| 1 | |
| 2 | |
| 3 | * CPU Microcode Update Driver for Linux
|
| 4 | *
|
| 5 | * Copyright (C) 2000-2006 Tigran Aivazian <[email protected]>
|
| 6 | * 2006 Shaohua Li <[email protected]>
|
| 7 | * 2013-2016 Borislav Petkov <[email protected]>
|
| 8 | *
|
| 9 | * X86 CPU microcode early update for Linux:
|
| 10 | *
|
| 11 | * Copyright (C) 2012 Fenghua Yu <[email protected]>
|
| 12 | * H Peter Anvin" <[email protected]>
|
| 13 | * (C) 2015 Borislav Petkov <[email protected]>
|
| 14 | *
|
| 15 | * This driver allows to upgrade microcode on x86 processors.
|
| 16 |
|
| 17 |
|
| 18 | #define pr_fmt(fmt) "microcode: " fmt
|
| 19 |
|
| 20 | #include <linux/stop_machine.h>
|
| 21 | #include <linux/device/faux.h>
|
| 22 | #include <linux/syscore_ops.h>
|
| 23 | #include <linux/miscdevice.h>
|
| 24 | #include <linux/capability.h>
|
| 25 | #include <linux/firmware.h>
|
| 26 | #include <linux/cpumask.h>
|
| 27 | #include <linux/kernel.h>
|
| 28 | #include <linux/delay.h>
|
| 29 | #include <linux/mutex.h>
|
| 30 | #include <linux/cpu.h>
|
| 31 | #include <linux/nmi.h>
|
| 32 | #include <linux/fs.h>
|
| 33 | #include <linux/mm.h>
|
| 34 |
|
| 35 | #include <asm/apic.h>
|
| 36 | #include <asm/cpu_device_id.h>
|
| 37 | #include <asm/cpuid/api.h>
|
| 38 | #include <asm/perf_event.h>
|
| 39 | #include <asm/processor.h>
|
| 40 | #include <asm/cmdline.h>
|
| 41 | #include <asm/msr.h>
|
| 42 | #include <asm/setup.h>
|
| 43 |
|
| 44 | #include "internal.h"
|
| 45 |
|
| 46 | static struct microcode_ops *microcode_ops;
|
| 47 | static bool dis_ucode_ldr;
|
| 48 |
|
| 49 | bool force_minrev = IS_ENABLED(CONFIG_MICROCODE_LATE_FORCE_MINREV);
|
| 50 |
|
| 51 | |
| 52 | * Those below should be behind CONFIG_MICROCODE_DBG ifdeffery but in
|
| 53 | * order to not uglify the code with ifdeffery and use IS_ENABLED()
|
| 54 | * instead, leave them in. When microcode debugging is not enabled,
|
| 55 | * those are meaningless anyway.
|
| 56 |
|
| 57 |
|
| 58 | u32 base_rev;
|
| 59 | u32 microcode_rev[NR_CPUS] = {};
|
| 60 |
|
| 61 | bool __ro_after_init x86_hypervisor_present;
|
| 62 |
|
| 63 | |
| 64 | * Synchronization.
|
| 65 | *
|
| 66 | * All non cpu-hotplug-callback call sites use:
|
| 67 | *
|
| 68 | * - cpus_read_lock/unlock() to synchronize with
|
| 69 | * the cpu-hotplug-callback call sites.
|
| 70 | *
|
| 71 | * We guarantee that only a single cpu is being
|
| 72 | * updated at any particular moment of time.
|
| 73 |
|
| 74 | struct ucode_cpu_info ucode_cpu_info[NR_CPUS];
|
| 75 |
|
| 76 | |
| 77 | * Those patch levels cannot be updated to newer ones and thus should be final.
|
| 78 |
|
| 79 | static u32 final_levels[] = {
|
| 80 | 0x01000098,
|
| 81 | 0x0100009f,
|
| 82 | 0x010000af,
|
| 83 | 0,
|
| 84 | };
|
| 85 |
|
| 86 | struct early_load_data early_data;
|
| 87 |
|
| 88 | |
| 89 | * Check the current patch level on this CPU.
|
| 90 | *
|
| 91 | * Returns:
|
| 92 | * - true: if update should stop
|
| 93 | * - false: otherwise
|
| 94 |
|
| 95 | static bool amd_check_current_patch_level(void)
|
| 96 | {
|
| 97 | u32 lvl, dummy, i;
|
| 98 | u32 *levels;
|
| 99 |
|
| 100 | if (x86_cpuid_vendor() != X86_VENDOR_AMD)
|
| 101 | return false;
|
| 102 |
|
| 103 | native_rdmsr(MSR_AMD64_PATCH_LEVEL, lvl, dummy);
|
| 104 |
|
| 105 | levels = final_levels;
|
| 106 |
|
| 107 | for (i = 0; levels[i]; i++) {
|
| 108 | if (lvl == levels[i])
|
| 109 | return true;
|
| 110 | }
|
| 111 | return false;
|
| 112 | }
|
| 113 |
|
| 114 | bool __init microcode_loader_disabled(void)
|
| 115 | {
|
| 116 | if (dis_ucode_ldr)
|
| 117 | return true;
|
| 118 |
|
| 119 | |
| 120 | * Disable when:
|
| 121 | *
|
| 122 | * 1) The CPU does not support CPUID, detected below in
|
| 123 | * load_ucode_bsp().
|
| 124 | *
|
| 125 | * 2) Bit 31 in CPUID[1]:ECX is set
|
| 126 | * The bit is reserved for hypervisor use. This is still not
|
| 127 | * completely accurate as XEN PV guests don't see that CPUID bit
|
| 128 | * set, but that's good enough as they don't land on the BSP
|
| 129 | * path anyway.
|
| 130 | *
|
| 131 | * 3) Certain AMD patch levels are not allowed to be
|
| 132 | * overwritten.
|
| 133 |
|
| 134 | if ((x86_hypervisor_present && !IS_ENABLED(CONFIG_MICROCODE_DBG)) ||
|
| 135 | amd_check_current_patch_level())
|
| 136 | dis_ucode_ldr = true;
|
| 137 |
|
| 138 | return dis_ucode_ldr;
|
| 139 | }
|
| 140 |
|
| 141 | static void __init early_parse_cmdline(void)
|
| 142 | {
|
| 143 | char cmd_buf[64] = {};
|
| 144 | char *s, *p = cmd_buf;
|
| 145 |
|
| 146 | if (cmdline_find_option(boot_command_line, "microcode", cmd_buf, sizeof(cmd_buf)) > 0) {
|
| 147 | while ((s = strsep(&p, ","))) {
|
| 148 | if (IS_ENABLED(CONFIG_MICROCODE_DBG)) {
|
| 149 | if (strstr(s, "base_rev=")) {
|
| 150 |
|
| 151 | strsep(&s, "=");
|
| 152 | if (kstrtouint(s, 16, &base_rev)) { ; }
|
| 153 | }
|
| 154 | }
|
| 155 |
|
| 156 | if (!strcmp("force_minrev", s))
|
| 157 | force_minrev = true;
|
| 158 |
|
| 159 | if (!strcmp(s, "dis_ucode_ldr"))
|
| 160 | dis_ucode_ldr = true;
|
| 161 | }
|
| 162 | }
|
| 163 |
|
| 164 |
|
| 165 | if (cmdline_find_option_bool(boot_command_line, "dis_ucode_ldr") > 0)
|
| 166 | dis_ucode_ldr = true;
|
| 167 | }
|
| 168 |
|
| 169 | void __init load_ucode_bsp(void)
|
| 170 | {
|
| 171 | unsigned int cpuid_1_eax;
|
| 172 | bool intel = true;
|
| 173 |
|
| 174 | early_parse_cmdline();
|
| 175 |
|
| 176 | if (!cpuid_feature())
|
| 177 | dis_ucode_ldr = true;
|
| 178 | else
|
| 179 | x86_hypervisor_present = native_cpuid_ecx(1) & BIT(31);
|
| 180 |
|
| 181 | if (microcode_loader_disabled())
|
| 182 | return;
|
| 183 |
|
| 184 | cpuid_1_eax = native_cpuid_eax(1);
|
| 185 |
|
| 186 | switch (x86_cpuid_vendor()) {
|
| 187 | case X86_VENDOR_INTEL:
|
| 188 | if (x86_family(cpuid_1_eax) < 6)
|
| 189 | return;
|
| 190 | break;
|
| 191 |
|
| 192 | case X86_VENDOR_AMD:
|
| 193 | if (x86_family(cpuid_1_eax) < 0x10)
|
| 194 | return;
|
| 195 | intel = false;
|
| 196 | break;
|
| 197 |
|
| 198 | default:
|
| 199 | return;
|
| 200 | }
|
| 201 |
|
| 202 | if (intel)
|
| 203 | load_ucode_intel_bsp(&early_data);
|
| 204 | else
|
| 205 | load_ucode_amd_bsp(&early_data, cpuid_1_eax);
|
| 206 | }
|
| 207 |
|
| 208 | void load_ucode_ap(void)
|
| 209 | {
|
| 210 | unsigned int cpuid_1_eax;
|
| 211 |
|
| 212 | |
| 213 | * Can't use microcode_loader_disabled() here - .init section
|
| 214 | * hell. It doesn't have to either - the BSP variant must've
|
| 215 | * parsed cmdline already anyway.
|
| 216 |
|
| 217 | if (dis_ucode_ldr)
|
| 218 | return;
|
| 219 |
|
| 220 | cpuid_1_eax = native_cpuid_eax(1);
|
| 221 |
|
| 222 | switch (x86_cpuid_vendor()) {
|
| 223 | case X86_VENDOR_INTEL:
|
| 224 | if (x86_family(cpuid_1_eax) >= 6)
|
| 225 | load_ucode_intel_ap();
|
| 226 | break;
|
| 227 | case X86_VENDOR_AMD:
|
| 228 | if (x86_family(cpuid_1_eax) >= 0x10)
|
| 229 | load_ucode_amd_ap(cpuid_1_eax);
|
| 230 | break;
|
| 231 | default:
|
| 232 | break;
|
| 233 | }
|
| 234 | }
|
| 235 |
|
| 236 | struct cpio_data __init find_microcode_in_initrd(const char *path)
|
| 237 | {
|
| 238 | #ifdef CONFIG_BLK_DEV_INITRD
|
| 239 | unsigned long start = 0;
|
| 240 | size_t size;
|
| 241 |
|
| 242 | #ifdef CONFIG_X86_32
|
| 243 | size = boot_params.hdr.ramdisk_size;
|
| 244 |
|
| 245 | if (size)
|
| 246 | start = initrd_start_early;
|
| 247 |
|
| 248 | #else
|
| 249 | size = (unsigned long)boot_params.ext_ramdisk_size << 32;
|
| 250 | size |= boot_params.hdr.ramdisk_size;
|
| 251 |
|
| 252 | if (size) {
|
| 253 | start = (unsigned long)boot_params.ext_ramdisk_image << 32;
|
| 254 | start |= boot_params.hdr.ramdisk_image;
|
| 255 | start += PAGE_OFFSET;
|
| 256 | }
|
| 257 | #endif
|
| 258 |
|
| 259 | |
| 260 | * Fixup the start address: after reserve_initrd() runs, initrd_start
|
| 261 | * has the virtual address of the beginning of the initrd. It also
|
| 262 | * possibly relocates the ramdisk. In either case, initrd_start contains
|
| 263 | * the updated address so use that instead.
|
| 264 |
|
| 265 | if (initrd_start)
|
| 266 | start = initrd_start;
|
| 267 |
|
| 268 | return find_cpio_data(path, (void *)start, size, NULL);
|
| 269 | #else
|
| 270 | return (struct cpio_data){ NULL, 0, "" };
|
| 271 | #endif
|
| 272 | }
|
| 273 |
|
| 274 | static void reload_early_microcode(unsigned int cpu)
|
| 275 | {
|
| 276 | int vendor, family;
|
| 277 |
|
| 278 | vendor = x86_cpuid_vendor();
|
| 279 | family = x86_cpuid_family();
|
| 280 |
|
| 281 | switch (vendor) {
|
| 282 | case X86_VENDOR_INTEL:
|
| 283 | if (family >= 6)
|
| 284 | reload_ucode_intel();
|
| 285 | break;
|
| 286 | case X86_VENDOR_AMD:
|
| 287 | if (family >= 0x10)
|
| 288 | reload_ucode_amd(cpu);
|
| 289 | break;
|
| 290 | default:
|
| 291 | break;
|
| 292 | }
|
| 293 | }
|
| 294 |
|
| 295 |
|
| 296 | static struct faux_device *microcode_fdev;
|
| 297 |
|
| 298 | #ifdef CONFIG_MICROCODE_LATE_LOADING
|
| 299 | |
| 300 | * Late loading dance. Why the heavy-handed stomp_machine effort?
|
| 301 | *
|
| 302 | * - HT siblings must be idle and not execute other code while the other sibling
|
| 303 | * is loading microcode in order to avoid any negative interactions caused by
|
| 304 | * the loading.
|
| 305 | *
|
| 306 | * - In addition, microcode update on the cores must be serialized until this
|
| 307 | * requirement can be relaxed in the future. Right now, this is conservative
|
| 308 | * and good.
|
| 309 |
|
| 310 | enum sibling_ctrl {
|
| 311 |
|
| 312 | SCTRL_WAIT,
|
| 313 |
|
| 314 | SCTRL_APPLY,
|
| 315 |
|
| 316 | SCTRL_DONE,
|
| 317 | };
|
| 318 |
|
| 319 | struct microcode_ctrl {
|
| 320 | enum sibling_ctrl ctrl;
|
| 321 | enum ucode_state result;
|
| 322 | unsigned int ctrl_cpu;
|
| 323 | bool nmi_enabled;
|
| 324 | };
|
| 325 |
|
| 326 | DEFINE_STATIC_KEY_FALSE(microcode_nmi_handler_enable);
|
| 327 | static DEFINE_PER_CPU(struct microcode_ctrl, ucode_ctrl);
|
| 328 | static atomic_t late_cpus_in, offline_in_nmi;
|
| 329 | static unsigned int loops_per_usec;
|
| 330 | static cpumask_t cpu_offline_mask;
|
| 331 |
|
| 332 | static noinstr bool wait_for_cpus(atomic_t *cnt)
|
| 333 | {
|
| 334 | unsigned int timeout, loops;
|
| 335 |
|
| 336 | WARN_ON_ONCE(raw_atomic_dec_return(cnt) < 0);
|
| 337 |
|
| 338 | for (timeout = 0; timeout < USEC_PER_SEC; timeout++) {
|
| 339 | if (!raw_atomic_read(cnt))
|
| 340 | return true;
|
| 341 |
|
| 342 | for (loops = 0; loops < loops_per_usec; loops++)
|
| 343 | cpu_relax();
|
| 344 |
|
| 345 |
|
| 346 | if (!microcode_ops->use_nmi && !(timeout % USEC_PER_MSEC)) {
|
| 347 | instrumentation_begin();
|
| 348 | touch_nmi_watchdog();
|
| 349 | instrumentation_end();
|
| 350 | }
|
| 351 | }
|
| 352 |
|
| 353 | raw_atomic_inc(cnt);
|
| 354 | return false;
|
| 355 | }
|
| 356 |
|
| 357 | static noinstr bool wait_for_ctrl(void)
|
| 358 | {
|
| 359 | unsigned int timeout, loops;
|
| 360 |
|
| 361 | for (timeout = 0; timeout < USEC_PER_SEC; timeout++) {
|
| 362 | if (raw_cpu_read(ucode_ctrl.ctrl) != SCTRL_WAIT)
|
| 363 | return true;
|
| 364 |
|
| 365 | for (loops = 0; loops < loops_per_usec; loops++)
|
| 366 | cpu_relax();
|
| 367 |
|
| 368 |
|
| 369 | if (!microcode_ops->use_nmi && !(timeout % USEC_PER_MSEC)) {
|
| 370 | instrumentation_begin();
|
| 371 | touch_nmi_watchdog();
|
| 372 | instrumentation_end();
|
| 373 | }
|
| 374 | }
|
| 375 | return false;
|
| 376 | }
|
| 377 |
|
| 378 | |
| 379 | * Protected against instrumentation up to the point where the primary
|
| 380 | * thread completed the update. See microcode_nmi_handler() for details.
|
| 381 |
|
| 382 | static noinstr bool load_secondary_wait(unsigned int ctrl_cpu)
|
| 383 | {
|
| 384 |
|
| 385 | if (!wait_for_cpus(&late_cpus_in)) {
|
| 386 | raw_cpu_write(ucode_ctrl.result, UCODE_TIMEOUT);
|
| 387 | return false;
|
| 388 | }
|
| 389 |
|
| 390 | |
| 391 | * Wait for primary threads to complete. If one of them hangs due
|
| 392 | * to the update, there is no way out. This is non-recoverable
|
| 393 | * because the CPU might hold locks or resources and confuse the
|
| 394 | * scheduler, watchdogs etc. There is no way to safely evacuate the
|
| 395 | * machine.
|
| 396 |
|
| 397 | if (wait_for_ctrl())
|
| 398 | return true;
|
| 399 |
|
| 400 | instrumentation_begin();
|
| 401 | panic("Microcode load: Primary CPU %d timed out\n", ctrl_cpu);
|
| 402 | instrumentation_end();
|
| 403 | }
|
| 404 |
|
| 405 | |
| 406 | * Protected against instrumentation up to the point where the primary
|
| 407 | * thread completed the update. See microcode_nmi_handler() for details.
|
| 408 |
|
| 409 | static noinstr void load_secondary(unsigned int cpu)
|
| 410 | {
|
| 411 | unsigned int ctrl_cpu = raw_cpu_read(ucode_ctrl.ctrl_cpu);
|
| 412 | enum ucode_state ret;
|
| 413 |
|
| 414 | if (!load_secondary_wait(ctrl_cpu)) {
|
| 415 | instrumentation_begin();
|
| 416 | pr_err_once("load: %d CPUs timed out\n",
|
| 417 | atomic_read(&late_cpus_in) - 1);
|
| 418 | instrumentation_end();
|
| 419 | return;
|
| 420 | }
|
| 421 |
|
| 422 |
|
| 423 | instrumentation_begin();
|
| 424 | |
| 425 | * If the primary succeeded then invoke the apply() callback,
|
| 426 | * otherwise copy the state from the primary thread.
|
| 427 |
|
| 428 | if (this_cpu_read(ucode_ctrl.ctrl) == SCTRL_APPLY)
|
| 429 | ret = microcode_ops->apply_microcode(cpu);
|
| 430 | else
|
| 431 | ret = per_cpu(ucode_ctrl.result, ctrl_cpu);
|
| 432 |
|
| 433 | this_cpu_write(ucode_ctrl.result, ret);
|
| 434 | this_cpu_write(ucode_ctrl.ctrl, SCTRL_DONE);
|
| 435 | instrumentation_end();
|
| 436 | }
|
| 437 |
|
| 438 | static void __load_primary(unsigned int cpu)
|
| 439 | {
|
| 440 | struct cpumask *secondaries = topology_sibling_cpumask(cpu);
|
| 441 | enum sibling_ctrl ctrl;
|
| 442 | enum ucode_state ret;
|
| 443 | unsigned int sibling;
|
| 444 |
|
| 445 |
|
| 446 | if (!wait_for_cpus(&late_cpus_in)) {
|
| 447 | this_cpu_write(ucode_ctrl.result, UCODE_TIMEOUT);
|
| 448 | pr_err_once("load: %d CPUs timed out\n", atomic_read(&late_cpus_in) - 1);
|
| 449 | return;
|
| 450 | }
|
| 451 |
|
| 452 | ret = microcode_ops->apply_microcode(cpu);
|
| 453 | this_cpu_write(ucode_ctrl.result, ret);
|
| 454 | this_cpu_write(ucode_ctrl.ctrl, SCTRL_DONE);
|
| 455 |
|
| 456 | |
| 457 | * If the update was successful, let the siblings run the apply()
|
| 458 | * callback. If not, tell them it's done. This also covers the
|
| 459 | * case where the CPU has uniform loading at package or system
|
| 460 | * scope implemented but does not advertise it.
|
| 461 |
|
| 462 | if (ret == UCODE_UPDATED || ret == UCODE_OK)
|
| 463 | ctrl = SCTRL_APPLY;
|
| 464 | else
|
| 465 | ctrl = SCTRL_DONE;
|
| 466 |
|
| 467 | for_each_cpu(sibling, secondaries) {
|
| 468 | if (sibling != cpu)
|
| 469 | per_cpu(ucode_ctrl.ctrl, sibling) = ctrl;
|
| 470 | }
|
| 471 | }
|
| 472 |
|
| 473 | static bool kick_offline_cpus(unsigned int nr_offl)
|
| 474 | {
|
| 475 | unsigned int cpu, timeout;
|
| 476 |
|
| 477 | for_each_cpu(cpu, &cpu_offline_mask) {
|
| 478 |
|
| 479 | per_cpu(ucode_ctrl.nmi_enabled, cpu) = true;
|
| 480 | apic_send_nmi_to_offline_cpu(cpu);
|
| 481 | }
|
| 482 |
|
| 483 |
|
| 484 | for (timeout = 0; timeout < (USEC_PER_SEC / 2); timeout++) {
|
| 485 | if (atomic_read(&offline_in_nmi) == nr_offl)
|
| 486 | return true;
|
| 487 | udelay(1);
|
| 488 | }
|
| 489 |
|
| 490 | return false;
|
| 491 | }
|
| 492 |
|
| 493 | static void release_offline_cpus(void)
|
| 494 | {
|
| 495 | unsigned int cpu;
|
| 496 |
|
| 497 | for_each_cpu(cpu, &cpu_offline_mask)
|
| 498 | per_cpu(ucode_ctrl.ctrl, cpu) = SCTRL_DONE;
|
| 499 | }
|
| 500 |
|
| 501 | static void load_primary(unsigned int cpu)
|
| 502 | {
|
| 503 | unsigned int nr_offl = cpumask_weight(&cpu_offline_mask);
|
| 504 | bool proceed = true;
|
| 505 |
|
| 506 |
|
| 507 | if (!cpu && nr_offl)
|
| 508 | proceed = kick_offline_cpus(nr_offl);
|
| 509 |
|
| 510 |
|
| 511 | if (proceed)
|
| 512 | __load_primary(cpu);
|
| 513 |
|
| 514 |
|
| 515 | if (!cpu && nr_offl)
|
| 516 | release_offline_cpus();
|
| 517 | }
|
| 518 |
|
| 519 | |
| 520 | * Minimal stub rendezvous handler for soft-offlined CPUs which participate
|
| 521 | * in the NMI rendezvous to protect against a concurrent NMI on affected
|
| 522 | * CPUs.
|
| 523 |
|
| 524 | void noinstr microcode_offline_nmi_handler(void)
|
| 525 | {
|
| 526 | if (!raw_cpu_read(ucode_ctrl.nmi_enabled))
|
| 527 | return;
|
| 528 | raw_cpu_write(ucode_ctrl.nmi_enabled, false);
|
| 529 | raw_cpu_write(ucode_ctrl.result, UCODE_OFFLINE);
|
| 530 | raw_atomic_inc(&offline_in_nmi);
|
| 531 | wait_for_ctrl();
|
| 532 | }
|
| 533 |
|
| 534 | static noinstr bool microcode_update_handler(void)
|
| 535 | {
|
| 536 | unsigned int cpu = raw_smp_processor_id();
|
| 537 |
|
| 538 | if (raw_cpu_read(ucode_ctrl.ctrl_cpu) == cpu) {
|
| 539 | instrumentation_begin();
|
| 540 | load_primary(cpu);
|
| 541 | instrumentation_end();
|
| 542 | } else {
|
| 543 | load_secondary(cpu);
|
| 544 | }
|
| 545 |
|
| 546 | instrumentation_begin();
|
| 547 | touch_nmi_watchdog();
|
| 548 | instrumentation_end();
|
| 549 |
|
| 550 | return true;
|
| 551 | }
|
| 552 |
|
| 553 | |
| 554 | * Protection against instrumentation is required for CPUs which are not
|
| 555 | * safe against an NMI which is delivered to the secondary SMT sibling
|
| 556 | * while the primary thread updates the microcode. Instrumentation can end
|
| 557 | * up in #INT3, #DB and #PF. The IRET from those exceptions reenables NMI
|
| 558 | * which is the opposite of what the NMI rendezvous is trying to achieve.
|
| 559 | *
|
| 560 | * The primary thread is safe versus instrumentation as the actual
|
| 561 | * microcode update handles this correctly. It's only the sibling code
|
| 562 | * path which must be NMI safe until the primary thread completed the
|
| 563 | * update.
|
| 564 |
|
| 565 | bool noinstr microcode_nmi_handler(void)
|
| 566 | {
|
| 567 | if (!raw_cpu_read(ucode_ctrl.nmi_enabled))
|
| 568 | return false;
|
| 569 |
|
| 570 | raw_cpu_write(ucode_ctrl.nmi_enabled, false);
|
| 571 | return microcode_update_handler();
|
| 572 | }
|
| 573 |
|
| 574 | static int load_cpus_stopped(void *unused)
|
| 575 | {
|
| 576 | if (microcode_ops->use_nmi) {
|
| 577 |
|
| 578 | this_cpu_write(ucode_ctrl.nmi_enabled, true);
|
| 579 | apic->send_IPI(smp_processor_id(), NMI_VECTOR);
|
| 580 | } else {
|
| 581 |
|
| 582 | microcode_update_handler();
|
| 583 | }
|
| 584 | return 0;
|
| 585 | }
|
| 586 |
|
| 587 | static int load_late_stop_cpus(bool is_safe)
|
| 588 | {
|
| 589 | unsigned int cpu, updated = 0, failed = 0, timedout = 0, siblings = 0;
|
| 590 | unsigned int nr_offl, offline = 0;
|
| 591 | int old_rev = boot_cpu_data.microcode;
|
| 592 | struct cpuinfo_x86 prev_info;
|
| 593 |
|
| 594 | if (!is_safe) {
|
| 595 | pr_err("Late microcode loading without minimal revision check.\n");
|
| 596 | pr_err("You should switch to early loading, if possible.\n");
|
| 597 | }
|
| 598 |
|
| 599 | |
| 600 | * Pre-load the microcode image into a staging device. This
|
| 601 | * process is preemptible and does not require stopping CPUs.
|
| 602 | * Successful staging simplifies the subsequent late-loading
|
| 603 | * process, reducing rendezvous time.
|
| 604 | *
|
| 605 | * Even if the transfer fails, the update will proceed as usual.
|
| 606 |
|
| 607 | if (microcode_ops->use_staging)
|
| 608 | microcode_ops->stage_microcode();
|
| 609 |
|
| 610 | atomic_set(&late_cpus_in, num_online_cpus());
|
| 611 | atomic_set(&offline_in_nmi, 0);
|
| 612 | loops_per_usec = loops_per_jiffy / (TICK_NSEC / 1000);
|
| 613 |
|
| 614 | |
| 615 | * Take a snapshot before the microcode update in order to compare and
|
| 616 | * check whether any bits changed after an update.
|
| 617 |
|
| 618 | store_cpu_caps(&prev_info);
|
| 619 |
|
| 620 | if (microcode_ops->use_nmi)
|
| 621 | static_branch_enable_cpuslocked(µcode_nmi_handler_enable);
|
| 622 |
|
| 623 | stop_machine_cpuslocked(load_cpus_stopped, NULL, cpu_online_mask);
|
| 624 |
|
| 625 | if (microcode_ops->use_nmi)
|
| 626 | static_branch_disable_cpuslocked(µcode_nmi_handler_enable);
|
| 627 |
|
| 628 |
|
| 629 | for_each_cpu_and(cpu, cpu_present_mask, &cpus_booted_once_mask) {
|
| 630 | switch (per_cpu(ucode_ctrl.result, cpu)) {
|
| 631 | case UCODE_UPDATED: updated++; break;
|
| 632 | case UCODE_TIMEOUT: timedout++; break;
|
| 633 | case UCODE_OK: siblings++; break;
|
| 634 | case UCODE_OFFLINE: offline++; break;
|
| 635 | default: failed++; break;
|
| 636 | }
|
| 637 | }
|
| 638 |
|
| 639 | if (microcode_ops->finalize_late_load)
|
| 640 | microcode_ops->finalize_late_load(!updated);
|
| 641 |
|
| 642 | if (!updated) {
|
| 643 |
|
| 644 | if (!failed && !timedout)
|
| 645 | return 0;
|
| 646 |
|
| 647 | nr_offl = cpumask_weight(&cpu_offline_mask);
|
| 648 | if (offline < nr_offl) {
|
| 649 | pr_warn("%u offline siblings did not respond.\n",
|
| 650 | nr_offl - atomic_read(&offline_in_nmi));
|
| 651 | return -EIO;
|
| 652 | }
|
| 653 | pr_err("update failed: %u CPUs failed %u CPUs timed out\n",
|
| 654 | failed, timedout);
|
| 655 | return -EIO;
|
| 656 | }
|
| 657 |
|
| 658 | if (!is_safe || failed || timedout)
|
| 659 | add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK);
|
| 660 |
|
| 661 | pr_info("load: updated on %u primary CPUs with %u siblings\n", updated, siblings);
|
| 662 | if (failed || timedout) {
|
| 663 | pr_err("load incomplete. %u CPUs timed out or failed\n",
|
| 664 | num_online_cpus() - (updated + siblings));
|
| 665 | }
|
| 666 | pr_info("revision: 0x%x -> 0x%x\n", old_rev, boot_cpu_data.microcode);
|
| 667 | microcode_check(&prev_info);
|
| 668 |
|
| 669 | return updated + siblings == num_online_cpus() ? 0 : -EIO;
|
| 670 | }
|
| 671 |
|
| 672 | |
| 673 | * This function does two things:
|
| 674 | *
|
| 675 | * 1) Ensure that all required CPUs which are present and have been booted
|
| 676 | * once are online.
|
| 677 | *
|
| 678 | * To pass this check, all primary threads must be online.
|
| 679 | *
|
| 680 | * If the microcode load is not safe against NMI then all SMT threads
|
| 681 | * must be online as well because they still react to NMIs when they are
|
| 682 | * soft-offlined and parked in one of the play_dead() variants. So if a
|
| 683 | * NMI hits while the primary thread updates the microcode the resulting
|
| 684 | * behaviour is undefined. The default play_dead() implementation on
|
| 685 | * modern CPUs uses MWAIT, which is also not guaranteed to be safe
|
| 686 | * against a microcode update which affects MWAIT.
|
| 687 | *
|
| 688 | * As soft-offlined CPUs still react on NMIs, the SMT sibling
|
| 689 | * restriction can be lifted when the vendor driver signals to use NMI
|
| 690 | * for rendezvous and the APIC provides a mechanism to send an NMI to a
|
| 691 | * soft-offlined CPU. The soft-offlined CPUs are then able to
|
| 692 | * participate in the rendezvous in a trivial stub handler.
|
| 693 | *
|
| 694 | * 2) Initialize the per CPU control structure and create a cpumask
|
| 695 | * which contains "offline"; secondary threads, so they can be handled
|
| 696 | * correctly by a control CPU.
|
| 697 |
|
| 698 | static bool setup_cpus(void)
|
| 699 | {
|
| 700 | struct microcode_ctrl ctrl = { .ctrl = SCTRL_WAIT, .result = -1, };
|
| 701 | bool allow_smt_offline;
|
| 702 | unsigned int cpu;
|
| 703 |
|
| 704 | allow_smt_offline = microcode_ops->nmi_safe ||
|
| 705 | (microcode_ops->use_nmi && apic->nmi_to_offline_cpu);
|
| 706 |
|
| 707 | cpumask_clear(&cpu_offline_mask);
|
| 708 |
|
| 709 | for_each_cpu_and(cpu, cpu_present_mask, &cpus_booted_once_mask) {
|
| 710 | |
| 711 | * Offline CPUs sit in one of the play_dead() functions
|
| 712 | * with interrupts disabled, but they still react on NMIs
|
| 713 | * and execute arbitrary code. Also MWAIT being updated
|
| 714 | * while the offline CPU sits there is not necessarily safe
|
| 715 | * on all CPU variants.
|
| 716 | *
|
| 717 | * Mark them in the offline_cpus mask which will be handled
|
| 718 | * by CPU0 later in the update process.
|
| 719 | *
|
| 720 | * Ensure that the primary thread is online so that it is
|
| 721 | * guaranteed that all cores are updated.
|
| 722 |
|
| 723 | if (!cpu_online(cpu)) {
|
| 724 | if (topology_is_primary_thread(cpu) || !allow_smt_offline) {
|
| 725 | pr_err("CPU %u not online, loading aborted\n", cpu);
|
| 726 | return false;
|
| 727 | }
|
| 728 | cpumask_set_cpu(cpu, &cpu_offline_mask);
|
| 729 | per_cpu(ucode_ctrl, cpu) = ctrl;
|
| 730 | continue;
|
| 731 | }
|
| 732 |
|
| 733 | |
| 734 | * Initialize the per CPU state. This is core scope for now,
|
| 735 | * but prepared to take package or system scope into account.
|
| 736 |
|
| 737 | ctrl.ctrl_cpu = cpumask_first(topology_sibling_cpumask(cpu));
|
| 738 | per_cpu(ucode_ctrl, cpu) = ctrl;
|
| 739 | }
|
| 740 | return true;
|
| 741 | }
|
| 742 |
|
| 743 | static int load_late_locked(void)
|
| 744 | {
|
| 745 | if (!setup_cpus())
|
| 746 | return -EBUSY;
|
| 747 |
|
| 748 | switch (microcode_ops->request_microcode_fw(0, µcode_fdev->dev)) {
|
| 749 | case UCODE_NEW:
|
| 750 | return load_late_stop_cpus(false);
|
| 751 | case UCODE_NEW_SAFE:
|
| 752 | return load_late_stop_cpus(true);
|
| 753 | case UCODE_NFOUND:
|
| 754 | return -ENOENT;
|
| 755 | case UCODE_OK:
|
| 756 | return 0;
|
| 757 | default:
|
| 758 | return -EBADFD;
|
| 759 | }
|
| 760 | }
|
| 761 |
|
| 762 | static ssize_t reload_store(struct device *dev,
|
| 763 | struct device_attribute *attr,
|
| 764 | const char *buf, size_t size)
|
| 765 | {
|
| 766 | unsigned long val;
|
| 767 | ssize_t ret;
|
| 768 |
|
| 769 | ret = kstrtoul(buf, 0, &val);
|
| 770 | if (ret || val != 1)
|
| 771 | return -EINVAL;
|
| 772 |
|
| 773 | cpus_read_lock();
|
| 774 | ret = load_late_locked();
|
| 775 | cpus_read_unlock();
|
| 776 |
|
| 777 | return ret ? : size;
|
| 778 | }
|
| 779 |
|
| 780 | static DEVICE_ATTR_WO(reload);
|
| 781 | #endif
|
| 782 |
|
| 783 | static ssize_t version_show(struct device *dev,
|
| 784 | struct device_attribute *attr, char *buf)
|
| 785 | {
|
| 786 | struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
|
| 787 |
|
| 788 | return sprintf(buf, "0x%x\n", uci->cpu_sig.rev);
|
| 789 | }
|
| 790 |
|
| 791 | static ssize_t processor_flags_show(struct device *dev,
|
| 792 | struct device_attribute *attr, char *buf)
|
| 793 | {
|
| 794 | struct ucode_cpu_info *uci = ucode_cpu_info + dev->id;
|
| 795 |
|
| 796 | return sprintf(buf, "0x%x\n", uci->cpu_sig.pf);
|
| 797 | }
|
| 798 |
|
| 799 | static DEVICE_ATTR_RO(version);
|
| 800 | static DEVICE_ATTR_RO(processor_flags);
|
| 801 |
|
| 802 | static struct attribute *mc_default_attrs[] = {
|
| 803 | &dev_attr_version.attr,
|
| 804 | &dev_attr_processor_flags.attr,
|
| 805 | NULL
|
| 806 | };
|
| 807 |
|
| 808 | static const struct attribute_group mc_attr_group = {
|
| 809 | .attrs = mc_default_attrs,
|
| 810 | .name = "microcode",
|
| 811 | };
|
| 812 |
|
| 813 | static void microcode_fini_cpu(int cpu)
|
| 814 | {
|
| 815 | if (microcode_ops->microcode_fini_cpu)
|
| 816 | microcode_ops->microcode_fini_cpu(cpu);
|
| 817 | }
|
| 818 |
|
| 819 | |
| 820 | * microcode_bsp_resume - Update boot CPU microcode during resume.
|
| 821 |
|
| 822 | void microcode_bsp_resume(void)
|
| 823 | {
|
| 824 | int cpu = smp_processor_id();
|
| 825 | struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
|
| 826 |
|
| 827 | if (uci->mc)
|
| 828 | microcode_ops->apply_microcode(cpu);
|
| 829 | else
|
| 830 | reload_early_microcode(cpu);
|
| 831 | }
|
| 832 |
|
| 833 | static void microcode_bsp_syscore_resume(void *data)
|
| 834 | {
|
| 835 | microcode_bsp_resume();
|
| 836 | }
|
| 837 |
|
| 838 | static const struct syscore_ops mc_syscore_ops = {
|
| 839 | .resume = microcode_bsp_syscore_resume,
|
| 840 | };
|
| 841 |
|
| 842 | static struct syscore mc_syscore = {
|
| 843 | .ops = &mc_syscore_ops,
|
| 844 | };
|
| 845 |
|
| 846 | static int mc_cpu_online(unsigned int cpu)
|
| 847 | {
|
| 848 | struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
|
| 849 | struct device *dev = get_cpu_device(cpu);
|
| 850 |
|
| 851 | memset(uci, 0, sizeof(*uci));
|
| 852 |
|
| 853 | microcode_ops->collect_cpu_info(cpu, &uci->cpu_sig);
|
| 854 | cpu_data(cpu).microcode = uci->cpu_sig.rev;
|
| 855 | if (!cpu)
|
| 856 | boot_cpu_data.microcode = uci->cpu_sig.rev;
|
| 857 |
|
| 858 | if (sysfs_create_group(&dev->kobj, &mc_attr_group))
|
| 859 | pr_err("Failed to create group for CPU%d\n", cpu);
|
| 860 | return 0;
|
| 861 | }
|
| 862 |
|
| 863 | static int mc_cpu_down_prep(unsigned int cpu)
|
| 864 | {
|
| 865 | struct device *dev = get_cpu_device(cpu);
|
| 866 |
|
| 867 | microcode_fini_cpu(cpu);
|
| 868 | sysfs_remove_group(&dev->kobj, &mc_attr_group);
|
| 869 | return 0;
|
| 870 | }
|
| 871 |
|
| 872 | static struct attribute *cpu_root_microcode_attrs[] = {
|
| 873 | #ifdef CONFIG_MICROCODE_LATE_LOADING
|
| 874 | &dev_attr_reload.attr,
|
| 875 | #endif
|
| 876 | NULL
|
| 877 | };
|
| 878 |
|
| 879 | static const struct attribute_group cpu_root_microcode_group = {
|
| 880 | .name = "microcode",
|
| 881 | .attrs = cpu_root_microcode_attrs,
|
| 882 | };
|
| 883 |
|
| 884 | static int __init microcode_init(void)
|
| 885 | {
|
| 886 | struct device *dev_root;
|
| 887 | struct cpuinfo_x86 *c = &boot_cpu_data;
|
| 888 | int error;
|
| 889 |
|
| 890 | if (microcode_loader_disabled())
|
| 891 | return -EINVAL;
|
| 892 |
|
| 893 | if (c->x86_vendor == X86_VENDOR_INTEL)
|
| 894 | microcode_ops = init_intel_microcode();
|
| 895 | else if (c->x86_vendor == X86_VENDOR_AMD)
|
| 896 | microcode_ops = init_amd_microcode();
|
| 897 | else
|
| 898 | pr_err("no support for this CPU vendor\n");
|
| 899 |
|
| 900 | if (!microcode_ops)
|
| 901 | return -ENODEV;
|
| 902 |
|
| 903 | pr_info_once("Current revision: 0x%08x\n", (early_data.new_rev ?: early_data.old_rev));
|
| 904 |
|
| 905 | if (early_data.new_rev)
|
| 906 | pr_info_once("Updated early from: 0x%08x\n", early_data.old_rev);
|
| 907 |
|
| 908 | microcode_fdev = faux_device_create("microcode", NULL, NULL);
|
| 909 | if (!microcode_fdev)
|
| 910 | return -ENODEV;
|
| 911 |
|
| 912 | dev_root = bus_get_dev_root(&cpu_subsys);
|
| 913 | if (dev_root) {
|
| 914 | error = sysfs_create_group(&dev_root->kobj, &cpu_root_microcode_group);
|
| 915 | put_device(dev_root);
|
| 916 | if (error) {
|
| 917 | pr_err("Error creating microcode group!\n");
|
| 918 | goto out_pdev;
|
| 919 | }
|
| 920 | }
|
| 921 |
|
| 922 | register_syscore(&mc_syscore);
|
| 923 | cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/microcode:online",
|
| 924 | mc_cpu_online, mc_cpu_down_prep);
|
| 925 |
|
| 926 | return 0;
|
| 927 |
|
| 928 | out_pdev:
|
| 929 | faux_device_destroy(microcode_fdev);
|
| 930 | return error;
|
| 931 |
|
| 932 | }
|
| 933 | late_initcall(microcode_init);
|