1 /*-
2 * Copyright (c) 1990 The Regents of the University of California.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_kstack_pages.h"
36 #include "opt_mac.h"
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/lock.h>
41 #include <sys/mac.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/smp.h>
46 #include <sys/sysproto.h>
47
48 #include <vm/vm.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_map.h>
51 #include <vm/vm_extern.h>
52
53 #include <machine/cpu.h>
54 #include <machine/pcb.h>
55 #include <machine/pcb_ext.h>
56 #include <machine/proc.h>
57 #include <machine/sysarch.h>
58
59 #include <vm/vm_kern.h> /* for kernel_map */
60
61 #define MAX_LD 8192
62 #define LD_PER_PAGE 512
63 #define NEW_MAX_LD(num) ((num + LD_PER_PAGE) & ~(LD_PER_PAGE-1))
64 #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3)
65
66
67
68 static int i386_get_ldt(struct thread *, char *);
69 static int i386_set_ldt(struct thread *, char *);
70 static int i386_set_ldt_data(struct thread *, int start, int num,
71 union descriptor *descs);
72 static int i386_ldt_grow(struct thread *td, int len);
73 static int i386_get_ioperm(struct thread *, char *);
74 static int i386_set_ioperm(struct thread *, char *);
75 #ifdef SMP
76 static void set_user_ldt_rv(struct thread *);
77 #endif
78
79 #ifndef _SYS_SYSPROTO_H_
80 struct sysarch_args {
81 int op;
82 char *parms;
83 };
84 #endif
85
86 int
87 sysarch(td, uap)
88 struct thread *td;
89 register struct sysarch_args *uap;
90 {
91 int error;
92 uint32_t base;
93 struct segment_descriptor sd, *sdp;
94
95
96 mtx_lock(&Giant);
97 switch(uap->op) {
98 case I386_GET_LDT:
99 error = i386_get_ldt(td, uap->parms);
100 break;
101
102 case I386_SET_LDT:
103 error = i386_set_ldt(td, uap->parms);
104 break;
105 case I386_GET_IOPERM:
106 error = i386_get_ioperm(td, uap->parms);
107 break;
108 case I386_SET_IOPERM:
109 error = i386_set_ioperm(td, uap->parms);
110 break;
111 case I386_VM86:
112 error = vm86_sysarch(td, uap->parms);
113 break;
114 case I386_GET_FSBASE:
115 sdp = (struct segment_descriptor *)&td->td_pcb->pcb_fsd;
116 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
117 error = copyout(&base, uap->parms, sizeof(base));
118 break;
119 case I386_SET_FSBASE:
120 error = copyin(uap->parms, &base, sizeof(base));
121 if (!error) {
122 /*
123 * Construct a descriptor and store it in the pcb for
124 * the next context switch. Also store it in the gdt
125 * so that the load of tf_fs into %fs will activate it
126 * at return to userland.
127 */
128 sd.sd_lobase = base & 0xffffff;
129 sd.sd_hibase = (base >> 24) & 0xff;
130 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
131 sd.sd_hilimit = 0xf;
132 sd.sd_type = SDT_MEMRWA;
133 sd.sd_dpl = SEL_UPL;
134 sd.sd_p = 1;
135 sd.sd_xx = 0;
136 sd.sd_def32 = 1;
137 sd.sd_gran = 1;
138 critical_enter();
139 *(struct segment_descriptor *)&td->td_pcb->pcb_fsd = sd;
140 PCPU_GET(fsgs_gdt)[0] = sd;
141 critical_exit();
142 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
143 }
144 break;
145 case I386_GET_GSBASE:
146 sdp = (struct segment_descriptor *)&td->td_pcb->pcb_gsd;
147 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
148 error = copyout(&base, uap->parms, sizeof(base));
149 break;
150 case I386_SET_GSBASE:
151 error = copyin(uap->parms, &base, sizeof(base));
152 if (!error) {
153 /*
154 * Construct a descriptor and store it in the pcb for
155 * the next context switch. Also store it in the gdt
156 * because we have to do a load_gs() right now.
157 */
158 sd.sd_lobase = base & 0xffffff;
159 sd.sd_hibase = (base >> 24) & 0xff;
160 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
161 sd.sd_hilimit = 0xf;
162 sd.sd_type = SDT_MEMRWA;
163 sd.sd_dpl = SEL_UPL;
164 sd.sd_p = 1;
165 sd.sd_xx = 0;
166 sd.sd_def32 = 1;
167 sd.sd_gran = 1;
168 critical_enter();
169 *(struct segment_descriptor *)&td->td_pcb->pcb_gsd = sd;
170 PCPU_GET(fsgs_gdt)[1] = sd;
171 critical_exit();
172 load_gs(GSEL(GUGS_SEL, SEL_UPL));
173 }
174 break;
175 default:
176 error = EINVAL;
177 break;
178 }
179 mtx_unlock(&Giant);
180 return (error);
181 }
182
183 int
184 i386_extend_pcb(struct thread *td)
185 {
186 int i, offset;
187 u_long *addr;
188 struct pcb_ext *ext;
189 struct soft_segment_descriptor ssd = {
190 0, /* segment base address (overwritten) */
191 ctob(IOPAGES + 1) - 1, /* length */
192 SDT_SYS386TSS, /* segment type */
193 0, /* priority level */
194 1, /* descriptor present */
195 0, 0,
196 0, /* default 32 size */
197 0 /* granularity */
198 };
199
200 if (td->td_proc->p_flag & P_SA)
201 return (EINVAL); /* XXXKSE */
202 /* XXXKSE All the code below only works in 1:1 needs changing */
203 ext = (struct pcb_ext *)kmem_alloc(kernel_map, ctob(IOPAGES+1));
204 if (ext == 0)
205 return (ENOMEM);
206 bzero(ext, sizeof(struct pcb_ext));
207 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
208 ext->ext_tss.tss_esp0 = td->td_kstack + ctob(KSTACK_PAGES) -
209 sizeof(struct pcb) - 16;
210 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
211 /*
212 * The last byte of the i/o map must be followed by an 0xff byte.
213 * We arbitrarily allocate 16 bytes here, to keep the starting
214 * address on a doubleword boundary.
215 */
216 offset = PAGE_SIZE - 16;
217 ext->ext_tss.tss_ioopt =
218 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
219 ext->ext_iomap = (caddr_t)ext + offset;
220 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
221
222 addr = (u_long *)ext->ext_vm86.vm86_intmap;
223 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
224 *addr++ = ~0;
225
226 ssd.ssd_base = (unsigned)&ext->ext_tss;
227 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
228 ssdtosd(&ssd, &ext->ext_tssd);
229
230 KASSERT(td == curthread, ("giving TSS to !curthread"));
231 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
232
233 /* Switch to the new TSS. */
234 mtx_lock_spin(&sched_lock);
235 td->td_pcb->pcb_ext = ext;
236 private_tss |= PCPU_GET(cpumask);
237 *PCPU_GET(tss_gdt) = ext->ext_tssd;
238 ltr(GSEL(GPROC0_SEL, SEL_KPL));
239 mtx_unlock_spin(&sched_lock);
240
241 return 0;
242 }
243
244 static int
245 i386_set_ioperm(td, args)
246 struct thread *td;
247 char *args;
248 {
249 int i, error;
250 struct i386_ioperm_args ua;
251 char *iomap;
252
253 if ((error = copyin(args, &ua, sizeof(struct i386_ioperm_args))) != 0)
254 return (error);
255
256 #ifdef MAC
257 if ((error = mac_check_sysarch_ioperm(td->td_ucred)) != 0)
258 return (error);
259 #endif
260 if ((error = suser(td)) != 0)
261 return (error);
262 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
263 return (error);
264 /*
265 * XXX
266 * While this is restricted to root, we should probably figure out
267 * whether any other driver is using this i/o address, as so not to
268 * cause confusion. This probably requires a global 'usage registry'.
269 */
270
271 if (td->td_pcb->pcb_ext == 0)
272 if ((error = i386_extend_pcb(td)) != 0)
273 return (error);
274 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
275
276 if (ua.start + ua.length > IOPAGES * PAGE_SIZE * NBBY)
277 return (EINVAL);
278
279 for (i = ua.start; i < ua.start + ua.length; i++) {
280 if (ua.enable)
281 iomap[i >> 3] &= ~(1 << (i & 7));
282 else
283 iomap[i >> 3] |= (1 << (i & 7));
284 }
285 return (error);
286 }
287
288 static int
289 i386_get_ioperm(td, args)
290 struct thread *td;
291 char *args;
292 {
293 int i, state, error;
294 struct i386_ioperm_args ua;
295 char *iomap;
296
297 if ((error = copyin(args, &ua, sizeof(struct i386_ioperm_args))) != 0)
298 return (error);
299 if (ua.start >= IOPAGES * PAGE_SIZE * NBBY)
300 return (EINVAL);
301
302 if (td->td_pcb->pcb_ext == 0) {
303 ua.length = 0;
304 goto done;
305 }
306
307 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
308
309 i = ua.start;
310 state = (iomap[i >> 3] >> (i & 7)) & 1;
311 ua.enable = !state;
312 ua.length = 1;
313
314 for (i = ua.start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
315 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
316 break;
317 ua.length++;
318 }
319
320 done:
321 error = copyout(&ua, args, sizeof(struct i386_ioperm_args));
322 return (error);
323 }
324
325 /*
326 * Update the GDT entry pointing to the LDT to point to the LDT of the
327 * current process.
328 *
329 * This must be called with sched_lock held. Unfortunately, we can't use a
330 * mtx_assert() here because cpu_switch() calls this function after changing
331 * curproc but before sched_lock's owner is updated in mi_switch().
332 */
333 void
334 set_user_ldt(struct mdproc *mdp)
335 {
336 struct proc_ldt *pldt;
337
338 pldt = mdp->md_ldt;
339 #ifdef SMP
340 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd;
341 #else
342 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd;
343 #endif
344 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
345 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
346 }
347
348 #ifdef SMP
349 static void
350 set_user_ldt_rv(struct thread *td)
351 {
352
353 if (td->td_proc != curthread->td_proc)
354 return;
355
356 set_user_ldt(&td->td_proc->p_md);
357 }
358 #endif
359
360 /*
361 * Must be called with either sched_lock free or held but not recursed.
362 * If it does not return NULL, it will return with it owned.
363 */
364 struct proc_ldt *
365 user_ldt_alloc(struct mdproc *mdp, int len)
366 {
367 struct proc_ldt *pldt, *new_ldt;
368
369 if (mtx_owned(&sched_lock))
370 mtx_unlock_spin(&sched_lock);
371 mtx_assert(&sched_lock, MA_NOTOWNED);
372 MALLOC(new_ldt, struct proc_ldt *, sizeof(struct proc_ldt),
373 M_SUBPROC, M_WAITOK);
374
375 new_ldt->ldt_len = len = NEW_MAX_LD(len);
376 new_ldt->ldt_base = (caddr_t)kmem_alloc(kernel_map,
377 len * sizeof(union descriptor));
378 if (new_ldt->ldt_base == NULL) {
379 FREE(new_ldt, M_SUBPROC);
380 return NULL;
381 }
382 new_ldt->ldt_refcnt = 1;
383 new_ldt->ldt_active = 0;
384
385 mtx_lock_spin(&sched_lock);
386 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
387 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
388 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
389
390 if ((pldt = mdp->md_ldt)) {
391 if (len > pldt->ldt_len)
392 len = pldt->ldt_len;
393 bcopy(pldt->ldt_base, new_ldt->ldt_base,
394 len * sizeof(union descriptor));
395 } else {
396 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt));
397 }
398 return new_ldt;
399 }
400
401 /*
402 * Must be called either with sched_lock free or held but not recursed.
403 * If md_ldt is not NULL, it will return with sched_lock released.
404 */
405 void
406 user_ldt_free(struct thread *td)
407 {
408 struct mdproc *mdp = &td->td_proc->p_md;
409 struct proc_ldt *pldt = mdp->md_ldt;
410
411 if (pldt == NULL)
412 return;
413
414 if (!mtx_owned(&sched_lock))
415 mtx_lock_spin(&sched_lock);
416 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
417 if (td == PCPU_GET(curthread)) {
418 lldt(_default_ldt);
419 PCPU_SET(currentldt, _default_ldt);
420 }
421
422 mdp->md_ldt = NULL;
423 if (--pldt->ldt_refcnt == 0) {
424 mtx_unlock_spin(&sched_lock);
425 kmem_free(kernel_map, (vm_offset_t)pldt->ldt_base,
426 pldt->ldt_len * sizeof(union descriptor));
427 FREE(pldt, M_SUBPROC);
428 } else
429 mtx_unlock_spin(&sched_lock);
430 }
431
432 static int
433 i386_get_ldt(td, args)
434 struct thread *td;
435 char *args;
436 {
437 int error = 0;
438 struct proc_ldt *pldt = td->td_proc->p_md.md_ldt;
439 int nldt, num;
440 union descriptor *lp;
441 struct i386_ldt_args ua, *uap = &ua;
442
443 if ((error = copyin(args, uap, sizeof(struct i386_ldt_args))) < 0)
444 return(error);
445
446 #ifdef DEBUG
447 printf("i386_get_ldt: start=%d num=%d descs=%p\n",
448 uap->start, uap->num, (void *)uap->descs);
449 #endif
450
451 if (pldt) {
452 nldt = pldt->ldt_len;
453 num = min(uap->num, nldt);
454 lp = &((union descriptor *)(pldt->ldt_base))[uap->start];
455 } else {
456 nldt = sizeof(ldt)/sizeof(ldt[0]);
457 num = min(uap->num, nldt);
458 lp = &ldt[uap->start];
459 }
460
461 if ((uap->start > (unsigned int)nldt) ||
462 ((unsigned int)num > (unsigned int)nldt) ||
463 ((unsigned int)(uap->start + num) > (unsigned int)nldt))
464 return(EINVAL);
465
466 error = copyout(lp, uap->descs, num * sizeof(union descriptor));
467 if (!error)
468 td->td_retval[0] = num;
469
470 return(error);
471 }
472
473 static int ldt_warnings;
474 #define NUM_LDT_WARNINGS 10
475
476 static int
477 i386_set_ldt(td, args)
478 struct thread *td;
479 char *args;
480 {
481 int error = 0, i;
482 int largest_ld;
483 struct mdproc *mdp = &td->td_proc->p_md;
484 struct proc_ldt *pldt = 0;
485 struct i386_ldt_args ua, *uap = &ua;
486 union descriptor *descs, *dp;
487 int descs_size;
488
489 if ((error = copyin(args, uap, sizeof(struct i386_ldt_args))) < 0)
490 return(error);
491
492 #ifdef DEBUG
493 printf("i386_set_ldt: start=%d num=%d descs=%p\n",
494 uap->start, uap->num, (void *)uap->descs);
495 #endif
496
497 if (uap->descs == NULL) {
498 /* Free descriptors */
499 if (uap->start == 0 && uap->num == 0) {
500 /*
501 * Treat this as a special case, so userland needn't
502 * know magic number NLDT.
503 */
504 uap->start = NLDT;
505 uap->num = MAX_LD - NLDT;
506 }
507 if (uap->start <= LUDATA_SEL || uap->num <= 0)
508 return (EINVAL);
509 mtx_lock_spin(&sched_lock);
510 pldt = mdp->md_ldt;
511 if (pldt == NULL || uap->start >= pldt->ldt_len) {
512 mtx_unlock_spin(&sched_lock);
513 return (0);
514 }
515 largest_ld = uap->start + uap->num;
516 if (largest_ld > pldt->ldt_len)
517 largest_ld = pldt->ldt_len;
518 i = largest_ld - uap->start;
519 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start],
520 sizeof(union descriptor) * i);
521 mtx_unlock_spin(&sched_lock);
522 return (0);
523 }
524
525 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
526 /* complain a for a while if using old methods */
527 if (ldt_warnings++ < NUM_LDT_WARNINGS) {
528 printf("Warning: pid %d used static ldt allocation.\n",
529 td->td_proc->p_pid);
530 printf("See the i386_set_ldt man page for more info\n");
531 }
532 /* verify range of descriptors to modify */
533 largest_ld = uap->start + uap->num;
534 if (uap->start >= MAX_LD ||
535 uap->num < 0 || largest_ld > MAX_LD) {
536 return (EINVAL);
537 }
538 }
539
540 descs_size = uap->num * sizeof(union descriptor);
541 descs = (union descriptor *)kmem_alloc(kernel_map, descs_size);
542 if (descs == NULL)
543 return (ENOMEM);
544 error = copyin(uap->descs, descs, descs_size);
545 if (error) {
546 kmem_free(kernel_map, (vm_offset_t)descs, descs_size);
547 return (error);
548 }
549
550 /* Check descriptors for access violations */
551 for (i = 0; i < uap->num; i++) {
552 dp = &descs[i];
553
554 switch (dp->sd.sd_type) {
555 case SDT_SYSNULL: /* system null */
556 dp->sd.sd_p = 0;
557 break;
558 case SDT_SYS286TSS: /* system 286 TSS available */
559 case SDT_SYSLDT: /* system local descriptor table */
560 case SDT_SYS286BSY: /* system 286 TSS busy */
561 case SDT_SYSTASKGT: /* system task gate */
562 case SDT_SYS286IGT: /* system 286 interrupt gate */
563 case SDT_SYS286TGT: /* system 286 trap gate */
564 case SDT_SYSNULL2: /* undefined by Intel */
565 case SDT_SYS386TSS: /* system 386 TSS available */
566 case SDT_SYSNULL3: /* undefined by Intel */
567 case SDT_SYS386BSY: /* system 386 TSS busy */
568 case SDT_SYSNULL4: /* undefined by Intel */
569 case SDT_SYS386IGT: /* system 386 interrupt gate */
570 case SDT_SYS386TGT: /* system 386 trap gate */
571 case SDT_SYS286CGT: /* system 286 call gate */
572 case SDT_SYS386CGT: /* system 386 call gate */
573 /* I can't think of any reason to allow a user proc
574 * to create a segment of these types. They are
575 * for OS use only.
576 */
577 kmem_free(kernel_map, (vm_offset_t)descs, descs_size);
578 return (EACCES);
579 /*NOTREACHED*/
580
581 /* memory segment types */
582 case SDT_MEMEC: /* memory execute only conforming */
583 case SDT_MEMEAC: /* memory execute only accessed conforming */
584 case SDT_MEMERC: /* memory execute read conforming */
585 case SDT_MEMERAC: /* memory execute read accessed conforming */
586 /* Must be "present" if executable and conforming. */
587 if (dp->sd.sd_p == 0) {
588 kmem_free(kernel_map, (vm_offset_t)descs,
589 descs_size);
590 return (EACCES);
591 }
592 break;
593 case SDT_MEMRO: /* memory read only */
594 case SDT_MEMROA: /* memory read only accessed */
595 case SDT_MEMRW: /* memory read write */
596 case SDT_MEMRWA: /* memory read write accessed */
597 case SDT_MEMROD: /* memory read only expand dwn limit */
598 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
599 case SDT_MEMRWD: /* memory read write expand dwn limit */
600 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
601 case SDT_MEME: /* memory execute only */
602 case SDT_MEMEA: /* memory execute only accessed */
603 case SDT_MEMER: /* memory execute read */
604 case SDT_MEMERA: /* memory execute read accessed */
605 break;
606 default:
607 kmem_free(kernel_map, (vm_offset_t)descs, descs_size);
608 return(EINVAL);
609 /*NOTREACHED*/
610 }
611
612 /* Only user (ring-3) descriptors may be present. */
613 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL)) {
614 kmem_free(kernel_map, (vm_offset_t)descs, descs_size);
615 return (EACCES);
616 }
617 }
618
619 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
620 /* Allocate a free slot */
621 pldt = mdp->md_ldt;
622 if (pldt == NULL) {
623 error = i386_ldt_grow(td, NLDT+1);
624 if (error) {
625 kmem_free(kernel_map, (vm_offset_t)descs,
626 descs_size);
627 return (error);
628 }
629 pldt = mdp->md_ldt;
630 }
631 again:
632 mtx_lock_spin(&sched_lock);
633 /*
634 * start scanning a bit up to leave room for NVidia and
635 * Wine, which still user the "Blat" method of allocation.
636 */
637 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
638 for (i = NLDT; i < pldt->ldt_len; ++i) {
639 if (dp->sd.sd_type == SDT_SYSNULL)
640 break;
641 dp++;
642 }
643 if (i >= pldt->ldt_len) {
644 mtx_unlock_spin(&sched_lock);
645 error = i386_ldt_grow(td, pldt->ldt_len+1);
646 if (error) {
647 kmem_free(kernel_map, (vm_offset_t)descs,
648 descs_size);
649 return (error);
650 }
651 goto again;
652 }
653 uap->start = i;
654 error = i386_set_ldt_data(td, i, 1, descs);
655 mtx_unlock_spin(&sched_lock);
656 } else {
657 largest_ld = uap->start + uap->num;
658 error = i386_ldt_grow(td, largest_ld);
659 if (error == 0) {
660 mtx_lock_spin(&sched_lock);
661 error = i386_set_ldt_data(td, uap->start, uap->num,
662 descs);
663 mtx_unlock_spin(&sched_lock);
664 }
665 }
666 kmem_free(kernel_map, (vm_offset_t)descs, descs_size);
667 if (error == 0)
668 td->td_retval[0] = uap->start;
669 return (error);
670 }
671
672 static int
673 i386_set_ldt_data(struct thread *td, int start, int num,
674 union descriptor *descs)
675 {
676 struct mdproc *mdp = &td->td_proc->p_md;
677 struct proc_ldt *pldt = mdp->md_ldt;
678
679 mtx_assert(&sched_lock, MA_OWNED);
680
681 /* Fill in range */
682 bcopy(descs,
683 &((union descriptor *)(pldt->ldt_base))[start],
684 num * sizeof(union descriptor));
685 return (0);
686 }
687
688 static int
689 i386_ldt_grow(struct thread *td, int len)
690 {
691 struct mdproc *mdp = &td->td_proc->p_md;
692 struct proc_ldt *pldt;
693 caddr_t old_ldt_base;
694 int old_ldt_len;
695
696 if (len > MAX_LD)
697 return (ENOMEM);
698 if (len < NLDT+1)
699 len = NLDT+1;
700 pldt = mdp->md_ldt;
701 /* allocate user ldt */
702 if (!pldt || len > pldt->ldt_len) {
703 struct proc_ldt *new_ldt = user_ldt_alloc(mdp, len);
704 if (new_ldt == NULL)
705 return (ENOMEM);
706 pldt = mdp->md_ldt;
707 /* sched_lock was held by user_ldt_alloc */
708 if (pldt) {
709 if (new_ldt->ldt_len > pldt->ldt_len) {
710 old_ldt_base = pldt->ldt_base;
711 old_ldt_len = pldt->ldt_len;
712 pldt->ldt_sd = new_ldt->ldt_sd;
713 pldt->ldt_base = new_ldt->ldt_base;
714 pldt->ldt_len = new_ldt->ldt_len;
715 mtx_unlock_spin(&sched_lock);
716 kmem_free(kernel_map, (vm_offset_t)old_ldt_base,
717 old_ldt_len * sizeof(union descriptor));
718 FREE(new_ldt, M_SUBPROC);
719 mtx_lock_spin(&sched_lock);
720 } else {
721 /*
722 * If other threads already did the work,
723 * do nothing
724 */
725 mtx_unlock_spin(&sched_lock);
726 kmem_free(kernel_map,
727 (vm_offset_t)new_ldt->ldt_base,
728 new_ldt->ldt_len * sizeof(union descriptor));
729 FREE(new_ldt, M_SUBPROC);
730 return (0);
731 }
732 } else {
733 mdp->md_ldt = pldt = new_ldt;
734 }
735 #ifdef SMP
736 mtx_unlock_spin(&sched_lock);
737 /* signal other cpus to reload ldt */
738 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv,
739 NULL, td);
740 #else
741 set_user_ldt(mdp);
742 mtx_unlock_spin(&sched_lock);
743 #endif
744 }
745 return (0);
746 }
Cache object: dc8ebabf00722075ba699221c2be294f
|