xref: /dragonfly/sys/platform/pc64/x86_64/efirt.c (revision b990a6be)
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/proc.h>
45 #include <sys/thread.h>
46 #include <sys/globaldata.h>
47 
48 #include <vm/vm.h>
49 #include <vm/pmap.h>
50 #include <vm/vm_map.h>
51 #include <vm/vm_object.h>
52 #include <vm/vm_param.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_extern.h>
56 
57 #include <vm/vm_page2.h>
58 #include <sys/thread2.h>
59 
60 #include <machine/efi.h>
61 #include <machine/metadata.h>
62 #include <machine/md_var.h>
63 #include <machine/smp.h>
64 #include <machine/vmparam.h>
65 
66 static struct efi_systbl *efi_systbl;
67 static struct efi_cfgtbl *efi_cfgtbl;
68 static struct efi_rt *efi_runtime;
69 
70 static int efi_status2err[25] = {
71 	0,		/* EFI_SUCCESS */
72 	ENOEXEC,	/* EFI_LOAD_ERROR */
73 	EINVAL,		/* EFI_INVALID_PARAMETER */
74 	ENOSYS,		/* EFI_UNSUPPORTED */
75 	EMSGSIZE, 	/* EFI_BAD_BUFFER_SIZE */
76 	EOVERFLOW,	/* EFI_BUFFER_TOO_SMALL */
77 	EBUSY,		/* EFI_NOT_READY */
78 	EIO,		/* EFI_DEVICE_ERROR */
79 	EROFS,		/* EFI_WRITE_PROTECTED */
80 	EAGAIN,		/* EFI_OUT_OF_RESOURCES */
81 	EIO,		/* EFI_VOLUME_CORRUPTED */
82 	ENOSPC,		/* EFI_VOLUME_FULL */
83 	ENXIO,		/* EFI_NO_MEDIA */
84 	ESTALE,		/* EFI_MEDIA_CHANGED */
85 	ENOENT,		/* EFI_NOT_FOUND */
86 	EACCES,		/* EFI_ACCESS_DENIED */
87 	ETIMEDOUT,	/* EFI_NO_RESPONSE */
88 	EADDRNOTAVAIL,	/* EFI_NO_MAPPING */
89 	ETIMEDOUT,	/* EFI_TIMEOUT */
90 	EDOOFUS,	/* EFI_NOT_STARTED */
91 	EALREADY,	/* EFI_ALREADY_STARTED */
92 	ECANCELED,	/* EFI_ABORTED */
93 	EPROTO,		/* EFI_ICMP_ERROR */
94 	EPROTO,		/* EFI_TFTP_ERROR */
95 	EPROTO		/* EFI_PROTOCOL_ERROR */
96 };
97 
98 MALLOC_DEFINE(M_EFI, "efi", "EFI BIOS");
99 
100 static int
101 efi_status_to_errno(efi_status status)
102 {
103 	u_long code;
104 
105 	code = status & 0x3ffffffffffffffful;
106 	return (code < nitems(efi_status2err) ? efi_status2err[code] : EDOOFUS);
107 }
108 
109 static struct lock efi_lock;
110 static struct lock resettodr_lock;
111 static mcontext_t efi_ctx;
112 static struct vmspace *efi_savevm;
113 static struct vmspace *efi_vmspace;
114 static vm_object_t efi_obj;
115 static struct efi_md *efi_map;
116 static int efi_ndesc;
117 static int efi_descsz;
118 
119 static void
120 efi_destroy_1t1_map(void)
121 {
122 	vm_object_t obj;
123 	vm_page_t m;
124 
125 	if ((obj = efi_obj) != NULL) {
126 		efi_obj = NULL;
127 		vm_object_hold(obj);
128 		vm_object_reference_locked(obj);	/* match deallocate */
129 	}
130 	if (efi_vmspace) {
131 		pmap_remove_pages(vmspace_pmap(efi_vmspace),
132 				  VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
133 		vm_map_remove(&efi_vmspace->vm_map,
134 			      VM_MIN_USER_ADDRESS,
135 			      VM_MAX_USER_ADDRESS);
136 		vmspace_rel(efi_vmspace);
137 		efi_vmspace = NULL;
138 	}
139 	if (obj) {
140 		while ((m = RB_ROOT(&obj->rb_memq)) != NULL) {
141 			vm_page_busy_wait(m, FALSE, "efipg");
142 			vm_page_unwire(m, 1);
143 			m->flags &= ~(PG_MAPPED | PG_WRITEABLE);
144 			cdev_pager_free_page(obj, m);
145 			kfree(m, M_EFI);
146 		}
147 		vm_object_drop(obj);
148 		vm_object_deallocate(obj);
149 	}
150 }
151 
152 static int
153 efi_pg_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
154 	    vm_ooffset_t foff, struct ucred *cred, u_short *color)
155 {
156 	*color = 0;
157 	return 0;
158 }
159 
160 static void
161 efi_pg_dtor(void *handle)
162 {
163 }
164 
165 static int
166 efi_pg_fault(vm_object_t obj, vm_ooffset_t offset, int prot, vm_page_t *mres)
167 {
168 	vm_page_t m;
169 
170 	m = *mres;
171 	if ((m->flags & PG_FICTITIOUS) == 0) {
172 		*mres = NULL;
173 		vm_page_remove(m);
174 		vm_page_free(m);
175 		m = NULL;
176 	}
177 	if (m == NULL) {
178 		kprintf("efi_pg_fault: unmapped pg @%016jx\n", offset);
179 		return VM_PAGER_ERROR;
180 	}
181 
182 	/*
183 	 * Shouldn't get hit, we pre-loaded all the pages.
184 	 */
185 	kprintf("efi_pg_fault: ok %p/%p @%016jx m=%016jx,%016jx\n",
186 		obj, efi_obj, offset, m->pindex, m->phys_addr);
187 
188 	return VM_PAGER_OK;
189 }
190 
191 static struct cdev_pager_ops efi_pager_ops = {
192 	.cdev_pg_fault	= efi_pg_fault,
193 	.cdev_pg_ctor	= efi_pg_ctor,
194 	.cdev_pg_dtor	= efi_pg_dtor
195 };
196 
197 static bool
198 efi_create_1t1_map(struct efi_md *map, int ndesc, int descsz)
199 {
200 	vm_page_t m;
201 	struct efi_md *p;
202 	int i;
203 	int count;
204 	int result;
205 
206 	efi_map = map;
207 	efi_ndesc = ndesc;
208 	efi_descsz = descsz;
209 
210 	efi_vmspace = vmspace_alloc(VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
211 	pmap_pinit2(vmspace_pmap(efi_vmspace));
212 	efi_obj = cdev_pager_allocate(NULL, OBJT_MGTDEVICE, &efi_pager_ops,
213 				  VM_MAX_USER_ADDRESS,
214 				  VM_PROT_READ | VM_PROT_WRITE,
215 				  0, proc0.p_ucred);
216 	vm_object_hold(efi_obj);
217 
218 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
219 	vm_map_lock(&efi_vmspace->vm_map);
220 	result = vm_map_insert(&efi_vmspace->vm_map, &count, efi_obj, NULL,
221 			      0, 0, VM_MAX_USER_ADDRESS,
222 			      VM_MAPTYPE_NORMAL,
223 			      VM_SUBSYS_EFI,
224 			      VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE,
225 			      VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE,
226 			      0);
227 	vm_map_unlock(&efi_vmspace->vm_map);
228 	if (result != KERN_SUCCESS)
229 		goto fail;
230 
231 	for (i = 0, p = map;
232 	     i < ndesc; i++, p = efi_next_descriptor(p, descsz)) {
233 		vm_offset_t va;
234 		uint64_t idx;
235 		int mode;
236 
237 		if ((p->md_attr & EFI_MD_ATTR_RT) == 0)
238 			continue;
239 		if (p->md_virt != NULL) {
240 			if (bootverbose)
241 				kprintf("EFI Runtime entry %d is mapped\n", i);
242 			goto fail;
243 		}
244 		if ((p->md_phys & EFI_PAGE_MASK) != 0) {
245 			if (bootverbose)
246 				kprintf("EFI Runtime entry %d is not aligned\n",
247 				    i);
248 			goto fail;
249 		}
250 		if (p->md_phys + p->md_pages * EFI_PAGE_SIZE < p->md_phys ||
251 		    p->md_phys + p->md_pages * EFI_PAGE_SIZE >=
252 		    VM_MAX_USER_ADDRESS) {
253 			kprintf("EFI Runtime entry %d is not in mappable for RT:"
254 			    "base %#016jx %#jx pages\n",
255 			    i, (uintmax_t)p->md_phys,
256 			    (uintmax_t)p->md_pages);
257 			goto fail;
258 		}
259 
260 		if ((p->md_attr & EFI_MD_ATTR_WB) != 0)
261 			mode = VM_MEMATTR_WRITE_BACK;
262 		else if ((p->md_attr & EFI_MD_ATTR_WT) != 0)
263 			mode = VM_MEMATTR_WRITE_THROUGH;
264 		else if ((p->md_attr & EFI_MD_ATTR_WC) != 0)
265 			mode = VM_MEMATTR_WRITE_COMBINING;
266 		else if ((p->md_attr & EFI_MD_ATTR_WP) != 0)
267 			mode = VM_MEMATTR_WRITE_PROTECTED;
268 		else if ((p->md_attr & EFI_MD_ATTR_UC) != 0)
269 			mode = VM_MEMATTR_UNCACHEABLE;
270 		else {
271 			if (bootverbose)
272 				kprintf("EFI Runtime entry %d mapping "
273 				    "attributes unsupported\n", i);
274 			mode = VM_MEMATTR_UNCACHEABLE;
275 		}
276 
277 		if (bootverbose) {
278 			kprintf("efirt: map %016jx-%016jx\n",
279 				p->md_phys,
280 				p->md_phys + IDX_TO_OFF(p->md_pages));
281 		}
282 
283 		for (va = p->md_phys, idx = 0; idx < p->md_pages; idx++,
284 		    va += PAGE_SIZE) {
285 			m = kmalloc(sizeof(*m), M_EFI, M_WAITOK | M_ZERO);
286 			/*m->flags |= PG_WRITEABLE;*/
287 			vm_page_initfake(m, va, mode);	/* va is phys addr */
288 			m->valid = VM_PAGE_BITS_ALL;
289 			m->dirty = m->valid;
290 			vm_page_insert(m, efi_obj, OFF_TO_IDX(va));
291 			vm_page_wakeup(m);
292 		}
293 	}
294 	vm_object_drop(efi_obj);
295 	vm_map_entry_release(count);
296 
297 	return true;
298 
299 fail:
300 	vm_object_drop(efi_obj);
301 	vm_map_entry_release(count);
302 	efi_destroy_1t1_map();
303 
304 	return false;
305 }
306 
307 /*
308  * Create an environment for the EFI runtime code call.  The most
309  * important part is creating the required 1:1 physical->virtual
310  * mappings for the runtime segments.  To do that, we manually create
311  * page table which unmap userspace but gives correct kernel mapping.
312  * The 1:1 mappings for runtime segments usually occupy low 4G of the
313  * physical address map.
314  *
315  * The 1:1 mappings were chosen over the SetVirtualAddressMap() EFI RT
316  * service, because there are some BIOSes which fail to correctly
317  * relocate itself on the call, requiring both 1:1 and virtual
318  * mapping.  As result, we must provide 1:1 mapping anyway, so no
319  * reason to bother with the virtual map, and no need to add a
320  * complexity into loader.
321  *
322  * The fpu_kern_enter() call allows firmware to use FPU, as mandated
323  * by the specification.  In particular, CR0.TS bit is cleared.  Also
324  * it enters critical section, giving us neccessary protection against
325  * context switch.
326  *
327  * There is no need to disable interrupts around the change of %cr3,
328  * the kernel mappings are correct, while we only grabbed the
329  * userspace portion of VA.  Interrupts handlers must not access
330  * userspace.  Having interrupts enabled fixes the issue with
331  * firmware/SMM long operation, which would negatively affect IPIs,
332  * esp. TLB shootdown requests.
333  */
334 static int
335 efi_enter(void)
336 {
337 	thread_t td = curthread;
338 
339 	if (efi_runtime == NULL)
340 		return (ENXIO);
341 	lockmgr(&efi_lock, LK_EXCLUSIVE);
342 	efi_savevm = td->td_lwp->lwp_vmspace;
343 	pmap_setlwpvm(td->td_lwp, efi_vmspace);
344 	npxpush(&efi_ctx);
345 	cpu_invltlb();
346 
347 	return (0);
348 }
349 
350 static void
351 efi_leave(void)
352 {
353 	thread_t td = curthread;
354 
355 	pmap_setlwpvm(td->td_lwp, efi_savevm);
356 	npxpop(&efi_ctx);
357 	cpu_invltlb();
358 	efi_savevm = NULL;
359 	lockmgr(&efi_lock, LK_RELEASE);
360 }
361 
362 static int
363 efi_init(void)
364 {
365 	struct efi_map_header *efihdr;
366 	struct efi_md *map;
367 	caddr_t kmdp;
368 	size_t efisz;
369 
370 	lockinit(&efi_lock, "efi", 0, LK_CANRECURSE);
371 	lockinit(&resettodr_lock, "efitodr", 0, LK_CANRECURSE);
372 
373 	if (efi_systbl_phys == 0) {
374 		if (bootverbose)
375 			kprintf("EFI systbl not available\n");
376 		return (ENXIO);
377 	}
378 	efi_systbl = (struct efi_systbl *)PHYS_TO_DMAP(efi_systbl_phys);
379 	if (efi_systbl->st_hdr.th_sig != EFI_SYSTBL_SIG) {
380 		efi_systbl = NULL;
381 		if (bootverbose)
382 			kprintf("EFI systbl signature invalid\n");
383 		return (ENXIO);
384 	}
385 	efi_cfgtbl = (efi_systbl->st_cfgtbl == 0) ? NULL :
386 	    (struct efi_cfgtbl *)efi_systbl->st_cfgtbl;
387 	if (efi_cfgtbl == NULL) {
388 		if (bootverbose)
389 			kprintf("EFI config table is not present\n");
390 	}
391 
392 	kmdp = preload_search_by_type("elf kernel");
393 	if (kmdp == NULL)
394 		kmdp = preload_search_by_type("elf64 kernel");
395 	efihdr = (struct efi_map_header *)preload_search_info(kmdp,
396 	    MODINFO_METADATA | MODINFOMD_EFI_MAP);
397 	if (efihdr == NULL) {
398 		if (bootverbose)
399 			kprintf("EFI map is not present\n");
400 		return (ENXIO);
401 	}
402 	efisz = (sizeof(struct efi_map_header) + 0xf) & ~0xf;
403 	map = (struct efi_md *)((uint8_t *)efihdr + efisz);
404 	if (efihdr->descriptor_size == 0)
405 		return (ENOMEM);
406 
407 	if (!efi_create_1t1_map(map, efihdr->memory_size /
408 	    efihdr->descriptor_size, efihdr->descriptor_size)) {
409 		if (bootverbose)
410 			kprintf("EFI cannot create runtime map\n");
411 		return (ENOMEM);
412 	}
413 
414 	efi_runtime = (efi_systbl->st_rt == 0) ? NULL :
415 			(struct efi_rt *)efi_systbl->st_rt;
416 	if (efi_runtime == NULL) {
417 		if (bootverbose)
418 			kprintf("EFI runtime services table is not present\n");
419 		efi_destroy_1t1_map();
420 		return (ENXIO);
421 	}
422 
423 	return (0);
424 }
425 
426 static void
427 efi_uninit(void)
428 {
429 	efi_destroy_1t1_map();
430 
431 	efi_systbl = NULL;
432 	efi_cfgtbl = NULL;
433 	efi_runtime = NULL;
434 
435 	lockuninit(&efi_lock);
436 	lockuninit(&resettodr_lock);
437 }
438 
439 int
440 efi_get_table(struct uuid *uuid, void **ptr)
441 {
442 	struct efi_cfgtbl *ct;
443 	u_long count;
444 
445 	if (efi_cfgtbl == NULL)
446 		return (ENXIO);
447 	count = efi_systbl->st_entries;
448 	ct = efi_cfgtbl;
449 	while (count--) {
450 		if (!bcmp(&ct->ct_uuid, uuid, sizeof(*uuid))) {
451 			*ptr = (void *)PHYS_TO_DMAP(ct->ct_data);
452 			return (0);
453 		}
454 		ct++;
455 	}
456 	return (ENOENT);
457 }
458 
459 char SaveCode[1024];
460 
461 int
462 efi_get_time_locked(struct efi_tm *tm)
463 {
464 	efi_status status;
465 	int error;
466 
467 	KKASSERT(lockowned(&resettodr_lock) != 0);
468 	error = efi_enter();
469 	if (error != 0)
470 		return (error);
471 	status = efi_runtime->rt_gettime(tm, NULL);
472 	efi_leave();
473 	error = efi_status_to_errno(status);
474 
475 	return (error);
476 }
477 
478 int
479 efi_get_time(struct efi_tm *tm)
480 {
481 	int error;
482 
483 	if (efi_runtime == NULL)
484 		return (ENXIO);
485 	lockmgr(&resettodr_lock, LK_EXCLUSIVE);
486 	error = efi_get_time_locked(tm);
487 	lockmgr(&resettodr_lock, LK_RELEASE);
488 
489 	return (error);
490 }
491 
492 int
493 efi_reset_system(void)
494 {
495 	int error;
496 
497 	error = efi_enter();
498 	if (error != 0)
499 		return (error);
500 	efi_runtime->rt_reset(EFI_RESET_WARM, 0, 0, NULL);
501 	efi_leave();
502 	return (EIO);
503 }
504 
505 int
506 efi_set_time_locked(struct efi_tm *tm)
507 {
508 	efi_status status;
509 	int error;
510 
511 	KKASSERT(lockowned(&resettodr_lock) != 0);
512 	error = efi_enter();
513 	if (error != 0)
514 		return (error);
515 	status = efi_runtime->rt_settime(tm);
516 	efi_leave();
517 	error = efi_status_to_errno(status);
518 	return (error);
519 }
520 
521 int
522 efi_set_time(struct efi_tm *tm)
523 {
524 	int error;
525 
526 	if (efi_runtime == NULL)
527 		return (ENXIO);
528 	lockmgr(&resettodr_lock, LK_EXCLUSIVE);
529 	error = efi_set_time_locked(tm);
530 	lockmgr(&resettodr_lock, LK_RELEASE);
531 	return (error);
532 }
533 
534 int
535 efi_var_get(efi_char *name, struct uuid *vendor, uint32_t *attrib,
536     size_t *datasize, void *data)
537 {
538 	efi_status status;
539 	int error;
540 
541 	error = efi_enter();
542 	if (error != 0)
543 		return (error);
544 	status = efi_runtime->rt_getvar(name, vendor, attrib, datasize, data);
545 	efi_leave();
546 	error = efi_status_to_errno(status);
547 	return (error);
548 }
549 
550 int
551 efi_var_nextname(size_t *namesize, efi_char *name, struct uuid *vendor)
552 {
553 	efi_status status;
554 	int error;
555 
556 	error = efi_enter();
557 	if (error != 0)
558 		return (error);
559 	status = efi_runtime->rt_scanvar(namesize, name, vendor);
560 	efi_leave();
561 	error = efi_status_to_errno(status);
562 	return (error);
563 }
564 
565 int
566 efi_var_set(efi_char *name, struct uuid *vendor, uint32_t attrib,
567     size_t datasize, void *data)
568 {
569 	efi_status status;
570 	int error;
571 
572 	error = efi_enter();
573 	if (error != 0)
574 		return (error);
575 	status = efi_runtime->rt_setvar(name, vendor, attrib, datasize, data);
576 	efi_leave();
577 	error = efi_status_to_errno(status);
578 	return (error);
579 }
580 
581 static int
582 efirt_modevents(module_t m, int event, void *arg __unused)
583 {
584 
585 	switch (event) {
586 	case MOD_LOAD:
587 		return (efi_init());
588 
589 	case MOD_UNLOAD:
590 		efi_uninit();
591 		return (0);
592 
593 	case MOD_SHUTDOWN:
594 		return (0);
595 
596 	default:
597 		return (EOPNOTSUPP);
598 	}
599 }
600 
601 static moduledata_t efirt_moddata = {
602 	.name = "efirt",
603 	.evhand = efirt_modevents,
604 	.priv = NULL,
605 };
606 
607 DECLARE_MODULE(efirt, efirt_moddata, SI_SUB_DRIVERS, SI_ORDER_ANY);
608 MODULE_VERSION(efirt, 1);
609 
610 
611 /* XXX debug stuff */
612 static int
613 efi_time_sysctl_handler(SYSCTL_HANDLER_ARGS)
614 {
615 	struct efi_tm tm;
616 	int error, val;
617 
618 	val = 0;
619 	error = sysctl_handle_int(oidp, &val, 0, req);
620 	if (error != 0 || req->newptr == NULL)
621 		return (error);
622 	error = efi_get_time(&tm);
623 	if (error == 0) {
624 		uprintf("EFI reports: Year %d Month %d Day %d Hour %d Min %d "
625 		    "Sec %d\n", tm.tm_year, tm.tm_mon, tm.tm_mday, tm.tm_hour,
626 		    tm.tm_min, tm.tm_sec);
627 	}
628 	return (error);
629 }
630 
631 SYSCTL_PROC(_debug, OID_AUTO, efi_time, CTLTYPE_INT | CTLFLAG_RW, NULL, 0,
632 	    efi_time_sysctl_handler, "I", "");
633