xref: /freebsd/sys/amd64/amd64/exec_machdep.c (revision eea3e4dd9703a252509d75814049aa8da5007ebb)
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 
41 #include <sys/cdefs.h>
42 #include "opt_cpu.h"
43 #include "opt_ddb.h"
44 #include "opt_kstack_pages.h"
45 
46 #include <sys/param.h>
47 #include <sys/proc.h>
48 #include <sys/systm.h>
49 #include <sys/exec.h>
50 #include <sys/imgact.h>
51 #include <sys/kdb.h>
52 #include <sys/kernel.h>
53 #include <sys/ktr.h>
54 #include <sys/linker.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
58 #include <sys/pcpu.h>
59 #include <sys/reg.h>
60 #include <sys/rwlock.h>
61 #include <sys/signalvar.h>
62 #ifdef SMP
63 #include <sys/smp.h>
64 #endif
65 #include <sys/syscallsubr.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/sysproto.h>
69 #include <sys/ucontext.h>
70 #include <sys/vmmeter.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/vm_extern.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 
78 #ifdef DDB
79 #ifndef KDB
80 #error KDB must be enabled in order for DDB to work!
81 #endif
82 #include <ddb/ddb.h>
83 #include <ddb/db_sym.h>
84 #endif
85 
86 #include <machine/vmparam.h>
87 #include <machine/frame.h>
88 #include <machine/md_var.h>
89 #include <machine/pcb.h>
90 #include <machine/proc.h>
91 #include <machine/sigframe.h>
92 #include <machine/specialreg.h>
93 #include <machine/trap.h>
94 
95 _Static_assert(sizeof(mcontext_t) == 800, "mcontext_t size incorrect");
96 _Static_assert(sizeof(ucontext_t) == 880, "ucontext_t size incorrect");
97 _Static_assert(sizeof(siginfo_t) == 80, "siginfo_t size incorrect");
98 
99 /*
100  * Send an interrupt to process.
101  *
102  * Stack is set up to allow sigcode stored at top to call routine,
103  * followed by call to sigreturn routine below.  After sigreturn
104  * resets the signal mask, the stack, and the frame pointer, it
105  * returns to the user specified pc, psl.
106  */
107 void
sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)108 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
109 {
110 	struct sigframe sf, *sfp;
111 	struct pcb *pcb;
112 	struct proc *p;
113 	struct thread *td;
114 	struct sigacts *psp;
115 	char *sp;
116 	struct trapframe *regs;
117 	char *xfpusave;
118 	size_t xfpusave_len;
119 	int sig;
120 	int oonstack;
121 
122 	td = curthread;
123 	pcb = td->td_pcb;
124 	p = td->td_proc;
125 	PROC_LOCK_ASSERT(p, MA_OWNED);
126 	sig = ksi->ksi_signo;
127 	psp = p->p_sigacts;
128 	mtx_assert(&psp->ps_mtx, MA_OWNED);
129 	regs = td->td_frame;
130 	oonstack = sigonstack(regs->tf_rsp);
131 
132 	/* Save user context. */
133 	bzero(&sf, sizeof(sf));
134 	sf.sf_uc.uc_sigmask = *mask;
135 	sf.sf_uc.uc_stack = td->td_sigstk;
136 	sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
137 	    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
138 	sf.sf_uc.uc_mcontext.mc_onstack = (oonstack) ? 1 : 0;
139 	bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(*regs));
140 	sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); /* magic */
141 	get_fpcontext(td, &sf.sf_uc.uc_mcontext, &xfpusave, &xfpusave_len);
142 	update_pcb_bases(pcb);
143 	sf.sf_uc.uc_mcontext.mc_fsbase = pcb->pcb_fsbase;
144 	sf.sf_uc.uc_mcontext.mc_gsbase = pcb->pcb_gsbase;
145 	bzero(sf.sf_uc.uc_mcontext.mc_spare,
146 	    sizeof(sf.sf_uc.uc_mcontext.mc_spare));
147 
148 	/* Allocate space for the signal handler context. */
149 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
150 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
151 		sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
152 #if defined(COMPAT_43)
153 		td->td_sigstk.ss_flags |= SS_ONSTACK;
154 #endif
155 	} else
156 		sp = (char *)regs->tf_rsp - 128;
157 	if (xfpusave != NULL) {
158 		sp -= xfpusave_len;
159 		sp = (char *)((unsigned long)sp & ~0x3Ful);
160 		sf.sf_uc.uc_mcontext.mc_xfpustate = (register_t)sp;
161 	}
162 	sp -= sizeof(struct sigframe);
163 	/* Align to 16 bytes. */
164 	sfp = (struct sigframe *)((unsigned long)sp & ~0xFul);
165 
166 	/* Build the argument list for the signal handler. */
167 	regs->tf_rdi = sig;			/* arg 1 in %rdi */
168 	regs->tf_rdx = (register_t)&sfp->sf_uc;	/* arg 3 in %rdx */
169 	bzero(&sf.sf_si, sizeof(sf.sf_si));
170 	if (SIGISMEMBER(psp->ps_siginfo, sig)) {
171 		/* Signal handler installed with SA_SIGINFO. */
172 		regs->tf_rsi = (register_t)&sfp->sf_si;	/* arg 2 in %rsi */
173 		sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
174 
175 		/* Fill in POSIX parts */
176 		sf.sf_si = ksi->ksi_info;
177 		sf.sf_si.si_signo = sig; /* maybe a translated signal */
178 		regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */
179 	} else {
180 		/* Old FreeBSD-style arguments. */
181 		regs->tf_rsi = ksi->ksi_code;	/* arg 2 in %rsi */
182 		regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */
183 		sf.sf_ahu.sf_handler = catcher;
184 	}
185 	mtx_unlock(&psp->ps_mtx);
186 	PROC_UNLOCK(p);
187 
188 	/*
189 	 * Copy the sigframe out to the user's stack.
190 	 */
191 	if (copyout(&sf, sfp, sizeof(*sfp)) != 0 ||
192 	    (xfpusave != NULL && copyout(xfpusave,
193 	    (void *)sf.sf_uc.uc_mcontext.mc_xfpustate, xfpusave_len)
194 	    != 0)) {
195 		uprintf("pid %d comm %s has trashed its stack, killing\n",
196 		    p->p_pid, p->p_comm);
197 		PROC_LOCK(p);
198 		sigexit(td, SIGILL);
199 	}
200 
201 	fpstate_drop(td);
202 	regs->tf_rsp = (long)sfp;
203 	regs->tf_rip = PROC_SIGCODE(p);
204 	regs->tf_rflags &= ~(PSL_T | PSL_D);
205 	regs->tf_cs = _ucodesel;
206 	regs->tf_ds = _udatasel;
207 	regs->tf_ss = _udatasel;
208 	regs->tf_es = _udatasel;
209 	regs->tf_fs = _ufssel;
210 	regs->tf_gs = _ugssel;
211 	regs->tf_flags = TF_HASSEGS;
212 	if ((pcb->pcb_flags & PCB_TLSBASE) != 0)
213 		pcb->pcb_fsbase = pcb->pcb_tlsbase;
214 	PROC_LOCK(p);
215 	mtx_lock(&psp->ps_mtx);
216 }
217 
218 /*
219  * System call to cleanup state after a signal
220  * has been taken.  Reset signal mask and
221  * stack state from context left by sendsig (above).
222  * Return to previous pc and psl as specified by
223  * context left by sendsig. Check carefully to
224  * make sure that the user has not modified the
225  * state to gain improper privileges.
226  */
227 int
sys_sigreturn(struct thread * td,struct sigreturn_args * uap)228 sys_sigreturn(struct thread *td, struct sigreturn_args *uap)
229 {
230 	ucontext_t uc;
231 	struct pcb *pcb;
232 	struct proc *p;
233 	struct trapframe *regs;
234 	ucontext_t *ucp;
235 	char *xfpustate;
236 	size_t xfpustate_len;
237 	long rflags;
238 	int cs, error, ret;
239 	ksiginfo_t ksi;
240 
241 	pcb = td->td_pcb;
242 	p = td->td_proc;
243 
244 	error = copyin(uap->sigcntxp, &uc, sizeof(uc));
245 	if (error != 0) {
246 		uprintf("pid %d (%s): sigreturn copyin failed\n",
247 		    p->p_pid, td->td_name);
248 		return (error);
249 	}
250 	ucp = &uc;
251 	if ((ucp->uc_mcontext.mc_flags & ~_MC_FLAG_MASK) != 0) {
252 		uprintf("pid %d (%s): sigreturn mc_flags %x\n", p->p_pid,
253 		    td->td_name, ucp->uc_mcontext.mc_flags);
254 		return (EINVAL);
255 	}
256 	regs = td->td_frame;
257 	rflags = ucp->uc_mcontext.mc_rflags;
258 	/*
259 	 * Don't allow users to change privileged or reserved flags.
260 	 */
261 	if (!EFL_SECURE(rflags, regs->tf_rflags)) {
262 		uprintf("pid %d (%s): sigreturn rflags = 0x%lx\n", p->p_pid,
263 		    td->td_name, rflags);
264 		return (EINVAL);
265 	}
266 
267 	/*
268 	 * Don't allow users to load a valid privileged %cs.  Let the
269 	 * hardware check for invalid selectors, excess privilege in
270 	 * other selectors, invalid %eip's and invalid %esp's.
271 	 */
272 	cs = ucp->uc_mcontext.mc_cs;
273 	if (!CS_SECURE(cs)) {
274 		uprintf("pid %d (%s): sigreturn cs = 0x%x\n", p->p_pid,
275 		    td->td_name, cs);
276 		ksiginfo_init_trap(&ksi);
277 		ksi.ksi_signo = SIGBUS;
278 		ksi.ksi_code = BUS_OBJERR;
279 		ksi.ksi_trapno = T_PROTFLT;
280 		ksi.ksi_addr = (void *)regs->tf_rip;
281 		trapsignal(td, &ksi);
282 		return (EINVAL);
283 	}
284 
285 	if ((uc.uc_mcontext.mc_flags & _MC_HASFPXSTATE) != 0) {
286 		xfpustate_len = uc.uc_mcontext.mc_xfpustate_len;
287 		if (xfpustate_len > cpu_max_ext_state_size -
288 		    sizeof(struct savefpu)) {
289 			uprintf("pid %d (%s): sigreturn xfpusave_len = 0x%zx\n",
290 			    p->p_pid, td->td_name, xfpustate_len);
291 			return (EINVAL);
292 		}
293 		xfpustate = (char *)fpu_save_area_alloc();
294 		error = copyin((const void *)uc.uc_mcontext.mc_xfpustate,
295 		    xfpustate, xfpustate_len);
296 		if (error != 0) {
297 			fpu_save_area_free((struct savefpu *)xfpustate);
298 			uprintf(
299 	"pid %d (%s): sigreturn copying xfpustate failed\n",
300 			    p->p_pid, td->td_name);
301 			return (error);
302 		}
303 	} else {
304 		xfpustate = NULL;
305 		xfpustate_len = 0;
306 	}
307 	ret = set_fpcontext(td, &ucp->uc_mcontext, xfpustate, xfpustate_len);
308 	fpu_save_area_free((struct savefpu *)xfpustate);
309 	if (ret != 0) {
310 		uprintf("pid %d (%s): sigreturn set_fpcontext err %d\n",
311 		    p->p_pid, td->td_name, ret);
312 		return (ret);
313 	}
314 	bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(*regs));
315 	update_pcb_bases(pcb);
316 	pcb->pcb_fsbase = ucp->uc_mcontext.mc_fsbase;
317 	pcb->pcb_gsbase = ucp->uc_mcontext.mc_gsbase;
318 
319 #if defined(COMPAT_43)
320 	if (ucp->uc_mcontext.mc_onstack & 1)
321 		td->td_sigstk.ss_flags |= SS_ONSTACK;
322 	else
323 		td->td_sigstk.ss_flags &= ~SS_ONSTACK;
324 #endif
325 
326 	kern_sigprocmask(td, SIG_SETMASK, &ucp->uc_sigmask, NULL, 0);
327 	return (EJUSTRETURN);
328 }
329 
330 #ifdef COMPAT_FREEBSD4
331 int
freebsd4_sigreturn(struct thread * td,struct freebsd4_sigreturn_args * uap)332 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
333 {
334 
335 	return sys_sigreturn(td, (struct sigreturn_args *)uap);
336 }
337 #endif
338 
339 /*
340  * Reset the hardware debug registers if they were in use.
341  * They won't have any meaning for the newly exec'd process.
342  */
343 void
x86_clear_dbregs(struct pcb * pcb)344 x86_clear_dbregs(struct pcb *pcb)
345 {
346 	if ((pcb->pcb_flags & PCB_DBREGS) == 0)
347 		return;
348 
349 	pcb->pcb_dr0 = 0;
350 	pcb->pcb_dr1 = 0;
351 	pcb->pcb_dr2 = 0;
352 	pcb->pcb_dr3 = 0;
353 	pcb->pcb_dr6 = 0;
354 	pcb->pcb_dr7 = 0;
355 
356 	if (pcb == curpcb) {
357 		/*
358 		 * Clear the debug registers on the running CPU,
359 		 * otherwise they will end up affecting the next
360 		 * process we switch to.
361 		 */
362 		reset_dbregs();
363 	}
364 	clear_pcb_flags(pcb, PCB_DBREGS);
365 }
366 
367 /*
368  * Reset registers to default values on exec.
369  */
370 void
exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)371 exec_setregs(struct thread *td, struct image_params *imgp, uintptr_t stack)
372 {
373 	struct trapframe *regs;
374 	struct pcb *pcb;
375 	register_t saved_rflags;
376 
377 	regs = td->td_frame;
378 	pcb = td->td_pcb;
379 
380 	if (td->td_proc->p_md.md_ldt != NULL)
381 		user_ldt_free(td);
382 
383 	update_pcb_bases(pcb);
384 	pcb->pcb_fsbase = pcb->pcb_tlsbase = 0;
385 	pcb->pcb_gsbase = 0;
386 	clear_pcb_flags(pcb, PCB_32BIT | PCB_TLSBASE);
387 	pcb->pcb_initial_fpucw = __INITIAL_FPUCW__;
388 
389 	saved_rflags = regs->tf_rflags & PSL_T;
390 	bzero((char *)regs, sizeof(struct trapframe));
391 	regs->tf_rip = imgp->entry_addr;
392 	regs->tf_rsp = ((stack - 8) & ~0xFul) + 8;
393 	regs->tf_rdi = stack;		/* argv */
394 	regs->tf_rflags = PSL_USER | saved_rflags;
395 	regs->tf_ss = _udatasel;
396 	regs->tf_cs = _ucodesel;
397 	regs->tf_ds = _udatasel;
398 	regs->tf_es = _udatasel;
399 	regs->tf_fs = _ufssel;
400 	regs->tf_gs = _ugssel;
401 	regs->tf_flags = TF_HASSEGS;
402 
403 	x86_clear_dbregs(pcb);
404 
405 	/*
406 	 * Drop the FP state if we hold it, so that the process gets a
407 	 * clean FP state if it uses the FPU again.
408 	 */
409 	fpstate_drop(td);
410 }
411 
412 int
fill_regs(struct thread * td,struct reg * regs)413 fill_regs(struct thread *td, struct reg *regs)
414 {
415 	struct trapframe *tp;
416 
417 	tp = td->td_frame;
418 	return (fill_frame_regs(tp, regs));
419 }
420 
421 int
fill_frame_regs(struct trapframe * tp,struct reg * regs)422 fill_frame_regs(struct trapframe *tp, struct reg *regs)
423 {
424 
425 	regs->r_r15 = tp->tf_r15;
426 	regs->r_r14 = tp->tf_r14;
427 	regs->r_r13 = tp->tf_r13;
428 	regs->r_r12 = tp->tf_r12;
429 	regs->r_r11 = tp->tf_r11;
430 	regs->r_r10 = tp->tf_r10;
431 	regs->r_r9  = tp->tf_r9;
432 	regs->r_r8  = tp->tf_r8;
433 	regs->r_rdi = tp->tf_rdi;
434 	regs->r_rsi = tp->tf_rsi;
435 	regs->r_rbp = tp->tf_rbp;
436 	regs->r_rbx = tp->tf_rbx;
437 	regs->r_rdx = tp->tf_rdx;
438 	regs->r_rcx = tp->tf_rcx;
439 	regs->r_rax = tp->tf_rax;
440 	regs->r_rip = tp->tf_rip;
441 	regs->r_cs = tp->tf_cs;
442 	regs->r_rflags = tp->tf_rflags;
443 	regs->r_rsp = tp->tf_rsp;
444 	regs->r_ss = tp->tf_ss;
445 	if (tp->tf_flags & TF_HASSEGS) {
446 		regs->r_ds = tp->tf_ds;
447 		regs->r_es = tp->tf_es;
448 		regs->r_fs = tp->tf_fs;
449 		regs->r_gs = tp->tf_gs;
450 	} else {
451 		regs->r_ds = 0;
452 		regs->r_es = 0;
453 		regs->r_fs = 0;
454 		regs->r_gs = 0;
455 	}
456 	regs->r_err = 0;
457 	regs->r_trapno = 0;
458 	return (0);
459 }
460 
461 int
set_regs(struct thread * td,struct reg * regs)462 set_regs(struct thread *td, struct reg *regs)
463 {
464 	struct trapframe *tp;
465 	register_t rflags;
466 
467 	tp = td->td_frame;
468 	rflags = regs->r_rflags & 0xffffffff;
469 	if (!EFL_SECURE(rflags, tp->tf_rflags) || !CS_SECURE(regs->r_cs))
470 		return (EINVAL);
471 	tp->tf_r15 = regs->r_r15;
472 	tp->tf_r14 = regs->r_r14;
473 	tp->tf_r13 = regs->r_r13;
474 	tp->tf_r12 = regs->r_r12;
475 	tp->tf_r11 = regs->r_r11;
476 	tp->tf_r10 = regs->r_r10;
477 	tp->tf_r9  = regs->r_r9;
478 	tp->tf_r8  = regs->r_r8;
479 	tp->tf_rdi = regs->r_rdi;
480 	tp->tf_rsi = regs->r_rsi;
481 	tp->tf_rbp = regs->r_rbp;
482 	tp->tf_rbx = regs->r_rbx;
483 	tp->tf_rdx = regs->r_rdx;
484 	tp->tf_rcx = regs->r_rcx;
485 	tp->tf_rax = regs->r_rax;
486 	tp->tf_rip = regs->r_rip;
487 	tp->tf_cs = regs->r_cs;
488 	tp->tf_rflags = rflags;
489 	tp->tf_rsp = regs->r_rsp;
490 	tp->tf_ss = regs->r_ss;
491 	if (0) {	/* XXXKIB */
492 		tp->tf_ds = regs->r_ds;
493 		tp->tf_es = regs->r_es;
494 		tp->tf_fs = regs->r_fs;
495 		tp->tf_gs = regs->r_gs;
496 		tp->tf_flags = TF_HASSEGS;
497 	}
498 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
499 	return (0);
500 }
501 
502 /* XXX check all this stuff! */
503 /* externalize from sv_xmm */
504 static void
fill_fpregs_xmm(struct savefpu * sv_xmm,struct fpreg * fpregs)505 fill_fpregs_xmm(struct savefpu *sv_xmm, struct fpreg *fpregs)
506 {
507 	struct envxmm *penv_fpreg = (struct envxmm *)&fpregs->fpr_env;
508 	struct envxmm *penv_xmm = &sv_xmm->sv_env;
509 	int i;
510 
511 	/* pcb -> fpregs */
512 	bzero(fpregs, sizeof(*fpregs));
513 
514 	/* FPU control/status */
515 	penv_fpreg->en_cw = penv_xmm->en_cw;
516 	penv_fpreg->en_sw = penv_xmm->en_sw;
517 	penv_fpreg->en_tw = penv_xmm->en_tw;
518 	penv_fpreg->en_opcode = penv_xmm->en_opcode;
519 	penv_fpreg->en_rip = penv_xmm->en_rip;
520 	penv_fpreg->en_rdp = penv_xmm->en_rdp;
521 	penv_fpreg->en_mxcsr = penv_xmm->en_mxcsr;
522 	penv_fpreg->en_mxcsr_mask = penv_xmm->en_mxcsr_mask;
523 
524 	/* FPU registers */
525 	for (i = 0; i < 8; ++i)
526 		bcopy(sv_xmm->sv_fp[i].fp_acc.fp_bytes, fpregs->fpr_acc[i], 10);
527 
528 	/* SSE registers */
529 	for (i = 0; i < 16; ++i)
530 		bcopy(sv_xmm->sv_xmm[i].xmm_bytes, fpregs->fpr_xacc[i], 16);
531 }
532 
533 /* internalize from fpregs into sv_xmm */
534 static void
set_fpregs_xmm(struct fpreg * fpregs,struct savefpu * sv_xmm)535 set_fpregs_xmm(struct fpreg *fpregs, struct savefpu *sv_xmm)
536 {
537 	struct envxmm *penv_xmm = &sv_xmm->sv_env;
538 	struct envxmm *penv_fpreg = (struct envxmm *)&fpregs->fpr_env;
539 	int i;
540 
541 	/* fpregs -> pcb */
542 	/* FPU control/status */
543 	penv_xmm->en_cw = penv_fpreg->en_cw;
544 	penv_xmm->en_sw = penv_fpreg->en_sw;
545 	penv_xmm->en_tw = penv_fpreg->en_tw;
546 	penv_xmm->en_opcode = penv_fpreg->en_opcode;
547 	penv_xmm->en_rip = penv_fpreg->en_rip;
548 	penv_xmm->en_rdp = penv_fpreg->en_rdp;
549 	penv_xmm->en_mxcsr = penv_fpreg->en_mxcsr;
550 	penv_xmm->en_mxcsr_mask = penv_fpreg->en_mxcsr_mask & cpu_mxcsr_mask;
551 
552 	/* FPU registers */
553 	for (i = 0; i < 8; ++i)
554 		bcopy(fpregs->fpr_acc[i], sv_xmm->sv_fp[i].fp_acc.fp_bytes, 10);
555 
556 	/* SSE registers */
557 	for (i = 0; i < 16; ++i)
558 		bcopy(fpregs->fpr_xacc[i], sv_xmm->sv_xmm[i].xmm_bytes, 16);
559 }
560 
561 /* externalize from td->pcb */
562 int
fill_fpregs(struct thread * td,struct fpreg * fpregs)563 fill_fpregs(struct thread *td, struct fpreg *fpregs)
564 {
565 
566 	KASSERT(td == curthread || TD_IS_SUSPENDED(td) ||
567 	    P_SHOULDSTOP(td->td_proc),
568 	    ("not suspended thread %p", td));
569 	fpugetregs(td);
570 	fill_fpregs_xmm(get_pcb_user_save_td(td), fpregs);
571 	return (0);
572 }
573 
574 /* internalize to td->pcb */
575 int
set_fpregs(struct thread * td,struct fpreg * fpregs)576 set_fpregs(struct thread *td, struct fpreg *fpregs)
577 {
578 
579 	critical_enter();
580 	set_fpregs_xmm(fpregs, get_pcb_user_save_td(td));
581 	fpuuserinited(td);
582 	critical_exit();
583 	return (0);
584 }
585 
586 /*
587  * Get machine context.
588  */
589 int
get_mcontext(struct thread * td,mcontext_t * mcp,int flags)590 get_mcontext(struct thread *td, mcontext_t *mcp, int flags)
591 {
592 	struct pcb *pcb;
593 	struct trapframe *tp;
594 
595 	pcb = td->td_pcb;
596 	tp = td->td_frame;
597 	PROC_LOCK(curthread->td_proc);
598 	mcp->mc_onstack = sigonstack(tp->tf_rsp);
599 	PROC_UNLOCK(curthread->td_proc);
600 	mcp->mc_r15 = tp->tf_r15;
601 	mcp->mc_r14 = tp->tf_r14;
602 	mcp->mc_r13 = tp->tf_r13;
603 	mcp->mc_r12 = tp->tf_r12;
604 	mcp->mc_r11 = tp->tf_r11;
605 	mcp->mc_r10 = tp->tf_r10;
606 	mcp->mc_r9  = tp->tf_r9;
607 	mcp->mc_r8  = tp->tf_r8;
608 	mcp->mc_rdi = tp->tf_rdi;
609 	mcp->mc_rsi = tp->tf_rsi;
610 	mcp->mc_rbp = tp->tf_rbp;
611 	mcp->mc_rbx = tp->tf_rbx;
612 	mcp->mc_rcx = tp->tf_rcx;
613 	mcp->mc_rflags = tp->tf_rflags;
614 	if (flags & GET_MC_CLEAR_RET) {
615 		mcp->mc_rax = 0;
616 		mcp->mc_rdx = 0;
617 		mcp->mc_rflags &= ~PSL_C;
618 	} else {
619 		mcp->mc_rax = tp->tf_rax;
620 		mcp->mc_rdx = tp->tf_rdx;
621 	}
622 	mcp->mc_rip = tp->tf_rip;
623 	mcp->mc_cs = tp->tf_cs;
624 	mcp->mc_rsp = tp->tf_rsp;
625 	mcp->mc_ss = tp->tf_ss;
626 	mcp->mc_ds = tp->tf_ds;
627 	mcp->mc_es = tp->tf_es;
628 	mcp->mc_fs = tp->tf_fs;
629 	mcp->mc_gs = tp->tf_gs;
630 	mcp->mc_flags = tp->tf_flags;
631 	mcp->mc_len = sizeof(*mcp);
632 	get_fpcontext(td, mcp, NULL, NULL);
633 	update_pcb_bases(pcb);
634 	mcp->mc_fsbase = pcb->pcb_fsbase;
635 	mcp->mc_gsbase = pcb->pcb_gsbase;
636 	mcp->mc_xfpustate = 0;
637 	mcp->mc_xfpustate_len = 0;
638 	mcp->mc_tlsbase = (pcb->pcb_flags & PCB_TLSBASE) != 0 ?
639 	    pcb->pcb_tlsbase : 0;
640 	bzero(mcp->mc_spare, sizeof(mcp->mc_spare));
641 	return (0);
642 }
643 
644 /*
645  * Set machine context.
646  *
647  * However, we don't set any but the user modifiable flags, and we won't
648  * touch the cs selector.
649  */
650 int
set_mcontext(struct thread * td,mcontext_t * mcp)651 set_mcontext(struct thread *td, mcontext_t *mcp)
652 {
653 	struct pcb *pcb;
654 	struct trapframe *tp;
655 	char *xfpustate;
656 	long rflags;
657 	int ret;
658 
659 	pcb = td->td_pcb;
660 	tp = td->td_frame;
661 	if (mcp->mc_len != sizeof(*mcp) ||
662 	    (mcp->mc_flags & ~_MC_FLAG_MASK) != 0)
663 		return (EINVAL);
664 	rflags = (mcp->mc_rflags & PSL_USERCHANGE) |
665 	    (tp->tf_rflags & ~PSL_USERCHANGE);
666 	if (mcp->mc_flags & _MC_HASFPXSTATE) {
667 		if (mcp->mc_xfpustate_len > cpu_max_ext_state_size -
668 		    sizeof(struct savefpu))
669 			return (EINVAL);
670 		xfpustate = (char *)fpu_save_area_alloc();
671 		ret = copyin((void *)mcp->mc_xfpustate, xfpustate,
672 		    mcp->mc_xfpustate_len);
673 		if (ret != 0) {
674 			fpu_save_area_free((struct savefpu *)xfpustate);
675 			return (ret);
676 		}
677 	} else
678 		xfpustate = NULL;
679 	ret = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len);
680 	fpu_save_area_free((struct savefpu *)xfpustate);
681 	if (ret != 0)
682 		return (ret);
683 	tp->tf_r15 = mcp->mc_r15;
684 	tp->tf_r14 = mcp->mc_r14;
685 	tp->tf_r13 = mcp->mc_r13;
686 	tp->tf_r12 = mcp->mc_r12;
687 	tp->tf_r11 = mcp->mc_r11;
688 	tp->tf_r10 = mcp->mc_r10;
689 	tp->tf_r9  = mcp->mc_r9;
690 	tp->tf_r8  = mcp->mc_r8;
691 	tp->tf_rdi = mcp->mc_rdi;
692 	tp->tf_rsi = mcp->mc_rsi;
693 	tp->tf_rbp = mcp->mc_rbp;
694 	tp->tf_rbx = mcp->mc_rbx;
695 	tp->tf_rdx = mcp->mc_rdx;
696 	tp->tf_rcx = mcp->mc_rcx;
697 	tp->tf_rax = mcp->mc_rax;
698 	tp->tf_rip = mcp->mc_rip;
699 	tp->tf_rflags = rflags;
700 	tp->tf_rsp = mcp->mc_rsp;
701 	tp->tf_ss = mcp->mc_ss;
702 	tp->tf_flags = mcp->mc_flags;
703 	if (tp->tf_flags & TF_HASSEGS) {
704 		tp->tf_ds = mcp->mc_ds;
705 		tp->tf_es = mcp->mc_es;
706 		tp->tf_fs = mcp->mc_fs;
707 		tp->tf_gs = mcp->mc_gs;
708 	}
709 	set_pcb_flags(pcb, PCB_FULL_IRET);
710 	if (mcp->mc_flags & _MC_HASBASES) {
711 		pcb->pcb_fsbase = mcp->mc_fsbase;
712 		pcb->pcb_gsbase = mcp->mc_gsbase;
713 	}
714 	if ((mcp->mc_flags & _MC_HASTLSBASE) != 0) {
715 		pcb->pcb_tlsbase = mcp->mc_tlsbase;
716 		set_pcb_flags(pcb, PCB_TLSBASE);
717 	}
718 	return (0);
719 }
720 
721 void
get_fpcontext(struct thread * td,mcontext_t * mcp,char ** xfpusave,size_t * xfpusave_len)722 get_fpcontext(struct thread *td, mcontext_t *mcp, char **xfpusave,
723     size_t *xfpusave_len)
724 {
725 	mcp->mc_ownedfp = fpugetregs(td);
726 	bcopy(get_pcb_user_save_td(td), &mcp->mc_fpstate[0],
727 	    sizeof(mcp->mc_fpstate));
728 	mcp->mc_fpformat = fpuformat();
729 	if (xfpusave == NULL)
730 		return;
731 	if (!use_xsave || cpu_max_ext_state_size <= sizeof(struct savefpu)) {
732 		*xfpusave_len = 0;
733 		*xfpusave = NULL;
734 	} else {
735 		mcp->mc_flags |= _MC_HASFPXSTATE;
736 		*xfpusave_len = mcp->mc_xfpustate_len =
737 		    cpu_max_ext_state_size - sizeof(struct savefpu);
738 		*xfpusave = (char *)(get_pcb_user_save_td(td) + 1);
739 	}
740 }
741 
742 int
set_fpcontext(struct thread * td,mcontext_t * mcp,char * xfpustate,size_t xfpustate_len)743 set_fpcontext(struct thread *td, mcontext_t *mcp, char *xfpustate,
744     size_t xfpustate_len)
745 {
746 	int error;
747 
748 	if (mcp->mc_fpformat == _MC_FPFMT_NODEV)
749 		return (0);
750 	else if (mcp->mc_fpformat != _MC_FPFMT_XMM)
751 		return (EINVAL);
752 	else if (mcp->mc_ownedfp == _MC_FPOWNED_NONE) {
753 		/* We don't care what state is left in the FPU or PCB. */
754 		fpstate_drop(td);
755 		error = 0;
756 	} else if (mcp->mc_ownedfp == _MC_FPOWNED_FPU ||
757 	    mcp->mc_ownedfp == _MC_FPOWNED_PCB) {
758 		error = fpusetregs(td, (struct savefpu *)&mcp->mc_fpstate,
759 		    xfpustate, xfpustate_len);
760 	} else
761 		return (EINVAL);
762 	return (error);
763 }
764 
765 void
fpstate_drop(struct thread * td)766 fpstate_drop(struct thread *td)
767 {
768 
769 	KASSERT(PCB_USER_FPU(td->td_pcb), ("fpstate_drop: kernel-owned fpu"));
770 	critical_enter();
771 	if (PCPU_GET(fpcurthread) == td)
772 		fpudrop();
773 	/*
774 	 * XXX force a full drop of the fpu.  The above only drops it if we
775 	 * owned it.
776 	 *
777 	 * XXX I don't much like fpugetuserregs()'s semantics of doing a full
778 	 * drop.  Dropping only to the pcb matches fnsave's behaviour.
779 	 * We only need to drop to !PCB_INITDONE in sendsig().  But
780 	 * sendsig() is the only caller of fpugetuserregs()... perhaps we just
781 	 * have too many layers.
782 	 */
783 	clear_pcb_flags(curthread->td_pcb,
784 	    PCB_FPUINITDONE | PCB_USERFPUINITDONE);
785 	critical_exit();
786 }
787 
788 int
fill_dbregs(struct thread * td,struct dbreg * dbregs)789 fill_dbregs(struct thread *td, struct dbreg *dbregs)
790 {
791 	struct pcb *pcb;
792 
793 	if (td == NULL) {
794 		dbregs->dr[0] = rdr0();
795 		dbregs->dr[1] = rdr1();
796 		dbregs->dr[2] = rdr2();
797 		dbregs->dr[3] = rdr3();
798 		dbregs->dr[6] = rdr6();
799 		dbregs->dr[7] = rdr7();
800 	} else {
801 		pcb = td->td_pcb;
802 		dbregs->dr[0] = pcb->pcb_dr0;
803 		dbregs->dr[1] = pcb->pcb_dr1;
804 		dbregs->dr[2] = pcb->pcb_dr2;
805 		dbregs->dr[3] = pcb->pcb_dr3;
806 		dbregs->dr[6] = pcb->pcb_dr6;
807 		dbregs->dr[7] = pcb->pcb_dr7;
808 	}
809 	dbregs->dr[4] = 0;
810 	dbregs->dr[5] = 0;
811 	dbregs->dr[8] = 0;
812 	dbregs->dr[9] = 0;
813 	dbregs->dr[10] = 0;
814 	dbregs->dr[11] = 0;
815 	dbregs->dr[12] = 0;
816 	dbregs->dr[13] = 0;
817 	dbregs->dr[14] = 0;
818 	dbregs->dr[15] = 0;
819 	return (0);
820 }
821 
822 int
set_dbregs(struct thread * td,struct dbreg * dbregs)823 set_dbregs(struct thread *td, struct dbreg *dbregs)
824 {
825 	struct pcb *pcb;
826 	int i;
827 
828 	if (td == NULL) {
829 		load_dr0(dbregs->dr[0]);
830 		load_dr1(dbregs->dr[1]);
831 		load_dr2(dbregs->dr[2]);
832 		load_dr3(dbregs->dr[3]);
833 		load_dr6(dbregs->dr[6]);
834 		load_dr7(dbregs->dr[7]);
835 	} else {
836 		/*
837 		 * Don't let an illegal value for dr7 get set.  Specifically,
838 		 * check for undefined settings.  Setting these bit patterns
839 		 * result in undefined behaviour and can lead to an unexpected
840 		 * TRCTRAP or a general protection fault right here.
841 		 * Upper bits of dr6 and dr7 must not be set
842 		 */
843 		for (i = 0; i < 4; i++) {
844 			if (DBREG_DR7_ACCESS(dbregs->dr[7], i) == 0x02)
845 				return (EINVAL);
846 			if (td->td_frame->tf_cs == _ucode32sel &&
847 			    DBREG_DR7_LEN(dbregs->dr[7], i) == DBREG_DR7_LEN_8)
848 				return (EINVAL);
849 		}
850 		if ((dbregs->dr[6] & 0xffffffff00000000ul) != 0 ||
851 		    (dbregs->dr[7] & 0xffffffff00000000ul) != 0)
852 			return (EINVAL);
853 
854 		pcb = td->td_pcb;
855 
856 		/*
857 		 * Don't let a process set a breakpoint that is not within the
858 		 * process's address space.  If a process could do this, it
859 		 * could halt the system by setting a breakpoint in the kernel
860 		 * (if ddb was enabled).  Thus, we need to check to make sure
861 		 * that no breakpoints are being enabled for addresses outside
862 		 * process's address space.
863 		 *
864 		 * XXX - what about when the watched area of the user's
865 		 * address space is written into from within the kernel
866 		 * ... wouldn't that still cause a breakpoint to be generated
867 		 * from within kernel mode?
868 		 */
869 
870 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 0)) {
871 			/* dr0 is enabled */
872 			if (dbregs->dr[0] >= VM_MAXUSER_ADDRESS)
873 				return (EINVAL);
874 		}
875 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 1)) {
876 			/* dr1 is enabled */
877 			if (dbregs->dr[1] >= VM_MAXUSER_ADDRESS)
878 				return (EINVAL);
879 		}
880 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 2)) {
881 			/* dr2 is enabled */
882 			if (dbregs->dr[2] >= VM_MAXUSER_ADDRESS)
883 				return (EINVAL);
884 		}
885 		if (DBREG_DR7_ENABLED(dbregs->dr[7], 3)) {
886 			/* dr3 is enabled */
887 			if (dbregs->dr[3] >= VM_MAXUSER_ADDRESS)
888 				return (EINVAL);
889 		}
890 
891 		pcb->pcb_dr0 = dbregs->dr[0];
892 		pcb->pcb_dr1 = dbregs->dr[1];
893 		pcb->pcb_dr2 = dbregs->dr[2];
894 		pcb->pcb_dr3 = dbregs->dr[3];
895 		pcb->pcb_dr6 = dbregs->dr[6];
896 		pcb->pcb_dr7 = dbregs->dr[7];
897 
898 		set_pcb_flags(pcb, PCB_DBREGS);
899 	}
900 
901 	return (0);
902 }
903 
904 void
reset_dbregs(void)905 reset_dbregs(void)
906 {
907 
908 	load_dr7(0);	/* Turn off the control bits first */
909 	load_dr0(0);
910 	load_dr1(0);
911 	load_dr2(0);
912 	load_dr3(0);
913 	load_dr6(0);
914 }
915 
916 /*
917  * Return > 0 if a hardware breakpoint has been hit, and the
918  * breakpoint was in user space.  Return 0, otherwise.
919  */
920 int
user_dbreg_trap(register_t dr6)921 user_dbreg_trap(register_t dr6)
922 {
923         u_int64_t dr7;
924         u_int64_t bp;       /* breakpoint bits extracted from dr6 */
925         int nbp;            /* number of breakpoints that triggered */
926         caddr_t addr[4];    /* breakpoint addresses */
927         int i;
928 
929         bp = dr6 & DBREG_DR6_BMASK;
930         if (bp == 0) {
931                 /*
932                  * None of the breakpoint bits are set meaning this
933                  * trap was not caused by any of the debug registers
934                  */
935                 return (0);
936         }
937 
938         dr7 = rdr7();
939         if ((dr7 & 0x000000ff) == 0) {
940                 /*
941                  * all GE and LE bits in the dr7 register are zero,
942                  * thus the trap couldn't have been caused by the
943                  * hardware debug registers
944                  */
945 		return (0);
946         }
947 
948         nbp = 0;
949 
950         /*
951          * at least one of the breakpoints were hit, check to see
952          * which ones and if any of them are user space addresses
953          */
954 
955         if (bp & 0x01) {
956                 addr[nbp++] = (caddr_t)rdr0();
957         }
958         if (bp & 0x02) {
959                 addr[nbp++] = (caddr_t)rdr1();
960         }
961         if (bp & 0x04) {
962                 addr[nbp++] = (caddr_t)rdr2();
963         }
964         if (bp & 0x08) {
965                 addr[nbp++] = (caddr_t)rdr3();
966         }
967 
968         for (i = 0; i < nbp; i++) {
969                 if (addr[i] < (caddr_t)VM_MAXUSER_ADDRESS) {
970                         /*
971                          * addr[i] is in user space
972                          */
973                         return (nbp);
974                 }
975         }
976 
977         /*
978          * None of the breakpoints are in user space.
979          */
980         return (0);
981 }
982