sys_machdep.c revision 1.35 1 1.35 maxv /* $NetBSD: sys_machdep.c,v 1.35 2017/03/10 14:54:12 maxv Exp $ */
2 1.1 ad
3 1.1 ad /*-
4 1.17 ad * Copyright (c) 1998, 2007, 2009 The NetBSD Foundation, Inc.
5 1.1 ad * All rights reserved.
6 1.1 ad *
7 1.1 ad * This code is derived from software contributed to The NetBSD Foundation
8 1.1 ad * by Charles M. Hannum, and by Andrew Doran.
9 1.1 ad *
10 1.1 ad * Redistribution and use in source and binary forms, with or without
11 1.1 ad * modification, are permitted provided that the following conditions
12 1.1 ad * are met:
13 1.1 ad * 1. Redistributions of source code must retain the above copyright
14 1.1 ad * notice, this list of conditions and the following disclaimer.
15 1.1 ad * 2. Redistributions in binary form must reproduce the above copyright
16 1.1 ad * notice, this list of conditions and the following disclaimer in the
17 1.1 ad * documentation and/or other materials provided with the distribution.
18 1.1 ad *
19 1.1 ad * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 1.1 ad * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 1.1 ad * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 1.1 ad * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 1.1 ad * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 1.1 ad * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 1.1 ad * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 1.1 ad * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 1.1 ad * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 1.1 ad * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 1.1 ad * POSSIBILITY OF SUCH DAMAGE.
30 1.1 ad */
31 1.1 ad
32 1.1 ad #include <sys/cdefs.h>
33 1.35 maxv __KERNEL_RCSID(0, "$NetBSD: sys_machdep.c,v 1.35 2017/03/10 14:54:12 maxv Exp $");
34 1.1 ad
35 1.1 ad #include "opt_mtrr.h"
36 1.1 ad #include "opt_user_ldt.h"
37 1.27 christos #include "opt_compat_netbsd.h"
38 1.26 dsl #ifdef i386
39 1.1 ad #include "opt_vm86.h"
40 1.26 dsl #endif
41 1.1 ad #include "opt_xen.h"
42 1.1 ad
43 1.1 ad #include <sys/param.h>
44 1.1 ad #include <sys/systm.h>
45 1.1 ad #include <sys/ioctl.h>
46 1.1 ad #include <sys/file.h>
47 1.1 ad #include <sys/time.h>
48 1.1 ad #include <sys/proc.h>
49 1.1 ad #include <sys/uio.h>
50 1.1 ad #include <sys/kernel.h>
51 1.1 ad #include <sys/buf.h>
52 1.1 ad #include <sys/signal.h>
53 1.1 ad #include <sys/malloc.h>
54 1.9 yamt #include <sys/kmem.h>
55 1.1 ad #include <sys/kauth.h>
56 1.17 ad #include <sys/cpu.h>
57 1.1 ad #include <sys/mount.h>
58 1.1 ad #include <sys/syscallargs.h>
59 1.1 ad
60 1.1 ad #include <uvm/uvm_extern.h>
61 1.1 ad
62 1.1 ad #include <machine/cpufunc.h>
63 1.1 ad #include <machine/gdt.h>
64 1.1 ad #include <machine/psl.h>
65 1.1 ad #include <machine/reg.h>
66 1.1 ad #include <machine/sysarch.h>
67 1.1 ad #include <machine/mtrr.h>
68 1.1 ad
69 1.1 ad #ifdef __x86_64__
70 1.32 maxv /*
71 1.32 maxv * The code for USER_LDT on amd64 is mostly functional, but it is still not
72 1.32 maxv * enabled.
73 1.32 maxv *
74 1.32 maxv * On amd64 we are allowing only 8-byte-sized entries in the LDT, and we are
75 1.32 maxv * not allowing the user to overwrite the existing entries (below LDT_SIZE).
76 1.32 maxv * Note that USER_LDT is used only by 32bit applications, under compat_netbsd32.
77 1.32 maxv * This is theoretically enough for Wine to work.
78 1.32 maxv *
79 1.32 maxv * However, letting segment registers have different location breaks amd64's
80 1.32 maxv * Thread Local Storage: %fs and %gs must be reloaded when returning to
81 1.32 maxv * userland. See the tech-kern@ archive from February 2017. A patch has been
82 1.32 maxv * proposed to fix that, but Wine still randomly crashes; it is not clear
83 1.32 maxv * whether the issues come from Wine, from netbsd32 or from the patch itself.
84 1.32 maxv */
85 1.32 maxv #undef USER_LDT
86 1.1 ad /* Need to be checked. */
87 1.1 ad #undef IOPERM
88 1.1 ad #else
89 1.9 yamt #if defined(XEN)
90 1.9 yamt #undef IOPERM
91 1.9 yamt #else /* defined(XEN) */
92 1.1 ad #define IOPERM
93 1.9 yamt #endif /* defined(XEN) */
94 1.1 ad #endif
95 1.1 ad
96 1.35 maxv #ifdef XEN
97 1.35 maxv #undef PMC
98 1.35 maxv #endif
99 1.35 maxv
100 1.1 ad #ifdef VM86
101 1.1 ad #include <machine/vm86.h>
102 1.1 ad #endif
103 1.1 ad
104 1.34 maxv #ifdef PMC
105 1.1 ad #include <machine/pmc.h>
106 1.1 ad #endif
107 1.1 ad
108 1.1 ad extern struct vm_map *kernel_map;
109 1.1 ad
110 1.1 ad int x86_get_ioperm(struct lwp *, void *, register_t *);
111 1.1 ad int x86_set_ioperm(struct lwp *, void *, register_t *);
112 1.1 ad int x86_get_mtrr(struct lwp *, void *, register_t *);
113 1.1 ad int x86_set_mtrr(struct lwp *, void *, register_t *);
114 1.24 chs int x86_set_sdbase32(void *, char, lwp_t *, bool);
115 1.18 ad int x86_set_sdbase(void *, char, lwp_t *, bool);
116 1.24 chs int x86_get_sdbase32(void *, char);
117 1.18 ad int x86_get_sdbase(void *, char);
118 1.1 ad
119 1.25 jakllsch #if defined(USER_LDT) && defined(LDT_DEBUG)
120 1.1 ad static void x86_print_ldt(int, const struct segment_descriptor *);
121 1.1 ad
122 1.1 ad static void
123 1.1 ad x86_print_ldt(int i, const struct segment_descriptor *d)
124 1.1 ad {
125 1.1 ad printf("[%d] lolimit=0x%x, lobase=0x%x, type=%u, dpl=%u, p=%u, "
126 1.1 ad "hilimit=0x%x, xx=%x, def32=%u, gran=%u, hibase=0x%x\n",
127 1.1 ad i, d->sd_lolimit, d->sd_lobase, d->sd_type, d->sd_dpl, d->sd_p,
128 1.1 ad d->sd_hilimit, d->sd_xx, d->sd_def32, d->sd_gran, d->sd_hibase);
129 1.1 ad }
130 1.1 ad #endif
131 1.1 ad
132 1.1 ad int
133 1.1 ad x86_get_ldt(struct lwp *l, void *args, register_t *retval)
134 1.1 ad {
135 1.2 dsl #ifndef USER_LDT
136 1.2 dsl return EINVAL;
137 1.2 dsl #else
138 1.2 dsl struct x86_get_ldt_args ua;
139 1.2 dsl union descriptor *cp;
140 1.2 dsl int error;
141 1.2 dsl
142 1.2 dsl if ((error = copyin(args, &ua, sizeof(ua))) != 0)
143 1.2 dsl return error;
144 1.2 dsl
145 1.2 dsl if (ua.num < 0 || ua.num > 8192)
146 1.2 dsl return EINVAL;
147 1.2 dsl
148 1.2 dsl cp = malloc(ua.num * sizeof(union descriptor), M_TEMP, M_WAITOK);
149 1.2 dsl if (cp == NULL)
150 1.2 dsl return ENOMEM;
151 1.2 dsl
152 1.2 dsl error = x86_get_ldt1(l, &ua, cp);
153 1.2 dsl *retval = ua.num;
154 1.2 dsl if (error == 0)
155 1.2 dsl error = copyout(cp, ua.desc, ua.num * sizeof(*cp));
156 1.2 dsl
157 1.2 dsl free(cp, M_TEMP);
158 1.2 dsl return error;
159 1.2 dsl #endif
160 1.2 dsl }
161 1.2 dsl
162 1.2 dsl int
163 1.2 dsl x86_get_ldt1(struct lwp *l, struct x86_get_ldt_args *ua, union descriptor *cp)
164 1.2 dsl {
165 1.2 dsl #ifndef USER_LDT
166 1.2 dsl return EINVAL;
167 1.2 dsl #else
168 1.1 ad int error;
169 1.1 ad struct proc *p = l->l_proc;
170 1.1 ad pmap_t pmap = p->p_vmspace->vm_map.pmap;
171 1.1 ad int nldt, num;
172 1.2 dsl union descriptor *lp;
173 1.1 ad
174 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_LDT_GET,
175 1.1 ad NULL, NULL, NULL, NULL);
176 1.1 ad if (error)
177 1.1 ad return (error);
178 1.1 ad
179 1.1 ad #ifdef LDT_DEBUG
180 1.2 dsl printf("x86_get_ldt: start=%d num=%d descs=%p\n", ua->start,
181 1.2 dsl ua->num, ua->desc);
182 1.1 ad #endif
183 1.1 ad
184 1.2 dsl if (ua->start < 0 || ua->num < 0 || ua->start > 8192 || ua->num > 8192 ||
185 1.2 dsl ua->start + ua->num > 8192)
186 1.1 ad return (EINVAL);
187 1.1 ad
188 1.32 maxv #ifdef __x86_64__
189 1.32 maxv if (ua->start * sizeof(union descriptor) < LDT_SIZE)
190 1.32 maxv return EINVAL;
191 1.32 maxv #endif
192 1.32 maxv
193 1.17 ad mutex_enter(&cpu_lock);
194 1.1 ad
195 1.17 ad if (pmap->pm_ldt != NULL) {
196 1.17 ad nldt = pmap->pm_ldt_len / sizeof(*lp);
197 1.1 ad lp = pmap->pm_ldt;
198 1.1 ad } else {
199 1.32 maxv #ifdef __x86_64__
200 1.32 maxv nldt = LDT_SIZE / sizeof(*lp);
201 1.32 maxv #else
202 1.1 ad nldt = NLDT;
203 1.32 maxv #endif
204 1.32 maxv lp = (union descriptor *)ldtstore;
205 1.1 ad }
206 1.1 ad
207 1.2 dsl if (ua->start > nldt) {
208 1.17 ad mutex_exit(&cpu_lock);
209 1.1 ad return (EINVAL);
210 1.1 ad }
211 1.1 ad
212 1.2 dsl lp += ua->start;
213 1.2 dsl num = min(ua->num, nldt - ua->start);
214 1.2 dsl ua->num = num;
215 1.1 ad #ifdef LDT_DEBUG
216 1.1 ad {
217 1.1 ad int i;
218 1.1 ad for (i = 0; i < num; i++)
219 1.1 ad x86_print_ldt(i, &lp[i].sd);
220 1.1 ad }
221 1.1 ad #endif
222 1.1 ad
223 1.1 ad memcpy(cp, lp, num * sizeof(union descriptor));
224 1.17 ad mutex_exit(&cpu_lock);
225 1.1 ad
226 1.2 dsl return 0;
227 1.2 dsl #endif
228 1.2 dsl }
229 1.2 dsl
230 1.2 dsl int
231 1.2 dsl x86_set_ldt(struct lwp *l, void *args, register_t *retval)
232 1.2 dsl {
233 1.2 dsl #ifndef USER_LDT
234 1.2 dsl return EINVAL;
235 1.2 dsl #else
236 1.2 dsl struct x86_set_ldt_args ua;
237 1.2 dsl union descriptor *descv;
238 1.2 dsl int error;
239 1.2 dsl
240 1.2 dsl if ((error = copyin(args, &ua, sizeof(ua))) != 0)
241 1.2 dsl return (error);
242 1.2 dsl
243 1.2 dsl if (ua.num < 0 || ua.num > 8192)
244 1.2 dsl return EINVAL;
245 1.2 dsl
246 1.2 dsl descv = malloc(sizeof (*descv) * ua.num, M_TEMP, M_NOWAIT);
247 1.2 dsl if (descv == NULL)
248 1.2 dsl return ENOMEM;
249 1.2 dsl
250 1.2 dsl error = copyin(ua.desc, descv, sizeof (*descv) * ua.num);
251 1.1 ad if (error == 0)
252 1.2 dsl error = x86_set_ldt1(l, &ua, descv);
253 1.2 dsl *retval = ua.start;
254 1.1 ad
255 1.2 dsl free(descv, M_TEMP);
256 1.2 dsl return error;
257 1.1 ad #endif
258 1.1 ad }
259 1.1 ad
260 1.1 ad int
261 1.2 dsl x86_set_ldt1(struct lwp *l, struct x86_set_ldt_args *ua,
262 1.2 dsl union descriptor *descv)
263 1.1 ad {
264 1.2 dsl #ifndef USER_LDT
265 1.2 dsl return EINVAL;
266 1.2 dsl #else
267 1.17 ad int error, i, n, old_sel, new_sel;
268 1.1 ad struct proc *p = l->l_proc;
269 1.1 ad pmap_t pmap = p->p_vmspace->vm_map.pmap;
270 1.17 ad size_t old_len, new_len;
271 1.17 ad union descriptor *old_ldt, *new_ldt;
272 1.1 ad
273 1.32 maxv #ifdef __x86_64__
274 1.32 maxv const size_t min_ldt_size = LDT_SIZE;
275 1.32 maxv #else
276 1.32 maxv const size_t min_ldt_size = NLDT * sizeof(union descriptor);
277 1.32 maxv #endif
278 1.32 maxv
279 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_LDT_SET,
280 1.1 ad NULL, NULL, NULL, NULL);
281 1.1 ad if (error)
282 1.1 ad return (error);
283 1.1 ad
284 1.2 dsl if (ua->start < 0 || ua->num < 0 || ua->start > 8192 || ua->num > 8192 ||
285 1.2 dsl ua->start + ua->num > 8192)
286 1.1 ad return (EINVAL);
287 1.1 ad
288 1.32 maxv #ifdef __x86_64__
289 1.32 maxv if (ua->start * sizeof(union descriptor) < LDT_SIZE)
290 1.32 maxv return EINVAL;
291 1.32 maxv #endif
292 1.32 maxv
293 1.1 ad /* Check descriptors for access violations. */
294 1.2 dsl for (i = 0; i < ua->num; i++) {
295 1.1 ad union descriptor *desc = &descv[i];
296 1.1 ad
297 1.1 ad switch (desc->sd.sd_type) {
298 1.1 ad case SDT_SYSNULL:
299 1.1 ad desc->sd.sd_p = 0;
300 1.1 ad break;
301 1.32 maxv #ifdef __x86_64__
302 1.32 maxv case SDT_SYS286CGT:
303 1.32 maxv case SDT_SYS386CGT:
304 1.32 maxv /* We don't allow these on amd64. */
305 1.32 maxv return EACCES;
306 1.32 maxv #else
307 1.1 ad case SDT_SYS286CGT:
308 1.1 ad case SDT_SYS386CGT:
309 1.1 ad /*
310 1.1 ad * Only allow call gates targeting a segment
311 1.1 ad * in the LDT or a user segment in the fixed
312 1.1 ad * part of the gdt. Segments in the LDT are
313 1.1 ad * constrained (below) to be user segments.
314 1.1 ad */
315 1.1 ad if (desc->gd.gd_p != 0 &&
316 1.1 ad !ISLDT(desc->gd.gd_selector) &&
317 1.1 ad ((IDXSEL(desc->gd.gd_selector) >= NGDT) ||
318 1.31 maxv (gdtstore[IDXSEL(desc->gd.gd_selector)].sd.sd_dpl !=
319 1.1 ad SEL_UPL))) {
320 1.2 dsl return EACCES;
321 1.1 ad }
322 1.1 ad break;
323 1.32 maxv #endif
324 1.1 ad case SDT_MEMEC:
325 1.1 ad case SDT_MEMEAC:
326 1.1 ad case SDT_MEMERC:
327 1.1 ad case SDT_MEMERAC:
328 1.1 ad /* Must be "present" if executable and conforming. */
329 1.2 dsl if (desc->sd.sd_p == 0)
330 1.2 dsl return EACCES;
331 1.1 ad break;
332 1.1 ad case SDT_MEMRO:
333 1.1 ad case SDT_MEMROA:
334 1.1 ad case SDT_MEMRW:
335 1.1 ad case SDT_MEMRWA:
336 1.1 ad case SDT_MEMROD:
337 1.1 ad case SDT_MEMRODA:
338 1.1 ad case SDT_MEMRWD:
339 1.1 ad case SDT_MEMRWDA:
340 1.1 ad case SDT_MEME:
341 1.1 ad case SDT_MEMEA:
342 1.1 ad case SDT_MEMER:
343 1.1 ad case SDT_MEMERA:
344 1.1 ad break;
345 1.1 ad default:
346 1.1 ad /*
347 1.1 ad * Make sure that unknown descriptor types are
348 1.1 ad * not marked present.
349 1.1 ad */
350 1.2 dsl if (desc->sd.sd_p != 0)
351 1.2 dsl return EACCES;
352 1.1 ad break;
353 1.1 ad }
354 1.1 ad
355 1.1 ad if (desc->sd.sd_p != 0) {
356 1.1 ad /* Only user (ring-3) descriptors may be present. */
357 1.2 dsl if (desc->sd.sd_dpl != SEL_UPL)
358 1.2 dsl return EACCES;
359 1.1 ad }
360 1.1 ad }
361 1.1 ad
362 1.17 ad /*
363 1.17 ad * Install selected changes. We perform a copy, write, swap dance
364 1.17 ad * here to ensure that all updates happen atomically.
365 1.17 ad */
366 1.17 ad
367 1.17 ad /* Allocate a new LDT. */
368 1.17 ad for (;;) {
369 1.17 ad new_len = (ua->start + ua->num) * sizeof(union descriptor);
370 1.17 ad new_len = max(new_len, pmap->pm_ldt_len);
371 1.32 maxv new_len = max(new_len, min_ldt_size);
372 1.17 ad new_len = round_page(new_len);
373 1.1 ad new_ldt = (union descriptor *)uvm_km_alloc(kernel_map,
374 1.30 dholland new_len, 0, UVM_KMF_WIRED | UVM_KMF_ZERO | UVM_KMF_WAITVA);
375 1.17 ad mutex_enter(&cpu_lock);
376 1.17 ad if (pmap->pm_ldt_len <= new_len) {
377 1.17 ad break;
378 1.1 ad }
379 1.17 ad mutex_exit(&cpu_lock);
380 1.17 ad uvm_km_free(kernel_map, (vaddr_t)new_ldt, new_len,
381 1.17 ad UVM_KMF_WIRED);
382 1.17 ad }
383 1.1 ad
384 1.17 ad /* Copy existing entries, if any. */
385 1.17 ad if (pmap->pm_ldt != NULL) {
386 1.1 ad old_ldt = pmap->pm_ldt;
387 1.17 ad old_len = pmap->pm_ldt_len;
388 1.17 ad old_sel = pmap->pm_ldt_sel;
389 1.1 ad memcpy(new_ldt, old_ldt, old_len);
390 1.17 ad } else {
391 1.17 ad old_ldt = NULL;
392 1.17 ad old_len = 0;
393 1.17 ad old_sel = -1;
394 1.32 maxv memcpy(new_ldt, ldtstore, min_ldt_size);
395 1.17 ad }
396 1.1 ad
397 1.17 ad /* Apply requested changes. */
398 1.17 ad for (i = 0, n = ua->start; i < ua->num; i++, n++) {
399 1.17 ad new_ldt[n] = descv[i];
400 1.17 ad }
401 1.1 ad
402 1.17 ad /* Allocate LDT selector. */
403 1.17 ad new_sel = ldt_alloc(new_ldt, new_len);
404 1.17 ad if (new_sel == -1) {
405 1.17 ad mutex_exit(&cpu_lock);
406 1.1 ad uvm_km_free(kernel_map, (vaddr_t)new_ldt, new_len,
407 1.1 ad UVM_KMF_WIRED);
408 1.17 ad return ENOMEM;
409 1.17 ad }
410 1.17 ad
411 1.17 ad /* All changes are now globally visible. Swap in the new LDT. */
412 1.17 ad pmap->pm_ldt_len = new_len;
413 1.17 ad pmap->pm_ldt_sel = new_sel;
414 1.30 dholland /* membar_store_store for pmap_fork() to read these unlocked safely */
415 1.30 dholland membar_producer();
416 1.30 dholland pmap->pm_ldt = new_ldt;
417 1.17 ad
418 1.17 ad /* Switch existing users onto new LDT. */
419 1.17 ad pmap_ldt_sync(pmap);
420 1.17 ad
421 1.17 ad /* Free existing LDT (if any). */
422 1.17 ad if (old_ldt != NULL) {
423 1.17 ad ldt_free(old_sel);
424 1.30 dholland /* exit the mutex before free */
425 1.30 dholland mutex_exit(&cpu_lock);
426 1.17 ad uvm_km_free(kernel_map, (vaddr_t)old_ldt, old_len,
427 1.1 ad UVM_KMF_WIRED);
428 1.30 dholland } else {
429 1.30 dholland mutex_exit(&cpu_lock);
430 1.17 ad }
431 1.2 dsl
432 1.17 ad return error;
433 1.1 ad #endif
434 1.1 ad }
435 1.1 ad
436 1.1 ad int
437 1.1 ad x86_iopl(struct lwp *l, void *args, register_t *retval)
438 1.1 ad {
439 1.1 ad int error;
440 1.1 ad struct x86_iopl_args ua;
441 1.1 ad #ifdef XEN
442 1.9 yamt int iopl;
443 1.1 ad #else
444 1.1 ad struct trapframe *tf = l->l_md.md_regs;
445 1.1 ad #endif
446 1.1 ad
447 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_IOPL,
448 1.1 ad NULL, NULL, NULL, NULL);
449 1.1 ad if (error)
450 1.1 ad return (error);
451 1.1 ad
452 1.1 ad if ((error = copyin(args, &ua, sizeof(ua))) != 0)
453 1.1 ad return error;
454 1.1 ad
455 1.1 ad #ifdef XEN
456 1.9 yamt if (ua.iopl)
457 1.9 yamt iopl = SEL_UPL;
458 1.9 yamt else
459 1.9 yamt iopl = SEL_KPL;
460 1.22 rmind
461 1.22 rmind {
462 1.22 rmind struct physdev_op physop;
463 1.22 rmind struct pcb *pcb;
464 1.22 rmind
465 1.22 rmind pcb = lwp_getpcb(l);
466 1.22 rmind pcb->pcb_iopl = iopl;
467 1.22 rmind
468 1.1 ad /* Force the change at ring 0. */
469 1.22 rmind physop.cmd = PHYSDEVOP_SET_IOPL;
470 1.22 rmind physop.u.set_iopl.iopl = iopl;
471 1.22 rmind HYPERVISOR_physdev_op(&physop);
472 1.22 rmind }
473 1.1 ad #elif defined(__x86_64__)
474 1.1 ad if (ua.iopl)
475 1.1 ad tf->tf_rflags |= PSL_IOPL;
476 1.1 ad else
477 1.1 ad tf->tf_rflags &= ~PSL_IOPL;
478 1.1 ad #else
479 1.1 ad if (ua.iopl)
480 1.1 ad tf->tf_eflags |= PSL_IOPL;
481 1.1 ad else
482 1.1 ad tf->tf_eflags &= ~PSL_IOPL;
483 1.1 ad #endif
484 1.1 ad
485 1.1 ad return 0;
486 1.1 ad }
487 1.1 ad
488 1.1 ad int
489 1.1 ad x86_get_ioperm(struct lwp *l, void *args, register_t *retval)
490 1.1 ad {
491 1.1 ad #ifdef IOPERM
492 1.1 ad int error;
493 1.22 rmind struct pcb *pcb = lwp_getpcb(l);
494 1.1 ad struct x86_get_ioperm_args ua;
495 1.9 yamt void *dummymap = NULL;
496 1.9 yamt void *iomap;
497 1.1 ad
498 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_IOPERM_GET,
499 1.1 ad NULL, NULL, NULL, NULL);
500 1.1 ad if (error)
501 1.1 ad return (error);
502 1.1 ad
503 1.1 ad if ((error = copyin(args, &ua, sizeof(ua))) != 0)
504 1.1 ad return (error);
505 1.1 ad
506 1.9 yamt iomap = pcb->pcb_iomap;
507 1.9 yamt if (iomap == NULL) {
508 1.9 yamt iomap = dummymap = kmem_alloc(IOMAPSIZE, KM_SLEEP);
509 1.9 yamt memset(dummymap, 0xff, IOMAPSIZE);
510 1.9 yamt }
511 1.9 yamt error = copyout(iomap, ua.iomap, IOMAPSIZE);
512 1.9 yamt if (dummymap != NULL) {
513 1.9 yamt kmem_free(dummymap, IOMAPSIZE);
514 1.9 yamt }
515 1.9 yamt return error;
516 1.1 ad #else
517 1.1 ad return EINVAL;
518 1.1 ad #endif
519 1.1 ad }
520 1.1 ad
521 1.1 ad int
522 1.1 ad x86_set_ioperm(struct lwp *l, void *args, register_t *retval)
523 1.1 ad {
524 1.1 ad #ifdef IOPERM
525 1.9 yamt struct cpu_info *ci;
526 1.1 ad int error;
527 1.22 rmind struct pcb *pcb = lwp_getpcb(l);
528 1.1 ad struct x86_set_ioperm_args ua;
529 1.9 yamt void *new;
530 1.9 yamt void *old;
531 1.1 ad
532 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_IOPERM_SET,
533 1.1 ad NULL, NULL, NULL, NULL);
534 1.1 ad if (error)
535 1.1 ad return (error);
536 1.1 ad
537 1.1 ad if ((error = copyin(args, &ua, sizeof(ua))) != 0)
538 1.1 ad return (error);
539 1.1 ad
540 1.9 yamt new = kmem_alloc(IOMAPSIZE, KM_SLEEP);
541 1.9 yamt error = copyin(ua.iomap, new, IOMAPSIZE);
542 1.9 yamt if (error) {
543 1.9 yamt kmem_free(new, IOMAPSIZE);
544 1.9 yamt return error;
545 1.9 yamt }
546 1.9 yamt old = pcb->pcb_iomap;
547 1.9 yamt pcb->pcb_iomap = new;
548 1.9 yamt if (old != NULL) {
549 1.9 yamt kmem_free(old, IOMAPSIZE);
550 1.9 yamt }
551 1.9 yamt
552 1.13 ad kpreempt_disable();
553 1.9 yamt ci = curcpu();
554 1.9 yamt memcpy(ci->ci_iomap, pcb->pcb_iomap, sizeof(ci->ci_iomap));
555 1.9 yamt ci->ci_tss.tss_iobase =
556 1.9 yamt ((uintptr_t)ci->ci_iomap - (uintptr_t)&ci->ci_tss) << 16;
557 1.13 ad kpreempt_enable();
558 1.9 yamt
559 1.9 yamt return error;
560 1.1 ad #else
561 1.1 ad return EINVAL;
562 1.1 ad #endif
563 1.1 ad }
564 1.1 ad
565 1.1 ad int
566 1.1 ad x86_get_mtrr(struct lwp *l, void *args, register_t *retval)
567 1.1 ad {
568 1.1 ad #ifdef MTRR
569 1.1 ad struct x86_get_mtrr_args ua;
570 1.1 ad int error, n;
571 1.1 ad
572 1.1 ad if (mtrr_funcs == NULL)
573 1.1 ad return ENOSYS;
574 1.1 ad
575 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_MTRR_GET,
576 1.1 ad NULL, NULL, NULL, NULL);
577 1.1 ad if (error)
578 1.1 ad return (error);
579 1.1 ad
580 1.1 ad error = copyin(args, &ua, sizeof ua);
581 1.1 ad if (error != 0)
582 1.1 ad return error;
583 1.1 ad
584 1.1 ad error = copyin(ua.n, &n, sizeof n);
585 1.1 ad if (error != 0)
586 1.1 ad return error;
587 1.1 ad
588 1.12 ad KERNEL_LOCK(1, NULL);
589 1.1 ad error = mtrr_get(ua.mtrrp, &n, l->l_proc, MTRR_GETSET_USER);
590 1.12 ad KERNEL_UNLOCK_ONE(NULL);
591 1.1 ad
592 1.1 ad copyout(&n, ua.n, sizeof (int));
593 1.1 ad
594 1.1 ad return error;
595 1.1 ad #else
596 1.1 ad return EINVAL;
597 1.1 ad #endif
598 1.1 ad }
599 1.1 ad
600 1.1 ad int
601 1.1 ad x86_set_mtrr(struct lwp *l, void *args, register_t *retval)
602 1.1 ad {
603 1.1 ad #ifdef MTRR
604 1.1 ad int error, n;
605 1.1 ad struct x86_set_mtrr_args ua;
606 1.1 ad
607 1.1 ad if (mtrr_funcs == NULL)
608 1.1 ad return ENOSYS;
609 1.1 ad
610 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_MTRR_SET,
611 1.1 ad NULL, NULL, NULL, NULL);
612 1.1 ad if (error)
613 1.1 ad return (error);
614 1.1 ad
615 1.1 ad error = copyin(args, &ua, sizeof ua);
616 1.1 ad if (error != 0)
617 1.1 ad return error;
618 1.1 ad
619 1.1 ad error = copyin(ua.n, &n, sizeof n);
620 1.1 ad if (error != 0)
621 1.1 ad return error;
622 1.1 ad
623 1.12 ad KERNEL_LOCK(1, NULL);
624 1.1 ad error = mtrr_set(ua.mtrrp, &n, l->l_proc, MTRR_GETSET_USER);
625 1.1 ad if (n != 0)
626 1.1 ad mtrr_commit();
627 1.12 ad KERNEL_UNLOCK_ONE(NULL);
628 1.1 ad
629 1.1 ad copyout(&n, ua.n, sizeof n);
630 1.1 ad
631 1.1 ad return error;
632 1.1 ad #else
633 1.1 ad return EINVAL;
634 1.1 ad #endif
635 1.1 ad }
636 1.1 ad
637 1.24 chs #ifdef __x86_64__
638 1.24 chs #define pcb_fsd pcb_fs
639 1.24 chs #define pcb_gsd pcb_gs
640 1.24 chs #define segment_descriptor mem_segment_descriptor
641 1.24 chs #endif
642 1.24 chs
643 1.1 ad int
644 1.24 chs x86_set_sdbase32(void *arg, char which, lwp_t *l, bool direct)
645 1.5 ad {
646 1.24 chs struct trapframe *tf = l->l_md.md_regs;
647 1.24 chs union descriptor usd;
648 1.18 ad struct pcb *pcb;
649 1.24 chs uint32_t base;
650 1.6 ad int error;
651 1.5 ad
652 1.18 ad if (direct) {
653 1.18 ad base = (vaddr_t)arg;
654 1.18 ad } else {
655 1.18 ad error = copyin(arg, &base, sizeof(base));
656 1.18 ad if (error != 0)
657 1.18 ad return error;
658 1.18 ad }
659 1.5 ad
660 1.24 chs memset(&usd, 0, sizeof(usd));
661 1.19 bouyer usd.sd.sd_lobase = base & 0xffffff;
662 1.19 bouyer usd.sd.sd_hibase = (base >> 24) & 0xff;
663 1.19 bouyer usd.sd.sd_lolimit = 0xffff;
664 1.19 bouyer usd.sd.sd_hilimit = 0xf;
665 1.19 bouyer usd.sd.sd_type = SDT_MEMRWA;
666 1.19 bouyer usd.sd.sd_dpl = SEL_UPL;
667 1.19 bouyer usd.sd.sd_p = 1;
668 1.19 bouyer usd.sd.sd_def32 = 1;
669 1.19 bouyer usd.sd.sd_gran = 1;
670 1.6 ad
671 1.24 chs pcb = lwp_getpcb(l);
672 1.13 ad kpreempt_disable();
673 1.6 ad if (which == 'f') {
674 1.19 bouyer memcpy(&pcb->pcb_fsd, &usd.sd,
675 1.19 bouyer sizeof(struct segment_descriptor));
676 1.18 ad if (l == curlwp) {
677 1.19 bouyer update_descriptor(&curcpu()->ci_gdt[GUFS_SEL], &usd);
678 1.24 chs #ifdef __x86_64__
679 1.24 chs setfs(GSEL(GUFS_SEL, SEL_UPL));
680 1.24 chs #endif
681 1.18 ad }
682 1.24 chs tf->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
683 1.6 ad } else /* which == 'g' */ {
684 1.19 bouyer memcpy(&pcb->pcb_gsd, &usd.sd,
685 1.19 bouyer sizeof(struct segment_descriptor));
686 1.18 ad if (l == curlwp) {
687 1.19 bouyer update_descriptor(&curcpu()->ci_gdt[GUGS_SEL], &usd);
688 1.24 chs #ifdef __x86_64__
689 1.24 chs #ifndef XEN
690 1.24 chs setusergs(GSEL(GUGS_SEL, SEL_UPL));
691 1.24 chs #else
692 1.24 chs HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL,
693 1.24 chs GSEL(GUGS_SEL, SEL_UPL));
694 1.24 chs #endif
695 1.24 chs #endif
696 1.18 ad }
697 1.24 chs tf->tf_gs = GSEL(GUGS_SEL, SEL_UPL);
698 1.6 ad }
699 1.13 ad kpreempt_enable();
700 1.24 chs return 0;
701 1.24 chs }
702 1.5 ad
703 1.24 chs int
704 1.24 chs x86_set_sdbase(void *arg, char which, lwp_t *l, bool direct)
705 1.24 chs {
706 1.24 chs #ifdef i386
707 1.24 chs return x86_set_sdbase32(arg, which, l, direct);
708 1.5 ad #else
709 1.24 chs struct pcb *pcb;
710 1.24 chs vaddr_t base;
711 1.24 chs
712 1.24 chs if (l->l_proc->p_flag & PK_32) {
713 1.24 chs return x86_set_sdbase32(arg, which, l, direct);
714 1.24 chs }
715 1.24 chs
716 1.24 chs if (direct) {
717 1.24 chs base = (vaddr_t)arg;
718 1.24 chs } else {
719 1.29 christos int error = copyin(arg, &base, sizeof(base));
720 1.24 chs if (error != 0)
721 1.24 chs return error;
722 1.24 chs }
723 1.24 chs
724 1.24 chs if (base >= VM_MAXUSER_ADDRESS)
725 1.24 chs return EINVAL;
726 1.24 chs
727 1.24 chs pcb = lwp_getpcb(l);
728 1.24 chs
729 1.24 chs kpreempt_disable();
730 1.24 chs switch(which) {
731 1.24 chs case 'f':
732 1.24 chs pcb->pcb_fs = base;
733 1.24 chs if (l == curlwp)
734 1.24 chs wrmsr(MSR_FSBASE, pcb->pcb_fs);
735 1.24 chs break;
736 1.24 chs case 'g':
737 1.24 chs pcb->pcb_gs = base;
738 1.24 chs if (l == curlwp)
739 1.24 chs wrmsr(MSR_KERNELGSBASE, pcb->pcb_gs);
740 1.24 chs break;
741 1.24 chs default:
742 1.28 dholland panic("x86_set_sdbase");
743 1.24 chs }
744 1.24 chs kpreempt_enable();
745 1.24 chs
746 1.29 christos return 0;
747 1.5 ad #endif
748 1.5 ad }
749 1.5 ad
750 1.5 ad int
751 1.24 chs x86_get_sdbase32(void *arg, char which)
752 1.5 ad {
753 1.5 ad struct segment_descriptor *sd;
754 1.24 chs uint32_t base;
755 1.5 ad
756 1.5 ad switch (which) {
757 1.5 ad case 'f':
758 1.24 chs sd = (void *)&curpcb->pcb_fsd;
759 1.5 ad break;
760 1.5 ad case 'g':
761 1.24 chs sd = (void *)&curpcb->pcb_gsd;
762 1.5 ad break;
763 1.5 ad default:
764 1.28 dholland panic("x86_get_sdbase32");
765 1.5 ad }
766 1.5 ad
767 1.5 ad base = sd->sd_hibase << 24 | sd->sd_lobase;
768 1.21 yamt return copyout(&base, arg, sizeof(base));
769 1.24 chs }
770 1.24 chs
771 1.24 chs int
772 1.24 chs x86_get_sdbase(void *arg, char which)
773 1.24 chs {
774 1.24 chs #ifdef i386
775 1.24 chs return x86_get_sdbase32(arg, which);
776 1.5 ad #else
777 1.24 chs vaddr_t base;
778 1.24 chs struct pcb *pcb;
779 1.24 chs
780 1.24 chs if (curproc->p_flag & PK_32) {
781 1.24 chs return x86_get_sdbase32(arg, which);
782 1.24 chs }
783 1.24 chs
784 1.24 chs pcb = lwp_getpcb(curlwp);
785 1.24 chs
786 1.24 chs switch(which) {
787 1.24 chs case 'f':
788 1.24 chs base = pcb->pcb_fs;
789 1.24 chs break;
790 1.24 chs case 'g':
791 1.24 chs base = pcb->pcb_gs;
792 1.24 chs break;
793 1.24 chs default:
794 1.24 chs panic("x86_get_sdbase");
795 1.24 chs }
796 1.24 chs
797 1.24 chs return copyout(&base, arg, sizeof(base));
798 1.5 ad #endif
799 1.5 ad }
800 1.5 ad
801 1.5 ad int
802 1.8 dsl sys_sysarch(struct lwp *l, const struct sys_sysarch_args *uap, register_t *retval)
803 1.1 ad {
804 1.8 dsl /* {
805 1.1 ad syscallarg(int) op;
806 1.1 ad syscallarg(void *) parms;
807 1.8 dsl } */
808 1.1 ad int error = 0;
809 1.1 ad
810 1.1 ad switch(SCARG(uap, op)) {
811 1.1 ad case X86_IOPL:
812 1.1 ad error = x86_iopl(l, SCARG(uap, parms), retval);
813 1.1 ad break;
814 1.1 ad
815 1.32 maxv #ifdef i386
816 1.32 maxv /*
817 1.32 maxv * On amd64, this is done via netbsd32_sysarch.
818 1.32 maxv */
819 1.1 ad case X86_GET_LDT:
820 1.1 ad error = x86_get_ldt(l, SCARG(uap, parms), retval);
821 1.1 ad break;
822 1.1 ad
823 1.1 ad case X86_SET_LDT:
824 1.1 ad error = x86_set_ldt(l, SCARG(uap, parms), retval);
825 1.1 ad break;
826 1.32 maxv #endif
827 1.1 ad
828 1.1 ad case X86_GET_IOPERM:
829 1.1 ad error = x86_get_ioperm(l, SCARG(uap, parms), retval);
830 1.1 ad break;
831 1.1 ad
832 1.1 ad case X86_SET_IOPERM:
833 1.1 ad error = x86_set_ioperm(l, SCARG(uap, parms), retval);
834 1.1 ad break;
835 1.1 ad
836 1.1 ad case X86_GET_MTRR:
837 1.1 ad error = x86_get_mtrr(l, SCARG(uap, parms), retval);
838 1.1 ad break;
839 1.1 ad case X86_SET_MTRR:
840 1.1 ad error = x86_set_mtrr(l, SCARG(uap, parms), retval);
841 1.1 ad break;
842 1.1 ad
843 1.1 ad #ifdef VM86
844 1.1 ad case X86_VM86:
845 1.1 ad error = x86_vm86(l, SCARG(uap, parms), retval);
846 1.1 ad break;
847 1.1 ad case X86_OLD_VM86:
848 1.1 ad error = compat_16_x86_vm86(l, SCARG(uap, parms), retval);
849 1.1 ad break;
850 1.1 ad #endif
851 1.1 ad
852 1.34 maxv #ifdef PMC
853 1.1 ad case X86_PMC_INFO:
854 1.33 maxv error = sys_pmc_info(l, SCARG(uap, parms), retval);
855 1.1 ad break;
856 1.1 ad
857 1.1 ad case X86_PMC_STARTSTOP:
858 1.33 maxv error = sys_pmc_startstop(l, SCARG(uap, parms), retval);
859 1.1 ad break;
860 1.1 ad
861 1.1 ad case X86_PMC_READ:
862 1.33 maxv error = sys_pmc_read(l, SCARG(uap, parms), retval);
863 1.1 ad break;
864 1.1 ad #endif
865 1.1 ad
866 1.5 ad case X86_SET_FSBASE:
867 1.18 ad error = x86_set_sdbase(SCARG(uap, parms), 'f', curlwp, false);
868 1.5 ad break;
869 1.5 ad
870 1.5 ad case X86_SET_GSBASE:
871 1.18 ad error = x86_set_sdbase(SCARG(uap, parms), 'g', curlwp, false);
872 1.5 ad break;
873 1.5 ad
874 1.5 ad case X86_GET_FSBASE:
875 1.5 ad error = x86_get_sdbase(SCARG(uap, parms), 'f');
876 1.5 ad break;
877 1.5 ad
878 1.5 ad case X86_GET_GSBASE:
879 1.5 ad error = x86_get_sdbase(SCARG(uap, parms), 'g');
880 1.5 ad break;
881 1.5 ad
882 1.1 ad default:
883 1.1 ad error = EINVAL;
884 1.1 ad break;
885 1.1 ad }
886 1.1 ad return (error);
887 1.1 ad }
888 1.18 ad
889 1.18 ad int
890 1.18 ad cpu_lwp_setprivate(lwp_t *l, void *addr)
891 1.18 ad {
892 1.18 ad
893 1.24 chs #ifdef __x86_64__
894 1.24 chs if ((l->l_proc->p_flag & PK_32) == 0) {
895 1.24 chs return x86_set_sdbase(addr, 'f', l, true);
896 1.24 chs }
897 1.24 chs #endif
898 1.18 ad return x86_set_sdbase(addr, 'g', l, true);
899 1.18 ad }
900