xref: /dragonfly/sys/platform/pc64/x86_64/efirt.c (revision bb8c85ff)
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  * $FreeBSD: head/sys/amd64/amd64/efirt.c 307391 2016-10-16 06:07:43Z kib $
32  */
33 
34 #include <sys/param.h>
35 #include <sys/efi.h>
36 #include <sys/kernel.h>
37 #include <sys/linker.h>
38 #include <sys/lock.h>
39 #include <sys/module.h>
40 #include <sys/proc.h>
41 #include <sys/sched.h>
42 #include <sys/sysctl.h>
43 #include <sys/systm.h>
44 #include <sys/thread.h>
45 #include <sys/globaldata.h>
46 
47 #include <vm/vm.h>
48 #include <vm/pmap.h>
49 #include <vm/vm_map.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_param.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_extern.h>
55 
56 #include <vm/vm_page2.h>
57 #include <sys/thread2.h>
58 
59 #include <machine/efi.h>
60 #include <machine/metadata.h>
61 #include <machine/md_var.h>
62 #include <machine/smp.h>
63 #include <machine/vmparam.h>
64 
65 static struct efi_systbl *efi_systbl;
66 static struct efi_cfgtbl *efi_cfgtbl;
67 static struct efi_rt *efi_runtime;
68 
69 static int efi_status2err[25] = {
70 	0,		/* EFI_SUCCESS */
71 	ENOEXEC,	/* EFI_LOAD_ERROR */
72 	EINVAL,		/* EFI_INVALID_PARAMETER */
73 	ENOSYS,		/* EFI_UNSUPPORTED */
74 	EMSGSIZE, 	/* EFI_BAD_BUFFER_SIZE */
75 	EOVERFLOW,	/* EFI_BUFFER_TOO_SMALL */
76 	EBUSY,		/* EFI_NOT_READY */
77 	EIO,		/* EFI_DEVICE_ERROR */
78 	EROFS,		/* EFI_WRITE_PROTECTED */
79 	EAGAIN,		/* EFI_OUT_OF_RESOURCES */
80 	EIO,		/* EFI_VOLUME_CORRUPTED */
81 	ENOSPC,		/* EFI_VOLUME_FULL */
82 	ENXIO,		/* EFI_NO_MEDIA */
83 	ESTALE,		/* EFI_MEDIA_CHANGED */
84 	ENOENT,		/* EFI_NOT_FOUND */
85 	EACCES,		/* EFI_ACCESS_DENIED */
86 	ETIMEDOUT,	/* EFI_NO_RESPONSE */
87 	EADDRNOTAVAIL,	/* EFI_NO_MAPPING */
88 	ETIMEDOUT,	/* EFI_TIMEOUT */
89 	EDOOFUS,	/* EFI_NOT_STARTED */
90 	EALREADY,	/* EFI_ALREADY_STARTED */
91 	ECANCELED,	/* EFI_ABORTED */
92 	EPROTO,		/* EFI_ICMP_ERROR */
93 	EPROTO,		/* EFI_TFTP_ERROR */
94 	EPROTO		/* EFI_PROTOCOL_ERROR */
95 };
96 
97 MALLOC_DEFINE(M_EFI, "efi", "EFI BIOS");
98 
99 static int
100 efi_status_to_errno(efi_status status)
101 {
102 	u_long code;
103 
104 	code = status & 0x3ffffffffffffffful;
105 	return (code < nitems(efi_status2err) ? efi_status2err[code] : EDOOFUS);
106 }
107 
108 static struct lock efi_lock;
109 static struct lock resettodr_lock;
110 static mcontext_t efi_ctx;
111 static struct vmspace *efi_savevm;
112 static struct vmspace *efi_vmspace;
113 static vm_object_t efi_obj;
114 static struct efi_md *efi_map;
115 static int efi_ndesc;
116 static int efi_descsz;
117 
118 static void
119 efi_destroy_1t1_map(void)
120 {
121 	vm_object_t obj;
122 	vm_page_t m;
123 
124 	if ((obj = efi_obj) != NULL) {
125 		efi_obj = NULL;
126 		vm_object_hold(obj);
127 		vm_object_reference_locked(obj);	/* match deallocate */
128 	}
129 	if (efi_vmspace) {
130 		pmap_remove_pages(vmspace_pmap(efi_vmspace),
131 				  VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
132 		vm_map_remove(&efi_vmspace->vm_map,
133 			      VM_MIN_USER_ADDRESS,
134 			      VM_MAX_USER_ADDRESS);
135 		vmspace_rel(efi_vmspace);
136 		efi_vmspace = NULL;
137 	}
138 	if (obj) {
139 		while ((m = RB_ROOT(&obj->rb_memq)) != NULL) {
140 			vm_page_busy_wait(m, FALSE, "efipg");
141 			vm_page_unwire(m, 1);
142 			m->flags &= ~(PG_MAPPED | PG_WRITEABLE);
143 			cdev_pager_free_page(obj, m);
144 			kfree(m, M_EFI);
145 		}
146 		vm_object_drop(obj);
147 		vm_object_deallocate(obj);
148 	}
149 }
150 
151 static int
152 efi_pg_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
153 	    vm_ooffset_t foff, struct ucred *cred, u_short *color)
154 {
155 	*color = 0;
156 	return 0;
157 }
158 
159 static void
160 efi_pg_dtor(void *handle)
161 {
162 }
163 
164 static int
165 efi_pg_fault(vm_object_t obj, vm_ooffset_t offset, int prot, vm_page_t *mres)
166 {
167 	vm_page_t m;
168 
169 	m = *mres;
170 	if ((m->flags & PG_FICTITIOUS) == 0) {
171 		*mres = NULL;
172 		vm_page_remove(m);
173 		vm_page_free(m);
174 		m = NULL;
175 	}
176 	if (m == NULL) {
177 		kprintf("efi_pg_fault: unmapped pg @%016jx\n", offset);
178 		return VM_PAGER_ERROR;
179 	}
180 
181 	/*
182 	 * Shouldn't get hit, we pre-loaded all the pages.
183 	 */
184 	kprintf("efi_pg_fault: ok %p/%p @%016jx m=%016jx,%016jx\n",
185 		obj, efi_obj, offset, m->pindex, m->phys_addr);
186 
187 	return VM_PAGER_OK;
188 }
189 
190 static struct cdev_pager_ops efi_pager_ops = {
191 	.cdev_pg_fault	= efi_pg_fault,
192 	.cdev_pg_ctor	= efi_pg_ctor,
193 	.cdev_pg_dtor	= efi_pg_dtor
194 };
195 
196 static bool
197 efi_create_1t1_map(struct efi_md *map, int ndesc, int descsz)
198 {
199 	vm_page_t m;
200 	struct efi_md *p;
201 	int i;
202 	int count;
203 	int result;
204 
205 	efi_map = map;
206 	efi_ndesc = ndesc;
207 	efi_descsz = descsz;
208 
209 	efi_vmspace = vmspace_alloc(VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
210 	pmap_pinit2(vmspace_pmap(efi_vmspace));
211 	efi_obj = cdev_pager_allocate(NULL, OBJT_MGTDEVICE, &efi_pager_ops,
212 				  VM_MAX_USER_ADDRESS,
213 				  VM_PROT_READ | VM_PROT_WRITE,
214 				  0, proc0.p_ucred);
215 	vm_object_hold(efi_obj);
216 
217 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
218 	vm_map_lock(&efi_vmspace->vm_map);
219 	result = vm_map_insert(&efi_vmspace->vm_map, &count, efi_obj, NULL,
220 			      0, 0, VM_MAX_USER_ADDRESS,
221 			      VM_MAPTYPE_NORMAL,
222 			      VM_SUBSYS_EFI,
223 			      VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE,
224 			      VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE,
225 			      0);
226 	vm_map_unlock(&efi_vmspace->vm_map);
227 	if (result != KERN_SUCCESS)
228 		goto fail;
229 
230 	for (i = 0, p = map;
231 	     i < ndesc; i++, p = efi_next_descriptor(p, descsz)) {
232 		vm_offset_t va;
233 		uint64_t idx;
234 		int mode;
235 
236 		if ((p->md_attr & EFI_MD_ATTR_RT) == 0)
237 			continue;
238 		if (p->md_virt != NULL) {
239 			if (bootverbose)
240 				kprintf("EFI Runtime entry %d is mapped\n", i);
241 			goto fail;
242 		}
243 		if ((p->md_phys & EFI_PAGE_MASK) != 0) {
244 			if (bootverbose)
245 				kprintf("EFI Runtime entry %d is not aligned\n",
246 				    i);
247 			goto fail;
248 		}
249 		if (p->md_phys + p->md_pages * EFI_PAGE_SIZE < p->md_phys ||
250 		    p->md_phys + p->md_pages * EFI_PAGE_SIZE >=
251 		    VM_MAX_USER_ADDRESS) {
252 			kprintf("EFI Runtime entry %d is not in mappable for RT:"
253 			    "base %#016jx %#jx pages\n",
254 			    i, (uintmax_t)p->md_phys,
255 			    (uintmax_t)p->md_pages);
256 			goto fail;
257 		}
258 
259 		if ((p->md_attr & EFI_MD_ATTR_WB) != 0)
260 			mode = VM_MEMATTR_WRITE_BACK;
261 		else if ((p->md_attr & EFI_MD_ATTR_WT) != 0)
262 			mode = VM_MEMATTR_WRITE_THROUGH;
263 		else if ((p->md_attr & EFI_MD_ATTR_WC) != 0)
264 			mode = VM_MEMATTR_WRITE_COMBINING;
265 		else if ((p->md_attr & EFI_MD_ATTR_WP) != 0)
266 			mode = VM_MEMATTR_WRITE_PROTECTED;
267 		else if ((p->md_attr & EFI_MD_ATTR_UC) != 0)
268 			mode = VM_MEMATTR_UNCACHEABLE;
269 		else {
270 			if (bootverbose)
271 				kprintf("EFI Runtime entry %d mapping "
272 				    "attributes unsupported\n", i);
273 			mode = VM_MEMATTR_UNCACHEABLE;
274 		}
275 
276 		if (bootverbose) {
277 			kprintf("efirt: map %016jx-%016jx\n",
278 				p->md_phys,
279 				p->md_phys + IDX_TO_OFF(p->md_pages));
280 		}
281 
282 		for (va = p->md_phys, idx = 0; idx < p->md_pages; idx++,
283 		    va += PAGE_SIZE) {
284 			m = kmalloc(sizeof(*m), M_EFI, M_WAITOK | M_ZERO);
285 			/*m->flags |= PG_WRITEABLE;*/
286 			vm_page_initfake(m, va, mode);	/* va is phys addr */
287 			m->valid = VM_PAGE_BITS_ALL;
288 			m->dirty = m->valid;
289 			vm_page_insert(m, efi_obj, OFF_TO_IDX(va));
290 			vm_page_wakeup(m);
291 		}
292 	}
293 	vm_object_drop(efi_obj);
294 	vm_map_entry_release(count);
295 
296 	return true;
297 
298 fail:
299 	vm_object_drop(efi_obj);
300 	vm_map_entry_release(count);
301 	efi_destroy_1t1_map();
302 
303 	return false;
304 }
305 
306 /*
307  * Create an environment for the EFI runtime code call.  The most
308  * important part is creating the required 1:1 physical->virtual
309  * mappings for the runtime segments.  To do that, we manually create
310  * page table which unmap userspace but gives correct kernel mapping.
311  * The 1:1 mappings for runtime segments usually occupy low 4G of the
312  * physical address map.
313  *
314  * The 1:1 mappings were chosen over the SetVirtualAddressMap() EFI RT
315  * service, because there are some BIOSes which fail to correctly
316  * relocate itself on the call, requiring both 1:1 and virtual
317  * mapping.  As result, we must provide 1:1 mapping anyway, so no
318  * reason to bother with the virtual map, and no need to add a
319  * complexity into loader.
320  *
321  * The fpu_kern_enter() call allows firmware to use FPU, as mandated
322  * by the specification.  In particular, CR0.TS bit is cleared.  Also
323  * it enters critical section, giving us neccessary protection against
324  * context switch.
325  *
326  * There is no need to disable interrupts around the change of %cr3,
327  * the kernel mappings are correct, while we only grabbed the
328  * userspace portion of VA.  Interrupts handlers must not access
329  * userspace.  Having interrupts enabled fixes the issue with
330  * firmware/SMM long operation, which would negatively affect IPIs,
331  * esp. TLB shootdown requests.
332  */
333 static int
334 efi_enter(void)
335 {
336 	thread_t td = curthread;
337 
338 	if (efi_runtime == NULL)
339 		return (ENXIO);
340 	lockmgr(&efi_lock, LK_EXCLUSIVE);
341 	efi_savevm = td->td_lwp->lwp_vmspace;
342 	pmap_setlwpvm(td->td_lwp, efi_vmspace);
343 	npxpush(&efi_ctx);
344 	cpu_invltlb();
345 
346 	return (0);
347 }
348 
349 static void
350 efi_leave(void)
351 {
352 	thread_t td = curthread;
353 
354 	pmap_setlwpvm(td->td_lwp, efi_savevm);
355 	npxpop(&efi_ctx);
356 	cpu_invltlb();
357 	efi_savevm = NULL;
358 	lockmgr(&efi_lock, LK_RELEASE);
359 }
360 
361 static int
362 efi_init(void)
363 {
364 	struct efi_map_header *efihdr;
365 	struct efi_md *map;
366 	caddr_t kmdp;
367 	size_t efisz;
368 
369 	lockinit(&efi_lock, "efi", 0, LK_CANRECURSE);
370 	lockinit(&resettodr_lock, "efitodr", 0, LK_CANRECURSE);
371 
372 	if (efi_systbl_phys == 0) {
373 		if (bootverbose)
374 			kprintf("EFI systbl not available\n");
375 		return (ENXIO);
376 	}
377 	efi_systbl = (struct efi_systbl *)PHYS_TO_DMAP(efi_systbl_phys);
378 	if (efi_systbl->st_hdr.th_sig != EFI_SYSTBL_SIG) {
379 		efi_systbl = NULL;
380 		if (bootverbose)
381 			kprintf("EFI systbl signature invalid\n");
382 		return (ENXIO);
383 	}
384 	efi_cfgtbl = (efi_systbl->st_cfgtbl == 0) ? NULL :
385 	    (struct efi_cfgtbl *)efi_systbl->st_cfgtbl;
386 	if (efi_cfgtbl == NULL) {
387 		if (bootverbose)
388 			kprintf("EFI config table is not present\n");
389 	}
390 
391 	kmdp = preload_search_by_type("elf kernel");
392 	if (kmdp == NULL)
393 		kmdp = preload_search_by_type("elf64 kernel");
394 	efihdr = (struct efi_map_header *)preload_search_info(kmdp,
395 	    MODINFO_METADATA | MODINFOMD_EFI_MAP);
396 	if (efihdr == NULL) {
397 		if (bootverbose)
398 			kprintf("EFI map is not present\n");
399 		return (ENXIO);
400 	}
401 	efisz = (sizeof(struct efi_map_header) + 0xf) & ~0xf;
402 	map = (struct efi_md *)((uint8_t *)efihdr + efisz);
403 	if (efihdr->descriptor_size == 0)
404 		return (ENOMEM);
405 
406 	if (!efi_create_1t1_map(map, efihdr->memory_size /
407 	    efihdr->descriptor_size, efihdr->descriptor_size)) {
408 		if (bootverbose)
409 			kprintf("EFI cannot create runtime map\n");
410 		return (ENOMEM);
411 	}
412 
413 	efi_runtime = (efi_systbl->st_rt == 0) ? NULL :
414 			(struct efi_rt *)efi_systbl->st_rt;
415 	if (efi_runtime == NULL) {
416 		if (bootverbose)
417 			kprintf("EFI runtime services table is not present\n");
418 		efi_destroy_1t1_map();
419 		return (ENXIO);
420 	}
421 
422 	return (0);
423 }
424 
425 static void
426 efi_uninit(void)
427 {
428 	efi_destroy_1t1_map();
429 
430 	efi_systbl = NULL;
431 	efi_cfgtbl = NULL;
432 	efi_runtime = NULL;
433 
434 	lockuninit(&efi_lock);
435 	lockuninit(&resettodr_lock);
436 }
437 
438 int
439 efi_get_table(struct uuid *uuid, void **ptr)
440 {
441 	struct efi_cfgtbl *ct;
442 	u_long count;
443 
444 	if (efi_cfgtbl == NULL)
445 		return (ENXIO);
446 	count = efi_systbl->st_entries;
447 	ct = efi_cfgtbl;
448 	while (count--) {
449 		if (!bcmp(&ct->ct_uuid, uuid, sizeof(*uuid))) {
450 			*ptr = (void *)PHYS_TO_DMAP(ct->ct_data);
451 			return (0);
452 		}
453 		ct++;
454 	}
455 	return (ENOENT);
456 }
457 
458 char SaveCode[1024];
459 
460 int
461 efi_get_time_locked(struct efi_tm *tm)
462 {
463 	efi_status status;
464 	int error;
465 
466 	KKASSERT(lockowned(&resettodr_lock) != 0);
467 	error = efi_enter();
468 	if (error != 0)
469 		return (error);
470 	status = efi_runtime->rt_gettime(tm, NULL);
471 	efi_leave();
472 	error = efi_status_to_errno(status);
473 
474 	return (error);
475 }
476 
477 int
478 efi_get_time(struct efi_tm *tm)
479 {
480 	int error;
481 
482 	if (efi_runtime == NULL)
483 		return (ENXIO);
484 	lockmgr(&resettodr_lock, LK_EXCLUSIVE);
485 	error = efi_get_time_locked(tm);
486 	lockmgr(&resettodr_lock, LK_RELEASE);
487 
488 	return (error);
489 }
490 
491 int
492 efi_reset_system(void)
493 {
494 	int error;
495 
496 	error = efi_enter();
497 	if (error != 0)
498 		return (error);
499 	efi_runtime->rt_reset(EFI_RESET_WARM, 0, 0, NULL);
500 	efi_leave();
501 	return (EIO);
502 }
503 
504 int
505 efi_set_time_locked(struct efi_tm *tm)
506 {
507 	efi_status status;
508 	int error;
509 
510 	KKASSERT(lockowned(&resettodr_lock) != 0);
511 	error = efi_enter();
512 	if (error != 0)
513 		return (error);
514 	status = efi_runtime->rt_settime(tm);
515 	efi_leave();
516 	error = efi_status_to_errno(status);
517 	return (error);
518 }
519 
520 int
521 efi_set_time(struct efi_tm *tm)
522 {
523 	int error;
524 
525 	if (efi_runtime == NULL)
526 		return (ENXIO);
527 	lockmgr(&resettodr_lock, LK_EXCLUSIVE);
528 	error = efi_set_time_locked(tm);
529 	lockmgr(&resettodr_lock, LK_RELEASE);
530 	return (error);
531 }
532 
533 int
534 efi_var_get(efi_char *name, struct uuid *vendor, uint32_t *attrib,
535     size_t *datasize, void *data)
536 {
537 	efi_status status;
538 	int error;
539 
540 	error = efi_enter();
541 	if (error != 0)
542 		return (error);
543 	status = efi_runtime->rt_getvar(name, vendor, attrib, datasize, data);
544 	efi_leave();
545 	error = efi_status_to_errno(status);
546 	return (error);
547 }
548 
549 int
550 efi_var_nextname(size_t *namesize, efi_char *name, struct uuid *vendor)
551 {
552 	efi_status status;
553 	int error;
554 
555 	error = efi_enter();
556 	if (error != 0)
557 		return (error);
558 	status = efi_runtime->rt_scanvar(namesize, name, vendor);
559 	efi_leave();
560 	error = efi_status_to_errno(status);
561 	return (error);
562 }
563 
564 int
565 efi_var_set(efi_char *name, struct uuid *vendor, uint32_t attrib,
566     size_t datasize, void *data)
567 {
568 	efi_status status;
569 	int error;
570 
571 	error = efi_enter();
572 	if (error != 0)
573 		return (error);
574 	status = efi_runtime->rt_setvar(name, vendor, attrib, datasize, data);
575 	efi_leave();
576 	error = efi_status_to_errno(status);
577 	return (error);
578 }
579 
580 static int
581 efirt_modevents(module_t m, int event, void *arg __unused)
582 {
583 
584 	switch (event) {
585 	case MOD_LOAD:
586 		return (efi_init());
587 
588 	case MOD_UNLOAD:
589 		efi_uninit();
590 		return (0);
591 
592 	case MOD_SHUTDOWN:
593 		return (0);
594 
595 	default:
596 		return (EOPNOTSUPP);
597 	}
598 }
599 
600 static moduledata_t efirt_moddata = {
601 	.name = "efirt",
602 	.evhand = efirt_modevents,
603 	.priv = NULL,
604 };
605 
606 DECLARE_MODULE(efirt, efirt_moddata, SI_SUB_DRIVERS, SI_ORDER_ANY);
607 MODULE_VERSION(efirt, 1);
608 
609 
610 /* XXX debug stuff */
611 static int
612 efi_time_sysctl_handler(SYSCTL_HANDLER_ARGS)
613 {
614 	struct efi_tm tm;
615 	int error, val;
616 
617 	val = 0;
618 	error = sysctl_handle_int(oidp, &val, 0, req);
619 	if (error != 0 || req->newptr == NULL)
620 		return (error);
621 	error = efi_get_time(&tm);
622 	if (error == 0) {
623 		uprintf("EFI reports: Year %d Month %d Day %d Hour %d Min %d "
624 		    "Sec %d\n", tm.tm_year, tm.tm_mon, tm.tm_mday, tm.tm_hour,
625 		    tm.tm_min, tm.tm_sec);
626 	}
627 	return (error);
628 }
629 
630 SYSCTL_PROC(_debug, OID_AUTO, efi_time, CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
631 	    efi_time_sysctl_handler, "I", "");
632