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 __FBSDID("$FreeBSD$"); 30 31 #include <sys/param.h> 32 33 #include <stand.h> 34 #include <bootstrap.h> 35 36 #include <efi.h> 37 #include <efilib.h> 38 39 #include "loader_efi.h" 40 41 #define M(x) ((x) * 1024 * 1024) 42 #define G(x) (1UL * (x) * 1024 * 1024 * 1024) 43 44 #if defined(__amd64__) 45 #include <machine/cpufunc.h> 46 #include <machine/specialreg.h> 47 #include <machine/vmparam.h> 48 49 /* 50 * The code is excerpted from sys/x86/x86/identcpu.c: identify_cpu(), 51 * identify_hypervisor(), and dev/hyperv/vmbus/hyperv.c: hyperv_identify(). 52 */ 53 #define CPUID_LEAF_HV_MAXLEAF 0x40000000 54 #define CPUID_LEAF_HV_INTERFACE 0x40000001 55 #define CPUID_LEAF_HV_FEATURES 0x40000003 56 #define CPUID_LEAF_HV_LIMITS 0x40000005 57 #define CPUID_HV_IFACE_HYPERV 0x31237648 /* HV#1 */ 58 #define CPUID_HV_MSR_TIME_REFCNT 0x0002 /* MSR_HV_TIME_REF_COUNT */ 59 #define CPUID_HV_MSR_HYPERCALL 0x0020 60 61 static int 62 running_on_hyperv(void) 63 { 64 char hv_vendor[16]; 65 uint32_t regs[4]; 66 67 do_cpuid(1, regs); 68 if ((regs[2] & CPUID2_HV) == 0) 69 return (0); 70 71 do_cpuid(CPUID_LEAF_HV_MAXLEAF, regs); 72 if (regs[0] < CPUID_LEAF_HV_LIMITS) 73 return (0); 74 75 ((uint32_t *)&hv_vendor)[0] = regs[1]; 76 ((uint32_t *)&hv_vendor)[1] = regs[2]; 77 ((uint32_t *)&hv_vendor)[2] = regs[3]; 78 hv_vendor[12] = '\0'; 79 if (strcmp(hv_vendor, "Microsoft Hv") != 0) 80 return (0); 81 82 do_cpuid(CPUID_LEAF_HV_INTERFACE, regs); 83 if (regs[0] != CPUID_HV_IFACE_HYPERV) 84 return (0); 85 86 do_cpuid(CPUID_LEAF_HV_FEATURES, regs); 87 if ((regs[0] & CPUID_HV_MSR_HYPERCALL) == 0) 88 return (0); 89 if ((regs[0] & CPUID_HV_MSR_TIME_REFCNT) == 0) 90 return (0); 91 92 return (1); 93 } 94 95 static void 96 efi_verify_staging_size(unsigned long *nr_pages) 97 { 98 UINTN sz; 99 EFI_MEMORY_DESCRIPTOR *map = NULL, *p; 100 EFI_PHYSICAL_ADDRESS start, end; 101 UINTN key, dsz; 102 UINT32 dver; 103 EFI_STATUS status; 104 int i, ndesc; 105 unsigned long available_pages = 0; 106 107 sz = 0; 108 109 for (;;) { 110 status = BS->GetMemoryMap(&sz, map, &key, &dsz, &dver); 111 if (!EFI_ERROR(status)) 112 break; 113 114 if (status != EFI_BUFFER_TOO_SMALL) { 115 printf("Can't read memory map: %lu\n", 116 EFI_ERROR_CODE(status)); 117 goto out; 118 } 119 120 free(map); 121 122 /* Allocate 10 descriptors more than the size reported, 123 * to allow for any fragmentation caused by calling 124 * malloc */ 125 map = malloc(sz + (10 * dsz)); 126 if (map == NULL) { 127 printf("Unable to allocate memory\n"); 128 goto out; 129 } 130 } 131 132 ndesc = sz / dsz; 133 for (i = 0, p = map; i < ndesc; 134 i++, p = NextMemoryDescriptor(p, dsz)) { 135 start = p->PhysicalStart; 136 end = start + p->NumberOfPages * EFI_PAGE_SIZE; 137 138 if (KERNLOAD < start || KERNLOAD >= end) 139 continue; 140 141 available_pages = p->NumberOfPages - 142 ((KERNLOAD - start) >> EFI_PAGE_SHIFT); 143 break; 144 } 145 146 if (available_pages == 0) { 147 printf("Can't find valid memory map for staging area!\n"); 148 goto out; 149 } 150 151 i++; 152 p = NextMemoryDescriptor(p, dsz); 153 154 for ( ; i < ndesc; 155 i++, p = NextMemoryDescriptor(p, dsz)) { 156 if (p->Type != EfiConventionalMemory && 157 p->Type != EfiLoaderData) 158 break; 159 160 if (p->PhysicalStart != end) 161 break; 162 163 end = p->PhysicalStart + p->NumberOfPages * EFI_PAGE_SIZE; 164 165 available_pages += p->NumberOfPages; 166 } 167 168 if (*nr_pages > available_pages) { 169 printf("Staging area's size is reduced: %ld -> %ld!\n", 170 *nr_pages, available_pages); 171 *nr_pages = available_pages; 172 } 173 out: 174 free(map); 175 } 176 #endif /* __amd64__ */ 177 178 #if defined(__arm__) 179 #define DEFAULT_EFI_STAGING_SIZE 32 180 #else 181 #define DEFAULT_EFI_STAGING_SIZE 64 182 #endif 183 #ifndef EFI_STAGING_SIZE 184 #define EFI_STAGING_SIZE DEFAULT_EFI_STAGING_SIZE 185 #endif 186 187 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \ 188 defined(__riscv) 189 #define EFI_STAGING_2M_ALIGN 1 190 #else 191 #define EFI_STAGING_2M_ALIGN 0 192 #endif 193 194 #if defined(__amd64__) 195 #define EFI_STAGING_SLOP M(8) 196 #else 197 #define EFI_STAGING_SLOP 0 198 #endif 199 200 static u_long staging_slop = EFI_STAGING_SLOP; 201 202 EFI_PHYSICAL_ADDRESS staging, staging_end, staging_base; 203 bool stage_offset_set = false; 204 ssize_t stage_offset; 205 206 static void 207 efi_copy_free(void) 208 { 209 BS->FreePages(staging_base, (staging_end - staging_base) / 210 EFI_PAGE_SIZE); 211 stage_offset_set = false; 212 stage_offset = 0; 213 } 214 215 #ifdef __amd64__ 216 int copy_staging = COPY_STAGING_AUTO; 217 218 static int 219 command_copy_staging(int argc, char *argv[]) 220 { 221 static const char *const mode[3] = { 222 [COPY_STAGING_ENABLE] = "enable", 223 [COPY_STAGING_DISABLE] = "disable", 224 [COPY_STAGING_AUTO] = "auto", 225 }; 226 int prev, res; 227 228 res = CMD_OK; 229 if (argc > 2) { 230 res = CMD_ERROR; 231 } else if (argc == 2) { 232 prev = copy_staging; 233 if (strcmp(argv[1], "enable") == 0) 234 copy_staging = COPY_STAGING_ENABLE; 235 else if (strcmp(argv[1], "disable") == 0) 236 copy_staging = COPY_STAGING_DISABLE; 237 else if (strcmp(argv[1], "auto") == 0) 238 copy_staging = COPY_STAGING_AUTO; 239 else { 240 printf("usage: copy_staging enable|disable|auto\n"); 241 res = CMD_ERROR; 242 } 243 if (res == CMD_OK && prev != copy_staging) { 244 printf("changed copy_staging, unloading kernel\n"); 245 unload(); 246 efi_copy_free(); 247 efi_copy_init(); 248 } 249 } else { 250 printf("copy staging: %s\n", mode[copy_staging]); 251 } 252 return (res); 253 } 254 COMMAND_SET(copy_staging, "copy_staging", "copy staging", command_copy_staging); 255 #endif 256 257 static int 258 command_staging_slop(int argc, char *argv[]) 259 { 260 char *endp; 261 u_long new, prev; 262 int res; 263 264 res = CMD_OK; 265 if (argc > 2) { 266 res = CMD_ERROR; 267 } else if (argc == 2) { 268 new = strtoul(argv[1], &endp, 0); 269 if (*endp != '\0') { 270 printf("invalid slop value\n"); 271 res = CMD_ERROR; 272 } 273 if (res == CMD_OK && staging_slop != new) { 274 printf("changed slop, unloading kernel\n"); 275 unload(); 276 efi_copy_free(); 277 efi_copy_init(); 278 } 279 } else { 280 printf("staging slop %#lx\n", staging_slop); 281 } 282 return (res); 283 } 284 COMMAND_SET(staging_slop, "staging_slop", "set staging slop", 285 command_staging_slop); 286 287 #if defined(__amd64__) 288 /* 289 * The staging area must reside in the first 1GB or 4GB physical 290 * memory: see elf64_exec() in 291 * boot/efi/loader/arch/amd64/elf64_freebsd.c. 292 */ 293 static EFI_PHYSICAL_ADDRESS 294 get_staging_max(void) 295 { 296 EFI_PHYSICAL_ADDRESS res; 297 298 res = copy_staging == COPY_STAGING_ENABLE ? G(1) : G(4); 299 return (res); 300 } 301 #define EFI_ALLOC_METHOD AllocateMaxAddress 302 #else 303 #define EFI_ALLOC_METHOD AllocateAnyPages 304 #endif 305 306 int 307 efi_copy_init(void) 308 { 309 EFI_STATUS status; 310 unsigned long nr_pages; 311 vm_offset_t ess; 312 313 ess = EFI_STAGING_SIZE; 314 if (ess < DEFAULT_EFI_STAGING_SIZE) 315 ess = DEFAULT_EFI_STAGING_SIZE; 316 nr_pages = EFI_SIZE_TO_PAGES(M(1) * ess); 317 318 #if defined(__amd64__) 319 /* 320 * We'll decrease nr_pages, if it's too big. Currently we only 321 * apply this to FreeBSD VM running on Hyper-V. Why? Please see 322 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=211746#c28 323 */ 324 if (running_on_hyperv()) 325 efi_verify_staging_size(&nr_pages); 326 327 staging = get_staging_max(); 328 #endif 329 status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode, 330 nr_pages, &staging); 331 if (EFI_ERROR(status)) { 332 printf("failed to allocate staging area: %lu\n", 333 EFI_ERROR_CODE(status)); 334 return (status); 335 } 336 staging_base = staging; 337 staging_end = staging + nr_pages * EFI_PAGE_SIZE; 338 339 #if EFI_STAGING_2M_ALIGN 340 /* 341 * Round the kernel load address to a 2MiB value. This is needed 342 * because the kernel builds a page table based on where it has 343 * been loaded in physical address space. As the kernel will use 344 * either a 1MiB or 2MiB page for this we need to make sure it 345 * is correctly aligned for both cases. 346 */ 347 staging = roundup2(staging, M(2)); 348 #endif 349 350 return (0); 351 } 352 353 static bool 354 efi_check_space(vm_offset_t end) 355 { 356 EFI_PHYSICAL_ADDRESS addr, new_base, new_staging; 357 EFI_STATUS status; 358 unsigned long nr_pages; 359 360 end = roundup2(end, EFI_PAGE_SIZE); 361 362 /* There is already enough space */ 363 if (end + staging_slop <= staging_end) 364 return (true); 365 366 if (!boot_services_active) { 367 if (end <= staging_end) 368 return (true); 369 panic("efi_check_space: cannot expand staging area " 370 "after boot services were exited\n"); 371 } 372 373 /* 374 * Add slop at the end: 375 * 1. amd64 kernel expects to do some very early allocations 376 * by carving out memory after kernend. Slop guarantees 377 * that it does not ovewrite anything useful. 378 * 2. It seems that initial calculation of the staging size 379 * could be somewhat smaller than actually copying in after 380 * boot services are exited. Slop avoids calling 381 * BS->AllocatePages() when it cannot work. 382 */ 383 end += staging_slop; 384 385 nr_pages = EFI_SIZE_TO_PAGES(end - staging_end); 386 #if defined(__amd64__) 387 /* 388 * amd64 needs all memory to be allocated under the 1G or 4G boundary. 389 */ 390 if (end > get_staging_max()) 391 goto before_staging; 392 #endif 393 394 /* Try to allocate more space after the previous allocation */ 395 addr = staging_end; 396 status = BS->AllocatePages(AllocateAddress, EfiLoaderCode, nr_pages, 397 &addr); 398 if (!EFI_ERROR(status)) { 399 staging_end = staging_end + nr_pages * EFI_PAGE_SIZE; 400 return (true); 401 } 402 403 before_staging: 404 /* Try allocating space before the previous allocation */ 405 if (staging < nr_pages * EFI_PAGE_SIZE) 406 goto expand; 407 addr = staging - nr_pages * EFI_PAGE_SIZE; 408 #if EFI_STAGING_2M_ALIGN 409 /* See efi_copy_init for why this is needed */ 410 addr = rounddown2(addr, M(2)); 411 #endif 412 nr_pages = EFI_SIZE_TO_PAGES(staging_base - addr); 413 status = BS->AllocatePages(AllocateAddress, EfiLoaderCode, nr_pages, 414 &addr); 415 if (!EFI_ERROR(status)) { 416 /* 417 * Move the old allocation and update the state so 418 * translation still works. 419 */ 420 staging_base = addr; 421 memmove((void *)(uintptr_t)staging_base, 422 (void *)(uintptr_t)staging, staging_end - staging); 423 stage_offset -= staging - staging_base; 424 staging = staging_base; 425 return (true); 426 } 427 428 expand: 429 nr_pages = EFI_SIZE_TO_PAGES(end - (vm_offset_t)staging); 430 #if EFI_STAGING_2M_ALIGN 431 nr_pages += M(2) / EFI_PAGE_SIZE; 432 #endif 433 #if defined(__amd64__) 434 new_base = get_staging_max(); 435 #endif 436 status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode, 437 nr_pages, &new_base); 438 if (!EFI_ERROR(status)) { 439 #if EFI_STAGING_2M_ALIGN 440 new_staging = roundup2(new_base, M(2)); 441 #else 442 new_staging = new_base; 443 #endif 444 /* 445 * Move the old allocation and update the state so 446 * translation still works. 447 */ 448 memcpy((void *)(uintptr_t)new_staging, 449 (void *)(uintptr_t)staging, staging_end - staging); 450 BS->FreePages(staging_base, (staging_end - staging_base) / 451 EFI_PAGE_SIZE); 452 stage_offset -= staging - new_staging; 453 staging = new_staging; 454 staging_end = new_base + nr_pages * EFI_PAGE_SIZE; 455 staging_base = new_base; 456 return (true); 457 } 458 459 printf("efi_check_space: Unable to expand staging area\n"); 460 return (false); 461 } 462 463 void * 464 efi_translate(vm_offset_t ptr) 465 { 466 467 return ((void *)(ptr + stage_offset)); 468 } 469 470 ssize_t 471 efi_copyin(const void *src, vm_offset_t dest, const size_t len) 472 { 473 474 if (!stage_offset_set) { 475 stage_offset = (vm_offset_t)staging - dest; 476 stage_offset_set = true; 477 } 478 479 /* XXX: Callers do not check for failure. */ 480 if (!efi_check_space(dest + stage_offset + len)) { 481 errno = ENOMEM; 482 return (-1); 483 } 484 bcopy(src, (void *)(dest + stage_offset), len); 485 return (len); 486 } 487 488 ssize_t 489 efi_copyout(const vm_offset_t src, void *dest, const size_t len) 490 { 491 492 /* XXX: Callers do not check for failure. */ 493 if (src + stage_offset + len > staging_end) { 494 errno = ENOMEM; 495 return (-1); 496 } 497 bcopy((void *)(src + stage_offset), dest, len); 498 return (len); 499 } 500 501 ssize_t 502 efi_readin(readin_handle_t fd, vm_offset_t dest, const size_t len) 503 { 504 505 if (!stage_offset_set) { 506 stage_offset = (vm_offset_t)staging - dest; 507 stage_offset_set = true; 508 } 509 510 if (!efi_check_space(dest + stage_offset + len)) { 511 errno = ENOMEM; 512 return (-1); 513 } 514 return (VECTX_READ(fd, (void *)(dest + stage_offset), len)); 515 } 516 517 void 518 efi_copy_finish(void) 519 { 520 uint64_t *src, *dst, *last; 521 522 src = (uint64_t *)(uintptr_t)staging; 523 dst = (uint64_t *)(uintptr_t)(staging - stage_offset); 524 last = (uint64_t *)(uintptr_t)staging_end; 525 526 while (src < last) 527 *dst++ = *src++; 528 } 529 530 void 531 efi_copy_finish_nop(void) 532 { 533 } 534