xref: /freebsd/sys/i386/i386/exec_machdep.c (revision d4f495fb)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 2018 The FreeBSD Foundation
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  * Portions of this software were developed by A. Joseph Koshy under
13  * sponsorship from the FreeBSD Foundation and Google, Inc.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgement:
25  *	This product includes software developed by the University of
26  *	California, Berkeley and its contributors.
27  * 4. Neither the name of the University nor the names of its contributors
28  *    may be used to endorse or promote products derived from this software
29  *    without specific prior written permission.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41  * SUCH DAMAGE.
42  *
43  *	from: @(#)machdep.c	7.4 (Berkeley) 6/3/91
44  */
45 
46 #include <sys/cdefs.h>
47 __FBSDID("$FreeBSD$");
48 
49 #include "opt_cpu.h"
50 #include "opt_ddb.h"
51 #include "opt_kstack_pages.h"
52 
53 #include <sys/param.h>
54 #include <sys/proc.h>
55 #include <sys/systm.h>
56 #include <sys/exec.h>
57 #include <sys/imgact.h>
58 #include <sys/kdb.h>
59 #include <sys/kernel.h>
60 #include <sys/ktr.h>
61 #include <sys/linker.h>
62 #include <sys/lock.h>
63 #include <sys/malloc.h>
64 #include <sys/mutex.h>
65 #include <sys/pcpu.h>
66 #include <sys/ptrace.h>
67 #include <sys/reg.h>
68 #include <sys/rwlock.h>
69 #include <sys/signalvar.h>
70 #include <sys/syscallsubr.h>
71 #include <sys/sysctl.h>
72 #include <sys/sysent.h>
73 #include <sys/sysproto.h>
74 #include <sys/ucontext.h>
75 #include <sys/vmmeter.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_object.h>
84 
85 #ifdef DDB
86 #ifndef KDB
87 #error KDB must be enabled in order for DDB to work!
88 #endif
89 #include <ddb/ddb.h>
90 #include <ddb/db_sym.h>
91 #endif
92 
93 #include <machine/cpu.h>
94 #include <machine/cputypes.h>
95 #include <machine/md_var.h>
96 #include <machine/pcb.h>
97 #include <machine/pcb_ext.h>
98 #include <machine/proc.h>
99 #include <machine/sigframe.h>
100 #include <machine/specialreg.h>
101 #include <machine/sysarch.h>
102 #include <machine/trap.h>
103 
104 static void fpstate_drop(struct thread *td);
105 static void get_fpcontext(struct thread *td, mcontext_t *mcp,
106     char *xfpusave, size_t xfpusave_len);
107 static int  set_fpcontext(struct thread *td, mcontext_t *mcp,
108     char *xfpustate, size_t xfpustate_len);
109 #ifdef COMPAT_43
110 static void osendsig(sig_t catcher, ksiginfo_t *, sigset_t *mask);
111 #endif
112 #ifdef COMPAT_FREEBSD4
113 static void freebsd4_sendsig(sig_t catcher, ksiginfo_t *, sigset_t *mask);
114 #endif
115 
116 extern struct sysentvec elf32_freebsd_sysvec;
117 
118 _Static_assert(sizeof(mcontext_t) == 640, "mcontext_t size incorrect");
119 _Static_assert(sizeof(ucontext_t) == 704, "ucontext_t size incorrect");
120 _Static_assert(sizeof(siginfo_t) == 64, "siginfo_t size incorrect");
121 
122 /*
123  * Send an interrupt to process.
124  *
125  * Stack is set up to allow sigcode stored at top to call routine,
126  * followed by call to sigreturn routine below.  After sigreturn
127  * resets the signal mask, the stack, and the frame pointer, it
128  * returns to the user specified pc, psl.
129  */
130 #ifdef COMPAT_43
131 static void
132 osendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
133 {
134 	struct osigframe sf, *fp;
135 	struct proc *p;
136 	struct thread *td;
137 	struct sigacts *psp;
138 	struct trapframe *regs;
139 	int sig;
140 	int oonstack;
141 
142 	td = curthread;
143 	p = td->td_proc;
144 	PROC_LOCK_ASSERT(p, MA_OWNED);
145 	sig = ksi->ksi_signo;
146 	psp = p->p_sigacts;
147 	mtx_assert(&psp->ps_mtx, MA_OWNED);
148 	regs = td->td_frame;
149 	oonstack = sigonstack(regs->tf_esp);
150 
151 	/* Allocate space for the signal handler context. */
152 	if ((td->td_pflags & TDP_ALTSTACK) && !oonstack &&
153 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
154 		fp = (struct osigframe *)((uintptr_t)td->td_sigstk.ss_sp +
155 		    td->td_sigstk.ss_size - sizeof(struct osigframe));
156 #if defined(COMPAT_43)
157 		td->td_sigstk.ss_flags |= SS_ONSTACK;
158 #endif
159 	} else
160 		fp = (struct osigframe *)regs->tf_esp - 1;
161 
162 	/* Build the argument list for the signal handler. */
163 	sf.sf_signum = sig;
164 	sf.sf_scp = (register_t)&fp->sf_siginfo.si_sc;
165 	bzero(&sf.sf_siginfo, sizeof(sf.sf_siginfo));
166 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
167 		/* Signal handler installed with SA_SIGINFO. */
168 		sf.sf_arg2 = (register_t)&fp->sf_siginfo;
169 		sf.sf_siginfo.si_signo = sig;
170 		sf.sf_siginfo.si_code = ksi->ksi_code;
171 		sf.sf_ahu.sf_action = (__osiginfohandler_t *)catcher;
172 		sf.sf_addr = 0;
173 	} else {
174 		/* Old FreeBSD-style arguments. */
175 		sf.sf_arg2 = ksi->ksi_code;
176 		sf.sf_addr = (register_t)ksi->ksi_addr;
177 		sf.sf_ahu.sf_handler = catcher;
178 	}
179 	mtx_unlock(&psp->ps_mtx);
180 	PROC_UNLOCK(p);
181 
182 	/* Save most if not all of trap frame. */
183 	sf.sf_siginfo.si_sc.sc_eax = regs->tf_eax;
184 	sf.sf_siginfo.si_sc.sc_ebx = regs->tf_ebx;
185 	sf.sf_siginfo.si_sc.sc_ecx = regs->tf_ecx;
186 	sf.sf_siginfo.si_sc.sc_edx = regs->tf_edx;
187 	sf.sf_siginfo.si_sc.sc_esi = regs->tf_esi;
188 	sf.sf_siginfo.si_sc.sc_edi = regs->tf_edi;
189 	sf.sf_siginfo.si_sc.sc_cs = regs->tf_cs;
190 	sf.sf_siginfo.si_sc.sc_ds = regs->tf_ds;
191 	sf.sf_siginfo.si_sc.sc_ss = regs->tf_ss;
192 	sf.sf_siginfo.si_sc.sc_es = regs->tf_es;
193 	sf.sf_siginfo.si_sc.sc_fs = regs->tf_fs;
194 	sf.sf_siginfo.si_sc.sc_gs = rgs();
195 	sf.sf_siginfo.si_sc.sc_isp = regs->tf_isp;
196 
197 	/* Build the signal context to be used by osigreturn(). */
198 	sf.sf_siginfo.si_sc.sc_onstack = (oonstack) ? 1 : 0;
199 	SIG2OSIG(*mask, sf.sf_siginfo.si_sc.sc_mask);
200 	sf.sf_siginfo.si_sc.sc_sp = regs->tf_esp;
201 	sf.sf_siginfo.si_sc.sc_fp = regs->tf_ebp;
202 	sf.sf_siginfo.si_sc.sc_pc = regs->tf_eip;
203 	sf.sf_siginfo.si_sc.sc_ps = regs->tf_eflags;
204 	sf.sf_siginfo.si_sc.sc_trapno = regs->tf_trapno;
205 	sf.sf_siginfo.si_sc.sc_err = regs->tf_err;
206 
207 	/*
208 	 * If we're a vm86 process, we want to save the segment registers.
209 	 * We also change eflags to be our emulated eflags, not the actual
210 	 * eflags.
211 	 */
212 	if (regs->tf_eflags & PSL_VM) {
213 		/* XXX confusing names: `tf' isn't a trapframe; `regs' is. */
214 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
215 		struct vm86_kernel *vm86 = &td->td_pcb->pcb_ext->ext_vm86;
216 
217 		sf.sf_siginfo.si_sc.sc_gs = tf->tf_vm86_gs;
218 		sf.sf_siginfo.si_sc.sc_fs = tf->tf_vm86_fs;
219 		sf.sf_siginfo.si_sc.sc_es = tf->tf_vm86_es;
220 		sf.sf_siginfo.si_sc.sc_ds = tf->tf_vm86_ds;
221 
222 		if (vm86->vm86_has_vme == 0)
223 			sf.sf_siginfo.si_sc.sc_ps =
224 			    (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
225 			    (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
226 
227 		/* See sendsig() for comments. */
228 		tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
229 	}
230 
231 	/*
232 	 * Copy the sigframe out to the user's stack.
233 	 */
234 	if (copyout(&sf, fp, sizeof(*fp)) != 0) {
235 		PROC_LOCK(p);
236 		sigexit(td, SIGILL);
237 	}
238 
239 	regs->tf_esp = (int)fp;
240 	if (p->p_sysent->sv_sigcode_base != 0) {
241 		regs->tf_eip = p->p_sysent->sv_sigcode_base + szsigcode -
242 		    szosigcode;
243 	} else {
244 		/* a.out sysentvec does not use shared page */
245 		regs->tf_eip = PROC_PS_STRINGS(p) - szosigcode;
246 	}
247 	regs->tf_eflags &= ~(PSL_T | PSL_D);
248 	regs->tf_cs = _ucodesel;
249 	regs->tf_ds = _udatasel;
250 	regs->tf_es = _udatasel;
251 	regs->tf_fs = _udatasel;
252 	load_gs(_udatasel);
253 	regs->tf_ss = _udatasel;
254 	PROC_LOCK(p);
255 	mtx_lock(&psp->ps_mtx);
256 }
257 #endif /* COMPAT_43 */
258 
259 #ifdef COMPAT_FREEBSD4
260 static void
261 freebsd4_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
262 {
263 	struct freebsd4_sigframe sf, *sfp;
264 	struct proc *p;
265 	struct thread *td;
266 	struct sigacts *psp;
267 	struct trapframe *regs;
268 	int sig;
269 	int oonstack;
270 
271 	td = curthread;
272 	p = td->td_proc;
273 	PROC_LOCK_ASSERT(p, MA_OWNED);
274 	sig = ksi->ksi_signo;
275 	psp = p->p_sigacts;
276 	mtx_assert(&psp->ps_mtx, MA_OWNED);
277 	regs = td->td_frame;
278 	oonstack = sigonstack(regs->tf_esp);
279 
280 	/* Save user context. */
281 	bzero(&sf, sizeof(sf));
282 	sf.sf_uc.uc_sigmask = *mask;
283 	sf.sf_uc.uc_stack = td->td_sigstk;
284 	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
285 	    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
286 	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
287 	sf.sf_uc.uc_mcontext.mc_gs = rgs();
288 	bcopy(regs, &sf.sf_uc.uc_mcontext.mc_fs, sizeof(*regs));
289 	bzero(sf.sf_uc.uc_mcontext.mc_fpregs,
290 	    sizeof(sf.sf_uc.uc_mcontext.mc_fpregs));
291 	bzero(sf.sf_uc.uc_mcontext.__spare__,
292 	    sizeof(sf.sf_uc.uc_mcontext.__spare__));
293 	bzero(sf.sf_uc.__spare__, sizeof(sf.sf_uc.__spare__));
294 
295 	/* Allocate space for the signal handler context. */
296 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
297 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
298 		sfp = (struct freebsd4_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
299 		    td->td_sigstk.ss_size - sizeof(struct freebsd4_sigframe));
300 #if defined(COMPAT_43)
301 		td->td_sigstk.ss_flags |= SS_ONSTACK;
302 #endif
303 	} else
304 		sfp = (struct freebsd4_sigframe *)regs->tf_esp - 1;
305 
306 	/* Build the argument list for the signal handler. */
307 	sf.sf_signum = sig;
308 	sf.sf_ucontext = (register_t)&sfp->sf_uc;
309 	bzero(&sf.sf_si, sizeof(sf.sf_si));
310 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
311 		/* Signal handler installed with SA_SIGINFO. */
312 		sf.sf_siginfo = (register_t)&sfp->sf_si;
313 		sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
314 
315 		/* Fill in POSIX parts */
316 		sf.sf_si.si_signo = sig;
317 		sf.sf_si.si_code = ksi->ksi_code;
318 		sf.sf_si.si_addr = ksi->ksi_addr;
319 	} else {
320 		/* Old FreeBSD-style arguments. */
321 		sf.sf_siginfo = ksi->ksi_code;
322 		sf.sf_addr = (register_t)ksi->ksi_addr;
323 		sf.sf_ahu.sf_handler = catcher;
324 	}
325 	mtx_unlock(&psp->ps_mtx);
326 	PROC_UNLOCK(p);
327 
328 	/*
329 	 * If we're a vm86 process, we want to save the segment registers.
330 	 * We also change eflags to be our emulated eflags, not the actual
331 	 * eflags.
332 	 */
333 	if (regs->tf_eflags & PSL_VM) {
334 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
335 		struct vm86_kernel *vm86 = &td->td_pcb->pcb_ext->ext_vm86;
336 
337 		sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
338 		sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
339 		sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
340 		sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
341 
342 		if (vm86->vm86_has_vme == 0)
343 			sf.sf_uc.uc_mcontext.mc_eflags =
344 			    (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
345 			    (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
346 
347 		/*
348 		 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
349 		 * syscalls made by the signal handler.  This just avoids
350 		 * wasting time for our lazy fixup of such faults.  PSL_NT
351 		 * does nothing in vm86 mode, but vm86 programs can set it
352 		 * almost legitimately in probes for old cpu types.
353 		 */
354 		tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
355 	}
356 
357 	/*
358 	 * Copy the sigframe out to the user's stack.
359 	 */
360 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
361 		PROC_LOCK(p);
362 		sigexit(td, SIGILL);
363 	}
364 
365 	regs->tf_esp = (int)sfp;
366 	regs->tf_eip = p->p_sysent->sv_sigcode_base + szsigcode -
367 	    szfreebsd4_sigcode;
368 	regs->tf_eflags &= ~(PSL_T | PSL_D);
369 	regs->tf_cs = _ucodesel;
370 	regs->tf_ds = _udatasel;
371 	regs->tf_es = _udatasel;
372 	regs->tf_fs = _udatasel;
373 	regs->tf_ss = _udatasel;
374 	PROC_LOCK(p);
375 	mtx_lock(&psp->ps_mtx);
376 }
377 #endif	/* COMPAT_FREEBSD4 */
378 
379 void
380 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
381 {
382 	struct sigframe sf, *sfp;
383 	struct proc *p;
384 	struct thread *td;
385 	struct sigacts *psp;
386 	char *sp;
387 	struct trapframe *regs;
388 	struct segment_descriptor *sdp;
389 	char *xfpusave;
390 	size_t xfpusave_len;
391 	int sig;
392 	int oonstack;
393 
394 	td = curthread;
395 	p = td->td_proc;
396 	PROC_LOCK_ASSERT(p, MA_OWNED);
397 	sig = ksi->ksi_signo;
398 	psp = p->p_sigacts;
399 	mtx_assert(&psp->ps_mtx, MA_OWNED);
400 #ifdef COMPAT_FREEBSD4
401 	if (SIGISMEMBER(psp->ps_freebsd4, sig)) {
402 		freebsd4_sendsig(catcher, ksi, mask);
403 		return;
404 	}
405 #endif
406 #ifdef COMPAT_43
407 	if (SIGISMEMBER(psp->ps_osigset, sig)) {
408 		osendsig(catcher, ksi, mask);
409 		return;
410 	}
411 #endif
412 	regs = td->td_frame;
413 	oonstack = sigonstack(regs->tf_esp);
414 
415 	if (cpu_max_ext_state_size > sizeof(union savefpu) && use_xsave) {
416 		xfpusave_len = cpu_max_ext_state_size - sizeof(union savefpu);
417 		xfpusave = __builtin_alloca(xfpusave_len);
418 	} else {
419 		xfpusave_len = 0;
420 		xfpusave = NULL;
421 	}
422 
423 	/* Save user context. */
424 	bzero(&sf, sizeof(sf));
425 	sf.sf_uc.uc_sigmask = *mask;
426 	sf.sf_uc.uc_stack = td->td_sigstk;
427 	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
428 	    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
429 	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
430 	sf.sf_uc.uc_mcontext.mc_gs = rgs();
431 	bcopy(regs, &sf.sf_uc.uc_mcontext.mc_fs, sizeof(*regs));
432 	sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */
433 	get_fpcontext(td, &sf.sf_uc.uc_mcontext, xfpusave, xfpusave_len);
434 	fpstate_drop(td);
435 	/*
436 	 * Unconditionally fill the fsbase and gsbase into the mcontext.
437 	 */
438 	sdp = &td->td_pcb->pcb_fsd;
439 	sf.sf_uc.uc_mcontext.mc_fsbase = sdp->sd_hibase << 24 |
440 	    sdp->sd_lobase;
441 	sdp = &td->td_pcb->pcb_gsd;
442 	sf.sf_uc.uc_mcontext.mc_gsbase = sdp->sd_hibase << 24 |
443 	    sdp->sd_lobase;
444 	bzero(sf.sf_uc.uc_mcontext.mc_spare2,
445 	    sizeof(sf.sf_uc.uc_mcontext.mc_spare2));
446 
447 	/* Allocate space for the signal handler context. */
448 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
449 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
450 		sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
451 #if defined(COMPAT_43)
452 		td->td_sigstk.ss_flags |= SS_ONSTACK;
453 #endif
454 	} else
455 		sp = (char *)regs->tf_esp - 128;
456 	if (xfpusave != NULL) {
457 		sp -= xfpusave_len;
458 		sp = (char *)((unsigned int)sp & ~0x3F);
459 		sf.sf_uc.uc_mcontext.mc_xfpustate = (register_t)sp;
460 	}
461 	sp -= sizeof(struct sigframe);
462 
463 	/* Align to 16 bytes. */
464 	sfp = (struct sigframe *)((unsigned int)sp & ~0xF);
465 
466 	/* Build the argument list for the signal handler. */
467 	sf.sf_signum = sig;
468 	sf.sf_ucontext = (register_t)&sfp->sf_uc;
469 	bzero(&sf.sf_si, sizeof(sf.sf_si));
470 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
471 		/* Signal handler installed with SA_SIGINFO. */
472 		sf.sf_siginfo = (register_t)&sfp->sf_si;
473 		sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
474 
475 		/* Fill in POSIX parts */
476 		sf.sf_si = ksi->ksi_info;
477 		sf.sf_si.si_signo = sig; /* maybe a translated signal */
478 	} else {
479 		/* Old FreeBSD-style arguments. */
480 		sf.sf_siginfo = ksi->ksi_code;
481 		sf.sf_addr = (register_t)ksi->ksi_addr;
482 		sf.sf_ahu.sf_handler = catcher;
483 	}
484 	mtx_unlock(&psp->ps_mtx);
485 	PROC_UNLOCK(p);
486 
487 	/*
488 	 * If we're a vm86 process, we want to save the segment registers.
489 	 * We also change eflags to be our emulated eflags, not the actual
490 	 * eflags.
491 	 */
492 	if (regs->tf_eflags & PSL_VM) {
493 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
494 		struct vm86_kernel *vm86 = &td->td_pcb->pcb_ext->ext_vm86;
495 
496 		sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
497 		sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
498 		sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
499 		sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
500 
501 		if (vm86->vm86_has_vme == 0)
502 			sf.sf_uc.uc_mcontext.mc_eflags =
503 			    (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
504 			    (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
505 
506 		/*
507 		 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
508 		 * syscalls made by the signal handler.  This just avoids
509 		 * wasting time for our lazy fixup of such faults.  PSL_NT
510 		 * does nothing in vm86 mode, but vm86 programs can set it
511 		 * almost legitimately in probes for old cpu types.
512 		 */
513 		tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
514 	}
515 
516 	/*
517 	 * Copy the sigframe out to the user's stack.
518 	 */
519 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0 ||
520 	    (xfpusave != NULL && copyout(xfpusave,
521 	    (void *)sf.sf_uc.uc_mcontext.mc_xfpustate, xfpusave_len)
522 	    != 0)) {
523 		PROC_LOCK(p);
524 		sigexit(td, SIGILL);
525 	}
526 
527 	regs->tf_esp = (int)sfp;
528 	regs->tf_eip = p->p_sysent->sv_sigcode_base;
529 	if (regs->tf_eip == 0)
530 		regs->tf_eip = PROC_PS_STRINGS(p) - szsigcode;
531 	regs->tf_eflags &= ~(PSL_T | PSL_D);
532 	regs->tf_cs = _ucodesel;
533 	regs->tf_ds = _udatasel;
534 	regs->tf_es = _udatasel;
535 	regs->tf_fs = _udatasel;
536 	regs->tf_ss = _udatasel;
537 	PROC_LOCK(p);
538 	mtx_lock(&psp->ps_mtx);
539 }
540 
541 /*
542  * System call to cleanup state after a signal has been taken.  Reset
543  * signal mask and stack state from context left by sendsig (above).
544  * Return to previous pc and psl as specified by context left by
545  * sendsig. Check carefully to make sure that the user has not
546  * modified the state to gain improper privileges.
547  */
548 #ifdef COMPAT_43
549 int
550 osigreturn(struct thread *td, struct osigreturn_args *uap)
551 {
552 	struct osigcontext sc;
553 	struct trapframe *regs;
554 	struct osigcontext *scp;
555 	int eflags, error;
556 	ksiginfo_t ksi;
557 
558 	regs = td->td_frame;
559 	error = copyin(uap->sigcntxp, &sc, sizeof(sc));
560 	if (error != 0)
561 		return (error);
562 	scp = &sc;
563 	eflags = scp->sc_ps;
564 	if (eflags & PSL_VM) {
565 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
566 		struct vm86_kernel *vm86;
567 
568 		/*
569 		 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
570 		 * set up the vm86 area, and we can't enter vm86 mode.
571 		 */
572 		if (td->td_pcb->pcb_ext == 0)
573 			return (EINVAL);
574 		vm86 = &td->td_pcb->pcb_ext->ext_vm86;
575 		if (vm86->vm86_inited == 0)
576 			return (EINVAL);
577 
578 		/* Go back to user mode if both flags are set. */
579 		if ((eflags & PSL_VIP) && (eflags & PSL_VIF)) {
580 			ksiginfo_init_trap(&ksi);
581 			ksi.ksi_signo = SIGBUS;
582 			ksi.ksi_code = BUS_OBJERR;
583 			ksi.ksi_addr = (void *)regs->tf_eip;
584 			trapsignal(td, &ksi);
585 		}
586 
587 		if (vm86->vm86_has_vme) {
588 			eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
589 			    (eflags & VME_USERCHANGE) | PSL_VM;
590 		} else {
591 			vm86->vm86_eflags = eflags;	/* save VIF, VIP */
592 			eflags = (tf->tf_eflags & ~VM_USERCHANGE) |
593 			    (eflags & VM_USERCHANGE) | PSL_VM;
594 		}
595 		tf->tf_vm86_ds = scp->sc_ds;
596 		tf->tf_vm86_es = scp->sc_es;
597 		tf->tf_vm86_fs = scp->sc_fs;
598 		tf->tf_vm86_gs = scp->sc_gs;
599 		tf->tf_ds = _udatasel;
600 		tf->tf_es = _udatasel;
601 		tf->tf_fs = _udatasel;
602 	} else {
603 		/*
604 		 * Don't allow users to change privileged or reserved flags.
605 		 */
606 		if (!EFL_SECURE(eflags, regs->tf_eflags)) {
607 			return (EINVAL);
608 		}
609 
610 		/*
611 		 * Don't allow users to load a valid privileged %cs.  Let the
612 		 * hardware check for invalid selectors, excess privilege in
613 		 * other selectors, invalid %eip's and invalid %esp's.
614 		 */
615 		if (!CS_SECURE(scp->sc_cs)) {
616 			ksiginfo_init_trap(&ksi);
617 			ksi.ksi_signo = SIGBUS;
618 			ksi.ksi_code = BUS_OBJERR;
619 			ksi.ksi_trapno = T_PROTFLT;
620 			ksi.ksi_addr = (void *)regs->tf_eip;
621 			trapsignal(td, &ksi);
622 			return (EINVAL);
623 		}
624 		regs->tf_ds = scp->sc_ds;
625 		regs->tf_es = scp->sc_es;
626 		regs->tf_fs = scp->sc_fs;
627 	}
628 
629 	/* Restore remaining registers. */
630 	regs->tf_eax = scp->sc_eax;
631 	regs->tf_ebx = scp->sc_ebx;
632 	regs->tf_ecx = scp->sc_ecx;
633 	regs->tf_edx = scp->sc_edx;
634 	regs->tf_esi = scp->sc_esi;
635 	regs->tf_edi = scp->sc_edi;
636 	regs->tf_cs = scp->sc_cs;
637 	regs->tf_ss = scp->sc_ss;
638 	regs->tf_isp = scp->sc_isp;
639 	regs->tf_ebp = scp->sc_fp;
640 	regs->tf_esp = scp->sc_sp;
641 	regs->tf_eip = scp->sc_pc;
642 	regs->tf_eflags = eflags;
643 
644 #if defined(COMPAT_43)
645 	if (scp->sc_onstack & 1)
646 		td->td_sigstk.ss_flags |= SS_ONSTACK;
647 	else
648 		td->td_sigstk.ss_flags &= ~SS_ONSTACK;
649 #endif
650 	kern_sigprocmask(td, SIG_SETMASK, (sigset_t *)&scp->sc_mask, NULL,
651 	    SIGPROCMASK_OLD);
652 	return (EJUSTRETURN);
653 }
654 #endif /* COMPAT_43 */
655 
656 #ifdef COMPAT_FREEBSD4
657 int
658 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
659 {
660 	struct freebsd4_ucontext uc;
661 	struct trapframe *regs;
662 	struct freebsd4_ucontext *ucp;
663 	int cs, eflags, error;
664 	ksiginfo_t ksi;
665 
666 	error = copyin(uap->sigcntxp, &uc, sizeof(uc));
667 	if (error != 0)
668 		return (error);
669 	ucp = &uc;
670 	regs = td->td_frame;
671 	eflags = ucp->uc_mcontext.mc_eflags;
672 	if (eflags & PSL_VM) {
673 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
674 		struct vm86_kernel *vm86;
675 
676 		/*
677 		 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
678 		 * set up the vm86 area, and we can't enter vm86 mode.
679 		 */
680 		if (td->td_pcb->pcb_ext == 0)
681 			return (EINVAL);
682 		vm86 = &td->td_pcb->pcb_ext->ext_vm86;
683 		if (vm86->vm86_inited == 0)
684 			return (EINVAL);
685 
686 		/* Go back to user mode if both flags are set. */
687 		if ((eflags & PSL_VIP) && (eflags & PSL_VIF)) {
688 			ksiginfo_init_trap(&ksi);
689 			ksi.ksi_signo = SIGBUS;
690 			ksi.ksi_code = BUS_OBJERR;
691 			ksi.ksi_addr = (void *)regs->tf_eip;
692 			trapsignal(td, &ksi);
693 		}
694 		if (vm86->vm86_has_vme) {
695 			eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
696 			    (eflags & VME_USERCHANGE) | PSL_VM;
697 		} else {
698 			vm86->vm86_eflags = eflags;	/* save VIF, VIP */
699 			eflags = (tf->tf_eflags & ~VM_USERCHANGE) |
700 			    (eflags & VM_USERCHANGE) | PSL_VM;
701 		}
702 		bcopy(&ucp->uc_mcontext.mc_fs, tf, sizeof(struct trapframe));
703 		tf->tf_eflags = eflags;
704 		tf->tf_vm86_ds = tf->tf_ds;
705 		tf->tf_vm86_es = tf->tf_es;
706 		tf->tf_vm86_fs = tf->tf_fs;
707 		tf->tf_vm86_gs = ucp->uc_mcontext.mc_gs;
708 		tf->tf_ds = _udatasel;
709 		tf->tf_es = _udatasel;
710 		tf->tf_fs = _udatasel;
711 	} else {
712 		/*
713 		 * Don't allow users to change privileged or reserved flags.
714 		 */
715 		if (!EFL_SECURE(eflags, regs->tf_eflags)) {
716 			uprintf(
717 			    "pid %d (%s): freebsd4_sigreturn eflags = 0x%x\n",
718 			    td->td_proc->p_pid, td->td_name, eflags);
719 			return (EINVAL);
720 		}
721 
722 		/*
723 		 * Don't allow users to load a valid privileged %cs.  Let the
724 		 * hardware check for invalid selectors, excess privilege in
725 		 * other selectors, invalid %eip's and invalid %esp's.
726 		 */
727 		cs = ucp->uc_mcontext.mc_cs;
728 		if (!CS_SECURE(cs)) {
729 			uprintf("pid %d (%s): freebsd4_sigreturn cs = 0x%x\n",
730 			    td->td_proc->p_pid, td->td_name, cs);
731 			ksiginfo_init_trap(&ksi);
732 			ksi.ksi_signo = SIGBUS;
733 			ksi.ksi_code = BUS_OBJERR;
734 			ksi.ksi_trapno = T_PROTFLT;
735 			ksi.ksi_addr = (void *)regs->tf_eip;
736 			trapsignal(td, &ksi);
737 			return (EINVAL);
738 		}
739 
740 		bcopy(&ucp->uc_mcontext.mc_fs, regs, sizeof(*regs));
741 	}
742 
743 #if defined(COMPAT_43)
744 	if (ucp->uc_mcontext.mc_onstack & 1)
745 		td->td_sigstk.ss_flags |= SS_ONSTACK;
746 	else
747 		td->td_sigstk.ss_flags &= ~SS_ONSTACK;
748 #endif
749 	kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0);
750 	return (EJUSTRETURN);
751 }
752 #endif	/* COMPAT_FREEBSD4 */
753 
754 int
755 sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
756 {
757 	ucontext_t uc;
758 	struct proc *p;
759 	struct trapframe *regs;
760 	ucontext_t *ucp;
761 	char *xfpustate;
762 	size_t xfpustate_len;
763 	int cs, eflags, error, ret;
764 	ksiginfo_t ksi;
765 
766 	p = td->td_proc;
767 
768 	error = copyin(uap->sigcntxp, &uc, sizeof(uc));
769 	if (error != 0)
770 		return (error);
771 	ucp = &uc;
772 	if ((ucp->uc_mcontext.mc_flags & ~_MC_FLAG_MASK) != 0) {
773 		uprintf("pid %d (%s): sigreturn mc_flags %x\n", p->p_pid,
774 		    td->td_name, ucp->uc_mcontext.mc_flags);
775 		return (EINVAL);
776 	}
777 	regs = td->td_frame;
778 	eflags = ucp->uc_mcontext.mc_eflags;
779 	if (eflags & PSL_VM) {
780 		struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
781 		struct vm86_kernel *vm86;
782 
783 		/*
784 		 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
785 		 * set up the vm86 area, and we can't enter vm86 mode.
786 		 */
787 		if (td->td_pcb->pcb_ext == 0)
788 			return (EINVAL);
789 		vm86 = &td->td_pcb->pcb_ext->ext_vm86;
790 		if (vm86->vm86_inited == 0)
791 			return (EINVAL);
792 
793 		/* Go back to user mode if both flags are set. */
794 		if ((eflags & PSL_VIP) && (eflags & PSL_VIF)) {
795 			ksiginfo_init_trap(&ksi);
796 			ksi.ksi_signo = SIGBUS;
797 			ksi.ksi_code = BUS_OBJERR;
798 			ksi.ksi_addr = (void *)regs->tf_eip;
799 			trapsignal(td, &ksi);
800 		}
801 
802 		if (vm86->vm86_has_vme) {
803 			eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
804 			    (eflags & VME_USERCHANGE) | PSL_VM;
805 		} else {
806 			vm86->vm86_eflags = eflags;	/* save VIF, VIP */
807 			eflags = (tf->tf_eflags & ~VM_USERCHANGE) |
808 			    (eflags & VM_USERCHANGE) | PSL_VM;
809 		}
810 		bcopy(&ucp->uc_mcontext.mc_fs, tf, sizeof(struct trapframe));
811 		tf->tf_eflags = eflags;
812 		tf->tf_vm86_ds = tf->tf_ds;
813 		tf->tf_vm86_es = tf->tf_es;
814 		tf->tf_vm86_fs = tf->tf_fs;
815 		tf->tf_vm86_gs = ucp->uc_mcontext.mc_gs;
816 		tf->tf_ds = _udatasel;
817 		tf->tf_es = _udatasel;
818 		tf->tf_fs = _udatasel;
819 	} else {
820 		/*
821 		 * Don't allow users to change privileged or reserved flags.
822 		 */
823 		if (!EFL_SECURE(eflags, regs->tf_eflags)) {
824 			uprintf("pid %d (%s): sigreturn eflags = 0x%x\n",
825 			    td->td_proc->p_pid, td->td_name, eflags);
826 			return (EINVAL);
827 		}
828 
829 		/*
830 		 * Don't allow users to load a valid privileged %cs.  Let the
831 		 * hardware check for invalid selectors, excess privilege in
832 		 * other selectors, invalid %eip's and invalid %esp's.
833 		 */
834 		cs = ucp->uc_mcontext.mc_cs;
835 		if (!CS_SECURE(cs)) {
836 			uprintf("pid %d (%s): sigreturn cs = 0x%x\n",
837 			    td->td_proc->p_pid, td->td_name, cs);
838 			ksiginfo_init_trap(&ksi);
839 			ksi.ksi_signo = SIGBUS;
840 			ksi.ksi_code = BUS_OBJERR;
841 			ksi.ksi_trapno = T_PROTFLT;
842 			ksi.ksi_addr = (void *)regs->tf_eip;
843 			trapsignal(td, &ksi);
844 			return (EINVAL);
845 		}
846 
847 		if ((uc.uc_mcontext.mc_flags & _MC_HASFPXSTATE) != 0) {
848 			xfpustate_len = uc.uc_mcontext.mc_xfpustate_len;
849 			if (xfpustate_len > cpu_max_ext_state_size -
850 			    sizeof(union savefpu)) {
851 				uprintf(
852 			    "pid %d (%s): sigreturn xfpusave_len = 0x%zx\n",
853 				    p->p_pid, td->td_name, xfpustate_len);
854 				return (EINVAL);
855 			}
856 			xfpustate = __builtin_alloca(xfpustate_len);
857 			error = copyin(
858 			    (const void *)uc.uc_mcontext.mc_xfpustate,
859 			    xfpustate, xfpustate_len);
860 			if (error != 0) {
861 				uprintf(
862 	"pid %d (%s): sigreturn copying xfpustate failed\n",
863 				    p->p_pid, td->td_name);
864 				return (error);
865 			}
866 		} else {
867 			xfpustate = NULL;
868 			xfpustate_len = 0;
869 		}
870 		ret = set_fpcontext(td, &ucp->uc_mcontext, xfpustate,
871 		    xfpustate_len);
872 		if (ret != 0)
873 			return (ret);
874 		bcopy(&ucp->uc_mcontext.mc_fs, regs, sizeof(*regs));
875 	}
876 
877 #if defined(COMPAT_43)
878 	if (ucp->uc_mcontext.mc_onstack & 1)
879 		td->td_sigstk.ss_flags |= SS_ONSTACK;
880 	else
881 		td->td_sigstk.ss_flags &= ~SS_ONSTACK;
882 #endif
883 
884 	kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0);
885 	return (EJUSTRETURN);
886 }
887 
888 /*
889  * Reset the hardware debug registers if they were in use.
890  * They won't have any meaning for the newly exec'd process.
891  */
892 void
893 x86_clear_dbregs(struct pcb *pcb)
894 {
895 	if ((pcb->pcb_flags & PCB_DBREGS) == 0)
896 		return;
897 
898 	pcb->pcb_dr0 = 0;
899 	pcb->pcb_dr1 = 0;
900 	pcb->pcb_dr2 = 0;
901 	pcb->pcb_dr3 = 0;
902 	pcb->pcb_dr6 = 0;
903 	pcb->pcb_dr7 = 0;
904 
905 	if (pcb == curpcb) {
906 		/*
907 		 * Clear the debug registers on the running CPU,
908 		 * otherwise they will end up affecting the next
909 		 * process we switch to.
910 		 */
911 		reset_dbregs();
912 	}
913 	pcb->pcb_flags &= ~PCB_DBREGS;
914 }
915 
916 #ifdef COMPAT_43
917 static void
918 setup_priv_lcall_gate(struct proc *p)
919 {
920 	struct i386_ldt_args uap;
921 	union descriptor desc;
922 	u_int lcall_addr;
923 
924 	bzero(&uap, sizeof(uap));
925 	uap.start = 0;
926 	uap.num = 1;
927 	lcall_addr = p->p_sysent->sv_psstrings - sz_lcall_tramp;
928 	bzero(&desc, sizeof(desc));
929 	desc.sd.sd_type = SDT_MEMERA;
930 	desc.sd.sd_dpl = SEL_UPL;
931 	desc.sd.sd_p = 1;
932 	desc.sd.sd_def32 = 1;
933 	desc.sd.sd_gran = 1;
934 	desc.sd.sd_lolimit = 0xffff;
935 	desc.sd.sd_hilimit = 0xf;
936 	desc.sd.sd_lobase = lcall_addr;
937 	desc.sd.sd_hibase = lcall_addr >> 24;
938 	i386_set_ldt(curthread, &uap, &desc);
939 }
940 #endif
941 
942 /*
943  * Reset registers to default values on exec.
944  */
945 void
946 exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack)
947 {
948 	struct trapframe *regs;
949 	struct pcb *pcb;
950 	register_t saved_eflags;
951 
952 	regs = td->td_frame;
953 	pcb = td->td_pcb;
954 
955 	/* Reset pc->pcb_gs and %gs before possibly invalidating it. */
956 	pcb->pcb_gs = _udatasel;
957 	load_gs(_udatasel);
958 
959 	mtx_lock_spin(&dt_lock);
960 	if (td->td_proc->p_md.md_ldt != NULL)
961 		user_ldt_free(td);
962 	else
963 		mtx_unlock_spin(&dt_lock);
964 
965 #ifdef COMPAT_43
966 	if (td->td_proc->p_sysent->sv_psstrings !=
967 	    elf32_freebsd_sysvec.sv_psstrings)
968 		setup_priv_lcall_gate(td->td_proc);
969 #endif
970 
971 	/*
972 	 * Reset the fs and gs bases.  The values from the old address
973 	 * space do not make sense for the new program.  In particular,
974 	 * gsbase might be the TLS base for the old program but the new
975 	 * program has no TLS now.
976 	 */
977 	set_fsbase(td, 0);
978 	set_gsbase(td, 0);
979 
980 	/* Make sure edx is 0x0 on entry. Linux binaries depend on it. */
981 	saved_eflags = regs->tf_eflags & PSL_T;
982 	bzero((char *)regs, sizeof(struct trapframe));
983 	regs->tf_eip = imgp->entry_addr;
984 	regs->tf_esp = stack;
985 	regs->tf_eflags = PSL_USER | saved_eflags;
986 	regs->tf_ss = _udatasel;
987 	regs->tf_ds = _udatasel;
988 	regs->tf_es = _udatasel;
989 	regs->tf_fs = _udatasel;
990 	regs->tf_cs = _ucodesel;
991 
992 	/* PS_STRINGS value for BSD/OS binaries.  It is 0 for non-BSD/OS. */
993 	regs->tf_ebx = (register_t)imgp->ps_strings;
994 
995 	x86_clear_dbregs(pcb);
996 
997 	pcb->pcb_initial_npxcw = __INITIAL_NPXCW__;
998 
999 	/*
1000 	 * Drop the FP state if we hold it, so that the process gets a
1001 	 * clean FP state if it uses the FPU again.
1002 	 */
1003 	fpstate_drop(td);
1004 }
1005 
1006 int
1007 fill_regs(struct thread *td, struct reg *regs)
1008 {
1009 	struct pcb *pcb;
1010 	struct trapframe *tp;
1011 
1012 	tp = td->td_frame;
1013 	pcb = td->td_pcb;
1014 	regs->r_gs = pcb->pcb_gs;
1015 	return (fill_frame_regs(tp, regs));
1016 }
1017 
1018 int
1019 fill_frame_regs(struct trapframe *tp, struct reg *regs)
1020 {
1021 
1022 	regs->r_fs = tp->tf_fs;
1023 	regs->r_es = tp->tf_es;
1024 	regs->r_ds = tp->tf_ds;
1025 	regs->r_edi = tp->tf_edi;
1026 	regs->r_esi = tp->tf_esi;
1027 	regs->r_ebp = tp->tf_ebp;
1028 	regs->r_ebx = tp->tf_ebx;
1029 	regs->r_edx = tp->tf_edx;
1030 	regs->r_ecx = tp->tf_ecx;
1031 	regs->r_eax = tp->tf_eax;
1032 	regs->r_eip = tp->tf_eip;
1033 	regs->r_cs = tp->tf_cs;
1034 	regs->r_eflags = tp->tf_eflags;
1035 	regs->r_esp = tp->tf_esp;
1036 	regs->r_ss = tp->tf_ss;
1037 	regs->r_err = 0;
1038 	regs->r_trapno = 0;
1039 	return (0);
1040 }
1041 
1042 int
1043 set_regs(struct thread *td, struct reg *regs)
1044 {
1045 	struct pcb *pcb;
1046 	struct trapframe *tp;
1047 
1048 	tp = td->td_frame;
1049 	if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
1050 	    !CS_SECURE(regs->r_cs))
1051 		return (EINVAL);
1052 	pcb = td->td_pcb;
1053 	tp->tf_fs = regs->r_fs;
1054 	tp->tf_es = regs->r_es;
1055 	tp->tf_ds = regs->r_ds;
1056 	tp->tf_edi = regs->r_edi;
1057 	tp->tf_esi = regs->r_esi;
1058 	tp->tf_ebp = regs->r_ebp;
1059 	tp->tf_ebx = regs->r_ebx;
1060 	tp->tf_edx = regs->r_edx;
1061 	tp->tf_ecx = regs->r_ecx;
1062 	tp->tf_eax = regs->r_eax;
1063 	tp->tf_eip = regs->r_eip;
1064 	tp->tf_cs = regs->r_cs;
1065 	tp->tf_eflags = regs->r_eflags;
1066 	tp->tf_esp = regs->r_esp;
1067 	tp->tf_ss = regs->r_ss;
1068 	pcb->pcb_gs = regs->r_gs;
1069 	return (0);
1070 }
1071 
1072 int
1073 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1074 {
1075 
1076 	KASSERT(td == curthread || TD_IS_SUSPENDED(td) ||
1077 	    P_SHOULDSTOP(td->td_proc),
1078 	    ("not suspended thread %p", td));
1079 	npxgetregs(td);
1080 	if (cpu_fxsr)
1081 		npx_fill_fpregs_xmm(&get_pcb_user_save_td(td)->sv_xmm,
1082 		    (struct save87 *)fpregs);
1083 	else
1084 		bcopy(&get_pcb_user_save_td(td)->sv_87, fpregs,
1085 		    sizeof(*fpregs));
1086 	return (0);
1087 }
1088 
1089 int
1090 set_fpregs(struct thread *td, struct fpreg *fpregs)
1091 {
1092 
1093 	critical_enter();
1094 	if (cpu_fxsr)
1095 		npx_set_fpregs_xmm((struct save87 *)fpregs,
1096 		    &get_pcb_user_save_td(td)->sv_xmm);
1097 	else
1098 		bcopy(fpregs, &get_pcb_user_save_td(td)->sv_87,
1099 		    sizeof(*fpregs));
1100 	npxuserinited(td);
1101 	critical_exit();
1102 	return (0);
1103 }
1104 
1105 /*
1106  * Get machine context.
1107  */
1108 int
1109 get_mcontext(struct thread *td, mcontext_t *mcp, int flags)
1110 {
1111 	struct trapframe *tp;
1112 	struct segment_descriptor *sdp;
1113 
1114 	tp = td->td_frame;
1115 
1116 	PROC_LOCK(curthread->td_proc);
1117 	mcp->mc_onstack = sigonstack(tp->tf_esp);
1118 	PROC_UNLOCK(curthread->td_proc);
1119 	mcp->mc_gs = td->td_pcb->pcb_gs;
1120 	mcp->mc_fs = tp->tf_fs;
1121 	mcp->mc_es = tp->tf_es;
1122 	mcp->mc_ds = tp->tf_ds;
1123 	mcp->mc_edi = tp->tf_edi;
1124 	mcp->mc_esi = tp->tf_esi;
1125 	mcp->mc_ebp = tp->tf_ebp;
1126 	mcp->mc_isp = tp->tf_isp;
1127 	mcp->mc_eflags = tp->tf_eflags;
1128 	if (flags & GET_MC_CLEAR_RET) {
1129 		mcp->mc_eax = 0;
1130 		mcp->mc_edx = 0;
1131 		mcp->mc_eflags &= ~PSL_C;
1132 	} else {
1133 		mcp->mc_eax = tp->tf_eax;
1134 		mcp->mc_edx = tp->tf_edx;
1135 	}
1136 	mcp->mc_ebx = tp->tf_ebx;
1137 	mcp->mc_ecx = tp->tf_ecx;
1138 	mcp->mc_eip = tp->tf_eip;
1139 	mcp->mc_cs = tp->tf_cs;
1140 	mcp->mc_esp = tp->tf_esp;
1141 	mcp->mc_ss = tp->tf_ss;
1142 	mcp->mc_len = sizeof(*mcp);
1143 	get_fpcontext(td, mcp, NULL, 0);
1144 	sdp = &td->td_pcb->pcb_fsd;
1145 	mcp->mc_fsbase = sdp->sd_hibase << 24 | sdp->sd_lobase;
1146 	sdp = &td->td_pcb->pcb_gsd;
1147 	mcp->mc_gsbase = sdp->sd_hibase << 24 | sdp->sd_lobase;
1148 	mcp->mc_flags = 0;
1149 	mcp->mc_xfpustate = 0;
1150 	mcp->mc_xfpustate_len = 0;
1151 	bzero(mcp->mc_spare2, sizeof(mcp->mc_spare2));
1152 	return (0);
1153 }
1154 
1155 /*
1156  * Set machine context.
1157  *
1158  * However, we don't set any but the user modifiable flags, and we won't
1159  * touch the cs selector.
1160  */
1161 int
1162 set_mcontext(struct thread *td, mcontext_t *mcp)
1163 {
1164 	struct trapframe *tp;
1165 	char *xfpustate;
1166 	int eflags, ret;
1167 
1168 	tp = td->td_frame;
1169 	if (mcp->mc_len != sizeof(*mcp) ||
1170 	    (mcp->mc_flags & ~_MC_FLAG_MASK) != 0)
1171 		return (EINVAL);
1172 	eflags = (mcp->mc_eflags & PSL_USERCHANGE) |
1173 	    (tp->tf_eflags & ~PSL_USERCHANGE);
1174 	if (mcp->mc_flags & _MC_HASFPXSTATE) {
1175 		if (mcp->mc_xfpustate_len > cpu_max_ext_state_size -
1176 		    sizeof(union savefpu))
1177 			return (EINVAL);
1178 		xfpustate = __builtin_alloca(mcp->mc_xfpustate_len);
1179 		ret = copyin((void *)mcp->mc_xfpustate, xfpustate,
1180 		    mcp->mc_xfpustate_len);
1181 		if (ret != 0)
1182 			return (ret);
1183 	} else
1184 		xfpustate = NULL;
1185 	ret = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len);
1186 	if (ret != 0)
1187 		return (ret);
1188 	tp->tf_fs = mcp->mc_fs;
1189 	tp->tf_es = mcp->mc_es;
1190 	tp->tf_ds = mcp->mc_ds;
1191 	tp->tf_edi = mcp->mc_edi;
1192 	tp->tf_esi = mcp->mc_esi;
1193 	tp->tf_ebp = mcp->mc_ebp;
1194 	tp->tf_ebx = mcp->mc_ebx;
1195 	tp->tf_edx = mcp->mc_edx;
1196 	tp->tf_ecx = mcp->mc_ecx;
1197 	tp->tf_eax = mcp->mc_eax;
1198 	tp->tf_eip = mcp->mc_eip;
1199 	tp->tf_eflags = eflags;
1200 	tp->tf_esp = mcp->mc_esp;
1201 	tp->tf_ss = mcp->mc_ss;
1202 	td->td_pcb->pcb_gs = mcp->mc_gs;
1203 	return (0);
1204 }
1205 
1206 static void
1207 get_fpcontext(struct thread *td, mcontext_t *mcp, char *xfpusave,
1208     size_t xfpusave_len)
1209 {
1210 	size_t max_len, len;
1211 
1212 	mcp->mc_ownedfp = npxgetregs(td);
1213 	bcopy(get_pcb_user_save_td(td), &mcp->mc_fpstate[0],
1214 	    sizeof(mcp->mc_fpstate));
1215 	mcp->mc_fpformat = npxformat();
1216 	if (!use_xsave || xfpusave_len == 0)
1217 		return;
1218 	max_len = cpu_max_ext_state_size - sizeof(union savefpu);
1219 	len = xfpusave_len;
1220 	if (len > max_len) {
1221 		len = max_len;
1222 		bzero(xfpusave + max_len, len - max_len);
1223 	}
1224 	mcp->mc_flags |= _MC_HASFPXSTATE;
1225 	mcp->mc_xfpustate_len = len;
1226 	bcopy(get_pcb_user_save_td(td) + 1, xfpusave, len);
1227 }
1228 
1229 static int
1230 set_fpcontext(struct thread *td, mcontext_t *mcp, char *xfpustate,
1231     size_t xfpustate_len)
1232 {
1233 	int error;
1234 
1235 	if (mcp->mc_fpformat == _MC_FPFMT_NODEV)
1236 		return (0);
1237 	else if (mcp->mc_fpformat != _MC_FPFMT_387 &&
1238 	    mcp->mc_fpformat != _MC_FPFMT_XMM)
1239 		return (EINVAL);
1240 	else if (mcp->mc_ownedfp == _MC_FPOWNED_NONE) {
1241 		/* We don't care what state is left in the FPU or PCB. */
1242 		fpstate_drop(td);
1243 		error = 0;
1244 	} else if (mcp->mc_ownedfp == _MC_FPOWNED_FPU ||
1245 	    mcp->mc_ownedfp == _MC_FPOWNED_PCB) {
1246 		error = npxsetregs(td, (union savefpu *)&mcp->mc_fpstate,
1247 		    xfpustate, xfpustate_len);
1248 	} else
1249 		return (EINVAL);
1250 	return (error);
1251 }
1252 
1253 static void
1254 fpstate_drop(struct thread *td)
1255 {
1256 
1257 	KASSERT(PCB_USER_FPU(td->td_pcb), ("fpstate_drop: kernel-owned fpu"));
1258 	critical_enter();
1259 	if (PCPU_GET(fpcurthread) == td)
1260 		npxdrop();
1261 	/*
1262 	 * XXX force a full drop of the npx.  The above only drops it if we
1263 	 * owned it.  npxgetregs() has the same bug in the !cpu_fxsr case.
1264 	 *
1265 	 * XXX I don't much like npxgetregs()'s semantics of doing a full
1266 	 * drop.  Dropping only to the pcb matches fnsave's behaviour.
1267 	 * We only need to drop to !PCB_INITDONE in sendsig().  But
1268 	 * sendsig() is the only caller of npxgetregs()... perhaps we just
1269 	 * have too many layers.
1270 	 */
1271 	curthread->td_pcb->pcb_flags &= ~(PCB_NPXINITDONE |
1272 	    PCB_NPXUSERINITDONE);
1273 	critical_exit();
1274 }
1275 
1276 int
1277 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1278 {
1279 	struct pcb *pcb;
1280 
1281 	if (td == NULL) {
1282 		dbregs->dr[0] = rdr0();
1283 		dbregs->dr[1] = rdr1();
1284 		dbregs->dr[2] = rdr2();
1285 		dbregs->dr[3] = rdr3();
1286 		dbregs->dr[6] = rdr6();
1287 		dbregs->dr[7] = rdr7();
1288 	} else {
1289 		pcb = td->td_pcb;
1290 		dbregs->dr[0] = pcb->pcb_dr0;
1291 		dbregs->dr[1] = pcb->pcb_dr1;
1292 		dbregs->dr[2] = pcb->pcb_dr2;
1293 		dbregs->dr[3] = pcb->pcb_dr3;
1294 		dbregs->dr[6] = pcb->pcb_dr6;
1295 		dbregs->dr[7] = pcb->pcb_dr7;
1296 	}
1297 	dbregs->dr[4] = 0;
1298 	dbregs->dr[5] = 0;
1299 	return (0);
1300 }
1301 
1302 int
1303 set_dbregs(struct thread *td, struct dbreg *dbregs)
1304 {
1305 	struct pcb *pcb;
1306 	int i;
1307 
1308 	if (td == NULL) {
1309 		load_dr0(dbregs->dr[0]);
1310 		load_dr1(dbregs->dr[1]);
1311 		load_dr2(dbregs->dr[2]);
1312 		load_dr3(dbregs->dr[3]);
1313 		load_dr6(dbregs->dr[6]);
1314 		load_dr7(dbregs->dr[7]);
1315 	} else {
1316 		/*
1317 		 * Don't let an illegal value for dr7 get set.	Specifically,
1318 		 * check for undefined settings.  Setting these bit patterns
1319 		 * result in undefined behaviour and can lead to an unexpected
1320 		 * TRCTRAP.
1321 		 */
1322 		for (i = 0; i < 4; i++) {
1323 			if (DBREG_DR7_ACCESS(dbregs->dr[7], i) == 0x02)
1324 				return (EINVAL);
1325 			if (DBREG_DR7_LEN(dbregs->dr[7], i) == 0x02)
1326 				return (EINVAL);
1327 		}
1328 
1329 		pcb = td->td_pcb;
1330 
1331 		/*
1332 		 * Don't let a process set a breakpoint that is not within the
1333 		 * process's address space.  If a process could do this, it
1334 		 * could halt the system by setting a breakpoint in the kernel
1335 		 * (if ddb was enabled).  Thus, we need to check to make sure
1336 		 * that no breakpoints are being enabled for addresses outside
1337 		 * process's address space.
1338 		 *
1339 		 * XXX - what about when the watched area of the user's
1340 		 * address space is written into from within the kernel
1341 		 * ... wouldn't that still cause a breakpoint to be generated
1342 		 * from within kernel mode?
1343 		 */
1344 
1345 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 0)) {
1346 			/* dr0 is enabled */
1347 			if (dbregs->dr[0] >= VM_MAXUSER_ADDRESS)
1348 				return (EINVAL);
1349 		}
1350 
1351 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 1)) {
1352 			/* dr1 is enabled */
1353 			if (dbregs->dr[1] >= VM_MAXUSER_ADDRESS)
1354 				return (EINVAL);
1355 		}
1356 
1357 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 2)) {
1358 			/* dr2 is enabled */
1359 			if (dbregs->dr[2] >= VM_MAXUSER_ADDRESS)
1360 				return (EINVAL);
1361 		}
1362 
1363 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 3)) {
1364 			/* dr3 is enabled */
1365 			if (dbregs->dr[3] >= VM_MAXUSER_ADDRESS)
1366 				return (EINVAL);
1367 		}
1368 
1369 		pcb->pcb_dr0 = dbregs->dr[0];
1370 		pcb->pcb_dr1 = dbregs->dr[1];
1371 		pcb->pcb_dr2 = dbregs->dr[2];
1372 		pcb->pcb_dr3 = dbregs->dr[3];
1373 		pcb->pcb_dr6 = dbregs->dr[6];
1374 		pcb->pcb_dr7 = dbregs->dr[7];
1375 
1376 		pcb->pcb_flags |= PCB_DBREGS;
1377 	}
1378 
1379 	return (0);
1380 }
1381 
1382 /*
1383  * Return > 0 if a hardware breakpoint has been hit, and the
1384  * breakpoint was in user space.  Return 0, otherwise.
1385  */
1386 int
1387 user_dbreg_trap(register_t dr6)
1388 {
1389 	u_int32_t dr7;
1390 	u_int32_t bp;       /* breakpoint bits extracted from dr6 */
1391 	int nbp;            /* number of breakpoints that triggered */
1392 	caddr_t addr[4];    /* breakpoint addresses */
1393 	int i;
1394 
1395 	bp = dr6 & DBREG_DR6_BMASK;
1396 	if (bp == 0) {
1397 		/*
1398 		 * None of the breakpoint bits are set meaning this
1399 		 * trap was not caused by any of the debug registers
1400 		 */
1401 		return (0);
1402 	}
1403 
1404 	dr7 = rdr7();
1405 	if ((dr7 & 0x000000ff) == 0) {
1406 		/*
1407 		 * all GE and LE bits in the dr7 register are zero,
1408 		 * thus the trap couldn't have been caused by the
1409 		 * hardware debug registers
1410 		 */
1411 		return (0);
1412 	}
1413 
1414 	nbp = 0;
1415 
1416 	/*
1417 	 * at least one of the breakpoints were hit, check to see
1418 	 * which ones and if any of them are user space addresses
1419 	 */
1420 
1421 	if (bp & 0x01) {
1422 		addr[nbp++] = (caddr_t)rdr0();
1423 	}
1424 	if (bp & 0x02) {
1425 		addr[nbp++] = (caddr_t)rdr1();
1426 	}
1427 	if (bp & 0x04) {
1428 		addr[nbp++] = (caddr_t)rdr2();
1429 	}
1430 	if (bp & 0x08) {
1431 		addr[nbp++] = (caddr_t)rdr3();
1432 	}
1433 
1434 	for (i = 0; i < nbp; i++) {
1435 		if (addr[i] < (caddr_t)VM_MAXUSER_ADDRESS) {
1436 			/*
1437 			 * addr[i] is in user space
1438 			 */
1439 			return (nbp);
1440 		}
1441 	}
1442 
1443 	/*
1444 	 * None of the breakpoints are in user space.
1445 	 */
1446 	return (0);
1447 }
1448