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: releng/6.0/sys/i386/i386/sys_machdep.c 150588 2005-09-26 19:38:12Z jhb $");
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_set_ldt_data(struct thread *, int start, int num,
69 union descriptor *descs);
70 static int i386_ldt_grow(struct thread *td, int len);
71 #ifdef SMP
72 static void set_user_ldt_rv(struct thread *);
73 #endif
74
75 #ifndef _SYS_SYSPROTO_H_
76 struct sysarch_args {
77 int op;
78 char *parms;
79 };
80 #endif
81
82 int
83 sysarch(td, uap)
84 struct thread *td;
85 register struct sysarch_args *uap;
86 {
87 int error;
88 union descriptor *lp;
89 union {
90 struct i386_ldt_args largs;
91 struct i386_ioperm_args iargs;
92 } kargs;
93 uint32_t base;
94 struct segment_descriptor sd, *sdp;
95
96 switch (uap->op) {
97 case I386_GET_IOPERM:
98 case I386_SET_IOPERM:
99 if ((error = copyin(uap->parms, &kargs.iargs,
100 sizeof(struct i386_ioperm_args))) != 0)
101 return (error);
102 break;
103 case I386_GET_LDT:
104 case I386_SET_LDT:
105 if ((error = copyin(uap->parms, &kargs.largs,
106 sizeof(struct i386_ldt_args))) != 0)
107 return (error);
108 if (kargs.largs.num > MAX_LD || kargs.largs.num <= 0)
109 return (EINVAL);
110 break;
111 default:
112 break;
113 }
114
115 mtx_lock(&Giant);
116 switch(uap->op) {
117 case I386_GET_LDT:
118 error = i386_get_ldt(td, &kargs.largs);
119 break;
120 case I386_SET_LDT:
121 if (kargs.largs.descs != NULL) {
122 lp = (union descriptor *)kmem_alloc(kernel_map,
123 kargs.largs.num * sizeof(union descriptor));
124 if (lp == NULL) {
125 error = ENOMEM;
126 break;
127 }
128 error = copyin(kargs.largs.descs, lp,
129 kargs.largs.num * sizeof(union descriptor));
130 if (error == 0)
131 error = i386_set_ldt(td, &kargs.largs, lp);
132 kmem_free(kernel_map, (vm_offset_t)lp,
133 kargs.largs.num * sizeof(union descriptor));
134 } else {
135 error = i386_set_ldt(td, &kargs.largs, NULL);
136 }
137 break;
138 case I386_GET_IOPERM:
139 error = i386_get_ioperm(td, &kargs.iargs);
140 if (error == 0)
141 error = copyout(&kargs.iargs, uap->parms,
142 sizeof(struct i386_ioperm_args));
143 break;
144 case I386_SET_IOPERM:
145 error = i386_set_ioperm(td, &kargs.iargs);
146 break;
147 case I386_VM86:
148 error = vm86_sysarch(td, uap->parms);
149 break;
150 case I386_GET_FSBASE:
151 sdp = &td->td_pcb->pcb_fsd;
152 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
153 error = copyout(&base, uap->parms, sizeof(base));
154 break;
155 case I386_SET_FSBASE:
156 error = copyin(uap->parms, &base, sizeof(base));
157 if (!error) {
158 /*
159 * Construct a descriptor and store it in the pcb for
160 * the next context switch. Also store it in the gdt
161 * so that the load of tf_fs into %fs will activate it
162 * at return to userland.
163 */
164 sd.sd_lobase = base & 0xffffff;
165 sd.sd_hibase = (base >> 24) & 0xff;
166 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
167 sd.sd_hilimit = 0xf;
168 sd.sd_type = SDT_MEMRWA;
169 sd.sd_dpl = SEL_UPL;
170 sd.sd_p = 1;
171 sd.sd_xx = 0;
172 sd.sd_def32 = 1;
173 sd.sd_gran = 1;
174 critical_enter();
175 td->td_pcb->pcb_fsd = sd;
176 PCPU_GET(fsgs_gdt)[0] = sd;
177 critical_exit();
178 td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL);
179 }
180 break;
181 case I386_GET_GSBASE:
182 sdp = &td->td_pcb->pcb_gsd;
183 base = sdp->sd_hibase << 24 | sdp->sd_lobase;
184 error = copyout(&base, uap->parms, sizeof(base));
185 break;
186 case I386_SET_GSBASE:
187 error = copyin(uap->parms, &base, sizeof(base));
188 if (!error) {
189 /*
190 * Construct a descriptor and store it in the pcb for
191 * the next context switch. Also store it in the gdt
192 * because we have to do a load_gs() right now.
193 */
194 sd.sd_lobase = base & 0xffffff;
195 sd.sd_hibase = (base >> 24) & 0xff;
196 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
197 sd.sd_hilimit = 0xf;
198 sd.sd_type = SDT_MEMRWA;
199 sd.sd_dpl = SEL_UPL;
200 sd.sd_p = 1;
201 sd.sd_xx = 0;
202 sd.sd_def32 = 1;
203 sd.sd_gran = 1;
204 critical_enter();
205 td->td_pcb->pcb_gsd = sd;
206 PCPU_GET(fsgs_gdt)[1] = sd;
207 critical_exit();
208 load_gs(GSEL(GUGS_SEL, SEL_UPL));
209 }
210 break;
211 default:
212 error = EINVAL;
213 break;
214 }
215 mtx_unlock(&Giant);
216 return (error);
217 }
218
219 int
220 i386_extend_pcb(struct thread *td)
221 {
222 int i, offset;
223 u_long *addr;
224 struct pcb_ext *ext;
225 struct soft_segment_descriptor ssd = {
226 0, /* segment base address (overwritten) */
227 ctob(IOPAGES + 1) - 1, /* length */
228 SDT_SYS386TSS, /* segment type */
229 0, /* priority level */
230 1, /* descriptor present */
231 0, 0,
232 0, /* default 32 size */
233 0 /* granularity */
234 };
235
236 if (td->td_proc->p_flag & P_SA)
237 return (EINVAL); /* XXXKSE */
238 /* XXXKSE All the code below only works in 1:1 needs changing */
239 ext = (struct pcb_ext *)kmem_alloc(kernel_map, ctob(IOPAGES+1));
240 if (ext == 0)
241 return (ENOMEM);
242 bzero(ext, sizeof(struct pcb_ext));
243 /* -16 is so we can convert a trapframe into vm86trapframe inplace */
244 ext->ext_tss.tss_esp0 = td->td_kstack + ctob(KSTACK_PAGES) -
245 sizeof(struct pcb) - 16;
246 ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
247 /*
248 * The last byte of the i/o map must be followed by an 0xff byte.
249 * We arbitrarily allocate 16 bytes here, to keep the starting
250 * address on a doubleword boundary.
251 */
252 offset = PAGE_SIZE - 16;
253 ext->ext_tss.tss_ioopt =
254 (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16;
255 ext->ext_iomap = (caddr_t)ext + offset;
256 ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32;
257
258 addr = (u_long *)ext->ext_vm86.vm86_intmap;
259 for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++)
260 *addr++ = ~0;
261
262 ssd.ssd_base = (unsigned)&ext->ext_tss;
263 ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext);
264 ssdtosd(&ssd, &ext->ext_tssd);
265
266 KASSERT(td == curthread, ("giving TSS to !curthread"));
267 KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!"));
268
269 /* Switch to the new TSS. */
270 mtx_lock_spin(&sched_lock);
271 td->td_pcb->pcb_ext = ext;
272 private_tss |= PCPU_GET(cpumask);
273 *PCPU_GET(tss_gdt) = ext->ext_tssd;
274 ltr(GSEL(GPROC0_SEL, SEL_KPL));
275 mtx_unlock_spin(&sched_lock);
276
277 return 0;
278 }
279
280 int
281 i386_set_ioperm(td, uap)
282 struct thread *td;
283 struct i386_ioperm_args *uap;
284 {
285 int i, error;
286 char *iomap;
287
288 #ifdef MAC
289 if ((error = mac_check_sysarch_ioperm(td->td_ucred)) != 0)
290 return (error);
291 #endif
292 if ((error = suser(td)) != 0)
293 return (error);
294 if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
295 return (error);
296 /*
297 * XXX
298 * While this is restricted to root, we should probably figure out
299 * whether any other driver is using this i/o address, as so not to
300 * cause confusion. This probably requires a global 'usage registry'.
301 */
302
303 if (td->td_pcb->pcb_ext == 0)
304 if ((error = i386_extend_pcb(td)) != 0)
305 return (error);
306 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
307
308 if (uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY)
309 return (EINVAL);
310
311 for (i = uap->start; i < uap->start + uap->length; i++) {
312 if (uap->enable)
313 iomap[i >> 3] &= ~(1 << (i & 7));
314 else
315 iomap[i >> 3] |= (1 << (i & 7));
316 }
317 return (error);
318 }
319
320 int
321 i386_get_ioperm(td, uap)
322 struct thread *td;
323 struct i386_ioperm_args *uap;
324 {
325 int i, state;
326 char *iomap;
327
328 if (uap->start >= IOPAGES * PAGE_SIZE * NBBY)
329 return (EINVAL);
330
331 if (td->td_pcb->pcb_ext == 0) {
332 uap->length = 0;
333 goto done;
334 }
335
336 iomap = (char *)td->td_pcb->pcb_ext->ext_iomap;
337
338 i = uap->start;
339 state = (iomap[i >> 3] >> (i & 7)) & 1;
340 uap->enable = !state;
341 uap->length = 1;
342
343 for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) {
344 if (state != ((iomap[i >> 3] >> (i & 7)) & 1))
345 break;
346 uap->length++;
347 }
348
349 done:
350 return (0);
351 }
352
353 /*
354 * Update the GDT entry pointing to the LDT to point to the LDT of the
355 * current process.
356 *
357 * This must be called with sched_lock held. Unfortunately, we can't use a
358 * mtx_assert() here because cpu_switch() calls this function after changing
359 * curproc but before sched_lock's owner is updated in mi_switch().
360 */
361 void
362 set_user_ldt(struct mdproc *mdp)
363 {
364 struct proc_ldt *pldt;
365
366 pldt = mdp->md_ldt;
367 #ifdef SMP
368 gdt[PCPU_GET(cpuid) * NGDT + GUSERLDT_SEL].sd = pldt->ldt_sd;
369 #else
370 gdt[GUSERLDT_SEL].sd = pldt->ldt_sd;
371 #endif
372 lldt(GSEL(GUSERLDT_SEL, SEL_KPL));
373 PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL));
374 }
375
376 #ifdef SMP
377 static void
378 set_user_ldt_rv(struct thread *td)
379 {
380
381 if (td->td_proc != curthread->td_proc)
382 return;
383
384 set_user_ldt(&td->td_proc->p_md);
385 }
386 #endif
387
388 /*
389 * Must be called with either sched_lock free or held but not recursed.
390 * If it does not return NULL, it will return with it owned.
391 */
392 struct proc_ldt *
393 user_ldt_alloc(struct mdproc *mdp, int len)
394 {
395 struct proc_ldt *pldt, *new_ldt;
396
397 if (mtx_owned(&sched_lock))
398 mtx_unlock_spin(&sched_lock);
399 mtx_assert(&sched_lock, MA_NOTOWNED);
400 MALLOC(new_ldt, struct proc_ldt *, sizeof(struct proc_ldt),
401 M_SUBPROC, M_WAITOK);
402
403 new_ldt->ldt_len = len = NEW_MAX_LD(len);
404 new_ldt->ldt_base = (caddr_t)kmem_alloc(kernel_map,
405 len * sizeof(union descriptor));
406 if (new_ldt->ldt_base == NULL) {
407 FREE(new_ldt, M_SUBPROC);
408 return NULL;
409 }
410 new_ldt->ldt_refcnt = 1;
411 new_ldt->ldt_active = 0;
412
413 mtx_lock_spin(&sched_lock);
414 gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base;
415 gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1;
416 ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd);
417
418 if ((pldt = mdp->md_ldt)) {
419 if (len > pldt->ldt_len)
420 len = pldt->ldt_len;
421 bcopy(pldt->ldt_base, new_ldt->ldt_base,
422 len * sizeof(union descriptor));
423 } else {
424 bcopy(ldt, new_ldt->ldt_base, sizeof(ldt));
425 }
426 return new_ldt;
427 }
428
429 /*
430 * Must be called either with sched_lock free or held but not recursed.
431 * If md_ldt is not NULL, it will return with sched_lock released.
432 */
433 void
434 user_ldt_free(struct thread *td)
435 {
436 struct mdproc *mdp = &td->td_proc->p_md;
437 struct proc_ldt *pldt = mdp->md_ldt;
438
439 if (pldt == NULL)
440 return;
441
442 if (!mtx_owned(&sched_lock))
443 mtx_lock_spin(&sched_lock);
444 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
445 if (td == PCPU_GET(curthread)) {
446 lldt(_default_ldt);
447 PCPU_SET(currentldt, _default_ldt);
448 }
449
450 mdp->md_ldt = NULL;
451 if (--pldt->ldt_refcnt == 0) {
452 mtx_unlock_spin(&sched_lock);
453 kmem_free(kernel_map, (vm_offset_t)pldt->ldt_base,
454 pldt->ldt_len * sizeof(union descriptor));
455 FREE(pldt, M_SUBPROC);
456 } else
457 mtx_unlock_spin(&sched_lock);
458 }
459
460 /*
461 * Note for the authors of compat layers (linux, etc): copyout() in
462 * the function below is not a problem since it presents data in
463 * arch-specific format (i.e. i386-specific in this case), not in
464 * the OS-specific one.
465 */
466 int
467 i386_get_ldt(td, uap)
468 struct thread *td;
469 struct i386_ldt_args *uap;
470 {
471 int error = 0;
472 struct proc_ldt *pldt = td->td_proc->p_md.md_ldt;
473 int nldt, num;
474 union descriptor *lp;
475
476 #ifdef DEBUG
477 printf("i386_get_ldt: start=%d num=%d descs=%p\n",
478 uap->start, uap->num, (void *)uap->descs);
479 #endif
480
481 if (pldt) {
482 nldt = pldt->ldt_len;
483 num = min(uap->num, nldt);
484 lp = &((union descriptor *)(pldt->ldt_base))[uap->start];
485 } else {
486 nldt = sizeof(ldt)/sizeof(ldt[0]);
487 num = min(uap->num, nldt);
488 lp = &ldt[uap->start];
489 }
490
491 if ((uap->start > (unsigned int)nldt) ||
492 ((unsigned int)num > (unsigned int)nldt) ||
493 ((unsigned int)(uap->start + num) > (unsigned int)nldt))
494 return(EINVAL);
495
496 error = copyout(lp, uap->descs, num * sizeof(union descriptor));
497 if (!error)
498 td->td_retval[0] = num;
499
500 return(error);
501 }
502
503 static int ldt_warnings;
504 #define NUM_LDT_WARNINGS 10
505
506 int
507 i386_set_ldt(td, uap, descs)
508 struct thread *td;
509 struct i386_ldt_args *uap;
510 union descriptor *descs;
511 {
512 int error = 0, i;
513 int largest_ld;
514 struct mdproc *mdp = &td->td_proc->p_md;
515 struct proc_ldt *pldt;
516 union descriptor *dp;
517
518 #ifdef DEBUG
519 printf("i386_set_ldt: start=%d num=%d descs=%p\n",
520 uap->start, uap->num, (void *)uap->descs);
521 #endif
522
523 if (descs == NULL) {
524 /* Free descriptors */
525 if (uap->start == 0 && uap->num == 0) {
526 /*
527 * Treat this as a special case, so userland needn't
528 * know magic number NLDT.
529 */
530 uap->start = NLDT;
531 uap->num = MAX_LD - NLDT;
532 }
533 if (uap->num <= 0)
534 return (EINVAL);
535 mtx_lock_spin(&sched_lock);
536 pldt = mdp->md_ldt;
537 if (pldt == NULL || uap->start >= pldt->ldt_len) {
538 mtx_unlock_spin(&sched_lock);
539 return (0);
540 }
541 largest_ld = uap->start + uap->num;
542 if (largest_ld > pldt->ldt_len)
543 largest_ld = pldt->ldt_len;
544 i = largest_ld - uap->start;
545 bzero(&((union descriptor *)(pldt->ldt_base))[uap->start],
546 sizeof(union descriptor) * i);
547 mtx_unlock_spin(&sched_lock);
548 return (0);
549 }
550
551 if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) {
552 /* complain a for a while if using old methods */
553 if (ldt_warnings++ < NUM_LDT_WARNINGS) {
554 printf("Warning: pid %d used static ldt allocation.\n",
555 td->td_proc->p_pid);
556 printf("See the i386_set_ldt man page for more info\n");
557 }
558 /* verify range of descriptors to modify */
559 largest_ld = uap->start + uap->num;
560 if (uap->start >= MAX_LD ||
561 uap->num < 0 || largest_ld > MAX_LD) {
562 return (EINVAL);
563 }
564 }
565
566 /* Check descriptors for access violations */
567 for (i = 0; i < uap->num; i++) {
568 dp = &descs[i];
569
570 switch (dp->sd.sd_type) {
571 case SDT_SYSNULL: /* system null */
572 dp->sd.sd_p = 0;
573 break;
574 case SDT_SYS286TSS: /* system 286 TSS available */
575 case SDT_SYSLDT: /* system local descriptor table */
576 case SDT_SYS286BSY: /* system 286 TSS busy */
577 case SDT_SYSTASKGT: /* system task gate */
578 case SDT_SYS286IGT: /* system 286 interrupt gate */
579 case SDT_SYS286TGT: /* system 286 trap gate */
580 case SDT_SYSNULL2: /* undefined by Intel */
581 case SDT_SYS386TSS: /* system 386 TSS available */
582 case SDT_SYSNULL3: /* undefined by Intel */
583 case SDT_SYS386BSY: /* system 386 TSS busy */
584 case SDT_SYSNULL4: /* undefined by Intel */
585 case SDT_SYS386IGT: /* system 386 interrupt gate */
586 case SDT_SYS386TGT: /* system 386 trap gate */
587 case SDT_SYS286CGT: /* system 286 call gate */
588 case SDT_SYS386CGT: /* system 386 call gate */
589 /* I can't think of any reason to allow a user proc
590 * to create a segment of these types. They are
591 * for OS use only.
592 */
593 return (EACCES);
594 /*NOTREACHED*/
595
596 /* memory segment types */
597 case SDT_MEMEC: /* memory execute only conforming */
598 case SDT_MEMEAC: /* memory execute only accessed conforming */
599 case SDT_MEMERC: /* memory execute read conforming */
600 case SDT_MEMERAC: /* memory execute read accessed conforming */
601 /* Must be "present" if executable and conforming. */
602 if (dp->sd.sd_p == 0)
603 return (EACCES);
604 break;
605 case SDT_MEMRO: /* memory read only */
606 case SDT_MEMROA: /* memory read only accessed */
607 case SDT_MEMRW: /* memory read write */
608 case SDT_MEMRWA: /* memory read write accessed */
609 case SDT_MEMROD: /* memory read only expand dwn limit */
610 case SDT_MEMRODA: /* memory read only expand dwn lim accessed */
611 case SDT_MEMRWD: /* memory read write expand dwn limit */
612 case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */
613 case SDT_MEME: /* memory execute only */
614 case SDT_MEMEA: /* memory execute only accessed */
615 case SDT_MEMER: /* memory execute read */
616 case SDT_MEMERA: /* memory execute read accessed */
617 break;
618 default:
619 return(EINVAL);
620 /*NOTREACHED*/
621 }
622
623 /* Only user (ring-3) descriptors may be present. */
624 if ((dp->sd.sd_p != 0) && (dp->sd.sd_dpl != SEL_UPL))
625 return (EACCES);
626 }
627
628 if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) {
629 /* Allocate a free slot */
630 pldt = mdp->md_ldt;
631 if (pldt == NULL) {
632 error = i386_ldt_grow(td, NLDT + 1);
633 if (error)
634 return (error);
635 pldt = mdp->md_ldt;
636 }
637 again:
638 mtx_lock_spin(&sched_lock);
639 /*
640 * start scanning a bit up to leave room for NVidia and
641 * Wine, which still user the "Blat" method of allocation.
642 */
643 dp = &((union descriptor *)(pldt->ldt_base))[NLDT];
644 for (i = NLDT; i < pldt->ldt_len; ++i) {
645 if (dp->sd.sd_type == SDT_SYSNULL)
646 break;
647 dp++;
648 }
649 if (i >= pldt->ldt_len) {
650 mtx_unlock_spin(&sched_lock);
651 error = i386_ldt_grow(td, pldt->ldt_len+1);
652 if (error)
653 return (error);
654 goto again;
655 }
656 uap->start = i;
657 error = i386_set_ldt_data(td, i, 1, descs);
658 mtx_unlock_spin(&sched_lock);
659 } else {
660 largest_ld = uap->start + uap->num;
661 error = i386_ldt_grow(td, largest_ld);
662 if (error == 0) {
663 mtx_lock_spin(&sched_lock);
664 error = i386_set_ldt_data(td, uap->start, uap->num,
665 descs);
666 mtx_unlock_spin(&sched_lock);
667 }
668 }
669 if (error == 0)
670 td->td_retval[0] = uap->start;
671 return (error);
672 }
673
674 static int
675 i386_set_ldt_data(struct thread *td, int start, int num,
676 union descriptor *descs)
677 {
678 struct mdproc *mdp = &td->td_proc->p_md;
679 struct proc_ldt *pldt = mdp->md_ldt;
680
681 mtx_assert(&sched_lock, MA_OWNED);
682
683 /* Fill in range */
684 bcopy(descs,
685 &((union descriptor *)(pldt->ldt_base))[start],
686 num * sizeof(union descriptor));
687 return (0);
688 }
689
690 static int
691 i386_ldt_grow(struct thread *td, int len)
692 {
693 struct mdproc *mdp = &td->td_proc->p_md;
694 struct proc_ldt *pldt;
695 caddr_t old_ldt_base;
696 int old_ldt_len;
697
698 if (len > MAX_LD)
699 return (ENOMEM);
700 if (len < NLDT + 1)
701 len = NLDT + 1;
702
703 /* Allocate a user ldt. */
704 pldt = mdp->md_ldt;
705 if (!pldt || len > pldt->ldt_len) {
706 struct proc_ldt *new_ldt;
707
708 new_ldt = user_ldt_alloc(mdp, len);
709 if (new_ldt == NULL)
710 return (ENOMEM);
711 pldt = mdp->md_ldt;
712
713 /* sched_lock was acquired by user_ldt_alloc. */
714 if (pldt) {
715 if (new_ldt->ldt_len > pldt->ldt_len) {
716 old_ldt_base = pldt->ldt_base;
717 old_ldt_len = pldt->ldt_len;
718 pldt->ldt_sd = new_ldt->ldt_sd;
719 pldt->ldt_base = new_ldt->ldt_base;
720 pldt->ldt_len = new_ldt->ldt_len;
721 mtx_unlock_spin(&sched_lock);
722 kmem_free(kernel_map, (vm_offset_t)old_ldt_base,
723 old_ldt_len * sizeof(union descriptor));
724 FREE(new_ldt, M_SUBPROC);
725 mtx_lock_spin(&sched_lock);
726 } else {
727 /*
728 * If other threads already did the work,
729 * do nothing.
730 */
731 mtx_unlock_spin(&sched_lock);
732 kmem_free(kernel_map,
733 (vm_offset_t)new_ldt->ldt_base,
734 new_ldt->ldt_len * sizeof(union descriptor));
735 FREE(new_ldt, M_SUBPROC);
736 return (0);
737 }
738 } else {
739 mdp->md_ldt = pldt = new_ldt;
740 }
741 #ifdef SMP
742 mtx_unlock_spin(&sched_lock);
743 /* signal other cpus to reload ldt */
744 smp_rendezvous(NULL, (void (*)(void *))set_user_ldt_rv,
745 NULL, td);
746 #else
747 set_user_ldt(mdp);
748 mtx_unlock_spin(&sched_lock);
749 #endif
750 }
751 return (0);
752 }
Cache object: 396ab346c9e7da2c6e29640d45e719ea
|