xref: /freebsd/stand/efi/loader/copy.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
1 /*-
2  * Copyright (c) 2013 The FreeBSD Foundation
3  *
4  * This software was developed by Benno Rice under sponsorship from
5  * the FreeBSD Foundation.
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 #include <sys/param.h>
30 
31 #include <stand.h>
32 #include <bootstrap.h>
33 
34 #include <efi.h>
35 #include <efilib.h>
36 
37 #include "loader_efi.h"
38 
39 #define	M(x)	((x) * 1024 * 1024)
40 #define	G(x)	(1UL * (x) * 1024 * 1024 * 1024)
41 
42 #if defined(__amd64__)
43 #include <machine/cpufunc.h>
44 #include <machine/specialreg.h>
45 #include <machine/vmparam.h>
46 
47 /*
48  * The code is excerpted from sys/x86/x86/identcpu.c: identify_cpu(),
49  * identify_hypervisor(), and dev/hyperv/vmbus/hyperv.c: hyperv_identify().
50  */
51 #define CPUID_LEAF_HV_MAXLEAF		0x40000000
52 #define CPUID_LEAF_HV_INTERFACE		0x40000001
53 #define CPUID_LEAF_HV_FEATURES		0x40000003
54 #define CPUID_LEAF_HV_LIMITS		0x40000005
55 #define CPUID_HV_IFACE_HYPERV		0x31237648	/* HV#1 */
56 #define CPUID_HV_MSR_TIME_REFCNT	0x0002	/* MSR_HV_TIME_REF_COUNT */
57 #define CPUID_HV_MSR_HYPERCALL		0x0020
58 
59 static int
60 running_on_hyperv(void)
61 {
62 	char hv_vendor[16];
63 	uint32_t regs[4];
64 
65 	do_cpuid(1, regs);
66 	if ((regs[2] & CPUID2_HV) == 0)
67 		return (0);
68 
69 	do_cpuid(CPUID_LEAF_HV_MAXLEAF, regs);
70 	if (regs[0] < CPUID_LEAF_HV_LIMITS)
71 		return (0);
72 
73 	((uint32_t *)&hv_vendor)[0] = regs[1];
74 	((uint32_t *)&hv_vendor)[1] = regs[2];
75 	((uint32_t *)&hv_vendor)[2] = regs[3];
76 	hv_vendor[12] = '\0';
77 	if (strcmp(hv_vendor, "Microsoft Hv") != 0)
78 		return (0);
79 
80 	do_cpuid(CPUID_LEAF_HV_INTERFACE, regs);
81 	if (regs[0] != CPUID_HV_IFACE_HYPERV)
82 		return (0);
83 
84 	do_cpuid(CPUID_LEAF_HV_FEATURES, regs);
85 	if ((regs[0] & CPUID_HV_MSR_HYPERCALL) == 0)
86 		return (0);
87 	if ((regs[0] & CPUID_HV_MSR_TIME_REFCNT) == 0)
88 		return (0);
89 
90 	return (1);
91 }
92 
93 static void
94 efi_verify_staging_size(unsigned long *nr_pages)
95 {
96 	UINTN sz;
97 	EFI_MEMORY_DESCRIPTOR *map = NULL, *p;
98 	EFI_PHYSICAL_ADDRESS start, end;
99 	UINTN key, dsz;
100 	UINT32 dver;
101 	EFI_STATUS status;
102 	int i, ndesc;
103 	unsigned long available_pages = 0;
104 
105 	sz = 0;
106 
107 	for (;;) {
108 		status = BS->GetMemoryMap(&sz, map, &key, &dsz, &dver);
109 		if (!EFI_ERROR(status))
110 			break;
111 
112 		if (status != EFI_BUFFER_TOO_SMALL) {
113 			printf("Can't read memory map: %lu\n",
114 			    EFI_ERROR_CODE(status));
115 			goto out;
116 		}
117 
118 		free(map);
119 
120 		/* Allocate 10 descriptors more than the size reported,
121 		 * to allow for any fragmentation caused by calling
122 		 * malloc */
123 		map = malloc(sz + (10 * dsz));
124 		if (map == NULL) {
125 			printf("Unable to allocate memory\n");
126 			goto out;
127 		}
128 	}
129 
130 	ndesc = sz / dsz;
131 	for (i = 0, p = map; i < ndesc;
132 	     i++, p = NextMemoryDescriptor(p, dsz)) {
133 		start = p->PhysicalStart;
134 		end = start + p->NumberOfPages * EFI_PAGE_SIZE;
135 
136 		if (KERNLOAD < start || KERNLOAD >= end)
137 			continue;
138 
139 		available_pages = p->NumberOfPages -
140 			((KERNLOAD - start) >> EFI_PAGE_SHIFT);
141 		break;
142 	}
143 
144 	if (available_pages == 0) {
145 		printf("Can't find valid memory map for staging area!\n");
146 		goto out;
147 	}
148 
149 	i++;
150 	p = NextMemoryDescriptor(p, dsz);
151 
152 	for ( ; i < ndesc;
153 	     i++, p = NextMemoryDescriptor(p, dsz)) {
154 		if (p->Type != EfiConventionalMemory &&
155 		    p->Type != EfiLoaderData)
156 			break;
157 
158 		if (p->PhysicalStart != end)
159 			break;
160 
161 		end = p->PhysicalStart + p->NumberOfPages * EFI_PAGE_SIZE;
162 
163 		available_pages += p->NumberOfPages;
164 	}
165 
166 	if (*nr_pages > available_pages) {
167 		printf("Staging area's size is reduced: %ld -> %ld!\n",
168 		    *nr_pages, available_pages);
169 		*nr_pages = available_pages;
170 	}
171 out:
172 	free(map);
173 }
174 #endif /* __amd64__ */
175 
176 #if defined(__arm__)
177 #define	DEFAULT_EFI_STAGING_SIZE	32
178 #else
179 #define	DEFAULT_EFI_STAGING_SIZE	64
180 #endif
181 #ifndef EFI_STAGING_SIZE
182 #define	EFI_STAGING_SIZE	DEFAULT_EFI_STAGING_SIZE
183 #endif
184 
185 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \
186     defined(__riscv)
187 #define	EFI_STAGING_2M_ALIGN	1
188 #else
189 #define	EFI_STAGING_2M_ALIGN	0
190 #endif
191 
192 #if defined(__amd64__)
193 #define	EFI_STAGING_SLOP	M(8)
194 #else
195 #define	EFI_STAGING_SLOP	0
196 #endif
197 
198 static u_long staging_slop = EFI_STAGING_SLOP;
199 
200 EFI_PHYSICAL_ADDRESS	staging, staging_end, staging_base;
201 bool			stage_offset_set = false;
202 ssize_t			stage_offset;
203 
204 static void
205 efi_copy_free(void)
206 {
207 	BS->FreePages(staging_base, (staging_end - staging_base) /
208 	    EFI_PAGE_SIZE);
209 	stage_offset_set = false;
210 	stage_offset = 0;
211 }
212 
213 #ifdef __amd64__
214 int copy_staging = COPY_STAGING_AUTO;
215 
216 static int
217 command_copy_staging(int argc, char *argv[])
218 {
219 	static const char *const mode[3] = {
220 		[COPY_STAGING_ENABLE] = "enable",
221 		[COPY_STAGING_DISABLE] = "disable",
222 		[COPY_STAGING_AUTO] = "auto",
223 	};
224 	int prev, res;
225 
226 	res = CMD_OK;
227 	if (argc > 2) {
228 		res = CMD_ERROR;
229 	} else if (argc == 2) {
230 		prev = copy_staging;
231 		if (strcmp(argv[1], "enable") == 0)
232 			copy_staging = COPY_STAGING_ENABLE;
233 		else if (strcmp(argv[1], "disable") == 0)
234 			copy_staging = COPY_STAGING_DISABLE;
235 		else if (strcmp(argv[1], "auto") == 0)
236 			copy_staging = COPY_STAGING_AUTO;
237 		else {
238 			printf("usage: copy_staging enable|disable|auto\n");
239 			res = CMD_ERROR;
240 		}
241 		if (res == CMD_OK && prev != copy_staging) {
242 			printf("changed copy_staging, unloading kernel\n");
243 			unload();
244 			efi_copy_free();
245 			efi_copy_init();
246 		}
247 	} else {
248 		printf("copy staging: %s\n", mode[copy_staging]);
249 	}
250 	return (res);
251 }
252 COMMAND_SET(copy_staging, "copy_staging", "copy staging", command_copy_staging);
253 #endif
254 
255 static int
256 command_staging_slop(int argc, char *argv[])
257 {
258 	char *endp;
259 	u_long new, prev;
260 	int res;
261 
262 	res = CMD_OK;
263 	if (argc > 2) {
264 		res = CMD_ERROR;
265 	} else if (argc == 2) {
266 		new = strtoul(argv[1], &endp, 0);
267 		if (*endp != '\0') {
268 			printf("invalid slop value\n");
269 			res = CMD_ERROR;
270 		}
271 		if (res == CMD_OK && staging_slop != new) {
272 			printf("changed slop, unloading kernel\n");
273 			unload();
274 			efi_copy_free();
275 			efi_copy_init();
276 		}
277 	} else {
278 		printf("staging slop %#lx\n", staging_slop);
279 	}
280 	return (res);
281 }
282 COMMAND_SET(staging_slop, "staging_slop", "set staging slop",
283     command_staging_slop);
284 
285 #if defined(__amd64__)
286 /*
287  * The staging area must reside in the first 1GB or 4GB physical
288  * memory: see elf64_exec() in
289  * boot/efi/loader/arch/amd64/elf64_freebsd.c.
290  */
291 static EFI_PHYSICAL_ADDRESS
292 get_staging_max(void)
293 {
294 	EFI_PHYSICAL_ADDRESS res;
295 
296 	res = copy_staging == COPY_STAGING_ENABLE ? G(1) : G(4);
297 	return (res);
298 }
299 #define	EFI_ALLOC_METHOD	AllocateMaxAddress
300 #else
301 #define	EFI_ALLOC_METHOD	AllocateAnyPages
302 #endif
303 
304 int
305 efi_copy_init(void)
306 {
307 	EFI_STATUS	status;
308 	unsigned long nr_pages;
309 	vm_offset_t ess;
310 
311 	ess = EFI_STAGING_SIZE;
312 	if (ess < DEFAULT_EFI_STAGING_SIZE)
313 		ess = DEFAULT_EFI_STAGING_SIZE;
314 	nr_pages = EFI_SIZE_TO_PAGES(M(1) * ess);
315 
316 #if defined(__amd64__)
317 	/*
318 	 * We'll decrease nr_pages, if it's too big. Currently we only
319 	 * apply this to FreeBSD VM running on Hyper-V. Why? Please see
320 	 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=211746#c28
321 	 */
322 	if (running_on_hyperv())
323 		efi_verify_staging_size(&nr_pages);
324 
325 	staging = get_staging_max();
326 #endif
327 	status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode,
328 	    nr_pages, &staging);
329 	if (EFI_ERROR(status)) {
330 		printf("failed to allocate staging area: %lu\n",
331 		    EFI_ERROR_CODE(status));
332 		return (status);
333 	}
334 	staging_base = staging;
335 	staging_end = staging + nr_pages * EFI_PAGE_SIZE;
336 
337 #if EFI_STAGING_2M_ALIGN
338 	/*
339 	 * Round the kernel load address to a 2MiB value. This is needed
340 	 * because the kernel builds a page table based on where it has
341 	 * been loaded in physical address space. As the kernel will use
342 	 * either a 1MiB or 2MiB page for this we need to make sure it
343 	 * is correctly aligned for both cases.
344 	 */
345 	staging = roundup2(staging, M(2));
346 #endif
347 
348 	return (0);
349 }
350 
351 static bool
352 efi_check_space(vm_offset_t end)
353 {
354 	EFI_PHYSICAL_ADDRESS addr, new_base, new_staging;
355 	EFI_STATUS status;
356 	unsigned long nr_pages;
357 
358 	end = roundup2(end, EFI_PAGE_SIZE);
359 
360 	/* There is already enough space */
361 	if (end + staging_slop <= staging_end)
362 		return (true);
363 
364 	if (!boot_services_active) {
365 		if (end <= staging_end)
366 			return (true);
367 		panic("efi_check_space: cannot expand staging area "
368 		    "after boot services were exited\n");
369 	}
370 
371 	/*
372 	 * Add slop at the end:
373 	 * 1. amd64 kernel expects to do some very early allocations
374 	 *    by carving out memory after kernend.  Slop guarantees
375 	 *    that it does not ovewrite anything useful.
376 	 * 2. It seems that initial calculation of the staging size
377 	 *    could be somewhat smaller than actually copying in after
378 	 *    boot services are exited.  Slop avoids calling
379 	 *    BS->AllocatePages() when it cannot work.
380 	 */
381 	end += staging_slop;
382 
383 	nr_pages = EFI_SIZE_TO_PAGES(end - staging_end);
384 #if defined(__amd64__)
385 	/*
386 	 * amd64 needs all memory to be allocated under the 1G or 4G boundary.
387 	 */
388 	if (end > get_staging_max())
389 		goto before_staging;
390 #endif
391 
392 	/* Try to allocate more space after the previous allocation */
393 	addr = staging_end;
394 	status = BS->AllocatePages(AllocateAddress, EfiLoaderCode, nr_pages,
395 	    &addr);
396 	if (!EFI_ERROR(status)) {
397 		staging_end = staging_end + nr_pages * EFI_PAGE_SIZE;
398 		return (true);
399 	}
400 
401 before_staging:
402 	/* Try allocating space before the previous allocation */
403 	if (staging < nr_pages * EFI_PAGE_SIZE)
404 		goto expand;
405 	addr = staging - nr_pages * EFI_PAGE_SIZE;
406 #if EFI_STAGING_2M_ALIGN
407 	/* See efi_copy_init for why this is needed */
408 	addr = rounddown2(addr, M(2));
409 #endif
410 	nr_pages = EFI_SIZE_TO_PAGES(staging_base - addr);
411 	status = BS->AllocatePages(AllocateAddress, EfiLoaderCode, nr_pages,
412 	    &addr);
413 	if (!EFI_ERROR(status)) {
414 		/*
415 		 * Move the old allocation and update the state so
416 		 * translation still works.
417 		 */
418 		staging_base = addr;
419 		memmove((void *)(uintptr_t)staging_base,
420 		    (void *)(uintptr_t)staging, staging_end - staging);
421 		stage_offset -= staging - staging_base;
422 		staging = staging_base;
423 		return (true);
424 	}
425 
426 expand:
427 	nr_pages = EFI_SIZE_TO_PAGES(end - (vm_offset_t)staging);
428 #if EFI_STAGING_2M_ALIGN
429 	nr_pages += M(2) / EFI_PAGE_SIZE;
430 #endif
431 #if defined(__amd64__)
432 	new_base = get_staging_max();
433 #endif
434 	status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode,
435 	    nr_pages, &new_base);
436 	if (!EFI_ERROR(status)) {
437 #if EFI_STAGING_2M_ALIGN
438 		new_staging = roundup2(new_base, M(2));
439 #else
440 		new_staging = new_base;
441 #endif
442 		/*
443 		 * Move the old allocation and update the state so
444 		 * translation still works.
445 		 */
446 		memcpy((void *)(uintptr_t)new_staging,
447 		    (void *)(uintptr_t)staging, staging_end - staging);
448 		BS->FreePages(staging_base, (staging_end - staging_base) /
449 		    EFI_PAGE_SIZE);
450 		stage_offset -= staging - new_staging;
451 		staging = new_staging;
452 		staging_end = new_base + nr_pages * EFI_PAGE_SIZE;
453 		staging_base = new_base;
454 		return (true);
455 	}
456 
457 	printf("efi_check_space: Unable to expand staging area\n");
458 	return (false);
459 }
460 
461 void *
462 efi_translate(vm_offset_t ptr)
463 {
464 
465 	return ((void *)(ptr + stage_offset));
466 }
467 
468 ssize_t
469 efi_copyin(const void *src, vm_offset_t dest, const size_t len)
470 {
471 
472 	if (!stage_offset_set) {
473 		stage_offset = (vm_offset_t)staging - dest;
474 		stage_offset_set = true;
475 	}
476 
477 	/* XXX: Callers do not check for failure. */
478 	if (!efi_check_space(dest + stage_offset + len)) {
479 		errno = ENOMEM;
480 		return (-1);
481 	}
482 	bcopy(src, (void *)(dest + stage_offset), len);
483 	return (len);
484 }
485 
486 ssize_t
487 efi_copyout(const vm_offset_t src, void *dest, const size_t len)
488 {
489 
490 	/* XXX: Callers do not check for failure. */
491 	if (src + stage_offset + len > staging_end) {
492 		errno = ENOMEM;
493 		return (-1);
494 	}
495 	bcopy((void *)(src + stage_offset), dest, len);
496 	return (len);
497 }
498 
499 ssize_t
500 efi_readin(readin_handle_t fd, vm_offset_t dest, const size_t len)
501 {
502 
503 	if (!stage_offset_set) {
504 		stage_offset = (vm_offset_t)staging - dest;
505 		stage_offset_set = true;
506 	}
507 
508 	if (!efi_check_space(dest + stage_offset + len)) {
509 		errno = ENOMEM;
510 		return (-1);
511 	}
512 	return (VECTX_READ(fd, (void *)(dest + stage_offset), len));
513 }
514 
515 void
516 efi_copy_finish(void)
517 {
518 	uint64_t	*src, *dst, *last;
519 
520 	src = (uint64_t *)(uintptr_t)staging;
521 	dst = (uint64_t *)(uintptr_t)(staging - stage_offset);
522 	last = (uint64_t *)(uintptr_t)staging_end;
523 
524 	while (src < last)
525 		*dst++ = *src++;
526 }
527 
528 void
529 efi_copy_finish_nop(void)
530 {
531 }
532