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/param.h>
29
30 #include <stand.h>
31 #include <bootstrap.h>
32
33 #include <efi.h>
34 #include <efilib.h>
35
36 #include "loader_efi.h"
37
38 #if defined(__amd64__)
39 #include <machine/cpufunc.h>
40 #include <machine/specialreg.h>
41 #include <machine/vmparam.h>
42
43 /*
44 * The code is excerpted from sys/x86/x86/identcpu.c: identify_cpu(),
45 * identify_hypervisor(), and dev/hyperv/vmbus/hyperv.c: hyperv_identify().
46 */
47 #define CPUID_LEAF_HV_MAXLEAF 0x40000000
48 #define CPUID_LEAF_HV_INTERFACE 0x40000001
49 #define CPUID_LEAF_HV_FEATURES 0x40000003
50 #define CPUID_LEAF_HV_LIMITS 0x40000005
51 #define CPUID_HV_IFACE_HYPERV 0x31237648 /* HV#1 */
52 #define CPUID_HV_MSR_TIME_REFCNT 0x0002 /* MSR_HV_TIME_REF_COUNT */
53 #define CPUID_HV_MSR_HYPERCALL 0x0020
54
55 static int
running_on_hyperv(void)56 running_on_hyperv(void)
57 {
58 char hv_vendor[16];
59 uint32_t regs[4];
60
61 do_cpuid(1, regs);
62 if ((regs[2] & CPUID2_HV) == 0)
63 return (0);
64
65 do_cpuid(CPUID_LEAF_HV_MAXLEAF, regs);
66 if (regs[0] < CPUID_LEAF_HV_LIMITS)
67 return (0);
68
69 ((uint32_t *)&hv_vendor)[0] = regs[1];
70 ((uint32_t *)&hv_vendor)[1] = regs[2];
71 ((uint32_t *)&hv_vendor)[2] = regs[3];
72 hv_vendor[12] = '\0';
73 if (strcmp(hv_vendor, "Microsoft Hv") != 0)
74 return (0);
75
76 do_cpuid(CPUID_LEAF_HV_INTERFACE, regs);
77 if (regs[0] != CPUID_HV_IFACE_HYPERV)
78 return (0);
79
80 do_cpuid(CPUID_LEAF_HV_FEATURES, regs);
81 if ((regs[0] & CPUID_HV_MSR_HYPERCALL) == 0)
82 return (0);
83 if ((regs[0] & CPUID_HV_MSR_TIME_REFCNT) == 0)
84 return (0);
85
86 return (1);
87 }
88
89 static void
efi_verify_staging_size(unsigned long * nr_pages)90 efi_verify_staging_size(unsigned long *nr_pages)
91 {
92 UINTN sz;
93 EFI_MEMORY_DESCRIPTOR *map = NULL, *p;
94 EFI_PHYSICAL_ADDRESS start, end;
95 UINTN key, dsz;
96 UINT32 dver;
97 EFI_STATUS status;
98 int i, ndesc;
99 unsigned long available_pages = 0;
100
101 sz = 0;
102
103 for (;;) {
104 status = BS->GetMemoryMap(&sz, map, &key, &dsz, &dver);
105 if (!EFI_ERROR(status))
106 break;
107
108 if (status != EFI_BUFFER_TOO_SMALL) {
109 printf("Can't read memory map: %lu\n",
110 EFI_ERROR_CODE(status));
111 goto out;
112 }
113
114 free(map);
115
116 /* Allocate 10 descriptors more than the size reported,
117 * to allow for any fragmentation caused by calling
118 * malloc */
119 map = malloc(sz + (10 * dsz));
120 if (map == NULL) {
121 printf("Unable to allocate memory\n");
122 goto out;
123 }
124 }
125
126 ndesc = sz / dsz;
127 for (i = 0, p = map; i < ndesc;
128 i++, p = NextMemoryDescriptor(p, dsz)) {
129 start = p->PhysicalStart;
130 end = start + p->NumberOfPages * EFI_PAGE_SIZE;
131
132 if (KERNLOAD < start || KERNLOAD >= end)
133 continue;
134
135 available_pages = p->NumberOfPages -
136 ((KERNLOAD - start) >> EFI_PAGE_SHIFT);
137 break;
138 }
139
140 if (available_pages == 0) {
141 printf("Can't find valid memory map for staging area!\n");
142 goto out;
143 }
144
145 i++;
146 p = NextMemoryDescriptor(p, dsz);
147
148 for ( ; i < ndesc;
149 i++, p = NextMemoryDescriptor(p, dsz)) {
150 if (p->Type != EfiConventionalMemory &&
151 p->Type != EfiLoaderData)
152 break;
153
154 if (p->PhysicalStart != end)
155 break;
156
157 end = p->PhysicalStart + p->NumberOfPages * EFI_PAGE_SIZE;
158
159 available_pages += p->NumberOfPages;
160 }
161
162 if (*nr_pages > available_pages) {
163 printf("Staging area's size is reduced: %ld -> %ld!\n",
164 *nr_pages, available_pages);
165 *nr_pages = available_pages;
166 }
167 out:
168 free(map);
169 }
170 #endif /* __amd64__ */
171
172 #if defined(__arm__)
173 #define DEFAULT_EFI_STAGING_SIZE 32
174 #else
175 #define DEFAULT_EFI_STAGING_SIZE 64
176 #endif
177 #ifndef EFI_STAGING_SIZE
178 #define EFI_STAGING_SIZE DEFAULT_EFI_STAGING_SIZE
179 #endif
180
181 #define EFI_STAGING_2M_ALIGN 1
182
183 #if defined(__amd64__) || defined(__i386__)
184 #define EFI_STAGING_SLOP M(8)
185 #else
186 #define EFI_STAGING_SLOP 0
187 #endif
188
189 static u_long staging_slop = EFI_STAGING_SLOP;
190
191 EFI_PHYSICAL_ADDRESS staging, staging_end, staging_base;
192 bool stage_offset_set = false;
193 ssize_t stage_offset;
194
195 static void
efi_copy_free(void)196 efi_copy_free(void)
197 {
198 BS->FreePages(staging_base, (staging_end - staging_base) /
199 EFI_PAGE_SIZE);
200 stage_offset_set = false;
201 stage_offset = 0;
202 }
203
204 #if defined(__amd64__) || defined(__i386__)
205 int copy_staging = COPY_STAGING_AUTO;
206
207 static int
command_copy_staging(int argc,char * argv[])208 command_copy_staging(int argc, char *argv[])
209 {
210 static const char *const mode[3] = {
211 [COPY_STAGING_ENABLE] = "enable",
212 [COPY_STAGING_DISABLE] = "disable",
213 [COPY_STAGING_AUTO] = "auto",
214 };
215 int prev;
216
217 if (argc > 2) {
218 goto usage;
219 } else if (argc == 2) {
220 prev = copy_staging;
221 if (strcmp(argv[1], "enable") == 0)
222 copy_staging = COPY_STAGING_ENABLE;
223 else if (strcmp(argv[1], "disable") == 0)
224 copy_staging = COPY_STAGING_DISABLE;
225 else if (strcmp(argv[1], "auto") == 0)
226 copy_staging = COPY_STAGING_AUTO;
227 else
228 goto usage;
229 if (prev != copy_staging) {
230 printf("changed copy_staging, unloading kernel\n");
231 unload();
232 efi_copy_free();
233 efi_copy_init();
234 }
235 } else {
236 printf("copy staging: %s\n", mode[copy_staging]);
237 }
238 return (CMD_OK);
239
240 usage:
241 command_errmsg = "usage: copy_staging enable|disable|auto";
242 return (CMD_ERROR);
243 }
244 COMMAND_SET(copy_staging, "copy_staging", "copy staging", command_copy_staging);
245 #endif
246
247 static int
command_staging_slop(int argc,char * argv[])248 command_staging_slop(int argc, char *argv[])
249 {
250 char *endp;
251 u_long new, prev;
252
253 if (argc > 2) {
254 goto err;
255 } else if (argc == 2) {
256 new = strtoul(argv[1], &endp, 0);
257 if (*endp != '\0')
258 goto err;
259 if (staging_slop != new) {
260 staging_slop = new;
261 printf("changed slop, unloading kernel\n");
262
263 unload();
264 efi_copy_free();
265 efi_copy_init();
266 }
267 } else {
268 printf("staging slop %#lx\n", staging_slop);
269 }
270 return (CMD_OK);
271
272 err:
273 command_errmsg = "invalid slop value";
274 return (CMD_ERROR);
275 }
276 COMMAND_SET(staging_slop, "staging_slop", "set staging slop",
277 command_staging_slop);
278
279 #if defined(__amd64__) || defined(__i386__)
280 /*
281 * The staging area must reside in the first 1GB or 4GB physical
282 * memory: see elf64_exec() in
283 * boot/efi/loader/arch/amd64/elf64_freebsd.c.
284 */
285 static EFI_PHYSICAL_ADDRESS
get_staging_max(void)286 get_staging_max(void)
287 {
288 EFI_PHYSICAL_ADDRESS res;
289
290 res = copy_staging == COPY_STAGING_ENABLE ? G(1) : G(4);
291 return (res);
292 }
293 #define EFI_ALLOC_MAX_ADDR
294 #elif defined(__aarch64__)
295 /*
296 * Older kernels only support a 48-bit physical address space, and locore.S
297 * only supports a 50-bit space. Limit to 48 bits so older kernels can boot
298 * even if FEAT_LPA2 is supported by the hardware.
299 */
300 #define get_staging_max() (1ul << 48)
301 #define EFI_ALLOC_MAX_ADDR
302 #endif
303 #ifdef EFI_ALLOC_MAX_ADDR
304 #define EFI_ALLOC_METHOD AllocateMaxAddress
305 #else
306 #define EFI_ALLOC_METHOD AllocateAnyPages
307 #endif
308
309 int
efi_copy_init(void)310 efi_copy_init(void)
311 {
312 EFI_STATUS status;
313 unsigned long nr_pages;
314 vm_offset_t ess;
315
316 ess = EFI_STAGING_SIZE;
317 if (ess < DEFAULT_EFI_STAGING_SIZE)
318 ess = DEFAULT_EFI_STAGING_SIZE;
319 nr_pages = EFI_SIZE_TO_PAGES(M(1) * ess);
320
321 #if defined(__amd64__)
322 /*
323 * We'll decrease nr_pages, if it's too big. Currently we only
324 * apply this to FreeBSD VM running on Hyper-V. Why? Please see
325 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=211746#c28
326 */
327 if (running_on_hyperv())
328 efi_verify_staging_size(&nr_pages);
329 #endif
330 #ifdef EFI_ALLOC_MAX_ADDR
331 staging = get_staging_max();
332 #endif
333 status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode,
334 nr_pages, &staging);
335 if (EFI_ERROR(status)) {
336 printf("failed to allocate staging area: %lu\n",
337 EFI_ERROR_CODE(status));
338 return (status);
339 }
340 staging_base = staging;
341 staging_end = staging + nr_pages * EFI_PAGE_SIZE;
342
343 #if EFI_STAGING_2M_ALIGN
344 /*
345 * Round the kernel load address to a 2MiB value. This is needed
346 * because the kernel builds a page table based on where it has
347 * been loaded in physical address space. As the kernel will use
348 * either a 1MiB or 2MiB page for this we need to make sure it
349 * is correctly aligned for both cases.
350 */
351 staging = roundup2(staging, M(2));
352 #endif
353
354 return (0);
355 }
356
357 static bool
efi_check_space(vm_offset_t end)358 efi_check_space(vm_offset_t end)
359 {
360 EFI_PHYSICAL_ADDRESS addr, new_base, new_staging;
361 EFI_STATUS status;
362 unsigned long nr_pages;
363
364 end = roundup2(end, EFI_PAGE_SIZE);
365
366 /* There is already enough space */
367 if (end + staging_slop <= staging_end)
368 return (true);
369
370 if (!boot_services_active) {
371 if (end <= staging_end)
372 return (true);
373 panic("efi_check_space: cannot expand staging area "
374 "after boot services were exited\n");
375 }
376
377 /*
378 * Add slop at the end:
379 * 1. amd64 kernel expects to do some very early allocations
380 * by carving out memory after kernend. Slop guarantees
381 * that it does not ovewrite anything useful.
382 * 2. It seems that initial calculation of the staging size
383 * could be somewhat smaller than actually copying in after
384 * boot services are exited. Slop avoids calling
385 * BS->AllocatePages() when it cannot work.
386 */
387 end += staging_slop;
388
389 nr_pages = EFI_SIZE_TO_PAGES(end - staging_end);
390 #if defined(__amd64__) || defined(__i386__)
391 /*
392 * The amd64 kernel needs all memory to be allocated under the 1G or
393 * 4G boundary.
394 */
395 if (end > get_staging_max())
396 goto before_staging;
397 #endif
398
399 /* Try to allocate more space after the previous allocation */
400 addr = staging_end;
401 status = BS->AllocatePages(AllocateAddress, EfiLoaderCode, nr_pages,
402 &addr);
403 if (!EFI_ERROR(status)) {
404 staging_end = staging_end + nr_pages * EFI_PAGE_SIZE;
405 return (true);
406 }
407
408 before_staging:
409 /* Try allocating space before the previous allocation */
410 if (staging < nr_pages * EFI_PAGE_SIZE)
411 goto expand;
412 addr = staging - nr_pages * EFI_PAGE_SIZE;
413 #if EFI_STAGING_2M_ALIGN
414 /* See efi_copy_init for why this is needed */
415 addr = rounddown2(addr, M(2));
416 #endif
417 nr_pages = EFI_SIZE_TO_PAGES(staging_base - addr);
418 status = BS->AllocatePages(AllocateAddress, EfiLoaderCode, nr_pages,
419 &addr);
420 if (!EFI_ERROR(status)) {
421 /*
422 * Move the old allocation and update the state so
423 * translation still works.
424 */
425 staging_base = addr;
426 memmove((void *)(uintptr_t)staging_base,
427 (void *)(uintptr_t)staging, staging_end - staging);
428 stage_offset -= staging - staging_base;
429 staging = staging_base;
430 return (true);
431 }
432
433 expand:
434 nr_pages = EFI_SIZE_TO_PAGES(end - (vm_offset_t)staging);
435 #if EFI_STAGING_2M_ALIGN
436 nr_pages += M(2) / EFI_PAGE_SIZE;
437 #endif
438 #ifdef EFI_ALLOC_MAX_ADDR
439 new_base = get_staging_max();
440 #endif
441 status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode,
442 nr_pages, &new_base);
443 if (!EFI_ERROR(status)) {
444 #if EFI_STAGING_2M_ALIGN
445 new_staging = roundup2(new_base, M(2));
446 #else
447 new_staging = new_base;
448 #endif
449 /*
450 * Move the old allocation and update the state so
451 * translation still works.
452 */
453 memcpy((void *)(uintptr_t)new_staging,
454 (void *)(uintptr_t)staging, staging_end - staging);
455 BS->FreePages(staging_base, (staging_end - staging_base) /
456 EFI_PAGE_SIZE);
457 stage_offset -= staging - new_staging;
458 staging = new_staging;
459 staging_end = new_base + nr_pages * EFI_PAGE_SIZE;
460 staging_base = new_base;
461 return (true);
462 }
463
464 printf("efi_check_space: Unable to expand staging area\n");
465 return (false);
466 }
467
468 void *
efi_translate(vm_offset_t ptr)469 efi_translate(vm_offset_t ptr)
470 {
471
472 return ((void *)(ptr + stage_offset));
473 }
474
475 ssize_t
efi_copyin(const void * src,vm_offset_t dest,const size_t len)476 efi_copyin(const void *src, vm_offset_t dest, const size_t len)
477 {
478
479 if (!stage_offset_set) {
480 stage_offset = (vm_offset_t)staging - dest;
481 stage_offset_set = true;
482 }
483
484 /* XXX: Callers do not check for failure. */
485 if (!efi_check_space(dest + stage_offset + len)) {
486 errno = ENOMEM;
487 return (-1);
488 }
489 bcopy(src, (void *)(dest + stage_offset), len);
490 return (len);
491 }
492
493 ssize_t
efi_copyout(const vm_offset_t src,void * dest,const size_t len)494 efi_copyout(const vm_offset_t src, void *dest, const size_t len)
495 {
496
497 /* XXX: Callers do not check for failure. */
498 if (src + stage_offset + len > staging_end) {
499 errno = ENOMEM;
500 return (-1);
501 }
502 bcopy((void *)(src + stage_offset), dest, len);
503 return (len);
504 }
505
506 ssize_t
efi_readin(readin_handle_t fd,vm_offset_t dest,const size_t len)507 efi_readin(readin_handle_t fd, vm_offset_t dest, const size_t len)
508 {
509
510 if (!stage_offset_set) {
511 stage_offset = (vm_offset_t)staging - dest;
512 stage_offset_set = true;
513 }
514
515 if (!efi_check_space(dest + stage_offset + len)) {
516 errno = ENOMEM;
517 return (-1);
518 }
519 return (VECTX_READ(fd, (void *)(dest + stage_offset), len));
520 }
521
522 void
efi_copy_finish(void)523 efi_copy_finish(void)
524 {
525 uint64_t *src, *dst, *last;
526
527 src = (uint64_t *)(uintptr_t)staging;
528 dst = (uint64_t *)(uintptr_t)(staging - stage_offset);
529 last = (uint64_t *)(uintptr_t)staging_end;
530
531 while (src < last)
532 *dst++ = *src++;
533 }
534
535 void
efi_copy_finish_nop(void)536 efi_copy_finish_nop(void)
537 {
538 }
539