
204 	regs->tf_cs = _ucodesel;
205 	regs->tf_ds = _udatasel;
206 	regs->tf_ss = _udatasel;
207 	regs->tf_es = _udatasel;
208 	regs->tf_fs = _ufssel;
209 	regs->tf_gs = _ugssel;
210 	regs->tf_flags = TF_HASSEGS;
211 	PROC_LOCK(p);
212 	mtx_lock(&psp->ps_mtx);
213 }
214 
215 /*
216  * System call to cleanup state after a signal
217  * has been taken.  Reset signal mask and
218  * stack state from context left by sendsig (above).
219  * Return to previous pc and psl as specified by
220  * context left by sendsig. Check carefully to
221  * make sure that the user has not modified the
222  * state to gain improper privileges.
223  */
224 int
225 sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
226 {
227 	ucontext_t uc;
228 	struct pcb *pcb;
229 	struct proc *p;
230 	struct trapframe *regs;
231 	ucontext_t *ucp;
232 	char *xfpustate;
233 	size_t xfpustate_len;
234 	long rflags;
235 	int cs, error, ret;
236 	ksiginfo_t ksi;
237 
238 	pcb = td->td_pcb;
239 	p = td->td_proc;
240 
241 	error = copyin(uap->sigcntxp, &uc, sizeof(uc));
242 	if (error != 0) {
243 		uprintf("pid %d (%s): sigreturn copyin failed\n",
244 		    p->p_pid, td->td_name);
245 		return (error);
246 	}
247 	ucp = &uc;
248 	if ((ucp->uc_mcontext.mc_flags & ~_MC_FLAG_MASK) != 0) {
249 		uprintf("pid %d (%s): sigreturn mc_flags %x\n", p->p_pid,
250 		    td->td_name, ucp->uc_mcontext.mc_flags);
251 		return (EINVAL);
252 	}
253 	regs = td->td_frame;
254 	rflags = ucp->uc_mcontext.mc_rflags;
255 	/*
256 	 * Don't allow users to change privileged or reserved flags.
257 	 */
258 	if (!EFL_SECURE(rflags, regs->tf_rflags)) {
259 		uprintf("pid %d (%s): sigreturn rflags = 0x%lx\n", p->p_pid,
260 		    td->td_name, rflags);
261 		return (EINVAL);
262 	}
263 
264 	/*
265 	 * Don't allow users to load a valid privileged %cs.  Let the
266 	 * hardware check for invalid selectors, excess privilege in
267 	 * other selectors, invalid %eip's and invalid %esp's.
268 	 */
269 	cs = ucp->uc_mcontext.mc_cs;
270 	if (!CS_SECURE(cs)) {
271 		uprintf("pid %d (%s): sigreturn cs = 0x%x\n", p->p_pid,
272 		    td->td_name, cs);
273 		ksiginfo_init_trap(&ksi);
274 		ksi.ksi_signo = SIGBUS;
275 		ksi.ksi_code = BUS_OBJERR;
276 		ksi.ksi_trapno = T_PROTFLT;
277 		ksi.ksi_addr = (void *)regs->tf_rip;
278 		trapsignal(td, &ksi);
279 		return (EINVAL);
280 	}
281 
282 	if ((uc.uc_mcontext.mc_flags & _MC_HASFPXSTATE) != 0) {
283 		xfpustate_len = uc.uc_mcontext.mc_xfpustate_len;
284 		if (xfpustate_len > cpu_max_ext_state_size -
285 		    sizeof(struct savefpu)) {
286 			uprintf("pid %d (%s): sigreturn xfpusave_len = 0x%zx\n",
287 			    p->p_pid, td->td_name, xfpustate_len);
288 			return (EINVAL);
289 		}
290 		xfpustate = (char *)fpu_save_area_alloc();
291 		error = copyin((const void *)uc.uc_mcontext.mc_xfpustate,
292 		    xfpustate, xfpustate_len);
293 		if (error != 0) {
294 			fpu_save_area_free((struct savefpu *)xfpustate);
295 			uprintf(
296 	"pid %d (%s): sigreturn copying xfpustate failed\n",
297 			    p->p_pid, td->td_name);
298 			return (error);
299 		}
300 	} else {
301 		xfpustate = NULL;
302 		xfpustate_len = 0;
303 	}
304 	ret = set_fpcontext(td, &ucp->uc_mcontext, xfpustate, xfpustate_len);
305 	fpu_save_area_free((struct savefpu *)xfpustate);
306 	if (ret != 0) {
307 		uprintf("pid %d (%s): sigreturn set_fpcontext err %d\n",
308 		    p->p_pid, td->td_name, ret);
309 		return (ret);
310 	}
311 	bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(*regs));
312 	update_pcb_bases(pcb);
313 	pcb->pcb_fsbase = ucp->uc_mcontext.mc_fsbase;
314 	pcb->pcb_gsbase = ucp->uc_mcontext.mc_gsbase;
315 
316 #if defined(COMPAT_43)
317 	if (ucp->uc_mcontext.mc_onstack & 1)
318 		td->td_sigstk.ss_flags |= SS_ONSTACK;
319 	else
320 		td->td_sigstk.ss_flags &= ~SS_ONSTACK;
321 #endif
322 
323 	kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0);
324 	return (EJUSTRETURN);
325 }
326 
327 #ifdef COMPAT_FREEBSD4
328 int
329 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
330 {
331 
332 	return sys_sigreturn(td, (struct sigreturn_args *)uap);
333 }
334 #endif
335 
336 /*
337  * Reset the hardware debug registers if they were in use.
338  * They won't have any meaning for the newly exec'd process.
339  */
340 void
341 x86_clear_dbregs(struct pcb *pcb)
342 {
343 	if ((pcb->pcb_flags & PCB_DBREGS) == 0)
344 		return;
345 
346 	pcb->pcb_dr0 = 0;
347 	pcb->pcb_dr1 = 0;
348 	pcb->pcb_dr2 = 0;
349 	pcb->pcb_dr3 = 0;
350 	pcb->pcb_dr6 = 0;
351 	pcb->pcb_dr7 = 0;
352 
353 	if (pcb == curpcb) {
354 		/*
355 		 * Clear the debug registers on the running CPU,
356 		 * otherwise they will end up affecting the next
357 		 * process we switch to.
358 		 */
359 		reset_dbregs();
360 	}
361 	clear_pcb_flags(pcb, PCB_DBREGS);
362 }
363 
364 /*
365  * Reset registers to default values on exec.
366  */
367 void
368 exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack)
369 {
370 	struct trapframe *regs;
371 	struct pcb *pcb;
372 	register_t saved_rflags;
373 
374 	regs = td->td_frame;
375 	pcb = td->td_pcb;
376 
377 	if (td->td_proc->p_md.md_ldt != NULL)
378 		user_ldt_free(td);
379 
380 	update_pcb_bases(pcb);
381 	pcb->pcb_fsbase = 0;
382 	pcb->pcb_gsbase = 0;
383 	clear_pcb_flags(pcb, PCB_32BIT);
384 	pcb->pcb_initial_fpucw = __INITIAL_FPUCW__;
385 
386 	saved_rflags = regs->tf_rflags & PSL_T;
387 	bzero((char *)regs, sizeof(struct trapframe));
388 	regs->tf_rip = imgp->entry_addr;
389 	regs->tf_rsp = ((stack - 8) & ~0xFul) + 8;
390 	regs->tf_rdi = stack;		/* argv */
391 	regs->tf_rflags = PSL_USER | saved_rflags;
392 	regs->tf_ss = _udatasel;
393 	regs->tf_cs = _ucodesel;
394 	regs->tf_ds = _udatasel;
395 	regs->tf_es = _udatasel;
396 	regs->tf_fs = _ufssel;
397 	regs->tf_gs = _ugssel;
398 	regs->tf_flags = TF_HASSEGS;
399 
400 	x86_clear_dbregs(pcb);
401 
402 	/*
403 	 * Drop the FP state if we hold it, so that the process gets a
404 	 * clean FP state if it uses the FPU again.
405 	 */
406 	fpstate_drop(td);
407 }
408 
409 int
410 fill_regs(struct thread *td, struct reg *regs)
411 {
412 	struct trapframe *tp;
413 
414 	tp = td->td_frame;
415 	return (fill_frame_regs(tp, regs));
416 }
417 
418 int
419 fill_frame_regs(struct trapframe *tp, struct reg *regs)
420 {
421 
422 	regs->r_r15 = tp->tf_r15;
423 	regs->r_r14 = tp->tf_r14;
424 	regs->r_r13 = tp->tf_r13;
425 	regs->r_r12 = tp->tf_r12;
426 	regs->r_r11 = tp->tf_r11;
427 	regs->r_r10 = tp->tf_r10;
428 	regs->r_r9  = tp->tf_r9;
429 	regs->r_r8  = tp->tf_r8;
430 	regs->r_rdi = tp->tf_rdi;
431 	regs->r_rsi = tp->tf_rsi;
432 	regs->r_rbp = tp->tf_rbp;
433 	regs->r_rbx = tp->tf_rbx;
434 	regs->r_rdx = tp->tf_rdx;
435 	regs->r_rcx = tp->tf_rcx;
436 	regs->r_rax = tp->tf_rax;
437 	regs->r_rip = tp->tf_rip;
438 	regs->r_cs = tp->tf_cs;
439 	regs->r_rflags = tp->tf_rflags;
440 	regs->r_rsp = tp->tf_rsp;
441 	regs->r_ss = tp->tf_ss;
442 	if (tp->tf_flags & TF_HASSEGS) {
443 		regs->r_ds = tp->tf_ds;
444 		regs->r_es = tp->tf_es;
445 		regs->r_fs = tp->tf_fs;
446 		regs->r_gs = tp->tf_gs;
447 	} else {
448 		regs->r_ds = 0;
449 		regs->r_es = 0;
450 		regs->r_fs = 0;
451 		regs->r_gs = 0;
452 	}
453 	regs->r_err = 0;
454 	regs->r_trapno = 0;
455 	return (0);
456 }
457 
458 int
459 set_regs(struct thread *td, struct reg *regs)
460 {
461 	struct trapframe *tp;
462 	register_t rflags;
463 
464 	tp = td->td_frame;
465 	rflags = regs->r_rflags & 0xffffffff;
466 	if (!EFL_SECURE(rflags, tp->tf_rflags) || !CS_SECURE(regs->r_cs))
467 		return (EINVAL);
468 	tp->tf_r15 = regs->r_r15;
469 	tp->tf_r14 = regs->r_r14;
470 	tp->tf_r13 = regs->r_r13;
471 	tp->tf_r12 = regs->r_r12;
472 	tp->tf_r11 = regs->r_r11;
473 	tp->tf_r10 = regs->r_r10;
474 	tp->tf_r9  = regs->r_r9;
475 	tp->tf_r8  = regs->r_r8;
476 	tp->tf_rdi = regs->r_rdi;
477 	tp->tf_rsi = regs->r_rsi;
478 	tp->tf_rbp = regs->r_rbp;
479 	tp->tf_rbx = regs->r_rbx;
480 	tp->tf_rdx = regs->r_rdx;
481 	tp->tf_rcx = regs->r_rcx;
482 	tp->tf_rax = regs->r_rax;
483 	tp->tf_rip = regs->r_rip;
484 	tp->tf_cs = regs->r_cs;
485 	tp->tf_rflags = rflags;
486 	tp->tf_rsp = regs->r_rsp;
487 	tp->tf_ss = regs->r_ss;
488 	if (0) {	/* XXXKIB */
489 		tp->tf_ds = regs->r_ds;
490 		tp->tf_es = regs->r_es;
491 		tp->tf_fs = regs->r_fs;
492 		tp->tf_gs = regs->r_gs;
493 		tp->tf_flags = TF_HASSEGS;
494 	}
495 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
496 	return (0);
497 }
498 
499 /* XXX check all this stuff! */
500 /* externalize from sv_xmm */
501 static void
502 fill_fpregs_xmm(struct savefpu *sv_xmm, struct fpreg *fpregs)
503 {
504 	struct envxmm *penv_fpreg = (struct envxmm *)&fpregs->fpr_env;
505 	struct envxmm *penv_xmm = &sv_xmm->sv_env;
506 	int i;
507 
508 	/* pcb -> fpregs */
509 	bzero(fpregs, sizeof(*fpregs));
510 
511 	/* FPU control/status */
512 	penv_fpreg->en_cw = penv_xmm->en_cw;
513 	penv_fpreg->en_sw = penv_xmm->en_sw;
514 	penv_fpreg->en_tw = penv_xmm->en_tw;
515 	penv_fpreg->en_opcode = penv_xmm->en_opcode;
516 	penv_fpreg->en_rip = penv_xmm->en_rip;
517 	penv_fpreg->en_rdp = penv_xmm->en_rdp;
518 	penv_fpreg->en_mxcsr = penv_xmm->en_mxcsr;
519 	penv_fpreg->en_mxcsr_mask = penv_xmm->en_mxcsr_mask;
520 
521 	/* FPU registers */
522 	for (i = 0; i < 8; ++i)
523 		bcopy(sv_xmm->sv_fp[i].fp_acc.fp_bytes, fpregs->fpr_acc[i], 10);
524 
525 	/* SSE registers */
526 	for (i = 0; i < 16; ++i)
527 		bcopy(sv_xmm->sv_xmm[i].xmm_bytes, fpregs->fpr_xacc[i], 16);
528 }
529 
530 /* internalize from fpregs into sv_xmm */
531 static void
532 set_fpregs_xmm(struct fpreg *fpregs, struct savefpu *sv_xmm)
533 {
534 	struct envxmm *penv_xmm = &sv_xmm->sv_env;
535 	struct envxmm *penv_fpreg = (struct envxmm *)&fpregs->fpr_env;
536 	int i;
537 
538 	/* fpregs -> pcb */
539 	/* FPU control/status */
540 	penv_xmm->en_cw = penv_fpreg->en_cw;
541 	penv_xmm->en_sw = penv_fpreg->en_sw;
542 	penv_xmm->en_tw = penv_fpreg->en_tw;
543 	penv_xmm->en_opcode = penv_fpreg->en_opcode;
544 	penv_xmm->en_rip = penv_fpreg->en_rip;
545 	penv_xmm->en_rdp = penv_fpreg->en_rdp;
546 	penv_xmm->en_mxcsr = penv_fpreg->en_mxcsr;
547 	penv_xmm->en_mxcsr_mask = penv_fpreg->en_mxcsr_mask & cpu_mxcsr_mask;
548 
549 	/* FPU registers */
550 	for (i = 0; i < 8; ++i)
551 		bcopy(fpregs->fpr_acc[i], sv_xmm->sv_fp[i].fp_acc.fp_bytes, 10);
552 
553 	/* SSE registers */
554 	for (i = 0; i < 16; ++i)
555 		bcopy(fpregs->fpr_xacc[i], sv_xmm->sv_xmm[i].xmm_bytes, 16);
556 }
557 
558 /* externalize from td->pcb */
559 int
560 fill_fpregs(struct thread *td, struct fpreg *fpregs)
561 {
562 
563 	KASSERT(td == curthread || TD_IS_SUSPENDED(td) ||
564 	    P_SHOULDSTOP(td->td_proc),
565 	    ("not suspended thread %p", td));
566 	fpugetregs(td);
567 	fill_fpregs_xmm(get_pcb_user_save_td(td), fpregs);
568 	return (0);
569 }
570 
571 /* internalize to td->pcb */
572 int
573 set_fpregs(struct thread *td, struct fpreg *fpregs)
574 {
575 
576 	critical_enter();
577 	set_fpregs_xmm(fpregs, get_pcb_user_save_td(td));
578 	fpuuserinited(td);
579 	critical_exit();
580 	return (0);
581 }
582 
583 /*
584  * Get machine context.
585  */
586 int
587 get_mcontext(struct thread *td, mcontext_t *mcp, int flags)
588 {
589 	struct pcb *pcb;
590 	struct trapframe *tp;
591 
592 	pcb = td->td_pcb;
593 	tp = td->td_frame;
594 	PROC_LOCK(curthread->td_proc);
595 	mcp->mc_onstack = sigonstack(tp->tf_rsp);
596 	PROC_UNLOCK(curthread->td_proc);
597 	mcp->mc_r15 = tp->tf_r15;
598 	mcp->mc_r14 = tp->tf_r14;
599 	mcp->mc_r13 = tp->tf_r13;
600 	mcp->mc_r12 = tp->tf_r12;
601 	mcp->mc_r11 = tp->tf_r11;
602 	mcp->mc_r10 = tp->tf_r10;
603 	mcp->mc_r9  = tp->tf_r9;
604 	mcp->mc_r8  = tp->tf_r8;
605 	mcp->mc_rdi = tp->tf_rdi;
606 	mcp->mc_rsi = tp->tf_rsi;
607 	mcp->mc_rbp = tp->tf_rbp;
608 	mcp->mc_rbx = tp->tf_rbx;
609 	mcp->mc_rcx = tp->tf_rcx;
610 	mcp->mc_rflags = tp->tf_rflags;
611 	if (flags & GET_MC_CLEAR_RET) {
612 		mcp->mc_rax = 0;
613 		mcp->mc_rdx = 0;
614 		mcp->mc_rflags &= ~PSL_C;
615 	} else {
616 		mcp->mc_rax = tp->tf_rax;
617 		mcp->mc_rdx = tp->tf_rdx;
618 	}
619 	mcp->mc_rip = tp->tf_rip;
620 	mcp->mc_cs = tp->tf_cs;
621 	mcp->mc_rsp = tp->tf_rsp;
622 	mcp->mc_ss = tp->tf_ss;
623 	mcp->mc_ds = tp->tf_ds;
624 	mcp->mc_es = tp->tf_es;
625 	mcp->mc_fs = tp->tf_fs;
626 	mcp->mc_gs = tp->tf_gs;
627 	mcp->mc_flags = tp->tf_flags;
628 	mcp->mc_len = sizeof(*mcp);
629 	get_fpcontext(td, mcp, NULL, NULL);
630 	update_pcb_bases(pcb);
631 	mcp->mc_fsbase = pcb->pcb_fsbase;
632 	mcp->mc_gsbase = pcb->pcb_gsbase;
633 	mcp->mc_xfpustate = 0;
634 	mcp->mc_xfpustate_len = 0;
635 	bzero(mcp->mc_spare, sizeof(mcp->mc_spare));
636 	return (0);
637 }
638 
639 /*
640  * Set machine context.
641  *
642  * However, we don't set any but the user modifiable flags, and we won't
643  * touch the cs selector.
644  */
645 int
646 set_mcontext(struct thread *td, mcontext_t *mcp)
647 {
648 	struct pcb *pcb;
649 	struct trapframe *tp;
650 	char *xfpustate;
651 	long rflags;
652 	int ret;
653 
654 	pcb = td->td_pcb;
655 	tp = td->td_frame;
656 	if (mcp->mc_len != sizeof(*mcp) ||
657 	    (mcp->mc_flags & ~_MC_FLAG_MASK) != 0)
658 		return (EINVAL);
659 	rflags = (mcp->mc_rflags & PSL_USERCHANGE) |
660 	    (tp->tf_rflags & ~PSL_USERCHANGE);
661 	if (mcp->mc_flags & _MC_HASFPXSTATE) {
662 		if (mcp->mc_xfpustate_len > cpu_max_ext_state_size -
663 		    sizeof(struct savefpu))
664 			return (EINVAL);
665 		xfpustate = (char *)fpu_save_area_alloc();
666 		ret = copyin((void *)mcp->mc_xfpustate, xfpustate,
667 		    mcp->mc_xfpustate_len);
668 		if (ret != 0) {
669 			fpu_save_area_free((struct savefpu *)xfpustate);
670 			return (ret);
671 		}
672 	} else
673 		xfpustate = NULL;
674 	ret = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len);
675 	fpu_save_area_free((struct savefpu *)xfpustate);
676 	if (ret != 0)
677 		return (ret);
678 	tp->tf_r15 = mcp->mc_r15;
679 	tp->tf_r14 = mcp->mc_r14;
680 	tp->tf_r13 = mcp->mc_r13;
681 	tp->tf_r12 = mcp->mc_r12;
682 	tp->tf_r11 = mcp->mc_r11;
683 	tp->tf_r10 = mcp->mc_r10;
684 	tp->tf_r9  = mcp->mc_r9;
685 	tp->tf_r8  = mcp->mc_r8;
686 	tp->tf_rdi = mcp->mc_rdi;
687 	tp->tf_rsi = mcp->mc_rsi;
688 	tp->tf_rbp = mcp->mc_rbp;
689 	tp->tf_rbx = mcp->mc_rbx;
690 	tp->tf_rdx = mcp->mc_rdx;
691 	tp->tf_rcx = mcp->mc_rcx;
692 	tp->tf_rax = mcp->mc_rax;
693 	tp->tf_rip = mcp->mc_rip;
694 	tp->tf_rflags = rflags;
695 	tp->tf_rsp = mcp->mc_rsp;
696 	tp->tf_ss = mcp->mc_ss;
697 	tp->tf_flags = mcp->mc_flags;
698 	if (tp->tf_flags & TF_HASSEGS) {
699 		tp->tf_ds = mcp->mc_ds;
700 		tp->tf_es = mcp->mc_es;
701 		tp->tf_fs = mcp->mc_fs;
702 		tp->tf_gs = mcp->mc_gs;
703 	}
704 	set_pcb_flags(pcb, PCB_FULL_IRET);
705 	if (mcp->mc_flags & _MC_HASBASES) {
706 		pcb->pcb_fsbase = mcp->mc_fsbase;
707 		pcb->pcb_gsbase = mcp->mc_gsbase;
708 	}
709 	return (0);
710 }
711 
712 void
713 get_fpcontext(struct thread *td, mcontext_t *mcp, char **xfpusave,
714     size_t *xfpusave_len)
715 {
716 	mcp->mc_ownedfp = fpugetregs(td);
717 	bcopy(get_pcb_user_save_td(td), &mcp->mc_fpstate[0],
718 	    sizeof(mcp->mc_fpstate));
719 	mcp->mc_fpformat = fpuformat();
720 	if (xfpusave == NULL)
721 		return;
722 	if (!use_xsave || cpu_max_ext_state_size <= sizeof(struct savefpu)) {
723 		*xfpusave_len = 0;
724 		*xfpusave = NULL;
725 	} else {
726 		mcp->mc_flags |= _MC_HASFPXSTATE;
727 		*xfpusave_len = mcp->mc_xfpustate_len =
728 		    cpu_max_ext_state_size - sizeof(struct savefpu);
729 		*xfpusave = (char *)(get_pcb_user_save_td(td) + 1);
730 	}
731 }
732 
733 int
734 set_fpcontext(struct thread *td, mcontext_t *mcp, char *xfpustate,
735     size_t xfpustate_len)
736 {
737 	int error;
738 
739 	if (mcp->mc_fpformat == _MC_FPFMT_NODEV)
740 		return (0);
741 	else if (mcp->mc_fpformat != _MC_FPFMT_XMM)
742 		return (EINVAL);
743 	else if (mcp->mc_ownedfp == _MC_FPOWNED_NONE) {
744 		/* We don't care what state is left in the FPU or PCB. */
745 		fpstate_drop(td);
746 		error = 0;
747 	} else if (mcp->mc_ownedfp == _MC_FPOWNED_FPU ||
748 	    mcp->mc_ownedfp == _MC_FPOWNED_PCB) {
749 		error = fpusetregs(td, (struct savefpu *)&mcp->mc_fpstate,
750 		    xfpustate, xfpustate_len);
751 	} else
752 		return (EINVAL);
753 	return (error);
754 }
755 
756 void
757 fpstate_drop(struct thread *td)
758 {
759 
760 	KASSERT(PCB_USER_FPU(td->td_pcb), ("fpstate_drop: kernel-owned fpu"));
761 	critical_enter();
762 	if (PCPU_GET(fpcurthread) == td)
763 		fpudrop();
764 	/*
765 	 * XXX force a full drop of the fpu.  The above only drops it if we
766 	 * owned it.
767 	 *
768 	 * XXX I don't much like fpugetuserregs()'s semantics of doing a full
769 	 * drop.  Dropping only to the pcb matches fnsave's behaviour.
770 	 * We only need to drop to !PCB_INITDONE in sendsig().  But
771 	 * sendsig() is the only caller of fpugetuserregs()... perhaps we just
772 	 * have too many layers.
773 	 */
774 	clear_pcb_flags(curthread->td_pcb,
775 	    PCB_FPUINITDONE | PCB_USERFPUINITDONE);
776 	critical_exit();
777 }
778 
779 int
780 fill_dbregs(struct thread *td, struct dbreg *dbregs)
781 {
782 	struct pcb *pcb;
783 
784 	if (td == NULL) {
785 		dbregs->dr[0] = rdr0();
786 		dbregs->dr[1] = rdr1();
787 		dbregs->dr[2] = rdr2();
788 		dbregs->dr[3] = rdr3();
789 		dbregs->dr[6] = rdr6();
790 		dbregs->dr[7] = rdr7();
791 	} else {
792 		pcb = td->td_pcb;
793 		dbregs->dr[0] = pcb->pcb_dr0;
794 		dbregs->dr[1] = pcb->pcb_dr1;
795 		dbregs->dr[2] = pcb->pcb_dr2;
796 		dbregs->dr[3] = pcb->pcb_dr3;
797 		dbregs->dr[6] = pcb->pcb_dr6;
798 		dbregs->dr[7] = pcb->pcb_dr7;
799 	}
800 	dbregs->dr[4] = 0;
801 	dbregs->dr[5] = 0;
802 	dbregs->dr[8] = 0;
803 	dbregs->dr[9] = 0;
804 	dbregs->dr[10] = 0;
805 	dbregs->dr[11] = 0;
806 	dbregs->dr[12] = 0;
807 	dbregs->dr[13] = 0;
808 	dbregs->dr[14] = 0;
809 	dbregs->dr[15] = 0;
810 	return (0);
811 }
812 
813 int
814 set_dbregs(struct thread *td, struct dbreg *dbregs)
815 {
816 	struct pcb *pcb;
817 	int i;
818 
819 	if (td == NULL) {
820 		load_dr0(dbregs->dr[0]);
821 		load_dr1(dbregs->dr[1]);
822 		load_dr2(dbregs->dr[2]);
823 		load_dr3(dbregs->dr[3]);
824 		load_dr6(dbregs->dr[6]);
825 		load_dr7(dbregs->dr[7]);
826 	} else {
827 		/*
828 		 * Don't let an illegal value for dr7 get set.  Specifically,
829 		 * check for undefined settings.  Setting these bit patterns
830 		 * result in undefined behaviour and can lead to an unexpected
831 		 * TRCTRAP or a general protection fault right here.
832 		 * Upper bits of dr6 and dr7 must not be set
833 		 */
834 		for (i = 0; i < 4; i++) {
835 			if (DBREG_DR7_ACCESS(dbregs->dr[7], i) == 0x02)
836 				return (EINVAL);
837 			if (td->td_frame->tf_cs == _ucode32sel &&
838 			    DBREG_DR7_LEN(dbregs->dr[7], i) == DBREG_DR7_LEN_8)
839 				return (EINVAL);
840 		}
841 		if ((dbregs->dr[6] & 0xffffffff00000000ul) != 0 ||
842 		    (dbregs->dr[7] & 0xffffffff00000000ul) != 0)
843 			return (EINVAL);
844 
845 		pcb = td->td_pcb;
846 
847 		/*
848 		 * Don't let a process set a breakpoint that is not within the
849 		 * process's address space.  If a process could do this, it
850 		 * could halt the system by setting a breakpoint in the kernel
851 		 * (if ddb was enabled).  Thus, we need to check to make sure
852 		 * that no breakpoints are being enabled for addresses outside
853 		 * process's address space.
854 		 *
855 		 * XXX - what about when the watched area of the user's
856 		 * address space is written into from within the kernel
857 		 * ... wouldn't that still cause a breakpoint to be generated
858 		 * from within kernel mode?
859 		 */
860 
861 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 0)) {
862 			/* dr0 is enabled */
863 			if (dbregs->dr[0] >= VM_MAXUSER_ADDRESS)
864 				return (EINVAL);
865 		}
866 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 1)) {
867 			/* dr1 is enabled */
868 			if (dbregs->dr[1] >= VM_MAXUSER_ADDRESS)
869 				return (EINVAL);
870 		}
871 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 2)) {
872 			/* dr2 is enabled */
873 			if (dbregs->dr[2] >= VM_MAXUSER_ADDRESS)
874 				return (EINVAL);
875 		}
876 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 3)) {
877 			/* dr3 is enabled */
878 			if (dbregs->dr[3] >= VM_MAXUSER_ADDRESS)
879 				return (EINVAL);
880 		}
881 
882 		pcb->pcb_dr0 = dbregs->dr[0];
883 		pcb->pcb_dr1 = dbregs->dr[1];
884 		pcb->pcb_dr2 = dbregs->dr[2];
885 		pcb->pcb_dr3 = dbregs->dr[3];
886 		pcb->pcb_dr6 = dbregs->dr[6];
887 		pcb->pcb_dr7 = dbregs->dr[7];
888 
889 		set_pcb_flags(pcb, PCB_DBREGS);
890 	}
891 
892 	return (0);
893 }
894 
895 void
896 reset_dbregs(void)
897 {
898 
899 	load_dr7(0);	/* Turn off the control bits first */
900 	load_dr0(0);
901 	load_dr1(0);
902 	load_dr2(0);
903 	load_dr3(0);
904 	load_dr6(0);
905 }
906 
907 /*
908  * Return > 0 if a hardware breakpoint has been hit, and the
909  * breakpoint was in user space.  Return 0, otherwise.
910  */
911 int
912 user_dbreg_trap(register_t dr6)
913 {
914         u_int64_t dr7;
915         u_int64_t bp;       /* breakpoint bits extracted from dr6 */
916         int nbp;            /* number of breakpoints that triggered */
917         caddr_t addr[4];    /* breakpoint addresses */
918         int i;
919 
920         bp = dr6 & DBREG_DR6_BMASK;
921         if (bp == 0) {
922                 /*
923                  * None of the breakpoint bits are set meaning this
924                  * trap was not caused by any of the debug registers
925                  */
926                 return (0);
927         }
928 
929         dr7 = rdr7();
930         if ((dr7 & 0x000000ff) == 0) {
931                 /*
932                  * all GE and LE bits in the dr7 register are zero,
933                  * thus the trap couldn't have been caused by the
934                  * hardware debug registers
935                  */
936 		return (0);
937         }
938 
939         nbp = 0;
940 
941         /*
942          * at least one of the breakpoints were hit, check to see
943          * which ones and if any of them are user space addresses
944          */
945 
946         if (bp & 0x01) {
947                 addr[nbp++] = (caddr_t)rdr0();
948         }
949         if (bp & 0x02) {
950                 addr[nbp++] = (caddr_t)rdr1();
951         }
952         if (bp & 0x04) {
953                 addr[nbp++] = (caddr_t)rdr2();
954         }
955         if (bp & 0x08) {
956                 addr[nbp++] = (caddr_t)rdr3();
957         }
958 
959         for (i = 0; i < nbp; i++) {
960                 if (addr[i] < (caddr_t)VM_MAXUSER_ADDRESS) {
961                         /*
962                          * addr[i] is in user space
963                          */
964                         return (nbp);
965                 }
966         }
967 
968         /*
969          * None of the breakpoints are in user space.
970          */
971         return (0);
972 }
973