1 /*-
2 * Copyright (c) 2004 Marcel Moolenaar
3 * Copyright (c) 2001 Doug Rabson
4 * Copyright (c) 2016 The FreeBSD Foundation
5 * All rights reserved.
6 *
7 * Portions of this software were developed by Konstantin Belousov
8 * under sponsorship from the FreeBSD Foundation.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/11.1/sys/amd64/amd64/efirt.c 318576 2017-05-20 16:12:44Z kib $");
34
35 #include <sys/param.h>
36 #include <sys/efi.h>
37 #include <sys/kernel.h>
38 #include <sys/linker.h>
39 #include <sys/lock.h>
40 #include <sys/module.h>
41 #include <sys/mutex.h>
42 #include <sys/clock.h>
43 #include <sys/proc.h>
44 #include <sys/rwlock.h>
45 #include <sys/sched.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <machine/fpu.h>
49 #include <machine/efi.h>
50 #include <machine/metadata.h>
51 #include <machine/md_var.h>
52 #include <machine/smp.h>
53 #include <machine/vmparam.h>
54 #include <vm/vm.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
60
61 static struct efi_systbl *efi_systbl;
62 static struct efi_cfgtbl *efi_cfgtbl;
63 static struct efi_rt *efi_runtime;
64
65 static int efi_status2err[25] = {
66 0, /* EFI_SUCCESS */
67 ENOEXEC, /* EFI_LOAD_ERROR */
68 EINVAL, /* EFI_INVALID_PARAMETER */
69 ENOSYS, /* EFI_UNSUPPORTED */
70 EMSGSIZE, /* EFI_BAD_BUFFER_SIZE */
71 EOVERFLOW, /* EFI_BUFFER_TOO_SMALL */
72 EBUSY, /* EFI_NOT_READY */
73 EIO, /* EFI_DEVICE_ERROR */
74 EROFS, /* EFI_WRITE_PROTECTED */
75 EAGAIN, /* EFI_OUT_OF_RESOURCES */
76 EIO, /* EFI_VOLUME_CORRUPTED */
77 ENOSPC, /* EFI_VOLUME_FULL */
78 ENXIO, /* EFI_NO_MEDIA */
79 ESTALE, /* EFI_MEDIA_CHANGED */
80 ENOENT, /* EFI_NOT_FOUND */
81 EACCES, /* EFI_ACCESS_DENIED */
82 ETIMEDOUT, /* EFI_NO_RESPONSE */
83 EADDRNOTAVAIL, /* EFI_NO_MAPPING */
84 ETIMEDOUT, /* EFI_TIMEOUT */
85 EDOOFUS, /* EFI_NOT_STARTED */
86 EALREADY, /* EFI_ALREADY_STARTED */
87 ECANCELED, /* EFI_ABORTED */
88 EPROTO, /* EFI_ICMP_ERROR */
89 EPROTO, /* EFI_TFTP_ERROR */
90 EPROTO /* EFI_PROTOCOL_ERROR */
91 };
92
93 static int
94 efi_status_to_errno(efi_status status)
95 {
96 u_long code;
97
98 code = status & 0x3ffffffffffffffful;
99 return (code < nitems(efi_status2err) ? efi_status2err[code] : EDOOFUS);
100 }
101
102 static struct mtx efi_lock;
103 static pml4_entry_t *efi_pml4;
104 static vm_object_t obj_1t1_pt;
105 static vm_page_t efi_pml4_page;
106
107 static void
108 efi_destroy_1t1_map(void)
109 {
110 vm_page_t m;
111
112 if (obj_1t1_pt != NULL) {
113 VM_OBJECT_RLOCK(obj_1t1_pt);
114 TAILQ_FOREACH(m, &obj_1t1_pt->memq, listq)
115 m->wire_count = 0;
116 atomic_subtract_int(&vm_cnt.v_wire_count,
117 obj_1t1_pt->resident_page_count);
118 VM_OBJECT_RUNLOCK(obj_1t1_pt);
119 vm_object_deallocate(obj_1t1_pt);
120 }
121
122 obj_1t1_pt = NULL;
123 efi_pml4 = NULL;
124 efi_pml4_page = NULL;
125 }
126
127 static vm_page_t
128 efi_1t1_page(vm_pindex_t idx)
129 {
130
131 return (vm_page_grab(obj_1t1_pt, idx, VM_ALLOC_NOBUSY |
132 VM_ALLOC_WIRED | VM_ALLOC_ZERO));
133 }
134
135 static pt_entry_t *
136 efi_1t1_pte(vm_offset_t va)
137 {
138 pml4_entry_t *pml4e;
139 pdp_entry_t *pdpe;
140 pd_entry_t *pde;
141 pt_entry_t *pte;
142 vm_page_t m;
143 vm_pindex_t pml4_idx, pdp_idx, pd_idx;
144 vm_paddr_t mphys;
145
146 pml4_idx = pmap_pml4e_index(va);
147 pml4e = &efi_pml4[pml4_idx];
148 if (*pml4e == 0) {
149 m = efi_1t1_page(1 + pml4_idx);
150 mphys = VM_PAGE_TO_PHYS(m);
151 *pml4e = mphys | X86_PG_RW | X86_PG_V;
152 } else {
153 mphys = *pml4e & ~PAGE_MASK;
154 }
155
156 pdpe = (pdp_entry_t *)PHYS_TO_DMAP(mphys);
157 pdp_idx = pmap_pdpe_index(va);
158 pdpe += pdp_idx;
159 if (*pdpe == 0) {
160 m = efi_1t1_page(1 + NPML4EPG + (pml4_idx + 1) * (pdp_idx + 1));
161 mphys = VM_PAGE_TO_PHYS(m);
162 *pdpe = mphys | X86_PG_RW | X86_PG_V;
163 } else {
164 mphys = *pdpe & ~PAGE_MASK;
165 }
166
167 pde = (pd_entry_t *)PHYS_TO_DMAP(mphys);
168 pd_idx = pmap_pde_index(va);
169 pde += pd_idx;
170 if (*pde == 0) {
171 m = efi_1t1_page(1 + NPML4EPG + NPML4EPG * NPDPEPG +
172 (pml4_idx + 1) * (pdp_idx + 1) * (pd_idx + 1));
173 mphys = VM_PAGE_TO_PHYS(m);
174 *pde = mphys | X86_PG_RW | X86_PG_V;
175 } else {
176 mphys = *pde & ~PAGE_MASK;
177 }
178
179 pte = (pt_entry_t *)PHYS_TO_DMAP(mphys);
180 pte += pmap_pte_index(va);
181 KASSERT(*pte == 0, ("va %#jx *pt %#jx", va, *pte));
182
183 return (pte);
184 }
185
186 static bool
187 efi_create_1t1_map(struct efi_md *map, int ndesc, int descsz)
188 {
189 struct efi_md *p;
190 pt_entry_t *pte;
191 vm_offset_t va;
192 uint64_t idx;
193 int bits, i, mode;
194
195 obj_1t1_pt = vm_pager_allocate(OBJT_PHYS, NULL, 1 + NPML4EPG +
196 NPML4EPG * NPDPEPG + NPML4EPG * NPDPEPG * NPDEPG,
197 VM_PROT_ALL, 0, NULL);
198 VM_OBJECT_WLOCK(obj_1t1_pt);
199 efi_pml4_page = efi_1t1_page(0);
200 VM_OBJECT_WUNLOCK(obj_1t1_pt);
201 efi_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(efi_pml4_page));
202 pmap_pinit_pml4(efi_pml4_page);
203
204 for (i = 0, p = map; i < ndesc; i++, p = efi_next_descriptor(p,
205 descsz)) {
206 if ((p->md_attr & EFI_MD_ATTR_RT) == 0)
207 continue;
208 if (p->md_virt != NULL) {
209 if (bootverbose)
210 printf("EFI Runtime entry %d is mapped\n", i);
211 goto fail;
212 }
213 if ((p->md_phys & EFI_PAGE_MASK) != 0) {
214 if (bootverbose)
215 printf("EFI Runtime entry %d is not aligned\n",
216 i);
217 goto fail;
218 }
219 if (p->md_phys + p->md_pages * EFI_PAGE_SIZE < p->md_phys ||
220 p->md_phys + p->md_pages * EFI_PAGE_SIZE >=
221 VM_MAXUSER_ADDRESS) {
222 printf("EFI Runtime entry %d is not in mappable for RT:"
223 "base %#016jx %#jx pages\n",
224 i, (uintmax_t)p->md_phys,
225 (uintmax_t)p->md_pages);
226 goto fail;
227 }
228 if ((p->md_attr & EFI_MD_ATTR_WB) != 0)
229 mode = VM_MEMATTR_WRITE_BACK;
230 else if ((p->md_attr & EFI_MD_ATTR_WT) != 0)
231 mode = VM_MEMATTR_WRITE_THROUGH;
232 else if ((p->md_attr & EFI_MD_ATTR_WC) != 0)
233 mode = VM_MEMATTR_WRITE_COMBINING;
234 else if ((p->md_attr & EFI_MD_ATTR_WP) != 0)
235 mode = VM_MEMATTR_WRITE_PROTECTED;
236 else if ((p->md_attr & EFI_MD_ATTR_UC) != 0)
237 mode = VM_MEMATTR_UNCACHEABLE;
238 else {
239 if (bootverbose)
240 printf("EFI Runtime entry %d mapping "
241 "attributes unsupported\n", i);
242 mode = VM_MEMATTR_UNCACHEABLE;
243 }
244 bits = pmap_cache_bits(kernel_pmap, mode, FALSE) | X86_PG_RW |
245 X86_PG_V;
246 VM_OBJECT_WLOCK(obj_1t1_pt);
247 for (va = p->md_phys, idx = 0; idx < p->md_pages; idx++,
248 va += PAGE_SIZE) {
249 pte = efi_1t1_pte(va);
250 pte_store(pte, va | bits);
251 }
252 VM_OBJECT_WUNLOCK(obj_1t1_pt);
253 }
254
255 return (true);
256
257 fail:
258 efi_destroy_1t1_map();
259 return (false);
260 }
261
262 /*
263 * Create an environment for the EFI runtime code call. The most
264 * important part is creating the required 1:1 physical->virtual
265 * mappings for the runtime segments. To do that, we manually create
266 * page table which unmap userspace but gives correct kernel mapping.
267 * The 1:1 mappings for runtime segments usually occupy low 4G of the
268 * physical address map.
269 *
270 * The 1:1 mappings were chosen over the SetVirtualAddressMap() EFI RT
271 * service, because there are some BIOSes which fail to correctly
272 * relocate itself on the call, requiring both 1:1 and virtual
273 * mapping. As result, we must provide 1:1 mapping anyway, so no
274 * reason to bother with the virtual map, and no need to add a
275 * complexity into loader.
276 *
277 * The fpu_kern_enter() call allows firmware to use FPU, as mandated
278 * by the specification. In particular, CR0.TS bit is cleared. Also
279 * it enters critical section, giving us neccessary protection against
280 * context switch.
281 *
282 * There is no need to disable interrupts around the change of %cr3,
283 * the kernel mappings are correct, while we only grabbed the
284 * userspace portion of VA. Interrupts handlers must not access
285 * userspace. Having interrupts enabled fixes the issue with
286 * firmware/SMM long operation, which would negatively affect IPIs,
287 * esp. TLB shootdown requests.
288 */
289 static int
290 efi_enter(void)
291 {
292 pmap_t curpmap;
293 int error;
294
295 if (efi_runtime == NULL)
296 return (ENXIO);
297 curpmap = PCPU_GET(curpmap);
298 PMAP_LOCK(curpmap);
299 mtx_lock(&efi_lock);
300 error = fpu_kern_enter(curthread, NULL, FPU_KERN_NOCTX);
301 if (error != 0) {
302 PMAP_UNLOCK(curpmap);
303 return (error);
304 }
305
306 /*
307 * IPI TLB shootdown handler invltlb_pcid_handler() reloads
308 * %cr3 from the curpmap->pm_cr3, which would disable runtime
309 * segments mappings. Block the handler's action by setting
310 * curpmap to impossible value. See also comment in
311 * pmap.c:pmap_activate_sw().
312 */
313 if (pmap_pcid_enabled && !invpcid_works)
314 PCPU_SET(curpmap, NULL);
315
316 load_cr3(VM_PAGE_TO_PHYS(efi_pml4_page) | (pmap_pcid_enabled ?
317 curpmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid : 0));
318 /*
319 * If PCID is enabled, the clear CR3_PCID_SAVE bit in the loaded %cr3
320 * causes TLB invalidation.
321 */
322 if (!pmap_pcid_enabled)
323 invltlb();
324 return (0);
325 }
326
327 static void
328 efi_leave(void)
329 {
330 pmap_t curpmap;
331
332 curpmap = &curproc->p_vmspace->vm_pmap;
333 if (pmap_pcid_enabled && !invpcid_works)
334 PCPU_SET(curpmap, curpmap);
335 load_cr3(curpmap->pm_cr3 | (pmap_pcid_enabled ?
336 curpmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid : 0));
337 if (!pmap_pcid_enabled)
338 invltlb();
339
340 fpu_kern_leave(curthread, NULL);
341 mtx_unlock(&efi_lock);
342 PMAP_UNLOCK(curpmap);
343 }
344
345 static int
346 efi_init(void)
347 {
348 struct efi_map_header *efihdr;
349 struct efi_md *map;
350 caddr_t kmdp;
351 size_t efisz;
352
353 mtx_init(&efi_lock, "efi", NULL, MTX_DEF);
354
355 if (efi_systbl_phys == 0) {
356 if (bootverbose)
357 printf("EFI systbl not available\n");
358 return (0);
359 }
360 efi_systbl = (struct efi_systbl *)PHYS_TO_DMAP(efi_systbl_phys);
361 if (efi_systbl->st_hdr.th_sig != EFI_SYSTBL_SIG) {
362 efi_systbl = NULL;
363 if (bootverbose)
364 printf("EFI systbl signature invalid\n");
365 return (0);
366 }
367 efi_cfgtbl = (efi_systbl->st_cfgtbl == 0) ? NULL :
368 (struct efi_cfgtbl *)efi_systbl->st_cfgtbl;
369 if (efi_cfgtbl == NULL) {
370 if (bootverbose)
371 printf("EFI config table is not present\n");
372 }
373
374 kmdp = preload_search_by_type("elf kernel");
375 if (kmdp == NULL)
376 kmdp = preload_search_by_type("elf64 kernel");
377 efihdr = (struct efi_map_header *)preload_search_info(kmdp,
378 MODINFO_METADATA | MODINFOMD_EFI_MAP);
379 if (efihdr == NULL) {
380 if (bootverbose)
381 printf("EFI map is not present\n");
382 return (0);
383 }
384 efisz = (sizeof(struct efi_map_header) + 0xf) & ~0xf;
385 map = (struct efi_md *)((uint8_t *)efihdr + efisz);
386 if (efihdr->descriptor_size == 0)
387 return (ENOMEM);
388
389 if (!efi_create_1t1_map(map, efihdr->memory_size /
390 efihdr->descriptor_size, efihdr->descriptor_size)) {
391 if (bootverbose)
392 printf("EFI cannot create runtime map\n");
393 return (ENOMEM);
394 }
395
396 efi_runtime = (efi_systbl->st_rt == 0) ? NULL :
397 (struct efi_rt *)efi_systbl->st_rt;
398 if (efi_runtime == NULL) {
399 if (bootverbose)
400 printf("EFI runtime services table is not present\n");
401 efi_destroy_1t1_map();
402 return (ENXIO);
403 }
404
405 return (0);
406 }
407
408 static void
409 efi_uninit(void)
410 {
411
412 efi_destroy_1t1_map();
413
414 efi_systbl = NULL;
415 efi_cfgtbl = NULL;
416 efi_runtime = NULL;
417
418 mtx_destroy(&efi_lock);
419 }
420
421 int
422 efi_get_table(struct uuid *uuid, void **ptr)
423 {
424 struct efi_cfgtbl *ct;
425 u_long count;
426
427 if (efi_cfgtbl == NULL)
428 return (ENXIO);
429 count = efi_systbl->st_entries;
430 ct = efi_cfgtbl;
431 while (count--) {
432 if (!bcmp(&ct->ct_uuid, uuid, sizeof(*uuid))) {
433 *ptr = (void *)PHYS_TO_DMAP(ct->ct_data);
434 return (0);
435 }
436 ct++;
437 }
438 return (ENOENT);
439 }
440
441 int
442 efi_get_time_locked(struct efi_tm *tm)
443 {
444 efi_status status;
445 int error;
446
447 mtx_assert(&resettodr_lock, MA_OWNED);
448 error = efi_enter();
449 if (error != 0)
450 return (error);
451 status = efi_runtime->rt_gettime(tm, NULL);
452 efi_leave();
453 error = efi_status_to_errno(status);
454 return (error);
455 }
456
457 int
458 efi_get_time(struct efi_tm *tm)
459 {
460 int error;
461
462 if (efi_runtime == NULL)
463 return (ENXIO);
464 mtx_lock(&resettodr_lock);
465 error = efi_get_time_locked(tm);
466 mtx_unlock(&resettodr_lock);
467 return (error);
468 }
469
470 int
471 efi_reset_system(void)
472 {
473 int error;
474
475 error = efi_enter();
476 if (error != 0)
477 return (error);
478 efi_runtime->rt_reset(EFI_RESET_WARM, 0, 0, NULL);
479 efi_leave();
480 return (EIO);
481 }
482
483 int
484 efi_set_time_locked(struct efi_tm *tm)
485 {
486 efi_status status;
487 int error;
488
489 mtx_assert(&resettodr_lock, MA_OWNED);
490 error = efi_enter();
491 if (error != 0)
492 return (error);
493 status = efi_runtime->rt_settime(tm);
494 efi_leave();
495 error = efi_status_to_errno(status);
496 return (error);
497 }
498
499 int
500 efi_set_time(struct efi_tm *tm)
501 {
502 int error;
503
504 if (efi_runtime == NULL)
505 return (ENXIO);
506 mtx_lock(&resettodr_lock);
507 error = efi_set_time_locked(tm);
508 mtx_unlock(&resettodr_lock);
509 return (error);
510 }
511
512 int
513 efi_var_get(efi_char *name, struct uuid *vendor, uint32_t *attrib,
514 size_t *datasize, void *data)
515 {
516 efi_status status;
517 int error;
518
519 error = efi_enter();
520 if (error != 0)
521 return (error);
522 status = efi_runtime->rt_getvar(name, vendor, attrib, datasize, data);
523 efi_leave();
524 error = efi_status_to_errno(status);
525 return (error);
526 }
527
528 int
529 efi_var_nextname(size_t *namesize, efi_char *name, struct uuid *vendor)
530 {
531 efi_status status;
532 int error;
533
534 error = efi_enter();
535 if (error != 0)
536 return (error);
537 status = efi_runtime->rt_scanvar(namesize, name, vendor);
538 efi_leave();
539 error = efi_status_to_errno(status);
540 return (error);
541 }
542
543 int
544 efi_var_set(efi_char *name, struct uuid *vendor, uint32_t attrib,
545 size_t datasize, void *data)
546 {
547 efi_status status;
548 int error;
549
550 error = efi_enter();
551 if (error != 0)
552 return (error);
553 status = efi_runtime->rt_setvar(name, vendor, attrib, datasize, data);
554 efi_leave();
555 error = efi_status_to_errno(status);
556 return (error);
557 }
558
559 static int
560 efirt_modevents(module_t m, int event, void *arg __unused)
561 {
562
563 switch (event) {
564 case MOD_LOAD:
565 return (efi_init());
566
567 case MOD_UNLOAD:
568 efi_uninit();
569 return (0);
570
571 case MOD_SHUTDOWN:
572 return (0);
573
574 default:
575 return (EOPNOTSUPP);
576 }
577 }
578
579 static moduledata_t efirt_moddata = {
580 .name = "efirt",
581 .evhand = efirt_modevents,
582 .priv = NULL,
583 };
584 DECLARE_MODULE(efirt, efirt_moddata, SI_SUB_VM_CONF, SI_ORDER_ANY);
585 MODULE_VERSION(efirt, 1);
586
587 /* XXX debug stuff */
588 static int
589 efi_time_sysctl_handler(SYSCTL_HANDLER_ARGS)
590 {
591 struct efi_tm tm;
592 int error, val;
593
594 val = 0;
595 error = sysctl_handle_int(oidp, &val, 0, req);
596 if (error != 0 || req->newptr == NULL)
597 return (error);
598 error = efi_get_time(&tm);
599 if (error == 0) {
600 uprintf("EFI reports: Year %d Month %d Day %d Hour %d Min %d "
601 "Sec %d\n", tm.tm_year, tm.tm_mon, tm.tm_mday, tm.tm_hour,
602 tm.tm_min, tm.tm_sec);
603 }
604 return (error);
605 }
606
607 SYSCTL_PROC(_debug, OID_AUTO, efi_time, CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
608 efi_time_sysctl_handler, "I", "");
Cache object: ca60b30fdc8b55c8ff677bdc079446df
|