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 #define M(x) ((x) * 1024 * 1024) 39 #define G(x) (1ULL * (x) * 1024 * 1024 * 1024) 40 41 #if defined(__amd64__) 42 #include <machine/cpufunc.h> 43 #include <machine/specialreg.h> 44 #include <machine/vmparam.h> 45 46 /* 47 * The code is excerpted from sys/x86/x86/identcpu.c: identify_cpu(), 48 * identify_hypervisor(), and dev/hyperv/vmbus/hyperv.c: hyperv_identify(). 49 */ 50 #define CPUID_LEAF_HV_MAXLEAF 0x40000000 51 #define CPUID_LEAF_HV_INTERFACE 0x40000001 52 #define CPUID_LEAF_HV_FEATURES 0x40000003 53 #define CPUID_LEAF_HV_LIMITS 0x40000005 54 #define CPUID_HV_IFACE_HYPERV 0x31237648 /* HV#1 */ 55 #define CPUID_HV_MSR_TIME_REFCNT 0x0002 /* MSR_HV_TIME_REF_COUNT */ 56 #define CPUID_HV_MSR_HYPERCALL 0x0020 57 58 static int 59 running_on_hyperv(void) 60 { 61 char hv_vendor[16]; 62 uint32_t regs[4]; 63 64 do_cpuid(1, regs); 65 if ((regs[2] & CPUID2_HV) == 0) 66 return (0); 67 68 do_cpuid(CPUID_LEAF_HV_MAXLEAF, regs); 69 if (regs[0] < CPUID_LEAF_HV_LIMITS) 70 return (0); 71 72 ((uint32_t *)&hv_vendor)[0] = regs[1]; 73 ((uint32_t *)&hv_vendor)[1] = regs[2]; 74 ((uint32_t *)&hv_vendor)[2] = regs[3]; 75 hv_vendor[12] = '\0'; 76 if (strcmp(hv_vendor, "Microsoft Hv") != 0) 77 return (0); 78 79 do_cpuid(CPUID_LEAF_HV_INTERFACE, regs); 80 if (regs[0] != CPUID_HV_IFACE_HYPERV) 81 return (0); 82 83 do_cpuid(CPUID_LEAF_HV_FEATURES, regs); 84 if ((regs[0] & CPUID_HV_MSR_HYPERCALL) == 0) 85 return (0); 86 if ((regs[0] & CPUID_HV_MSR_TIME_REFCNT) == 0) 87 return (0); 88 89 return (1); 90 } 91 92 static void 93 efi_verify_staging_size(unsigned long *nr_pages) 94 { 95 UINTN sz; 96 EFI_MEMORY_DESCRIPTOR *map = NULL, *p; 97 EFI_PHYSICAL_ADDRESS start, end; 98 UINTN key, dsz; 99 UINT32 dver; 100 EFI_STATUS status; 101 int i, ndesc; 102 unsigned long available_pages = 0; 103 104 sz = 0; 105 106 for (;;) { 107 status = BS->GetMemoryMap(&sz, map, &key, &dsz, &dver); 108 if (!EFI_ERROR(status)) 109 break; 110 111 if (status != EFI_BUFFER_TOO_SMALL) { 112 printf("Can't read memory map: %lu\n", 113 EFI_ERROR_CODE(status)); 114 goto out; 115 } 116 117 free(map); 118 119 /* Allocate 10 descriptors more than the size reported, 120 * to allow for any fragmentation caused by calling 121 * malloc */ 122 map = malloc(sz + (10 * dsz)); 123 if (map == NULL) { 124 printf("Unable to allocate memory\n"); 125 goto out; 126 } 127 } 128 129 ndesc = sz / dsz; 130 for (i = 0, p = map; i < ndesc; 131 i++, p = NextMemoryDescriptor(p, dsz)) { 132 start = p->PhysicalStart; 133 end = start + p->NumberOfPages * EFI_PAGE_SIZE; 134 135 if (KERNLOAD < start || KERNLOAD >= end) 136 continue; 137 138 available_pages = p->NumberOfPages - 139 ((KERNLOAD - start) >> EFI_PAGE_SHIFT); 140 break; 141 } 142 143 if (available_pages == 0) { 144 printf("Can't find valid memory map for staging area!\n"); 145 goto out; 146 } 147 148 i++; 149 p = NextMemoryDescriptor(p, dsz); 150 151 for ( ; i < ndesc; 152 i++, p = NextMemoryDescriptor(p, dsz)) { 153 if (p->Type != EfiConventionalMemory && 154 p->Type != EfiLoaderData) 155 break; 156 157 if (p->PhysicalStart != end) 158 break; 159 160 end = p->PhysicalStart + p->NumberOfPages * EFI_PAGE_SIZE; 161 162 available_pages += p->NumberOfPages; 163 } 164 165 if (*nr_pages > available_pages) { 166 printf("Staging area's size is reduced: %ld -> %ld!\n", 167 *nr_pages, available_pages); 168 *nr_pages = available_pages; 169 } 170 out: 171 free(map); 172 } 173 #endif /* __amd64__ */ 174 175 #if defined(__arm__) 176 #define DEFAULT_EFI_STAGING_SIZE 32 177 #else 178 #define DEFAULT_EFI_STAGING_SIZE 64 179 #endif 180 #ifndef EFI_STAGING_SIZE 181 #define EFI_STAGING_SIZE DEFAULT_EFI_STAGING_SIZE 182 #endif 183 184 #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \ 185 defined(__riscv) || defined(__i386__) 186 #define EFI_STAGING_2M_ALIGN 1 187 #else 188 #define EFI_STAGING_2M_ALIGN 0 189 #endif 190 191 #if defined(__amd64__) || defined(__i386__) 192 #define EFI_STAGING_SLOP M(8) 193 #else 194 #define EFI_STAGING_SLOP 0 195 #endif 196 197 static u_long staging_slop = EFI_STAGING_SLOP; 198 199 EFI_PHYSICAL_ADDRESS staging, staging_end, staging_base; 200 bool stage_offset_set = false; 201 ssize_t stage_offset; 202 203 static void 204 efi_copy_free(void) 205 { 206 BS->FreePages(staging_base, (staging_end - staging_base) / 207 EFI_PAGE_SIZE); 208 stage_offset_set = false; 209 stage_offset = 0; 210 } 211 212 #if defined(__amd64__) || defined(__i386__) 213 int copy_staging = COPY_STAGING_AUTO; 214 215 static int 216 command_copy_staging(int argc, char *argv[]) 217 { 218 static const char *const mode[3] = { 219 [COPY_STAGING_ENABLE] = "enable", 220 [COPY_STAGING_DISABLE] = "disable", 221 [COPY_STAGING_AUTO] = "auto", 222 }; 223 int prev; 224 225 if (argc > 2) { 226 return (CMD_ERROR); 227 } else if (argc == 2) { 228 prev = copy_staging; 229 if (strcmp(argv[1], "enable") == 0) 230 copy_staging = COPY_STAGING_ENABLE; 231 else if (strcmp(argv[1], "disable") == 0) 232 copy_staging = COPY_STAGING_DISABLE; 233 else if (strcmp(argv[1], "auto") == 0) 234 copy_staging = COPY_STAGING_AUTO; 235 else { 236 printf("usage: copy_staging enable|disable|auto\n"); 237 return (CMD_ERROR); 238 } 239 if (prev != copy_staging) { 240 printf("changed copy_staging, unloading kernel\n"); 241 unload(); 242 efi_copy_free(); 243 efi_copy_init(); 244 } 245 } else { 246 printf("copy staging: %s\n", mode[copy_staging]); 247 } 248 return (CMD_OK); 249 } 250 COMMAND_SET(copy_staging, "copy_staging", "copy staging", command_copy_staging); 251 #endif 252 253 static int 254 command_staging_slop(int argc, char *argv[]) 255 { 256 char *endp; 257 u_long new, prev; 258 259 if (argc > 2) { 260 return (CMD_ERROR); 261 } else if (argc == 2) { 262 new = strtoul(argv[1], &endp, 0); 263 if (*endp != '\0') { 264 printf("invalid slop value\n"); 265 return (CMD_ERROR); 266 } 267 if (staging_slop != new) { 268 staging_slop = new; 269 printf("changed slop, unloading kernel\n"); 270 271 unload(); 272 efi_copy_free(); 273 efi_copy_init(); 274 } 275 } else { 276 printf("staging slop %#lx\n", staging_slop); 277 } 278 return (CMD_OK); 279 } 280 COMMAND_SET(staging_slop, "staging_slop", "set staging slop", 281 command_staging_slop); 282 283 #if defined(__amd64__) || defined(__i386__) 284 /* 285 * The staging area must reside in the first 1GB or 4GB physical 286 * memory: see elf64_exec() in 287 * boot/efi/loader/arch/amd64/elf64_freebsd.c. 288 */ 289 static EFI_PHYSICAL_ADDRESS 290 get_staging_max(void) 291 { 292 EFI_PHYSICAL_ADDRESS res; 293 294 res = copy_staging == COPY_STAGING_ENABLE ? G(1) : G(4); 295 return (res); 296 } 297 #define EFI_ALLOC_METHOD AllocateMaxAddress 298 #else 299 #define EFI_ALLOC_METHOD AllocateAnyPages 300 #endif 301 302 int 303 efi_copy_init(void) 304 { 305 EFI_STATUS status; 306 unsigned long nr_pages; 307 vm_offset_t ess; 308 309 ess = EFI_STAGING_SIZE; 310 if (ess < DEFAULT_EFI_STAGING_SIZE) 311 ess = DEFAULT_EFI_STAGING_SIZE; 312 nr_pages = EFI_SIZE_TO_PAGES(M(1) * ess); 313 314 #if defined(__amd64__) 315 /* 316 * We'll decrease nr_pages, if it's too big. Currently we only 317 * apply this to FreeBSD VM running on Hyper-V. Why? Please see 318 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=211746#c28 319 */ 320 if (running_on_hyperv()) 321 efi_verify_staging_size(&nr_pages); 322 #endif 323 #if defined(__amd64__) || defined(__i386__) 324 staging = get_staging_max(); 325 #endif 326 status = BS->AllocatePages(EFI_ALLOC_METHOD, EfiLoaderCode, 327 nr_pages, &staging); 328 if (EFI_ERROR(status)) { 329 printf("failed to allocate staging area: %lu\n", 330 EFI_ERROR_CODE(status)); 331 return (status); 332 } 333 staging_base = staging; 334 staging_end = staging + nr_pages * EFI_PAGE_SIZE; 335 336 #if EFI_STAGING_2M_ALIGN 337 /* 338 * Round the kernel load address to a 2MiB value. This is needed 339 * because the kernel builds a page table based on where it has 340 * been loaded in physical address space. As the kernel will use 341 * either a 1MiB or 2MiB page for this we need to make sure it 342 * is correctly aligned for both cases. 343 */ 344 staging = roundup2(staging, M(2)); 345 #endif 346 347 return (0); 348 } 349 350 static bool 351 efi_check_space(vm_offset_t end) 352 { 353 EFI_PHYSICAL_ADDRESS addr, new_base, new_staging; 354 EFI_STATUS status; 355 unsigned long nr_pages; 356 357 end = roundup2(end, EFI_PAGE_SIZE); 358 359 /* There is already enough space */ 360 if (end + staging_slop <= staging_end) 361 return (true); 362 363 if (!boot_services_active) { 364 if (end <= staging_end) 365 return (true); 366 panic("efi_check_space: cannot expand staging area " 367 "after boot services were exited\n"); 368 } 369 370 /* 371 * Add slop at the end: 372 * 1. amd64 kernel expects to do some very early allocations 373 * by carving out memory after kernend. Slop guarantees 374 * that it does not ovewrite anything useful. 375 * 2. It seems that initial calculation of the staging size 376 * could be somewhat smaller than actually copying in after 377 * boot services are exited. Slop avoids calling 378 * BS->AllocatePages() when it cannot work. 379 */ 380 end += staging_slop; 381 382 nr_pages = EFI_SIZE_TO_PAGES(end - staging_end); 383 #if defined(__amd64__) || defined(__i386__) 384 /* 385 * The amd64 kernel needs all memory to be allocated under the 1G or 386 * 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__) || defined(__i386__) 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