1 #include <linux/kernel.h> 2 3 #include <linux/string.h> 4 #include <linux/bitops.h> 5 #include <linux/smp.h> 6 #include <linux/sched.h> 7 #include <linux/thread_info.h> 8 #include <linux/module.h> 9 #include <linux/uaccess.h> 10 11 #include <asm/processor.h> 12 #include <asm/pgtable.h> 13 #include <asm/msr.h> 14 #include <asm/bugs.h> 15 #include <asm/cpu.h> 16 17 #ifdef CONFIG_X86_64 18 #include <linux/topology.h> 19 #endif 20 21 #include "cpu.h" 22 23 #ifdef CONFIG_X86_LOCAL_APIC 24 #include <asm/mpspec.h> 25 #include <asm/apic.h> 26 #endif 27 28 static void early_init_intel(struct cpuinfo_x86 *c) 29 { 30 u64 misc_enable; 31 32 /* Unmask CPUID levels if masked: */ 33 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) { 34 if (msr_clear_bit(MSR_IA32_MISC_ENABLE, 35 MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) { 36 c->cpuid_level = cpuid_eax(0); 37 get_cpu_cap(c); 38 } 39 } 40 41 if ((c->x86 == 0xf && c->x86_model >= 0x03) || 42 (c->x86 == 0x6 && c->x86_model >= 0x0e)) 43 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 44 45 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) { 46 unsigned lower_word; 47 48 wrmsr(MSR_IA32_UCODE_REV, 0, 0); 49 /* Required by the SDM */ 50 sync_core(); 51 rdmsr(MSR_IA32_UCODE_REV, lower_word, c->microcode); 52 } 53 54 /* 55 * Atom erratum AAE44/AAF40/AAG38/AAH41: 56 * 57 * A race condition between speculative fetches and invalidating 58 * a large page. This is worked around in microcode, but we 59 * need the microcode to have already been loaded... so if it is 60 * not, recommend a BIOS update and disable large pages. 61 */ 62 if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 && 63 c->microcode < 0x20e) { 64 printk(KERN_WARNING "Atom PSE erratum detected, BIOS microcode update recommended\n"); 65 clear_cpu_cap(c, X86_FEATURE_PSE); 66 } 67 68 #ifdef CONFIG_X86_64 69 set_cpu_cap(c, X86_FEATURE_SYSENTER32); 70 #else 71 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */ 72 if (c->x86 == 15 && c->x86_cache_alignment == 64) 73 c->x86_cache_alignment = 128; 74 #endif 75 76 /* CPUID workaround for 0F33/0F34 CPU */ 77 if (c->x86 == 0xF && c->x86_model == 0x3 78 && (c->x86_mask == 0x3 || c->x86_mask == 0x4)) 79 c->x86_phys_bits = 36; 80 81 /* 82 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate 83 * with P/T states and does not stop in deep C-states. 84 * 85 * It is also reliable across cores and sockets. (but not across 86 * cabinets - we turn it off in that case explicitly.) 87 */ 88 if (c->x86_power & (1 << 8)) { 89 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 90 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC); 91 if (!check_tsc_unstable()) 92 set_sched_clock_stable(); 93 } 94 95 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */ 96 if (c->x86 == 6) { 97 switch (c->x86_model) { 98 case 0x27: /* Penwell */ 99 case 0x35: /* Cloverview */ 100 case 0x4a: /* Merrifield */ 101 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3); 102 break; 103 default: 104 break; 105 } 106 } 107 108 /* 109 * There is a known erratum on Pentium III and Core Solo 110 * and Core Duo CPUs. 111 * " Page with PAT set to WC while associated MTRR is UC 112 * may consolidate to UC " 113 * Because of this erratum, it is better to stick with 114 * setting WC in MTRR rather than using PAT on these CPUs. 115 * 116 * Enable PAT WC only on P4, Core 2 or later CPUs. 117 */ 118 if (c->x86 == 6 && c->x86_model < 15) 119 clear_cpu_cap(c, X86_FEATURE_PAT); 120 121 #ifdef CONFIG_KMEMCHECK 122 /* 123 * P4s have a "fast strings" feature which causes single- 124 * stepping REP instructions to only generate a #DB on 125 * cache-line boundaries. 126 * 127 * Ingo Molnar reported a Pentium D (model 6) and a Xeon 128 * (model 2) with the same problem. 129 */ 130 if (c->x86 == 15) 131 if (msr_clear_bit(MSR_IA32_MISC_ENABLE, 132 MSR_IA32_MISC_ENABLE_FAST_STRING_BIT) > 0) 133 pr_info("kmemcheck: Disabling fast string operations\n"); 134 #endif 135 136 /* 137 * If fast string is not enabled in IA32_MISC_ENABLE for any reason, 138 * clear the fast string and enhanced fast string CPU capabilities. 139 */ 140 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) { 141 rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable); 142 if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) { 143 printk(KERN_INFO "Disabled fast string operations\n"); 144 setup_clear_cpu_cap(X86_FEATURE_REP_GOOD); 145 setup_clear_cpu_cap(X86_FEATURE_ERMS); 146 } 147 } 148 149 /* 150 * Intel Quark Core DevMan_001.pdf section 6.4.11 151 * "The operating system also is required to invalidate (i.e., flush) 152 * the TLB when any changes are made to any of the page table entries. 153 * The operating system must reload CR3 to cause the TLB to be flushed" 154 * 155 * As a result cpu_has_pge() in arch/x86/include/asm/tlbflush.h should 156 * be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE 157 * to be modified 158 */ 159 if (c->x86 == 5 && c->x86_model == 9) { 160 pr_info("Disabling PGE capability bit\n"); 161 setup_clear_cpu_cap(X86_FEATURE_PGE); 162 } 163 } 164 165 #ifdef CONFIG_X86_32 166 /* 167 * Early probe support logic for ppro memory erratum #50 168 * 169 * This is called before we do cpu ident work 170 */ 171 172 int ppro_with_ram_bug(void) 173 { 174 /* Uses data from early_cpu_detect now */ 175 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && 176 boot_cpu_data.x86 == 6 && 177 boot_cpu_data.x86_model == 1 && 178 boot_cpu_data.x86_mask < 8) { 179 printk(KERN_INFO "Pentium Pro with Errata#50 detected. Taking evasive action.\n"); 180 return 1; 181 } 182 return 0; 183 } 184 185 static void intel_smp_check(struct cpuinfo_x86 *c) 186 { 187 /* calling is from identify_secondary_cpu() ? */ 188 if (!c->cpu_index) 189 return; 190 191 /* 192 * Mask B, Pentium, but not Pentium MMX 193 */ 194 if (c->x86 == 5 && 195 c->x86_mask >= 1 && c->x86_mask <= 4 && 196 c->x86_model <= 3) { 197 /* 198 * Remember we have B step Pentia with bugs 199 */ 200 WARN_ONCE(1, "WARNING: SMP operation may be unreliable" 201 "with B stepping processors.\n"); 202 } 203 } 204 205 static int forcepae; 206 static int __init forcepae_setup(char *__unused) 207 { 208 forcepae = 1; 209 return 1; 210 } 211 __setup("forcepae", forcepae_setup); 212 213 static void intel_workarounds(struct cpuinfo_x86 *c) 214 { 215 #ifdef CONFIG_X86_F00F_BUG 216 /* 217 * All models of Pentium and Pentium with MMX technology CPUs 218 * have the F0 0F bug, which lets nonprivileged users lock up the 219 * system. Announce that the fault handler will be checking for it. 220 * The Quark is also family 5, but does not have the same bug. 221 */ 222 clear_cpu_bug(c, X86_BUG_F00F); 223 if (!paravirt_enabled() && c->x86 == 5 && c->x86_model < 9) { 224 static int f00f_workaround_enabled; 225 226 set_cpu_bug(c, X86_BUG_F00F); 227 if (!f00f_workaround_enabled) { 228 printk(KERN_NOTICE "Intel Pentium with F0 0F bug - workaround enabled.\n"); 229 f00f_workaround_enabled = 1; 230 } 231 } 232 #endif 233 234 /* 235 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until 236 * model 3 mask 3 237 */ 238 if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633) 239 clear_cpu_cap(c, X86_FEATURE_SEP); 240 241 /* 242 * PAE CPUID issue: many Pentium M report no PAE but may have a 243 * functionally usable PAE implementation. 244 * Forcefully enable PAE if kernel parameter "forcepae" is present. 245 */ 246 if (forcepae) { 247 printk(KERN_WARNING "PAE forced!\n"); 248 set_cpu_cap(c, X86_FEATURE_PAE); 249 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE); 250 } 251 252 /* 253 * P4 Xeon errata 037 workaround. 254 * Hardware prefetcher may cause stale data to be loaded into the cache. 255 */ 256 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) { 257 if (msr_set_bit(MSR_IA32_MISC_ENABLE, 258 MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) 259 > 0) { 260 pr_info("CPU: C0 stepping P4 Xeon detected.\n"); 261 pr_info("CPU: Disabling hardware prefetching (Errata 037)\n"); 262 } 263 } 264 265 /* 266 * See if we have a good local APIC by checking for buggy Pentia, 267 * i.e. all B steppings and the C2 stepping of P54C when using their 268 * integrated APIC (see 11AP erratum in "Pentium Processor 269 * Specification Update"). 270 */ 271 if (cpu_has_apic && (c->x86<<8 | c->x86_model<<4) == 0x520 && 272 (c->x86_mask < 0x6 || c->x86_mask == 0xb)) 273 set_cpu_bug(c, X86_BUG_11AP); 274 275 276 #ifdef CONFIG_X86_INTEL_USERCOPY 277 /* 278 * Set up the preferred alignment for movsl bulk memory moves 279 */ 280 switch (c->x86) { 281 case 4: /* 486: untested */ 282 break; 283 case 5: /* Old Pentia: untested */ 284 break; 285 case 6: /* PII/PIII only like movsl with 8-byte alignment */ 286 movsl_mask.mask = 7; 287 break; 288 case 15: /* P4 is OK down to 8-byte alignment */ 289 movsl_mask.mask = 7; 290 break; 291 } 292 #endif 293 294 intel_smp_check(c); 295 } 296 #else 297 static void intel_workarounds(struct cpuinfo_x86 *c) 298 { 299 } 300 #endif 301 302 static void srat_detect_node(struct cpuinfo_x86 *c) 303 { 304 #ifdef CONFIG_NUMA 305 unsigned node; 306 int cpu = smp_processor_id(); 307 308 /* Don't do the funky fallback heuristics the AMD version employs 309 for now. */ 310 node = numa_cpu_node(cpu); 311 if (node == NUMA_NO_NODE || !node_online(node)) { 312 /* reuse the value from init_cpu_to_node() */ 313 node = cpu_to_node(cpu); 314 } 315 numa_set_node(cpu, node); 316 #endif 317 } 318 319 /* 320 * find out the number of processor cores on the die 321 */ 322 static int intel_num_cpu_cores(struct cpuinfo_x86 *c) 323 { 324 unsigned int eax, ebx, ecx, edx; 325 326 if (c->cpuid_level < 4) 327 return 1; 328 329 /* Intel has a non-standard dependency on %ecx for this CPUID level. */ 330 cpuid_count(4, 0, &eax, &ebx, &ecx, &edx); 331 if (eax & 0x1f) 332 return (eax >> 26) + 1; 333 else 334 return 1; 335 } 336 337 static void detect_vmx_virtcap(struct cpuinfo_x86 *c) 338 { 339 /* Intel VMX MSR indicated features */ 340 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW 0x00200000 341 #define X86_VMX_FEATURE_PROC_CTLS_VNMI 0x00400000 342 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS 0x80000000 343 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC 0x00000001 344 #define X86_VMX_FEATURE_PROC_CTLS2_EPT 0x00000002 345 #define X86_VMX_FEATURE_PROC_CTLS2_VPID 0x00000020 346 347 u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2; 348 349 clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW); 350 clear_cpu_cap(c, X86_FEATURE_VNMI); 351 clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY); 352 clear_cpu_cap(c, X86_FEATURE_EPT); 353 clear_cpu_cap(c, X86_FEATURE_VPID); 354 355 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high); 356 msr_ctl = vmx_msr_high | vmx_msr_low; 357 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW) 358 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW); 359 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI) 360 set_cpu_cap(c, X86_FEATURE_VNMI); 361 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) { 362 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2, 363 vmx_msr_low, vmx_msr_high); 364 msr_ctl2 = vmx_msr_high | vmx_msr_low; 365 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) && 366 (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)) 367 set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY); 368 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT) 369 set_cpu_cap(c, X86_FEATURE_EPT); 370 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID) 371 set_cpu_cap(c, X86_FEATURE_VPID); 372 } 373 } 374 375 static void init_intel_energy_perf(struct cpuinfo_x86 *c) 376 { 377 u64 epb; 378 379 /* 380 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized. 381 * (x86_energy_perf_policy(8) is available to change it at run-time.) 382 */ 383 if (!cpu_has(c, X86_FEATURE_EPB)) 384 return; 385 386 rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb); 387 if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE) 388 return; 389 390 pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n"); 391 pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n"); 392 epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL; 393 wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb); 394 } 395 396 static void intel_bsp_resume(struct cpuinfo_x86 *c) 397 { 398 /* 399 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume, 400 * so reinitialize it properly like during bootup: 401 */ 402 init_intel_energy_perf(c); 403 } 404 405 static void init_intel(struct cpuinfo_x86 *c) 406 { 407 unsigned int l2 = 0; 408 409 early_init_intel(c); 410 411 intel_workarounds(c); 412 413 /* 414 * Detect the extended topology information if available. This 415 * will reinitialise the initial_apicid which will be used 416 * in init_intel_cacheinfo() 417 */ 418 detect_extended_topology(c); 419 420 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) { 421 /* 422 * let's use the legacy cpuid vector 0x1 and 0x4 for topology 423 * detection. 424 */ 425 c->x86_max_cores = intel_num_cpu_cores(c); 426 #ifdef CONFIG_X86_32 427 detect_ht(c); 428 #endif 429 } 430 431 l2 = init_intel_cacheinfo(c); 432 433 /* Detect legacy cache sizes if init_intel_cacheinfo did not */ 434 if (l2 == 0) { 435 cpu_detect_cache_sizes(c); 436 l2 = c->x86_cache_size; 437 } 438 439 if (c->cpuid_level > 9) { 440 unsigned eax = cpuid_eax(10); 441 /* Check for version and the number of counters */ 442 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1)) 443 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON); 444 } 445 446 if (cpu_has_xmm2) 447 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC); 448 if (cpu_has_ds) { 449 unsigned int l1; 450 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2); 451 if (!(l1 & (1<<11))) 452 set_cpu_cap(c, X86_FEATURE_BTS); 453 if (!(l1 & (1<<12))) 454 set_cpu_cap(c, X86_FEATURE_PEBS); 455 } 456 457 if (c->x86 == 6 && cpu_has_clflush && 458 (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47)) 459 set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR); 460 461 #ifdef CONFIG_X86_64 462 if (c->x86 == 15) 463 c->x86_cache_alignment = c->x86_clflush_size * 2; 464 if (c->x86 == 6) 465 set_cpu_cap(c, X86_FEATURE_REP_GOOD); 466 #else 467 /* 468 * Names for the Pentium II/Celeron processors 469 * detectable only by also checking the cache size. 470 * Dixon is NOT a Celeron. 471 */ 472 if (c->x86 == 6) { 473 char *p = NULL; 474 475 switch (c->x86_model) { 476 case 5: 477 if (l2 == 0) 478 p = "Celeron (Covington)"; 479 else if (l2 == 256) 480 p = "Mobile Pentium II (Dixon)"; 481 break; 482 483 case 6: 484 if (l2 == 128) 485 p = "Celeron (Mendocino)"; 486 else if (c->x86_mask == 0 || c->x86_mask == 5) 487 p = "Celeron-A"; 488 break; 489 490 case 8: 491 if (l2 == 128) 492 p = "Celeron (Coppermine)"; 493 break; 494 } 495 496 if (p) 497 strcpy(c->x86_model_id, p); 498 } 499 500 if (c->x86 == 15) 501 set_cpu_cap(c, X86_FEATURE_P4); 502 if (c->x86 == 6) 503 set_cpu_cap(c, X86_FEATURE_P3); 504 #endif 505 506 /* Work around errata */ 507 srat_detect_node(c); 508 509 if (cpu_has(c, X86_FEATURE_VMX)) 510 detect_vmx_virtcap(c); 511 512 init_intel_energy_perf(c); 513 } 514 515 #ifdef CONFIG_X86_32 516 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size) 517 { 518 /* 519 * Intel PIII Tualatin. This comes in two flavours. 520 * One has 256kb of cache, the other 512. We have no way 521 * to determine which, so we use a boottime override 522 * for the 512kb model, and assume 256 otherwise. 523 */ 524 if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0)) 525 size = 256; 526 527 /* 528 * Intel Quark SoC X1000 contains a 4-way set associative 529 * 16K cache with a 16 byte cache line and 256 lines per tag 530 */ 531 if ((c->x86 == 5) && (c->x86_model == 9)) 532 size = 16; 533 return size; 534 } 535 #endif 536 537 #define TLB_INST_4K 0x01 538 #define TLB_INST_4M 0x02 539 #define TLB_INST_2M_4M 0x03 540 541 #define TLB_INST_ALL 0x05 542 #define TLB_INST_1G 0x06 543 544 #define TLB_DATA_4K 0x11 545 #define TLB_DATA_4M 0x12 546 #define TLB_DATA_2M_4M 0x13 547 #define TLB_DATA_4K_4M 0x14 548 549 #define TLB_DATA_1G 0x16 550 551 #define TLB_DATA0_4K 0x21 552 #define TLB_DATA0_4M 0x22 553 #define TLB_DATA0_2M_4M 0x23 554 555 #define STLB_4K 0x41 556 #define STLB_4K_2M 0x42 557 558 static const struct _tlb_table intel_tlb_table[] = { 559 { 0x01, TLB_INST_4K, 32, " TLB_INST 4 KByte pages, 4-way set associative" }, 560 { 0x02, TLB_INST_4M, 2, " TLB_INST 4 MByte pages, full associative" }, 561 { 0x03, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way set associative" }, 562 { 0x04, TLB_DATA_4M, 8, " TLB_DATA 4 MByte pages, 4-way set associative" }, 563 { 0x05, TLB_DATA_4M, 32, " TLB_DATA 4 MByte pages, 4-way set associative" }, 564 { 0x0b, TLB_INST_4M, 4, " TLB_INST 4 MByte pages, 4-way set associative" }, 565 { 0x4f, TLB_INST_4K, 32, " TLB_INST 4 KByte pages */" }, 566 { 0x50, TLB_INST_ALL, 64, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" }, 567 { 0x51, TLB_INST_ALL, 128, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" }, 568 { 0x52, TLB_INST_ALL, 256, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" }, 569 { 0x55, TLB_INST_2M_4M, 7, " TLB_INST 2-MByte or 4-MByte pages, fully associative" }, 570 { 0x56, TLB_DATA0_4M, 16, " TLB_DATA0 4 MByte pages, 4-way set associative" }, 571 { 0x57, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, 4-way associative" }, 572 { 0x59, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, fully associative" }, 573 { 0x5a, TLB_DATA0_2M_4M, 32, " TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" }, 574 { 0x5b, TLB_DATA_4K_4M, 64, " TLB_DATA 4 KByte and 4 MByte pages" }, 575 { 0x5c, TLB_DATA_4K_4M, 128, " TLB_DATA 4 KByte and 4 MByte pages" }, 576 { 0x5d, TLB_DATA_4K_4M, 256, " TLB_DATA 4 KByte and 4 MByte pages" }, 577 { 0x61, TLB_INST_4K, 48, " TLB_INST 4 KByte pages, full associative" }, 578 { 0x63, TLB_DATA_1G, 4, " TLB_DATA 1 GByte pages, 4-way set associative" }, 579 { 0x76, TLB_INST_2M_4M, 8, " TLB_INST 2-MByte or 4-MByte pages, fully associative" }, 580 { 0xb0, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 4-way set associative" }, 581 { 0xb1, TLB_INST_2M_4M, 4, " TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" }, 582 { 0xb2, TLB_INST_4K, 64, " TLB_INST 4KByte pages, 4-way set associative" }, 583 { 0xb3, TLB_DATA_4K, 128, " TLB_DATA 4 KByte pages, 4-way set associative" }, 584 { 0xb4, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 4-way associative" }, 585 { 0xb5, TLB_INST_4K, 64, " TLB_INST 4 KByte pages, 8-way set associative" }, 586 { 0xb6, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 8-way set associative" }, 587 { 0xba, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way associative" }, 588 { 0xc0, TLB_DATA_4K_4M, 8, " TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" }, 589 { 0xc1, STLB_4K_2M, 1024, " STLB 4 KByte and 2 MByte pages, 8-way associative" }, 590 { 0xc2, TLB_DATA_2M_4M, 16, " DTLB 2 MByte/4MByte pages, 4-way associative" }, 591 { 0xca, STLB_4K, 512, " STLB 4 KByte pages, 4-way associative" }, 592 { 0x00, 0, 0 } 593 }; 594 595 static void intel_tlb_lookup(const unsigned char desc) 596 { 597 unsigned char k; 598 if (desc == 0) 599 return; 600 601 /* look up this descriptor in the table */ 602 for (k = 0; intel_tlb_table[k].descriptor != desc && \ 603 intel_tlb_table[k].descriptor != 0; k++) 604 ; 605 606 if (intel_tlb_table[k].tlb_type == 0) 607 return; 608 609 switch (intel_tlb_table[k].tlb_type) { 610 case STLB_4K: 611 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries) 612 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries; 613 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries) 614 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries; 615 break; 616 case STLB_4K_2M: 617 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries) 618 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries; 619 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries) 620 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries; 621 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries) 622 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries; 623 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries) 624 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries; 625 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries) 626 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries; 627 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries) 628 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries; 629 break; 630 case TLB_INST_ALL: 631 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries) 632 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries; 633 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries) 634 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries; 635 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries) 636 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries; 637 break; 638 case TLB_INST_4K: 639 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries) 640 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries; 641 break; 642 case TLB_INST_4M: 643 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries) 644 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries; 645 break; 646 case TLB_INST_2M_4M: 647 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries) 648 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries; 649 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries) 650 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries; 651 break; 652 case TLB_DATA_4K: 653 case TLB_DATA0_4K: 654 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries) 655 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries; 656 break; 657 case TLB_DATA_4M: 658 case TLB_DATA0_4M: 659 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries) 660 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries; 661 break; 662 case TLB_DATA_2M_4M: 663 case TLB_DATA0_2M_4M: 664 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries) 665 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries; 666 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries) 667 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries; 668 break; 669 case TLB_DATA_4K_4M: 670 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries) 671 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries; 672 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries) 673 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries; 674 break; 675 case TLB_DATA_1G: 676 if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries) 677 tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries; 678 break; 679 } 680 } 681 682 static void intel_detect_tlb(struct cpuinfo_x86 *c) 683 { 684 int i, j, n; 685 unsigned int regs[4]; 686 unsigned char *desc = (unsigned char *)regs; 687 688 if (c->cpuid_level < 2) 689 return; 690 691 /* Number of times to iterate */ 692 n = cpuid_eax(2) & 0xFF; 693 694 for (i = 0 ; i < n ; i++) { 695 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]); 696 697 /* If bit 31 is set, this is an unknown format */ 698 for (j = 0 ; j < 3 ; j++) 699 if (regs[j] & (1 << 31)) 700 regs[j] = 0; 701 702 /* Byte 0 is level count, not a descriptor */ 703 for (j = 1 ; j < 16 ; j++) 704 intel_tlb_lookup(desc[j]); 705 } 706 } 707 708 static const struct cpu_dev intel_cpu_dev = { 709 .c_vendor = "Intel", 710 .c_ident = { "GenuineIntel" }, 711 #ifdef CONFIG_X86_32 712 .legacy_models = { 713 { .family = 4, .model_names = 714 { 715 [0] = "486 DX-25/33", 716 [1] = "486 DX-50", 717 [2] = "486 SX", 718 [3] = "486 DX/2", 719 [4] = "486 SL", 720 [5] = "486 SX/2", 721 [7] = "486 DX/2-WB", 722 [8] = "486 DX/4", 723 [9] = "486 DX/4-WB" 724 } 725 }, 726 { .family = 5, .model_names = 727 { 728 [0] = "Pentium 60/66 A-step", 729 [1] = "Pentium 60/66", 730 [2] = "Pentium 75 - 200", 731 [3] = "OverDrive PODP5V83", 732 [4] = "Pentium MMX", 733 [7] = "Mobile Pentium 75 - 200", 734 [8] = "Mobile Pentium MMX", 735 [9] = "Quark SoC X1000", 736 } 737 }, 738 { .family = 6, .model_names = 739 { 740 [0] = "Pentium Pro A-step", 741 [1] = "Pentium Pro", 742 [3] = "Pentium II (Klamath)", 743 [4] = "Pentium II (Deschutes)", 744 [5] = "Pentium II (Deschutes)", 745 [6] = "Mobile Pentium II", 746 [7] = "Pentium III (Katmai)", 747 [8] = "Pentium III (Coppermine)", 748 [10] = "Pentium III (Cascades)", 749 [11] = "Pentium III (Tualatin)", 750 } 751 }, 752 { .family = 15, .model_names = 753 { 754 [0] = "Pentium 4 (Unknown)", 755 [1] = "Pentium 4 (Willamette)", 756 [2] = "Pentium 4 (Northwood)", 757 [4] = "Pentium 4 (Foster)", 758 [5] = "Pentium 4 (Foster)", 759 } 760 }, 761 }, 762 .legacy_cache_size = intel_size_cache, 763 #endif 764 .c_detect_tlb = intel_detect_tlb, 765 .c_early_init = early_init_intel, 766 .c_init = init_intel, 767 .c_bsp_resume = intel_bsp_resume, 768 .c_x86_vendor = X86_VENDOR_INTEL, 769 }; 770 771 cpu_dev_register(intel_cpu_dev); 772 773