sys_machdep.c revision 1.34 1 1.34 maxv /* $NetBSD: sys_machdep.c,v 1.34 2017/02/18 14:36:32 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 1.34 maxv __KERNEL_RCSID(0, "$NetBSD: sys_machdep.c,v 1.34 2017/02/18 14:36:32 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 1.34 maxv #undef PMC
88 1.1 ad #undef IOPERM
89 1.1 ad #else
90 1.9 yamt #if defined(XEN)
91 1.9 yamt #undef IOPERM
92 1.9 yamt #else /* defined(XEN) */
93 1.1 ad #define IOPERM
94 1.9 yamt #endif /* defined(XEN) */
95 1.1 ad #endif
96 1.1 ad
97 1.1 ad #ifdef VM86
98 1.1 ad #include <machine/vm86.h>
99 1.1 ad #endif
100 1.1 ad
101 1.34 maxv #ifdef PMC
102 1.1 ad #include <machine/pmc.h>
103 1.1 ad #endif
104 1.1 ad
105 1.1 ad extern struct vm_map *kernel_map;
106 1.1 ad
107 1.1 ad int x86_get_ioperm(struct lwp *, void *, register_t *);
108 1.1 ad int x86_set_ioperm(struct lwp *, void *, register_t *);
109 1.1 ad int x86_get_mtrr(struct lwp *, void *, register_t *);
110 1.1 ad int x86_set_mtrr(struct lwp *, void *, register_t *);
111 1.24 chs int x86_set_sdbase32(void *, char, lwp_t *, bool);
112 1.18 ad int x86_set_sdbase(void *, char, lwp_t *, bool);
113 1.24 chs int x86_get_sdbase32(void *, char);
114 1.18 ad int x86_get_sdbase(void *, char);
115 1.1 ad
116 1.25 jakllsch #if defined(USER_LDT) && defined(LDT_DEBUG)
117 1.1 ad static void x86_print_ldt(int, const struct segment_descriptor *);
118 1.1 ad
119 1.1 ad static void
120 1.1 ad x86_print_ldt(int i, const struct segment_descriptor *d)
121 1.1 ad {
122 1.1 ad printf("[%d] lolimit=0x%x, lobase=0x%x, type=%u, dpl=%u, p=%u, "
123 1.1 ad "hilimit=0x%x, xx=%x, def32=%u, gran=%u, hibase=0x%x\n",
124 1.1 ad i, d->sd_lolimit, d->sd_lobase, d->sd_type, d->sd_dpl, d->sd_p,
125 1.1 ad d->sd_hilimit, d->sd_xx, d->sd_def32, d->sd_gran, d->sd_hibase);
126 1.1 ad }
127 1.1 ad #endif
128 1.1 ad
129 1.1 ad int
130 1.1 ad x86_get_ldt(struct lwp *l, void *args, register_t *retval)
131 1.1 ad {
132 1.2 dsl #ifndef USER_LDT
133 1.2 dsl return EINVAL;
134 1.2 dsl #else
135 1.2 dsl struct x86_get_ldt_args ua;
136 1.2 dsl union descriptor *cp;
137 1.2 dsl int error;
138 1.2 dsl
139 1.2 dsl if ((error = copyin(args, &ua, sizeof(ua))) != 0)
140 1.2 dsl return error;
141 1.2 dsl
142 1.2 dsl if (ua.num < 0 || ua.num > 8192)
143 1.2 dsl return EINVAL;
144 1.2 dsl
145 1.2 dsl cp = malloc(ua.num * sizeof(union descriptor), M_TEMP, M_WAITOK);
146 1.2 dsl if (cp == NULL)
147 1.2 dsl return ENOMEM;
148 1.2 dsl
149 1.2 dsl error = x86_get_ldt1(l, &ua, cp);
150 1.2 dsl *retval = ua.num;
151 1.2 dsl if (error == 0)
152 1.2 dsl error = copyout(cp, ua.desc, ua.num * sizeof(*cp));
153 1.2 dsl
154 1.2 dsl free(cp, M_TEMP);
155 1.2 dsl return error;
156 1.2 dsl #endif
157 1.2 dsl }
158 1.2 dsl
159 1.2 dsl int
160 1.2 dsl x86_get_ldt1(struct lwp *l, struct x86_get_ldt_args *ua, union descriptor *cp)
161 1.2 dsl {
162 1.2 dsl #ifndef USER_LDT
163 1.2 dsl return EINVAL;
164 1.2 dsl #else
165 1.1 ad int error;
166 1.1 ad struct proc *p = l->l_proc;
167 1.1 ad pmap_t pmap = p->p_vmspace->vm_map.pmap;
168 1.1 ad int nldt, num;
169 1.2 dsl union descriptor *lp;
170 1.1 ad
171 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_LDT_GET,
172 1.1 ad NULL, NULL, NULL, NULL);
173 1.1 ad if (error)
174 1.1 ad return (error);
175 1.1 ad
176 1.1 ad #ifdef LDT_DEBUG
177 1.2 dsl printf("x86_get_ldt: start=%d num=%d descs=%p\n", ua->start,
178 1.2 dsl ua->num, ua->desc);
179 1.1 ad #endif
180 1.1 ad
181 1.2 dsl if (ua->start < 0 || ua->num < 0 || ua->start > 8192 || ua->num > 8192 ||
182 1.2 dsl ua->start + ua->num > 8192)
183 1.1 ad return (EINVAL);
184 1.1 ad
185 1.32 maxv #ifdef __x86_64__
186 1.32 maxv if (ua->start * sizeof(union descriptor) < LDT_SIZE)
187 1.32 maxv return EINVAL;
188 1.32 maxv #endif
189 1.32 maxv
190 1.17 ad mutex_enter(&cpu_lock);
191 1.1 ad
192 1.17 ad if (pmap->pm_ldt != NULL) {
193 1.17 ad nldt = pmap->pm_ldt_len / sizeof(*lp);
194 1.1 ad lp = pmap->pm_ldt;
195 1.1 ad } else {
196 1.32 maxv #ifdef __x86_64__
197 1.32 maxv nldt = LDT_SIZE / sizeof(*lp);
198 1.32 maxv #else
199 1.1 ad nldt = NLDT;
200 1.32 maxv #endif
201 1.32 maxv lp = (union descriptor *)ldtstore;
202 1.1 ad }
203 1.1 ad
204 1.2 dsl if (ua->start > nldt) {
205 1.17 ad mutex_exit(&cpu_lock);
206 1.1 ad return (EINVAL);
207 1.1 ad }
208 1.1 ad
209 1.2 dsl lp += ua->start;
210 1.2 dsl num = min(ua->num, nldt - ua->start);
211 1.2 dsl ua->num = num;
212 1.1 ad #ifdef LDT_DEBUG
213 1.1 ad {
214 1.1 ad int i;
215 1.1 ad for (i = 0; i < num; i++)
216 1.1 ad x86_print_ldt(i, &lp[i].sd);
217 1.1 ad }
218 1.1 ad #endif
219 1.1 ad
220 1.1 ad memcpy(cp, lp, num * sizeof(union descriptor));
221 1.17 ad mutex_exit(&cpu_lock);
222 1.1 ad
223 1.2 dsl return 0;
224 1.2 dsl #endif
225 1.2 dsl }
226 1.2 dsl
227 1.2 dsl int
228 1.2 dsl x86_set_ldt(struct lwp *l, void *args, register_t *retval)
229 1.2 dsl {
230 1.2 dsl #ifndef USER_LDT
231 1.2 dsl return EINVAL;
232 1.2 dsl #else
233 1.2 dsl struct x86_set_ldt_args ua;
234 1.2 dsl union descriptor *descv;
235 1.2 dsl int error;
236 1.2 dsl
237 1.2 dsl if ((error = copyin(args, &ua, sizeof(ua))) != 0)
238 1.2 dsl return (error);
239 1.2 dsl
240 1.2 dsl if (ua.num < 0 || ua.num > 8192)
241 1.2 dsl return EINVAL;
242 1.2 dsl
243 1.2 dsl descv = malloc(sizeof (*descv) * ua.num, M_TEMP, M_NOWAIT);
244 1.2 dsl if (descv == NULL)
245 1.2 dsl return ENOMEM;
246 1.2 dsl
247 1.2 dsl error = copyin(ua.desc, descv, sizeof (*descv) * ua.num);
248 1.1 ad if (error == 0)
249 1.2 dsl error = x86_set_ldt1(l, &ua, descv);
250 1.2 dsl *retval = ua.start;
251 1.1 ad
252 1.2 dsl free(descv, M_TEMP);
253 1.2 dsl return error;
254 1.1 ad #endif
255 1.1 ad }
256 1.1 ad
257 1.1 ad int
258 1.2 dsl x86_set_ldt1(struct lwp *l, struct x86_set_ldt_args *ua,
259 1.2 dsl union descriptor *descv)
260 1.1 ad {
261 1.2 dsl #ifndef USER_LDT
262 1.2 dsl return EINVAL;
263 1.2 dsl #else
264 1.17 ad int error, i, n, old_sel, new_sel;
265 1.1 ad struct proc *p = l->l_proc;
266 1.1 ad pmap_t pmap = p->p_vmspace->vm_map.pmap;
267 1.17 ad size_t old_len, new_len;
268 1.17 ad union descriptor *old_ldt, *new_ldt;
269 1.1 ad
270 1.32 maxv #ifdef __x86_64__
271 1.32 maxv const size_t min_ldt_size = LDT_SIZE;
272 1.32 maxv #else
273 1.32 maxv const size_t min_ldt_size = NLDT * sizeof(union descriptor);
274 1.32 maxv #endif
275 1.32 maxv
276 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_LDT_SET,
277 1.1 ad NULL, NULL, NULL, NULL);
278 1.1 ad if (error)
279 1.1 ad return (error);
280 1.1 ad
281 1.2 dsl if (ua->start < 0 || ua->num < 0 || ua->start > 8192 || ua->num > 8192 ||
282 1.2 dsl ua->start + ua->num > 8192)
283 1.1 ad return (EINVAL);
284 1.1 ad
285 1.32 maxv #ifdef __x86_64__
286 1.32 maxv if (ua->start * sizeof(union descriptor) < LDT_SIZE)
287 1.32 maxv return EINVAL;
288 1.32 maxv #endif
289 1.32 maxv
290 1.1 ad /* Check descriptors for access violations. */
291 1.2 dsl for (i = 0; i < ua->num; i++) {
292 1.1 ad union descriptor *desc = &descv[i];
293 1.1 ad
294 1.1 ad switch (desc->sd.sd_type) {
295 1.1 ad case SDT_SYSNULL:
296 1.1 ad desc->sd.sd_p = 0;
297 1.1 ad break;
298 1.32 maxv #ifdef __x86_64__
299 1.32 maxv case SDT_SYS286CGT:
300 1.32 maxv case SDT_SYS386CGT:
301 1.32 maxv /* We don't allow these on amd64. */
302 1.32 maxv return EACCES;
303 1.32 maxv #else
304 1.1 ad case SDT_SYS286CGT:
305 1.1 ad case SDT_SYS386CGT:
306 1.1 ad /*
307 1.1 ad * Only allow call gates targeting a segment
308 1.1 ad * in the LDT or a user segment in the fixed
309 1.1 ad * part of the gdt. Segments in the LDT are
310 1.1 ad * constrained (below) to be user segments.
311 1.1 ad */
312 1.1 ad if (desc->gd.gd_p != 0 &&
313 1.1 ad !ISLDT(desc->gd.gd_selector) &&
314 1.1 ad ((IDXSEL(desc->gd.gd_selector) >= NGDT) ||
315 1.31 maxv (gdtstore[IDXSEL(desc->gd.gd_selector)].sd.sd_dpl !=
316 1.1 ad SEL_UPL))) {
317 1.2 dsl return EACCES;
318 1.1 ad }
319 1.1 ad break;
320 1.32 maxv #endif
321 1.1 ad case SDT_MEMEC:
322 1.1 ad case SDT_MEMEAC:
323 1.1 ad case SDT_MEMERC:
324 1.1 ad case SDT_MEMERAC:
325 1.1 ad /* Must be "present" if executable and conforming. */
326 1.2 dsl if (desc->sd.sd_p == 0)
327 1.2 dsl return EACCES;
328 1.1 ad break;
329 1.1 ad case SDT_MEMRO:
330 1.1 ad case SDT_MEMROA:
331 1.1 ad case SDT_MEMRW:
332 1.1 ad case SDT_MEMRWA:
333 1.1 ad case SDT_MEMROD:
334 1.1 ad case SDT_MEMRODA:
335 1.1 ad case SDT_MEMRWD:
336 1.1 ad case SDT_MEMRWDA:
337 1.1 ad case SDT_MEME:
338 1.1 ad case SDT_MEMEA:
339 1.1 ad case SDT_MEMER:
340 1.1 ad case SDT_MEMERA:
341 1.1 ad break;
342 1.1 ad default:
343 1.1 ad /*
344 1.1 ad * Make sure that unknown descriptor types are
345 1.1 ad * not marked present.
346 1.1 ad */
347 1.2 dsl if (desc->sd.sd_p != 0)
348 1.2 dsl return EACCES;
349 1.1 ad break;
350 1.1 ad }
351 1.1 ad
352 1.1 ad if (desc->sd.sd_p != 0) {
353 1.1 ad /* Only user (ring-3) descriptors may be present. */
354 1.2 dsl if (desc->sd.sd_dpl != SEL_UPL)
355 1.2 dsl return EACCES;
356 1.1 ad }
357 1.1 ad }
358 1.1 ad
359 1.17 ad /*
360 1.17 ad * Install selected changes. We perform a copy, write, swap dance
361 1.17 ad * here to ensure that all updates happen atomically.
362 1.17 ad */
363 1.17 ad
364 1.17 ad /* Allocate a new LDT. */
365 1.17 ad for (;;) {
366 1.17 ad new_len = (ua->start + ua->num) * sizeof(union descriptor);
367 1.17 ad new_len = max(new_len, pmap->pm_ldt_len);
368 1.32 maxv new_len = max(new_len, min_ldt_size);
369 1.17 ad new_len = round_page(new_len);
370 1.1 ad new_ldt = (union descriptor *)uvm_km_alloc(kernel_map,
371 1.30 dholland new_len, 0, UVM_KMF_WIRED | UVM_KMF_ZERO | UVM_KMF_WAITVA);
372 1.17 ad mutex_enter(&cpu_lock);
373 1.17 ad if (pmap->pm_ldt_len <= new_len) {
374 1.17 ad break;
375 1.1 ad }
376 1.17 ad mutex_exit(&cpu_lock);
377 1.17 ad uvm_km_free(kernel_map, (vaddr_t)new_ldt, new_len,
378 1.17 ad UVM_KMF_WIRED);
379 1.17 ad }
380 1.1 ad
381 1.17 ad /* Copy existing entries, if any. */
382 1.17 ad if (pmap->pm_ldt != NULL) {
383 1.1 ad old_ldt = pmap->pm_ldt;
384 1.17 ad old_len = pmap->pm_ldt_len;
385 1.17 ad old_sel = pmap->pm_ldt_sel;
386 1.1 ad memcpy(new_ldt, old_ldt, old_len);
387 1.17 ad } else {
388 1.17 ad old_ldt = NULL;
389 1.17 ad old_len = 0;
390 1.17 ad old_sel = -1;
391 1.32 maxv memcpy(new_ldt, ldtstore, min_ldt_size);
392 1.17 ad }
393 1.1 ad
394 1.17 ad /* Apply requested changes. */
395 1.17 ad for (i = 0, n = ua->start; i < ua->num; i++, n++) {
396 1.17 ad new_ldt[n] = descv[i];
397 1.17 ad }
398 1.1 ad
399 1.17 ad /* Allocate LDT selector. */
400 1.17 ad new_sel = ldt_alloc(new_ldt, new_len);
401 1.17 ad if (new_sel == -1) {
402 1.17 ad mutex_exit(&cpu_lock);
403 1.1 ad uvm_km_free(kernel_map, (vaddr_t)new_ldt, new_len,
404 1.1 ad UVM_KMF_WIRED);
405 1.17 ad return ENOMEM;
406 1.17 ad }
407 1.17 ad
408 1.17 ad /* All changes are now globally visible. Swap in the new LDT. */
409 1.17 ad pmap->pm_ldt_len = new_len;
410 1.17 ad pmap->pm_ldt_sel = new_sel;
411 1.30 dholland /* membar_store_store for pmap_fork() to read these unlocked safely */
412 1.30 dholland membar_producer();
413 1.30 dholland pmap->pm_ldt = new_ldt;
414 1.17 ad
415 1.17 ad /* Switch existing users onto new LDT. */
416 1.17 ad pmap_ldt_sync(pmap);
417 1.17 ad
418 1.17 ad /* Free existing LDT (if any). */
419 1.17 ad if (old_ldt != NULL) {
420 1.17 ad ldt_free(old_sel);
421 1.30 dholland /* exit the mutex before free */
422 1.30 dholland mutex_exit(&cpu_lock);
423 1.17 ad uvm_km_free(kernel_map, (vaddr_t)old_ldt, old_len,
424 1.1 ad UVM_KMF_WIRED);
425 1.30 dholland } else {
426 1.30 dholland mutex_exit(&cpu_lock);
427 1.17 ad }
428 1.2 dsl
429 1.17 ad return error;
430 1.1 ad #endif
431 1.1 ad }
432 1.1 ad
433 1.1 ad int
434 1.1 ad x86_iopl(struct lwp *l, void *args, register_t *retval)
435 1.1 ad {
436 1.1 ad int error;
437 1.1 ad struct x86_iopl_args ua;
438 1.1 ad #ifdef XEN
439 1.9 yamt int iopl;
440 1.1 ad #else
441 1.1 ad struct trapframe *tf = l->l_md.md_regs;
442 1.1 ad #endif
443 1.1 ad
444 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_IOPL,
445 1.1 ad NULL, NULL, NULL, NULL);
446 1.1 ad if (error)
447 1.1 ad return (error);
448 1.1 ad
449 1.1 ad if ((error = copyin(args, &ua, sizeof(ua))) != 0)
450 1.1 ad return error;
451 1.1 ad
452 1.1 ad #ifdef XEN
453 1.9 yamt if (ua.iopl)
454 1.9 yamt iopl = SEL_UPL;
455 1.9 yamt else
456 1.9 yamt iopl = SEL_KPL;
457 1.22 rmind
458 1.22 rmind {
459 1.22 rmind struct physdev_op physop;
460 1.22 rmind struct pcb *pcb;
461 1.22 rmind
462 1.22 rmind pcb = lwp_getpcb(l);
463 1.22 rmind pcb->pcb_iopl = iopl;
464 1.22 rmind
465 1.1 ad /* Force the change at ring 0. */
466 1.22 rmind physop.cmd = PHYSDEVOP_SET_IOPL;
467 1.22 rmind physop.u.set_iopl.iopl = iopl;
468 1.22 rmind HYPERVISOR_physdev_op(&physop);
469 1.22 rmind }
470 1.1 ad #elif defined(__x86_64__)
471 1.1 ad if (ua.iopl)
472 1.1 ad tf->tf_rflags |= PSL_IOPL;
473 1.1 ad else
474 1.1 ad tf->tf_rflags &= ~PSL_IOPL;
475 1.1 ad #else
476 1.1 ad if (ua.iopl)
477 1.1 ad tf->tf_eflags |= PSL_IOPL;
478 1.1 ad else
479 1.1 ad tf->tf_eflags &= ~PSL_IOPL;
480 1.1 ad #endif
481 1.1 ad
482 1.1 ad return 0;
483 1.1 ad }
484 1.1 ad
485 1.1 ad int
486 1.1 ad x86_get_ioperm(struct lwp *l, void *args, register_t *retval)
487 1.1 ad {
488 1.1 ad #ifdef IOPERM
489 1.1 ad int error;
490 1.22 rmind struct pcb *pcb = lwp_getpcb(l);
491 1.1 ad struct x86_get_ioperm_args ua;
492 1.9 yamt void *dummymap = NULL;
493 1.9 yamt void *iomap;
494 1.1 ad
495 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_IOPERM_GET,
496 1.1 ad NULL, NULL, NULL, NULL);
497 1.1 ad if (error)
498 1.1 ad return (error);
499 1.1 ad
500 1.1 ad if ((error = copyin(args, &ua, sizeof(ua))) != 0)
501 1.1 ad return (error);
502 1.1 ad
503 1.9 yamt iomap = pcb->pcb_iomap;
504 1.9 yamt if (iomap == NULL) {
505 1.9 yamt iomap = dummymap = kmem_alloc(IOMAPSIZE, KM_SLEEP);
506 1.9 yamt memset(dummymap, 0xff, IOMAPSIZE);
507 1.9 yamt }
508 1.9 yamt error = copyout(iomap, ua.iomap, IOMAPSIZE);
509 1.9 yamt if (dummymap != NULL) {
510 1.9 yamt kmem_free(dummymap, IOMAPSIZE);
511 1.9 yamt }
512 1.9 yamt return error;
513 1.1 ad #else
514 1.1 ad return EINVAL;
515 1.1 ad #endif
516 1.1 ad }
517 1.1 ad
518 1.1 ad int
519 1.1 ad x86_set_ioperm(struct lwp *l, void *args, register_t *retval)
520 1.1 ad {
521 1.1 ad #ifdef IOPERM
522 1.9 yamt struct cpu_info *ci;
523 1.1 ad int error;
524 1.22 rmind struct pcb *pcb = lwp_getpcb(l);
525 1.1 ad struct x86_set_ioperm_args ua;
526 1.9 yamt void *new;
527 1.9 yamt void *old;
528 1.1 ad
529 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_IOPERM_SET,
530 1.1 ad NULL, NULL, NULL, NULL);
531 1.1 ad if (error)
532 1.1 ad return (error);
533 1.1 ad
534 1.1 ad if ((error = copyin(args, &ua, sizeof(ua))) != 0)
535 1.1 ad return (error);
536 1.1 ad
537 1.9 yamt new = kmem_alloc(IOMAPSIZE, KM_SLEEP);
538 1.9 yamt error = copyin(ua.iomap, new, IOMAPSIZE);
539 1.9 yamt if (error) {
540 1.9 yamt kmem_free(new, IOMAPSIZE);
541 1.9 yamt return error;
542 1.9 yamt }
543 1.9 yamt old = pcb->pcb_iomap;
544 1.9 yamt pcb->pcb_iomap = new;
545 1.9 yamt if (old != NULL) {
546 1.9 yamt kmem_free(old, IOMAPSIZE);
547 1.9 yamt }
548 1.9 yamt
549 1.13 ad kpreempt_disable();
550 1.9 yamt ci = curcpu();
551 1.9 yamt memcpy(ci->ci_iomap, pcb->pcb_iomap, sizeof(ci->ci_iomap));
552 1.9 yamt ci->ci_tss.tss_iobase =
553 1.9 yamt ((uintptr_t)ci->ci_iomap - (uintptr_t)&ci->ci_tss) << 16;
554 1.13 ad kpreempt_enable();
555 1.9 yamt
556 1.9 yamt return error;
557 1.1 ad #else
558 1.1 ad return EINVAL;
559 1.1 ad #endif
560 1.1 ad }
561 1.1 ad
562 1.1 ad int
563 1.1 ad x86_get_mtrr(struct lwp *l, void *args, register_t *retval)
564 1.1 ad {
565 1.1 ad #ifdef MTRR
566 1.1 ad struct x86_get_mtrr_args ua;
567 1.1 ad int error, n;
568 1.1 ad
569 1.1 ad if (mtrr_funcs == NULL)
570 1.1 ad return ENOSYS;
571 1.1 ad
572 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_MTRR_GET,
573 1.1 ad NULL, NULL, NULL, NULL);
574 1.1 ad if (error)
575 1.1 ad return (error);
576 1.1 ad
577 1.1 ad error = copyin(args, &ua, sizeof ua);
578 1.1 ad if (error != 0)
579 1.1 ad return error;
580 1.1 ad
581 1.1 ad error = copyin(ua.n, &n, sizeof n);
582 1.1 ad if (error != 0)
583 1.1 ad return error;
584 1.1 ad
585 1.12 ad KERNEL_LOCK(1, NULL);
586 1.1 ad error = mtrr_get(ua.mtrrp, &n, l->l_proc, MTRR_GETSET_USER);
587 1.12 ad KERNEL_UNLOCK_ONE(NULL);
588 1.1 ad
589 1.1 ad copyout(&n, ua.n, sizeof (int));
590 1.1 ad
591 1.1 ad return error;
592 1.1 ad #else
593 1.1 ad return EINVAL;
594 1.1 ad #endif
595 1.1 ad }
596 1.1 ad
597 1.1 ad int
598 1.1 ad x86_set_mtrr(struct lwp *l, void *args, register_t *retval)
599 1.1 ad {
600 1.1 ad #ifdef MTRR
601 1.1 ad int error, n;
602 1.1 ad struct x86_set_mtrr_args ua;
603 1.1 ad
604 1.1 ad if (mtrr_funcs == NULL)
605 1.1 ad return ENOSYS;
606 1.1 ad
607 1.1 ad error = kauth_authorize_machdep(l->l_cred, KAUTH_MACHDEP_MTRR_SET,
608 1.1 ad NULL, NULL, NULL, NULL);
609 1.1 ad if (error)
610 1.1 ad return (error);
611 1.1 ad
612 1.1 ad error = copyin(args, &ua, sizeof ua);
613 1.1 ad if (error != 0)
614 1.1 ad return error;
615 1.1 ad
616 1.1 ad error = copyin(ua.n, &n, sizeof n);
617 1.1 ad if (error != 0)
618 1.1 ad return error;
619 1.1 ad
620 1.12 ad KERNEL_LOCK(1, NULL);
621 1.1 ad error = mtrr_set(ua.mtrrp, &n, l->l_proc, MTRR_GETSET_USER);
622 1.1 ad if (n != 0)
623 1.1 ad mtrr_commit();
624 1.12 ad KERNEL_UNLOCK_ONE(NULL);
625 1.1 ad
626 1.1 ad copyout(&n, ua.n, sizeof n);
627 1.1 ad
628 1.1 ad return error;
629 1.1 ad #else
630 1.1 ad return EINVAL;
631 1.1 ad #endif
632 1.1 ad }
633 1.1 ad
634 1.24 chs #ifdef __x86_64__
635 1.24 chs #define pcb_fsd pcb_fs
636 1.24 chs #define pcb_gsd pcb_gs
637 1.24 chs #define segment_descriptor mem_segment_descriptor
638 1.24 chs #endif
639 1.24 chs
640 1.1 ad int
641 1.24 chs x86_set_sdbase32(void *arg, char which, lwp_t *l, bool direct)
642 1.5 ad {
643 1.24 chs struct trapframe *tf = l->l_md.md_regs;
644 1.24 chs union descriptor usd;
645 1.18 ad struct pcb *pcb;
646 1.24 chs uint32_t base;
647 1.6 ad int error;
648 1.5 ad
649 1.18 ad if (direct) {
650 1.18 ad base = (vaddr_t)arg;
651 1.18 ad } else {
652 1.18 ad error = copyin(arg, &base, sizeof(base));
653 1.18 ad if (error != 0)
654 1.18 ad return error;
655 1.18 ad }
656 1.5 ad
657 1.24 chs memset(&usd, 0, sizeof(usd));
658 1.19 bouyer usd.sd.sd_lobase = base & 0xffffff;
659 1.19 bouyer usd.sd.sd_hibase = (base >> 24) & 0xff;
660 1.19 bouyer usd.sd.sd_lolimit = 0xffff;
661 1.19 bouyer usd.sd.sd_hilimit = 0xf;
662 1.19 bouyer usd.sd.sd_type = SDT_MEMRWA;
663 1.19 bouyer usd.sd.sd_dpl = SEL_UPL;
664 1.19 bouyer usd.sd.sd_p = 1;
665 1.19 bouyer usd.sd.sd_def32 = 1;
666 1.19 bouyer usd.sd.sd_gran = 1;
667 1.6 ad
668 1.24 chs pcb = lwp_getpcb(l);
669 1.13 ad kpreempt_disable();
670 1.6 ad if (which == 'f') {
671 1.19 bouyer memcpy(&pcb->pcb_fsd, &usd.sd,
672 1.19 bouyer sizeof(struct segment_descriptor));
673 1.18 ad if (l == curlwp) {
674 1.19 bouyer update_descriptor(&curcpu()->ci_gdt[GUFS_SEL], &usd);
675 1.24 chs #ifdef __x86_64__
676 1.24 chs setfs(GSEL(GUFS_SEL, SEL_UPL));
677 1.24 chs #endif
678 1.18 ad }
679 1.24 chs tf->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
680 1.6 ad } else /* which == 'g' */ {
681 1.19 bouyer memcpy(&pcb->pcb_gsd, &usd.sd,
682 1.19 bouyer sizeof(struct segment_descriptor));
683 1.18 ad if (l == curlwp) {
684 1.19 bouyer update_descriptor(&curcpu()->ci_gdt[GUGS_SEL], &usd);
685 1.24 chs #ifdef __x86_64__
686 1.24 chs #ifndef XEN
687 1.24 chs setusergs(GSEL(GUGS_SEL, SEL_UPL));
688 1.24 chs #else
689 1.24 chs HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL,
690 1.24 chs GSEL(GUGS_SEL, SEL_UPL));
691 1.24 chs #endif
692 1.24 chs #endif
693 1.18 ad }
694 1.24 chs tf->tf_gs = GSEL(GUGS_SEL, SEL_UPL);
695 1.6 ad }
696 1.13 ad kpreempt_enable();
697 1.24 chs return 0;
698 1.24 chs }
699 1.5 ad
700 1.24 chs int
701 1.24 chs x86_set_sdbase(void *arg, char which, lwp_t *l, bool direct)
702 1.24 chs {
703 1.24 chs #ifdef i386
704 1.24 chs return x86_set_sdbase32(arg, which, l, direct);
705 1.5 ad #else
706 1.24 chs struct pcb *pcb;
707 1.24 chs vaddr_t base;
708 1.24 chs
709 1.24 chs if (l->l_proc->p_flag & PK_32) {
710 1.24 chs return x86_set_sdbase32(arg, which, l, direct);
711 1.24 chs }
712 1.24 chs
713 1.24 chs if (direct) {
714 1.24 chs base = (vaddr_t)arg;
715 1.24 chs } else {
716 1.29 christos int error = copyin(arg, &base, sizeof(base));
717 1.24 chs if (error != 0)
718 1.24 chs return error;
719 1.24 chs }
720 1.24 chs
721 1.24 chs if (base >= VM_MAXUSER_ADDRESS)
722 1.24 chs return EINVAL;
723 1.24 chs
724 1.24 chs pcb = lwp_getpcb(l);
725 1.24 chs
726 1.24 chs kpreempt_disable();
727 1.24 chs switch(which) {
728 1.24 chs case 'f':
729 1.24 chs pcb->pcb_fs = base;
730 1.24 chs if (l == curlwp)
731 1.24 chs wrmsr(MSR_FSBASE, pcb->pcb_fs);
732 1.24 chs break;
733 1.24 chs case 'g':
734 1.24 chs pcb->pcb_gs = base;
735 1.24 chs if (l == curlwp)
736 1.24 chs wrmsr(MSR_KERNELGSBASE, pcb->pcb_gs);
737 1.24 chs break;
738 1.24 chs default:
739 1.28 dholland panic("x86_set_sdbase");
740 1.24 chs }
741 1.24 chs kpreempt_enable();
742 1.24 chs
743 1.29 christos return 0;
744 1.5 ad #endif
745 1.5 ad }
746 1.5 ad
747 1.5 ad int
748 1.24 chs x86_get_sdbase32(void *arg, char which)
749 1.5 ad {
750 1.5 ad struct segment_descriptor *sd;
751 1.24 chs uint32_t base;
752 1.5 ad
753 1.5 ad switch (which) {
754 1.5 ad case 'f':
755 1.24 chs sd = (void *)&curpcb->pcb_fsd;
756 1.5 ad break;
757 1.5 ad case 'g':
758 1.24 chs sd = (void *)&curpcb->pcb_gsd;
759 1.5 ad break;
760 1.5 ad default:
761 1.28 dholland panic("x86_get_sdbase32");
762 1.5 ad }
763 1.5 ad
764 1.5 ad base = sd->sd_hibase << 24 | sd->sd_lobase;
765 1.21 yamt return copyout(&base, arg, sizeof(base));
766 1.24 chs }
767 1.24 chs
768 1.24 chs int
769 1.24 chs x86_get_sdbase(void *arg, char which)
770 1.24 chs {
771 1.24 chs #ifdef i386
772 1.24 chs return x86_get_sdbase32(arg, which);
773 1.5 ad #else
774 1.24 chs vaddr_t base;
775 1.24 chs struct pcb *pcb;
776 1.24 chs
777 1.24 chs if (curproc->p_flag & PK_32) {
778 1.24 chs return x86_get_sdbase32(arg, which);
779 1.24 chs }
780 1.24 chs
781 1.24 chs pcb = lwp_getpcb(curlwp);
782 1.24 chs
783 1.24 chs switch(which) {
784 1.24 chs case 'f':
785 1.24 chs base = pcb->pcb_fs;
786 1.24 chs break;
787 1.24 chs case 'g':
788 1.24 chs base = pcb->pcb_gs;
789 1.24 chs break;
790 1.24 chs default:
791 1.24 chs panic("x86_get_sdbase");
792 1.24 chs }
793 1.24 chs
794 1.24 chs return copyout(&base, arg, sizeof(base));
795 1.5 ad #endif
796 1.5 ad }
797 1.5 ad
798 1.5 ad int
799 1.8 dsl sys_sysarch(struct lwp *l, const struct sys_sysarch_args *uap, register_t *retval)
800 1.1 ad {
801 1.8 dsl /* {
802 1.1 ad syscallarg(int) op;
803 1.1 ad syscallarg(void *) parms;
804 1.8 dsl } */
805 1.1 ad int error = 0;
806 1.1 ad
807 1.1 ad switch(SCARG(uap, op)) {
808 1.1 ad case X86_IOPL:
809 1.1 ad error = x86_iopl(l, SCARG(uap, parms), retval);
810 1.1 ad break;
811 1.1 ad
812 1.32 maxv #ifdef i386
813 1.32 maxv /*
814 1.32 maxv * On amd64, this is done via netbsd32_sysarch.
815 1.32 maxv */
816 1.1 ad case X86_GET_LDT:
817 1.1 ad error = x86_get_ldt(l, SCARG(uap, parms), retval);
818 1.1 ad break;
819 1.1 ad
820 1.1 ad case X86_SET_LDT:
821 1.1 ad error = x86_set_ldt(l, SCARG(uap, parms), retval);
822 1.1 ad break;
823 1.32 maxv #endif
824 1.1 ad
825 1.1 ad case X86_GET_IOPERM:
826 1.1 ad error = x86_get_ioperm(l, SCARG(uap, parms), retval);
827 1.1 ad break;
828 1.1 ad
829 1.1 ad case X86_SET_IOPERM:
830 1.1 ad error = x86_set_ioperm(l, SCARG(uap, parms), retval);
831 1.1 ad break;
832 1.1 ad
833 1.1 ad case X86_GET_MTRR:
834 1.1 ad error = x86_get_mtrr(l, SCARG(uap, parms), retval);
835 1.1 ad break;
836 1.1 ad case X86_SET_MTRR:
837 1.1 ad error = x86_set_mtrr(l, SCARG(uap, parms), retval);
838 1.1 ad break;
839 1.1 ad
840 1.1 ad #ifdef VM86
841 1.1 ad case X86_VM86:
842 1.1 ad error = x86_vm86(l, SCARG(uap, parms), retval);
843 1.1 ad break;
844 1.1 ad case X86_OLD_VM86:
845 1.1 ad error = compat_16_x86_vm86(l, SCARG(uap, parms), retval);
846 1.1 ad break;
847 1.1 ad #endif
848 1.1 ad
849 1.34 maxv #ifdef PMC
850 1.1 ad case X86_PMC_INFO:
851 1.33 maxv error = sys_pmc_info(l, SCARG(uap, parms), retval);
852 1.1 ad break;
853 1.1 ad
854 1.1 ad case X86_PMC_STARTSTOP:
855 1.33 maxv error = sys_pmc_startstop(l, SCARG(uap, parms), retval);
856 1.1 ad break;
857 1.1 ad
858 1.1 ad case X86_PMC_READ:
859 1.33 maxv error = sys_pmc_read(l, SCARG(uap, parms), retval);
860 1.1 ad break;
861 1.1 ad #endif
862 1.1 ad
863 1.5 ad case X86_SET_FSBASE:
864 1.18 ad error = x86_set_sdbase(SCARG(uap, parms), 'f', curlwp, false);
865 1.5 ad break;
866 1.5 ad
867 1.5 ad case X86_SET_GSBASE:
868 1.18 ad error = x86_set_sdbase(SCARG(uap, parms), 'g', curlwp, false);
869 1.5 ad break;
870 1.5 ad
871 1.5 ad case X86_GET_FSBASE:
872 1.5 ad error = x86_get_sdbase(SCARG(uap, parms), 'f');
873 1.5 ad break;
874 1.5 ad
875 1.5 ad case X86_GET_GSBASE:
876 1.5 ad error = x86_get_sdbase(SCARG(uap, parms), 'g');
877 1.5 ad break;
878 1.5 ad
879 1.1 ad default:
880 1.1 ad error = EINVAL;
881 1.1 ad break;
882 1.1 ad }
883 1.1 ad return (error);
884 1.1 ad }
885 1.18 ad
886 1.18 ad int
887 1.18 ad cpu_lwp_setprivate(lwp_t *l, void *addr)
888 1.18 ad {
889 1.18 ad
890 1.24 chs #ifdef __x86_64__
891 1.24 chs if ((l->l_proc->p_flag & PK_32) == 0) {
892 1.24 chs return x86_set_sdbase(addr, 'f', l, true);
893 1.24 chs }
894 1.24 chs #endif
895 1.18 ad return x86_set_sdbase(addr, 'g', l, true);
896 1.18 ad }
897