xref: /freebsd/sys/amd64/amd64/sys_machdep.c (revision a0ee8cc6)
1 /*-
2  * Copyright (c) 2003 Peter Wemm.
3  * Copyright (c) 1990 The Regents of the University of California.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	from: @(#)sys_machdep.c	5.5 (Berkeley) 1/19/91
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include "opt_capsicum.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/capsicum.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/sysproto.h>
48 #include <sys/uio.h>
49 
50 #include <vm/vm.h>
51 #include <vm/pmap.h>
52 #include <vm/vm_kern.h>		/* for kernel_map */
53 #include <vm/vm_extern.h>
54 
55 #include <machine/frame.h>
56 #include <machine/md_var.h>
57 #include <machine/pcb.h>
58 #include <machine/specialreg.h>
59 #include <machine/sysarch.h>
60 #include <machine/tss.h>
61 #include <machine/vmparam.h>
62 
63 #include <security/audit/audit.h>
64 
65 #define	MAX_LD		8192
66 
67 int max_ldt_segment = 1024;
68 SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN,
69     &max_ldt_segment, 0,
70     "Maximum number of allowed LDT segments in the single address space");
71 
72 static void
73 max_ldt_segment_init(void *arg __unused)
74 {
75 
76 	if (max_ldt_segment <= 0)
77 		max_ldt_segment = 1;
78 	if (max_ldt_segment > MAX_LD)
79 		max_ldt_segment = MAX_LD;
80 }
81 SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL);
82 
83 #ifdef notyet
84 #ifdef SMP
85 static void set_user_ldt_rv(struct vmspace *vmsp);
86 #endif
87 #endif
88 static void user_ldt_derefl(struct proc_ldt *pldt);
89 
90 #ifndef _SYS_SYSPROTO_H_
91 struct sysarch_args {
92 	int op;
93 	char *parms;
94 };
95 #endif
96 
97 int
98 sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space)
99 {
100 	struct i386_ldt_args *largs, la;
101 	struct user_segment_descriptor *lp;
102 	int error = 0;
103 
104 	/*
105 	 * XXXKIB check that the BSM generation code knows to encode
106 	 * the op argument.
107 	 */
108 	AUDIT_ARG_CMD(uap->op);
109 	if (uap_space == UIO_USERSPACE) {
110 		error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args));
111 		if (error != 0)
112 			return (error);
113 		largs = &la;
114 	} else
115 		largs = (struct i386_ldt_args *)uap->parms;
116 
117 	switch (uap->op) {
118 	case I386_GET_LDT:
119 		error = amd64_get_ldt(td, largs);
120 		break;
121 	case I386_SET_LDT:
122 		if (largs->descs != NULL && largs->num > max_ldt_segment)
123 			return (EINVAL);
124 		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
125 		if (largs->descs != NULL) {
126 			lp = malloc(largs->num * sizeof(struct
127 			    user_segment_descriptor), M_TEMP, M_WAITOK);
128 			error = copyin(largs->descs, lp, largs->num *
129 			    sizeof(struct user_segment_descriptor));
130 			if (error == 0)
131 				error = amd64_set_ldt(td, largs, lp);
132 			free(lp, M_TEMP);
133 		} else {
134 			error = amd64_set_ldt(td, largs, NULL);
135 		}
136 		break;
137 	}
138 	return (error);
139 }
140 
141 void
142 update_gdt_gsbase(struct thread *td, uint32_t base)
143 {
144 	struct user_segment_descriptor *sd;
145 
146 	if (td != curthread)
147 		return;
148 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
149 	critical_enter();
150 	sd = PCPU_GET(gs32p);
151 	sd->sd_lobase = base & 0xffffff;
152 	sd->sd_hibase = (base >> 24) & 0xff;
153 	critical_exit();
154 }
155 
156 void
157 update_gdt_fsbase(struct thread *td, uint32_t base)
158 {
159 	struct user_segment_descriptor *sd;
160 
161 	if (td != curthread)
162 		return;
163 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
164 	critical_enter();
165 	sd = PCPU_GET(fs32p);
166 	sd->sd_lobase = base & 0xffffff;
167 	sd->sd_hibase = (base >> 24) & 0xff;
168 	critical_exit();
169 }
170 
171 int
172 sysarch(td, uap)
173 	struct thread *td;
174 	register struct sysarch_args *uap;
175 {
176 	int error = 0;
177 	struct pcb *pcb = curthread->td_pcb;
178 	uint32_t i386base;
179 	uint64_t a64base;
180 	struct i386_ioperm_args iargs;
181 	struct i386_get_xfpustate i386xfpu;
182 	struct amd64_get_xfpustate a64xfpu;
183 
184 #ifdef CAPABILITY_MODE
185 	/*
186 	 * When adding new operations, add a new case statement here to
187 	 * explicitly indicate whether or not the operation is safe to
188 	 * perform in capability mode.
189 	 */
190 	if (IN_CAPABILITY_MODE(td)) {
191 		switch (uap->op) {
192 		case I386_GET_LDT:
193 		case I386_SET_LDT:
194 		case I386_GET_IOPERM:
195 		case I386_GET_FSBASE:
196 		case I386_SET_FSBASE:
197 		case I386_GET_GSBASE:
198 		case I386_SET_GSBASE:
199 		case I386_GET_XFPUSTATE:
200 		case AMD64_GET_FSBASE:
201 		case AMD64_SET_FSBASE:
202 		case AMD64_GET_GSBASE:
203 		case AMD64_SET_GSBASE:
204 		case AMD64_GET_XFPUSTATE:
205 			break;
206 
207 		case I386_SET_IOPERM:
208 		default:
209 #ifdef KTRACE
210 			if (KTRPOINT(td, KTR_CAPFAIL))
211 				ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL);
212 #endif
213 			return (ECAPMODE);
214 		}
215 	}
216 #endif
217 
218 	if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT)
219 		return (sysarch_ldt(td, uap, UIO_USERSPACE));
220 	/*
221 	 * XXXKIB check that the BSM generation code knows to encode
222 	 * the op argument.
223 	 */
224 	AUDIT_ARG_CMD(uap->op);
225 	switch (uap->op) {
226 	case I386_GET_IOPERM:
227 	case I386_SET_IOPERM:
228 		if ((error = copyin(uap->parms, &iargs,
229 		    sizeof(struct i386_ioperm_args))) != 0)
230 			return (error);
231 		break;
232 	case I386_GET_XFPUSTATE:
233 		if ((error = copyin(uap->parms, &i386xfpu,
234 		    sizeof(struct i386_get_xfpustate))) != 0)
235 			return (error);
236 		a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr;
237 		a64xfpu.len = i386xfpu.len;
238 		break;
239 	case AMD64_GET_XFPUSTATE:
240 		if ((error = copyin(uap->parms, &a64xfpu,
241 		    sizeof(struct amd64_get_xfpustate))) != 0)
242 			return (error);
243 		break;
244 	default:
245 		break;
246 	}
247 
248 	switch (uap->op) {
249 	case I386_GET_IOPERM:
250 		error = amd64_get_ioperm(td, &iargs);
251 		if (error == 0)
252 			error = copyout(&iargs, uap->parms,
253 			    sizeof(struct i386_ioperm_args));
254 		break;
255 	case I386_SET_IOPERM:
256 		error = amd64_set_ioperm(td, &iargs);
257 		break;
258 	case I386_GET_FSBASE:
259 		i386base = pcb->pcb_fsbase;
260 		error = copyout(&i386base, uap->parms, sizeof(i386base));
261 		break;
262 	case I386_SET_FSBASE:
263 		error = copyin(uap->parms, &i386base, sizeof(i386base));
264 		if (!error) {
265 			pcb->pcb_fsbase = i386base;
266 			td->td_frame->tf_fs = _ufssel;
267 			update_gdt_fsbase(td, i386base);
268 		}
269 		break;
270 	case I386_GET_GSBASE:
271 		i386base = pcb->pcb_gsbase;
272 		error = copyout(&i386base, uap->parms, sizeof(i386base));
273 		break;
274 	case I386_SET_GSBASE:
275 		error = copyin(uap->parms, &i386base, sizeof(i386base));
276 		if (!error) {
277 			pcb->pcb_gsbase = i386base;
278 			td->td_frame->tf_gs = _ugssel;
279 			update_gdt_gsbase(td, i386base);
280 		}
281 		break;
282 	case AMD64_GET_FSBASE:
283 		error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase));
284 		break;
285 
286 	case AMD64_SET_FSBASE:
287 		error = copyin(uap->parms, &a64base, sizeof(a64base));
288 		if (!error) {
289 			if (a64base < VM_MAXUSER_ADDRESS) {
290 				pcb->pcb_fsbase = a64base;
291 				set_pcb_flags(pcb, PCB_FULL_IRET);
292 				td->td_frame->tf_fs = _ufssel;
293 			} else
294 				error = EINVAL;
295 		}
296 		break;
297 
298 	case AMD64_GET_GSBASE:
299 		error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase));
300 		break;
301 
302 	case AMD64_SET_GSBASE:
303 		error = copyin(uap->parms, &a64base, sizeof(a64base));
304 		if (!error) {
305 			if (a64base < VM_MAXUSER_ADDRESS) {
306 				pcb->pcb_gsbase = a64base;
307 				set_pcb_flags(pcb, PCB_FULL_IRET);
308 				td->td_frame->tf_gs = _ugssel;
309 			} else
310 				error = EINVAL;
311 		}
312 		break;
313 
314 	case I386_GET_XFPUSTATE:
315 	case AMD64_GET_XFPUSTATE:
316 		if (a64xfpu.len > cpu_max_ext_state_size -
317 		    sizeof(struct savefpu))
318 			return (EINVAL);
319 		fpugetregs(td);
320 		error = copyout((char *)(get_pcb_user_save_td(td) + 1),
321 		    a64xfpu.addr, a64xfpu.len);
322 		break;
323 
324 	default:
325 		error = EINVAL;
326 		break;
327 	}
328 	return (error);
329 }
330 
331 int
332 amd64_set_ioperm(td, uap)
333 	struct thread *td;
334 	struct i386_ioperm_args *uap;
335 {
336 	int i, error;
337 	char *iomap;
338 	struct amd64tss *tssp;
339 	struct system_segment_descriptor *tss_sd;
340 	struct pcb *pcb;
341 
342 	if ((error = priv_check(td, PRIV_IO)) != 0)
343 		return (error);
344 	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
345 		return (error);
346 	if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
347 		return (EINVAL);
348 
349 	/*
350 	 * XXX
351 	 * While this is restricted to root, we should probably figure out
352 	 * whether any other driver is using this i/o address, as so not to
353 	 * cause confusion.  This probably requires a global 'usage registry'.
354 	 */
355 	pcb = td->td_pcb;
356 	if (pcb->pcb_tssp == NULL) {
357 		tssp = (struct amd64tss *)kmem_malloc(kernel_arena,
358 		    ctob(IOPAGES+1), M_WAITOK);
359 		iomap = (char *)&tssp[1];
360 		memset(iomap, 0xff, IOPERM_BITMAP_SIZE);
361 		critical_enter();
362 		/* Takes care of tss_rsp0. */
363 		memcpy(tssp, &common_tss[PCPU_GET(cpuid)],
364 		    sizeof(struct amd64tss));
365 		tssp->tss_iobase = sizeof(*tssp);
366 		pcb->pcb_tssp = tssp;
367 		tss_sd = PCPU_GET(tss);
368 		tss_sd->sd_lobase = (u_long)tssp & 0xffffff;
369 		tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful;
370 		tss_sd->sd_type = SDT_SYSTSS;
371 		ltr(GSEL(GPROC0_SEL, SEL_KPL));
372 		PCPU_SET(tssp, tssp);
373 		critical_exit();
374 	} else
375 		iomap = (char *)&pcb->pcb_tssp[1];
376 	for (i = uap->start; i < uap->start + uap->length; i++) {
377 		if (uap->enable)
378 			iomap[i >> 3] &= ~(1 << (i & 7));
379 		else
380 			iomap[i >> 3] |= (1 << (i & 7));
381 	}
382 	return (error);
383 }
384 
385 int
386 amd64_get_ioperm(td, uap)
387 	struct thread *td;
388 	struct i386_ioperm_args *uap;
389 {
390 	int i, state;
391 	char *iomap;
392 
393 	if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
394 		return (EINVAL);
395 	if (td->td_pcb->pcb_tssp == NULL) {
396 		uap->length = 0;
397 		goto done;
398 	}
399 
400 	iomap = (char *)&td->td_pcb->pcb_tssp[1];
401 
402 	i = uap->start;
403 	state = (iomap[i >> 3] >> (i & 7)) & 1;
404 	uap->enable = !state;
405 	uap->length = 1;
406 
407 	for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
408 		if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
409 			break;
410 		uap->length++;
411 	}
412 
413 done:
414 	return (0);
415 }
416 
417 /*
418  * Update the GDT entry pointing to the LDT to point to the LDT of the
419  * current process.
420  */
421 void
422 set_user_ldt(struct mdproc *mdp)
423 {
424 
425 	critical_enter();
426 	*PCPU_GET(ldt) = mdp->md_ldt_sd;
427 	lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
428 	critical_exit();
429 }
430 
431 #ifdef notyet
432 #ifdef SMP
433 static void
434 set_user_ldt_rv(struct vmspace *vmsp)
435 {
436 	struct thread *td;
437 
438 	td = curthread;
439 	if (vmsp != td->td_proc->p_vmspace)
440 		return;
441 
442 	set_user_ldt(&td->td_proc->p_md);
443 }
444 #endif
445 #endif
446 
447 struct proc_ldt *
448 user_ldt_alloc(struct proc *p, int force)
449 {
450 	struct proc_ldt *pldt, *new_ldt;
451 	struct mdproc *mdp;
452 	struct soft_segment_descriptor sldt;
453 
454 	mtx_assert(&dt_lock, MA_OWNED);
455 	mdp = &p->p_md;
456 	if (!force && mdp->md_ldt != NULL)
457 		return (mdp->md_ldt);
458 	mtx_unlock(&dt_lock);
459 	new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK);
460 	new_ldt->ldt_base = (caddr_t)kmem_malloc(kernel_arena,
461 	     max_ldt_segment * sizeof(struct user_segment_descriptor),
462 	     M_WAITOK | M_ZERO);
463 	new_ldt->ldt_refcnt = 1;
464 	sldt.ssd_base = (uint64_t)new_ldt->ldt_base;
465 	sldt.ssd_limit = max_ldt_segment *
466 	    sizeof(struct user_segment_descriptor) - 1;
467 	sldt.ssd_type = SDT_SYSLDT;
468 	sldt.ssd_dpl = SEL_KPL;
469 	sldt.ssd_p = 1;
470 	sldt.ssd_long = 0;
471 	sldt.ssd_def32 = 0;
472 	sldt.ssd_gran = 0;
473 	mtx_lock(&dt_lock);
474 	pldt = mdp->md_ldt;
475 	if (pldt != NULL && !force) {
476 		kmem_free(kernel_arena, (vm_offset_t)new_ldt->ldt_base,
477 		    max_ldt_segment * sizeof(struct user_segment_descriptor));
478 		free(new_ldt, M_SUBPROC);
479 		return (pldt);
480 	}
481 
482 	if (pldt != NULL) {
483 		bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment *
484 		    sizeof(struct user_segment_descriptor));
485 		user_ldt_derefl(pldt);
486 	}
487 	ssdtosyssd(&sldt, &p->p_md.md_ldt_sd);
488 	atomic_store_rel_ptr((volatile uintptr_t *)&mdp->md_ldt,
489 	    (uintptr_t)new_ldt);
490 	if (p == curproc)
491 		set_user_ldt(mdp);
492 
493 	return (mdp->md_ldt);
494 }
495 
496 void
497 user_ldt_free(struct thread *td)
498 {
499 	struct proc *p = td->td_proc;
500 	struct mdproc *mdp = &p->p_md;
501 	struct proc_ldt *pldt;
502 
503 	mtx_assert(&dt_lock, MA_OWNED);
504 	if ((pldt = mdp->md_ldt) == NULL) {
505 		mtx_unlock(&dt_lock);
506 		return;
507 	}
508 
509 	mdp->md_ldt = NULL;
510 	bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd));
511 	if (td == curthread)
512 		lldt(GSEL(GNULL_SEL, SEL_KPL));
513 	user_ldt_deref(pldt);
514 }
515 
516 static void
517 user_ldt_derefl(struct proc_ldt *pldt)
518 {
519 
520 	if (--pldt->ldt_refcnt == 0) {
521 		kmem_free(kernel_arena, (vm_offset_t)pldt->ldt_base,
522 		    max_ldt_segment * sizeof(struct user_segment_descriptor));
523 		free(pldt, M_SUBPROC);
524 	}
525 }
526 
527 void
528 user_ldt_deref(struct proc_ldt *pldt)
529 {
530 
531 	mtx_assert(&dt_lock, MA_OWNED);
532 	user_ldt_derefl(pldt);
533 	mtx_unlock(&dt_lock);
534 }
535 
536 /*
537  * Note for the authors of compat layers (linux, etc): copyout() in
538  * the function below is not a problem since it presents data in
539  * arch-specific format (i.e. i386-specific in this case), not in
540  * the OS-specific one.
541  */
542 int
543 amd64_get_ldt(td, uap)
544 	struct thread *td;
545 	struct i386_ldt_args *uap;
546 {
547 	int error = 0;
548 	struct proc_ldt *pldt;
549 	int num;
550 	struct user_segment_descriptor *lp;
551 
552 #ifdef	DEBUG
553 	printf("amd64_get_ldt: start=%d num=%d descs=%p\n",
554 	    uap->start, uap->num, (void *)uap->descs);
555 #endif
556 
557 	if ((pldt = td->td_proc->p_md.md_ldt) != NULL) {
558 		lp = &((struct user_segment_descriptor *)(pldt->ldt_base))
559 		    [uap->start];
560 		num = min(uap->num, max_ldt_segment);
561 	} else
562 		return (EINVAL);
563 
564 	if ((uap->start > (unsigned int)max_ldt_segment) ||
565 	    ((unsigned int)num > (unsigned int)max_ldt_segment) ||
566 	    ((unsigned int)(uap->start + num) > (unsigned int)max_ldt_segment))
567 		return(EINVAL);
568 
569 	error = copyout(lp, uap->descs, num *
570 	    sizeof(struct user_segment_descriptor));
571 	if (!error)
572 		td->td_retval[0] = num;
573 
574 	return(error);
575 }
576 
577 int
578 amd64_set_ldt(td, uap, descs)
579 	struct thread *td;
580 	struct i386_ldt_args *uap;
581 	struct user_segment_descriptor *descs;
582 {
583 	int error = 0, i;
584 	int largest_ld;
585 	struct mdproc *mdp = &td->td_proc->p_md;
586 	struct proc_ldt *pldt;
587 	struct user_segment_descriptor *dp;
588 	struct proc *p;
589 
590 #ifdef	DEBUG
591 	printf("amd64_set_ldt: start=%d num=%d descs=%p\n",
592 	    uap->start, uap->num, (void *)uap->descs);
593 #endif
594 
595 	set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
596 	p = td->td_proc;
597 	if (descs == NULL) {
598 		/* Free descriptors */
599 		if (uap->start == 0 && uap->num == 0)
600 			uap->num = max_ldt_segment;
601 		if (uap->num == 0)
602 			return (EINVAL);
603 		if ((pldt = mdp->md_ldt) == NULL ||
604 		    uap->start >= max_ldt_segment)
605 			return (0);
606 		largest_ld = uap->start + uap->num;
607 		if (largest_ld > max_ldt_segment)
608 			largest_ld = max_ldt_segment;
609 		i = largest_ld - uap->start;
610 		mtx_lock(&dt_lock);
611 		bzero(&((struct user_segment_descriptor *)(pldt->ldt_base))
612 		    [uap->start], sizeof(struct user_segment_descriptor) * i);
613 		mtx_unlock(&dt_lock);
614 		return (0);
615 	}
616 
617 	if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
618 		/* verify range of descriptors to modify */
619 		largest_ld = uap->start + uap->num;
620 		if (uap->start >= max_ldt_segment ||
621 		    largest_ld > max_ldt_segment)
622 			return (EINVAL);
623 	}
624 
625 	/* Check descriptors for access violations */
626 	for (i = 0; i < uap->num; i++) {
627 		dp = &descs[i];
628 
629 		switch (dp->sd_type) {
630 		case SDT_SYSNULL:	/* system null */
631 			dp->sd_p = 0;
632 			break;
633 		case SDT_SYS286TSS:
634 		case SDT_SYSLDT:
635 		case SDT_SYS286BSY:
636 		case SDT_SYS286CGT:
637 		case SDT_SYSTASKGT:
638 		case SDT_SYS286IGT:
639 		case SDT_SYS286TGT:
640 		case SDT_SYSNULL2:
641 		case SDT_SYSTSS:
642 		case SDT_SYSNULL3:
643 		case SDT_SYSBSY:
644 		case SDT_SYSCGT:
645 		case SDT_SYSNULL4:
646 		case SDT_SYSIGT:
647 		case SDT_SYSTGT:
648 			/* I can't think of any reason to allow a user proc
649 			 * to create a segment of these types.  They are
650 			 * for OS use only.
651 			 */
652 			return (EACCES);
653 			/*NOTREACHED*/
654 
655 		/* memory segment types */
656 		case SDT_MEMEC:   /* memory execute only conforming */
657 		case SDT_MEMEAC:  /* memory execute only accessed conforming */
658 		case SDT_MEMERC:  /* memory execute read conforming */
659 		case SDT_MEMERAC: /* memory execute read accessed conforming */
660 			 /* Must be "present" if executable and conforming. */
661 			if (dp->sd_p == 0)
662 				return (EACCES);
663 			break;
664 		case SDT_MEMRO:   /* memory read only */
665 		case SDT_MEMROA:  /* memory read only accessed */
666 		case SDT_MEMRW:   /* memory read write */
667 		case SDT_MEMRWA:  /* memory read write accessed */
668 		case SDT_MEMROD:  /* memory read only expand dwn limit */
669 		case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
670 		case SDT_MEMRWD:  /* memory read write expand dwn limit */
671 		case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
672 		case SDT_MEME:    /* memory execute only */
673 		case SDT_MEMEA:   /* memory execute only accessed */
674 		case SDT_MEMER:   /* memory execute read */
675 		case SDT_MEMERA:  /* memory execute read accessed */
676 			break;
677 		default:
678 			return(EINVAL);
679 			/*NOTREACHED*/
680 		}
681 
682 		/* Only user (ring-3) descriptors may be present. */
683 		if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL))
684 			return (EACCES);
685 	}
686 
687 	if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
688 		/* Allocate a free slot */
689 		mtx_lock(&dt_lock);
690 		pldt = user_ldt_alloc(p, 0);
691 		if (pldt == NULL) {
692 			mtx_unlock(&dt_lock);
693 			return (ENOMEM);
694 		}
695 
696 		/*
697 		 * start scanning a bit up to leave room for NVidia and
698 		 * Wine, which still user the "Blat" method of allocation.
699 		 */
700 		i = 16;
701 		dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i];
702 		for (; i < max_ldt_segment; ++i, ++dp) {
703 			if (dp->sd_type == SDT_SYSNULL)
704 				break;
705 		}
706 		if (i >= max_ldt_segment) {
707 			mtx_unlock(&dt_lock);
708 			return (ENOSPC);
709 		}
710 		uap->start = i;
711 		error = amd64_set_ldt_data(td, i, 1, descs);
712 		mtx_unlock(&dt_lock);
713 	} else {
714 		largest_ld = uap->start + uap->num;
715 		if (largest_ld > max_ldt_segment)
716 			return (EINVAL);
717 		mtx_lock(&dt_lock);
718 		if (user_ldt_alloc(p, 0) != NULL) {
719 			error = amd64_set_ldt_data(td, uap->start, uap->num,
720 			    descs);
721 		}
722 		mtx_unlock(&dt_lock);
723 	}
724 	if (error == 0)
725 		td->td_retval[0] = uap->start;
726 	return (error);
727 }
728 
729 int
730 amd64_set_ldt_data(struct thread *td, int start, int num,
731     struct user_segment_descriptor *descs)
732 {
733 	struct mdproc *mdp = &td->td_proc->p_md;
734 	struct proc_ldt *pldt = mdp->md_ldt;
735 
736 	mtx_assert(&dt_lock, MA_OWNED);
737 
738 	/* Fill in range */
739 	bcopy(descs,
740 	    &((struct user_segment_descriptor *)(pldt->ldt_base))[start],
741 	    num * sizeof(struct user_segment_descriptor));
742 	return (0);
743 }
744