1 /* 2 * (C) Copyright 2008-2011 3 * Graeme Russ, <graeme.russ@gmail.com> 4 * 5 * (C) Copyright 2002 6 * Daniel Engström, Omicron Ceti AB, <daniel@omicron.se> 7 * 8 * (C) Copyright 2002 9 * Sysgo Real-Time Solutions, GmbH <www.elinos.com> 10 * Marius Groeger <mgroeger@sysgo.de> 11 * 12 * (C) Copyright 2002 13 * Sysgo Real-Time Solutions, GmbH <www.elinos.com> 14 * Alex Zuepke <azu@sysgo.de> 15 * 16 * Part of this file is adapted from coreboot 17 * src/arch/x86/lib/cpu.c 18 * 19 * SPDX-License-Identifier: GPL-2.0+ 20 */ 21 22 #include <common.h> 23 #include <command.h> 24 #include <errno.h> 25 #include <malloc.h> 26 #include <asm/control_regs.h> 27 #include <asm/cpu.h> 28 #include <asm/post.h> 29 #include <asm/processor.h> 30 #include <asm/processor-flags.h> 31 #include <asm/interrupt.h> 32 #include <linux/compiler.h> 33 34 DECLARE_GLOBAL_DATA_PTR; 35 36 /* 37 * Constructor for a conventional segment GDT (or LDT) entry 38 * This is a macro so it can be used in initialisers 39 */ 40 #define GDT_ENTRY(flags, base, limit) \ 41 ((((base) & 0xff000000ULL) << (56-24)) | \ 42 (((flags) & 0x0000f0ffULL) << 40) | \ 43 (((limit) & 0x000f0000ULL) << (48-16)) | \ 44 (((base) & 0x00ffffffULL) << 16) | \ 45 (((limit) & 0x0000ffffULL))) 46 47 struct gdt_ptr { 48 u16 len; 49 u32 ptr; 50 } __packed; 51 52 struct cpu_device_id { 53 unsigned vendor; 54 unsigned device; 55 }; 56 57 struct cpuinfo_x86 { 58 uint8_t x86; /* CPU family */ 59 uint8_t x86_vendor; /* CPU vendor */ 60 uint8_t x86_model; 61 uint8_t x86_mask; 62 }; 63 64 /* 65 * List of cpu vendor strings along with their normalized 66 * id values. 67 */ 68 static struct { 69 int vendor; 70 const char *name; 71 } x86_vendors[] = { 72 { X86_VENDOR_INTEL, "GenuineIntel", }, 73 { X86_VENDOR_CYRIX, "CyrixInstead", }, 74 { X86_VENDOR_AMD, "AuthenticAMD", }, 75 { X86_VENDOR_UMC, "UMC UMC UMC ", }, 76 { X86_VENDOR_NEXGEN, "NexGenDriven", }, 77 { X86_VENDOR_CENTAUR, "CentaurHauls", }, 78 { X86_VENDOR_RISE, "RiseRiseRise", }, 79 { X86_VENDOR_TRANSMETA, "GenuineTMx86", }, 80 { X86_VENDOR_TRANSMETA, "TransmetaCPU", }, 81 { X86_VENDOR_NSC, "Geode by NSC", }, 82 { X86_VENDOR_SIS, "SiS SiS SiS ", }, 83 }; 84 85 static const char *const x86_vendor_name[] = { 86 [X86_VENDOR_INTEL] = "Intel", 87 [X86_VENDOR_CYRIX] = "Cyrix", 88 [X86_VENDOR_AMD] = "AMD", 89 [X86_VENDOR_UMC] = "UMC", 90 [X86_VENDOR_NEXGEN] = "NexGen", 91 [X86_VENDOR_CENTAUR] = "Centaur", 92 [X86_VENDOR_RISE] = "Rise", 93 [X86_VENDOR_TRANSMETA] = "Transmeta", 94 [X86_VENDOR_NSC] = "NSC", 95 [X86_VENDOR_SIS] = "SiS", 96 }; 97 98 static void load_ds(u32 segment) 99 { 100 asm volatile("movl %0, %%ds" : : "r" (segment * X86_GDT_ENTRY_SIZE)); 101 } 102 103 static void load_es(u32 segment) 104 { 105 asm volatile("movl %0, %%es" : : "r" (segment * X86_GDT_ENTRY_SIZE)); 106 } 107 108 static void load_fs(u32 segment) 109 { 110 asm volatile("movl %0, %%fs" : : "r" (segment * X86_GDT_ENTRY_SIZE)); 111 } 112 113 static void load_gs(u32 segment) 114 { 115 asm volatile("movl %0, %%gs" : : "r" (segment * X86_GDT_ENTRY_SIZE)); 116 } 117 118 static void load_ss(u32 segment) 119 { 120 asm volatile("movl %0, %%ss" : : "r" (segment * X86_GDT_ENTRY_SIZE)); 121 } 122 123 static void load_gdt(const u64 *boot_gdt, u16 num_entries) 124 { 125 struct gdt_ptr gdt; 126 127 gdt.len = (num_entries * 8) - 1; 128 gdt.ptr = (u32)boot_gdt; 129 130 asm volatile("lgdtl %0\n" : : "m" (gdt)); 131 } 132 133 void setup_gdt(gd_t *id, u64 *gdt_addr) 134 { 135 /* CS: code, read/execute, 4 GB, base 0 */ 136 gdt_addr[X86_GDT_ENTRY_32BIT_CS] = GDT_ENTRY(0xc09b, 0, 0xfffff); 137 138 /* DS: data, read/write, 4 GB, base 0 */ 139 gdt_addr[X86_GDT_ENTRY_32BIT_DS] = GDT_ENTRY(0xc093, 0, 0xfffff); 140 141 /* FS: data, read/write, 4 GB, base (Global Data Pointer) */ 142 id->arch.gd_addr = id; 143 gdt_addr[X86_GDT_ENTRY_32BIT_FS] = GDT_ENTRY(0xc093, 144 (ulong)&id->arch.gd_addr, 0xfffff); 145 146 /* 16-bit CS: code, read/execute, 64 kB, base 0 */ 147 gdt_addr[X86_GDT_ENTRY_16BIT_CS] = GDT_ENTRY(0x109b, 0, 0x0ffff); 148 149 /* 16-bit DS: data, read/write, 64 kB, base 0 */ 150 gdt_addr[X86_GDT_ENTRY_16BIT_DS] = GDT_ENTRY(0x1093, 0, 0x0ffff); 151 152 load_gdt(gdt_addr, X86_GDT_NUM_ENTRIES); 153 load_ds(X86_GDT_ENTRY_32BIT_DS); 154 load_es(X86_GDT_ENTRY_32BIT_DS); 155 load_gs(X86_GDT_ENTRY_32BIT_DS); 156 load_ss(X86_GDT_ENTRY_32BIT_DS); 157 load_fs(X86_GDT_ENTRY_32BIT_FS); 158 } 159 160 int __weak x86_cleanup_before_linux(void) 161 { 162 #ifdef CONFIG_BOOTSTAGE_STASH 163 bootstage_stash((void *)CONFIG_BOOTSTAGE_STASH, 164 CONFIG_BOOTSTAGE_STASH_SIZE); 165 #endif 166 167 return 0; 168 } 169 170 /* 171 * Cyrix CPUs without cpuid or with cpuid not yet enabled can be detected 172 * by the fact that they preserve the flags across the division of 5/2. 173 * PII and PPro exhibit this behavior too, but they have cpuid available. 174 */ 175 176 /* 177 * Perform the Cyrix 5/2 test. A Cyrix won't change 178 * the flags, while other 486 chips will. 179 */ 180 static inline int test_cyrix_52div(void) 181 { 182 unsigned int test; 183 184 __asm__ __volatile__( 185 "sahf\n\t" /* clear flags (%eax = 0x0005) */ 186 "div %b2\n\t" /* divide 5 by 2 */ 187 "lahf" /* store flags into %ah */ 188 : "=a" (test) 189 : "0" (5), "q" (2) 190 : "cc"); 191 192 /* AH is 0x02 on Cyrix after the divide.. */ 193 return (unsigned char) (test >> 8) == 0x02; 194 } 195 196 /* 197 * Detect a NexGen CPU running without BIOS hypercode new enough 198 * to have CPUID. (Thanks to Herbert Oppmann) 199 */ 200 201 static int deep_magic_nexgen_probe(void) 202 { 203 int ret; 204 205 __asm__ __volatile__ ( 206 " movw $0x5555, %%ax\n" 207 " xorw %%dx,%%dx\n" 208 " movw $2, %%cx\n" 209 " divw %%cx\n" 210 " movl $0, %%eax\n" 211 " jnz 1f\n" 212 " movl $1, %%eax\n" 213 "1:\n" 214 : "=a" (ret) : : "cx", "dx"); 215 return ret; 216 } 217 218 static bool has_cpuid(void) 219 { 220 return flag_is_changeable_p(X86_EFLAGS_ID); 221 } 222 223 static int build_vendor_name(char *vendor_name) 224 { 225 struct cpuid_result result; 226 result = cpuid(0x00000000); 227 unsigned int *name_as_ints = (unsigned int *)vendor_name; 228 229 name_as_ints[0] = result.ebx; 230 name_as_ints[1] = result.edx; 231 name_as_ints[2] = result.ecx; 232 233 return result.eax; 234 } 235 236 static void identify_cpu(struct cpu_device_id *cpu) 237 { 238 char vendor_name[16]; 239 int i; 240 241 vendor_name[0] = '\0'; /* Unset */ 242 cpu->device = 0; /* fix gcc 4.4.4 warning */ 243 244 /* Find the id and vendor_name */ 245 if (!has_cpuid()) { 246 /* Its a 486 if we can modify the AC flag */ 247 if (flag_is_changeable_p(X86_EFLAGS_AC)) 248 cpu->device = 0x00000400; /* 486 */ 249 else 250 cpu->device = 0x00000300; /* 386 */ 251 if ((cpu->device == 0x00000400) && test_cyrix_52div()) { 252 memcpy(vendor_name, "CyrixInstead", 13); 253 /* If we ever care we can enable cpuid here */ 254 } 255 /* Detect NexGen with old hypercode */ 256 else if (deep_magic_nexgen_probe()) 257 memcpy(vendor_name, "NexGenDriven", 13); 258 } 259 if (has_cpuid()) { 260 int cpuid_level; 261 262 cpuid_level = build_vendor_name(vendor_name); 263 vendor_name[12] = '\0'; 264 265 /* Intel-defined flags: level 0x00000001 */ 266 if (cpuid_level >= 0x00000001) { 267 cpu->device = cpuid_eax(0x00000001); 268 } else { 269 /* Have CPUID level 0 only unheard of */ 270 cpu->device = 0x00000400; 271 } 272 } 273 cpu->vendor = X86_VENDOR_UNKNOWN; 274 for (i = 0; i < ARRAY_SIZE(x86_vendors); i++) { 275 if (memcmp(vendor_name, x86_vendors[i].name, 12) == 0) { 276 cpu->vendor = x86_vendors[i].vendor; 277 break; 278 } 279 } 280 } 281 282 static inline void get_fms(struct cpuinfo_x86 *c, uint32_t tfms) 283 { 284 c->x86 = (tfms >> 8) & 0xf; 285 c->x86_model = (tfms >> 4) & 0xf; 286 c->x86_mask = tfms & 0xf; 287 if (c->x86 == 0xf) 288 c->x86 += (tfms >> 20) & 0xff; 289 if (c->x86 >= 0x6) 290 c->x86_model += ((tfms >> 16) & 0xF) << 4; 291 } 292 293 int x86_cpu_init_f(void) 294 { 295 const u32 em_rst = ~X86_CR0_EM; 296 const u32 mp_ne_set = X86_CR0_MP | X86_CR0_NE; 297 298 /* initialize FPU, reset EM, set MP and NE */ 299 asm ("fninit\n" \ 300 "movl %%cr0, %%eax\n" \ 301 "andl %0, %%eax\n" \ 302 "orl %1, %%eax\n" \ 303 "movl %%eax, %%cr0\n" \ 304 : : "i" (em_rst), "i" (mp_ne_set) : "eax"); 305 306 /* identify CPU via cpuid and store the decoded info into gd->arch */ 307 if (has_cpuid()) { 308 struct cpu_device_id cpu; 309 struct cpuinfo_x86 c; 310 311 identify_cpu(&cpu); 312 get_fms(&c, cpu.device); 313 gd->arch.x86 = c.x86; 314 gd->arch.x86_vendor = cpu.vendor; 315 gd->arch.x86_model = c.x86_model; 316 gd->arch.x86_mask = c.x86_mask; 317 gd->arch.x86_device = cpu.device; 318 } 319 320 return 0; 321 } 322 323 int x86_cpu_init_r(void) 324 { 325 /* Initialize core interrupt and exception functionality of CPU */ 326 cpu_init_interrupts(); 327 return 0; 328 } 329 int cpu_init_r(void) __attribute__((weak, alias("x86_cpu_init_r"))); 330 331 void x86_enable_caches(void) 332 { 333 unsigned long cr0; 334 335 cr0 = read_cr0(); 336 cr0 &= ~(X86_CR0_NW | X86_CR0_CD); 337 write_cr0(cr0); 338 wbinvd(); 339 } 340 void enable_caches(void) __attribute__((weak, alias("x86_enable_caches"))); 341 342 void x86_disable_caches(void) 343 { 344 unsigned long cr0; 345 346 cr0 = read_cr0(); 347 cr0 |= X86_CR0_NW | X86_CR0_CD; 348 wbinvd(); 349 write_cr0(cr0); 350 wbinvd(); 351 } 352 void disable_caches(void) __attribute__((weak, alias("x86_disable_caches"))); 353 354 int x86_init_cache(void) 355 { 356 enable_caches(); 357 358 return 0; 359 } 360 int init_cache(void) __attribute__((weak, alias("x86_init_cache"))); 361 362 int do_reset(cmd_tbl_t *cmdtp, int flag, int argc, char * const argv[]) 363 { 364 printf("resetting ...\n"); 365 366 /* wait 50 ms */ 367 udelay(50000); 368 disable_interrupts(); 369 reset_cpu(0); 370 371 /*NOTREACHED*/ 372 return 0; 373 } 374 375 void flush_cache(unsigned long dummy1, unsigned long dummy2) 376 { 377 asm("wbinvd\n"); 378 } 379 380 void __attribute__ ((regparm(0))) generate_gpf(void); 381 382 /* segment 0x70 is an arbitrary segment which does not exist */ 383 asm(".globl generate_gpf\n" 384 ".hidden generate_gpf\n" 385 ".type generate_gpf, @function\n" 386 "generate_gpf:\n" 387 "ljmp $0x70, $0x47114711\n"); 388 389 __weak void reset_cpu(ulong addr) 390 { 391 printf("Resetting using x86 Triple Fault\n"); 392 set_vector(13, generate_gpf); /* general protection fault handler */ 393 set_vector(8, generate_gpf); /* double fault handler */ 394 generate_gpf(); /* start the show */ 395 } 396 397 int dcache_status(void) 398 { 399 return !(read_cr0() & 0x40000000); 400 } 401 402 /* Define these functions to allow ehch-hcd to function */ 403 void flush_dcache_range(unsigned long start, unsigned long stop) 404 { 405 } 406 407 void invalidate_dcache_range(unsigned long start, unsigned long stop) 408 { 409 } 410 411 void dcache_enable(void) 412 { 413 enable_caches(); 414 } 415 416 void dcache_disable(void) 417 { 418 disable_caches(); 419 } 420 421 void icache_enable(void) 422 { 423 } 424 425 void icache_disable(void) 426 { 427 } 428 429 int icache_status(void) 430 { 431 return 1; 432 } 433 434 void cpu_enable_paging_pae(ulong cr3) 435 { 436 __asm__ __volatile__( 437 /* Load the page table address */ 438 "movl %0, %%cr3\n" 439 /* Enable pae */ 440 "movl %%cr4, %%eax\n" 441 "orl $0x00000020, %%eax\n" 442 "movl %%eax, %%cr4\n" 443 /* Enable paging */ 444 "movl %%cr0, %%eax\n" 445 "orl $0x80000000, %%eax\n" 446 "movl %%eax, %%cr0\n" 447 : 448 : "r" (cr3) 449 : "eax"); 450 } 451 452 void cpu_disable_paging_pae(void) 453 { 454 /* Turn off paging */ 455 __asm__ __volatile__ ( 456 /* Disable paging */ 457 "movl %%cr0, %%eax\n" 458 "andl $0x7fffffff, %%eax\n" 459 "movl %%eax, %%cr0\n" 460 /* Disable pae */ 461 "movl %%cr4, %%eax\n" 462 "andl $0xffffffdf, %%eax\n" 463 "movl %%eax, %%cr4\n" 464 : 465 : 466 : "eax"); 467 } 468 469 static bool can_detect_long_mode(void) 470 { 471 return cpuid_eax(0x80000000) > 0x80000000UL; 472 } 473 474 static bool has_long_mode(void) 475 { 476 return cpuid_edx(0x80000001) & (1 << 29) ? true : false; 477 } 478 479 int cpu_has_64bit(void) 480 { 481 return has_cpuid() && can_detect_long_mode() && 482 has_long_mode(); 483 } 484 485 const char *cpu_vendor_name(int vendor) 486 { 487 const char *name; 488 name = "<invalid cpu vendor>"; 489 if ((vendor < (ARRAY_SIZE(x86_vendor_name))) && 490 (x86_vendor_name[vendor] != 0)) 491 name = x86_vendor_name[vendor]; 492 493 return name; 494 } 495 496 char *cpu_get_name(char *name) 497 { 498 unsigned int *name_as_ints = (unsigned int *)name; 499 struct cpuid_result regs; 500 char *ptr; 501 int i; 502 503 /* This bit adds up to 48 bytes */ 504 for (i = 0; i < 3; i++) { 505 regs = cpuid(0x80000002 + i); 506 name_as_ints[i * 4 + 0] = regs.eax; 507 name_as_ints[i * 4 + 1] = regs.ebx; 508 name_as_ints[i * 4 + 2] = regs.ecx; 509 name_as_ints[i * 4 + 3] = regs.edx; 510 } 511 name[CPU_MAX_NAME_LEN - 1] = '\0'; 512 513 /* Skip leading spaces. */ 514 ptr = name; 515 while (*ptr == ' ') 516 ptr++; 517 518 return ptr; 519 } 520 521 int default_print_cpuinfo(void) 522 { 523 printf("CPU: %s, vendor %s, device %xh\n", 524 cpu_has_64bit() ? "x86_64" : "x86", 525 cpu_vendor_name(gd->arch.x86_vendor), gd->arch.x86_device); 526 527 return 0; 528 } 529 530 #define PAGETABLE_SIZE (6 * 4096) 531 532 /** 533 * build_pagetable() - build a flat 4GiB page table structure for 64-bti mode 534 * 535 * @pgtable: Pointer to a 24iKB block of memory 536 */ 537 static void build_pagetable(uint32_t *pgtable) 538 { 539 uint i; 540 541 memset(pgtable, '\0', PAGETABLE_SIZE); 542 543 /* Level 4 needs a single entry */ 544 pgtable[0] = (uint32_t)&pgtable[1024] + 7; 545 546 /* Level 3 has one 64-bit entry for each GiB of memory */ 547 for (i = 0; i < 4; i++) { 548 pgtable[1024 + i * 2] = (uint32_t)&pgtable[2048] + 549 0x1000 * i + 7; 550 } 551 552 /* Level 2 has 2048 64-bit entries, each repesenting 2MiB */ 553 for (i = 0; i < 2048; i++) 554 pgtable[2048 + i * 2] = 0x183 + (i << 21UL); 555 } 556 557 int cpu_jump_to_64bit(ulong setup_base, ulong target) 558 { 559 uint32_t *pgtable; 560 561 pgtable = memalign(4096, PAGETABLE_SIZE); 562 if (!pgtable) 563 return -ENOMEM; 564 565 build_pagetable(pgtable); 566 cpu_call64((ulong)pgtable, setup_base, target); 567 free(pgtable); 568 569 return -EFAULT; 570 } 571 572 void show_boot_progress(int val) 573 { 574 #if MIN_PORT80_KCLOCKS_DELAY 575 /* 576 * Scale the time counter reading to avoid using 64 bit arithmetics. 577 * Can't use get_timer() here becuase it could be not yet 578 * initialized or even implemented. 579 */ 580 if (!gd->arch.tsc_prev) { 581 gd->arch.tsc_base_kclocks = rdtsc() / 1000; 582 gd->arch.tsc_prev = 0; 583 } else { 584 uint32_t now; 585 586 do { 587 now = rdtsc() / 1000 - gd->arch.tsc_base_kclocks; 588 } while (now < (gd->arch.tsc_prev + MIN_PORT80_KCLOCKS_DELAY)); 589 gd->arch.tsc_prev = now; 590 } 591 #endif 592 outb(val, POST_PORT); 593 } 594