| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * Intel CPU Microcode Update Driver for Linux |
| 4 | * |
| 5 | * Copyright (C) 2000-2006 Tigran Aivazian <aivazian.tigran@gmail.com> |
| 6 | * 2006 Shaohua Li <shaohua.li@intel.com> |
| 7 | * |
| 8 | * Intel CPU microcode early update for Linux |
| 9 | * |
| 10 | * Copyright (C) 2012 Fenghua Yu <fenghua.yu@intel.com> |
| 11 | * H Peter Anvin" <hpa@zytor.com> |
| 12 | */ |
| 13 | #define pr_fmt(fmt) "microcode: " fmt |
| 14 | #include <linux/earlycpio.h> |
| 15 | #include <linux/firmware.h> |
| 16 | #include <linux/pci_ids.h> |
| 17 | #include <linux/uaccess.h> |
| 18 | #include <linux/initrd.h> |
| 19 | #include <linux/kernel.h> |
| 20 | #include <linux/delay.h> |
| 21 | #include <linux/slab.h> |
| 22 | #include <linux/cpu.h> |
| 23 | #include <linux/uio.h> |
| 24 | #include <linux/io.h> |
| 25 | #include <linux/mm.h> |
| 26 | |
| 27 | #include <asm/cpu_device_id.h> |
| 28 | #include <asm/processor.h> |
| 29 | #include <asm/tlbflush.h> |
| 30 | #include <asm/setup.h> |
| 31 | #include <asm/msr.h> |
| 32 | |
| 33 | #include "internal.h" |
| 34 | |
| 35 | static const char ucode_path[] = "kernel/x86/microcode/GenuineIntel.bin" ; |
| 36 | |
| 37 | #define UCODE_BSP_LOADED ((struct microcode_intel *)0x1UL) |
| 38 | |
| 39 | /* Defines for the microcode staging mailbox interface */ |
| 40 | #define MBOX_REG_NUM 4 |
| 41 | #define MBOX_REG_SIZE sizeof(u32) |
| 42 | |
| 43 | #define MBOX_CONTROL_OFFSET 0x0 |
| 44 | #define MBOX_STATUS_OFFSET 0x4 |
| 45 | #define MBOX_WRDATA_OFFSET 0x8 |
| 46 | #define MBOX_RDDATA_OFFSET 0xc |
| 47 | |
| 48 | #define MASK_MBOX_CTRL_ABORT BIT(0) |
| 49 | #define MASK_MBOX_CTRL_GO BIT(31) |
| 50 | |
| 51 | #define MASK_MBOX_STATUS_ERROR BIT(2) |
| 52 | #define MASK_MBOX_STATUS_READY BIT(31) |
| 53 | |
| 54 | #define MASK_MBOX_RESP_SUCCESS BIT(0) |
| 55 | #define MASK_MBOX_RESP_PROGRESS BIT(1) |
| 56 | #define MASK_MBOX_RESP_ERROR BIT(2) |
| 57 | |
| 58 | #define MBOX_CMD_LOAD 0x3 |
| 59 | #define MBOX_OBJ_STAGING 0xb |
| 60 | #define (size) ((PCI_VENDOR_ID_INTEL) | \ |
| 61 | (MBOX_OBJ_STAGING << 16) | \ |
| 62 | ((u64)((size) / sizeof(u32)) << 32)) |
| 63 | |
| 64 | /* The size of each mailbox header */ |
| 65 | #define sizeof(u64) |
| 66 | /* The size of staging hardware response */ |
| 67 | #define MBOX_RESPONSE_SIZE sizeof(u64) |
| 68 | |
| 69 | #define MBOX_XACTION_TIMEOUT_MS (10 * MSEC_PER_SEC) |
| 70 | |
| 71 | /* Current microcode patch used in early patching on the APs. */ |
| 72 | static struct microcode_intel *ucode_patch_va __read_mostly; |
| 73 | static struct microcode_intel *ucode_patch_late __read_mostly; |
| 74 | |
| 75 | /* last level cache size per core */ |
| 76 | static unsigned int llc_size_per_core __ro_after_init; |
| 77 | |
| 78 | /* microcode format is extended from prescott processors */ |
| 79 | struct extended_signature { |
| 80 | unsigned int sig; |
| 81 | unsigned int pf; |
| 82 | unsigned int cksum; |
| 83 | }; |
| 84 | |
| 85 | struct extended_sigtable { |
| 86 | unsigned int count; |
| 87 | unsigned int cksum; |
| 88 | unsigned int reserved[3]; |
| 89 | struct extended_signature sigs[]; |
| 90 | }; |
| 91 | |
| 92 | /** |
| 93 | * struct staging_state - Track the current staging process state |
| 94 | * |
| 95 | * @mmio_base: MMIO base address for staging |
| 96 | * @ucode_len: Total size of the microcode image |
| 97 | * @chunk_size: Size of each data piece |
| 98 | * @bytes_sent: Total bytes transmitted so far |
| 99 | * @offset: Current offset in the microcode image |
| 100 | */ |
| 101 | struct staging_state { |
| 102 | void __iomem *mmio_base; |
| 103 | unsigned int ucode_len; |
| 104 | unsigned int chunk_size; |
| 105 | unsigned int bytes_sent; |
| 106 | unsigned int offset; |
| 107 | }; |
| 108 | |
| 109 | #define DEFAULT_UCODE_TOTALSIZE (DEFAULT_UCODE_DATASIZE + MC_HEADER_SIZE) |
| 110 | #define (sizeof(struct extended_sigtable)) |
| 111 | #define EXT_SIGNATURE_SIZE (sizeof(struct extended_signature)) |
| 112 | |
| 113 | static inline unsigned int get_totalsize(struct microcode_header_intel *hdr) |
| 114 | { |
| 115 | return hdr->datasize ? hdr->totalsize : DEFAULT_UCODE_TOTALSIZE; |
| 116 | } |
| 117 | |
| 118 | static inline unsigned int exttable_size(struct extended_sigtable *et) |
| 119 | { |
| 120 | return et->count * EXT_SIGNATURE_SIZE + EXT_HEADER_SIZE; |
| 121 | } |
| 122 | |
| 123 | void intel_collect_cpu_info(struct cpu_signature *sig) |
| 124 | { |
| 125 | sig->sig = cpuid_eax(op: 1); |
| 126 | sig->pf = 0; |
| 127 | sig->rev = intel_get_microcode_revision(); |
| 128 | |
| 129 | if (IFM(x86_family(sig->sig), x86_model(sig->sig)) >= INTEL_PENTIUM_III_DESCHUTES) { |
| 130 | unsigned int val[2]; |
| 131 | |
| 132 | /* get processor flags from MSR 0x17 */ |
| 133 | native_rdmsr(MSR_IA32_PLATFORM_ID, val[0], val[1]); |
| 134 | sig->pf = 1 << ((val[1] >> 18) & 7); |
| 135 | } |
| 136 | } |
| 137 | EXPORT_SYMBOL_GPL(intel_collect_cpu_info); |
| 138 | |
| 139 | static inline bool cpu_signatures_match(struct cpu_signature *s1, unsigned int sig2, |
| 140 | unsigned int pf2) |
| 141 | { |
| 142 | if (s1->sig != sig2) |
| 143 | return false; |
| 144 | |
| 145 | /* Processor flags are either both 0 or they intersect. */ |
| 146 | return ((!s1->pf && !pf2) || (s1->pf & pf2)); |
| 147 | } |
| 148 | |
| 149 | bool intel_find_matching_signature(void *mc, struct cpu_signature *sig) |
| 150 | { |
| 151 | struct microcode_header_intel *mc_hdr = mc; |
| 152 | struct extended_signature *ext_sig; |
| 153 | struct extended_sigtable *ext_hdr; |
| 154 | int i; |
| 155 | |
| 156 | if (cpu_signatures_match(s1: sig, sig2: mc_hdr->sig, pf2: mc_hdr->pf)) |
| 157 | return true; |
| 158 | |
| 159 | /* Look for ext. headers: */ |
| 160 | if (get_totalsize(hdr: mc_hdr) <= intel_microcode_get_datasize(hdr: mc_hdr) + MC_HEADER_SIZE) |
| 161 | return false; |
| 162 | |
| 163 | ext_hdr = mc + intel_microcode_get_datasize(hdr: mc_hdr) + MC_HEADER_SIZE; |
| 164 | ext_sig = (void *)ext_hdr + EXT_HEADER_SIZE; |
| 165 | |
| 166 | for (i = 0; i < ext_hdr->count; i++) { |
| 167 | if (cpu_signatures_match(s1: sig, sig2: ext_sig->sig, pf2: ext_sig->pf)) |
| 168 | return true; |
| 169 | ext_sig++; |
| 170 | } |
| 171 | return 0; |
| 172 | } |
| 173 | EXPORT_SYMBOL_GPL(intel_find_matching_signature); |
| 174 | |
| 175 | /** |
| 176 | * intel_microcode_sanity_check() - Sanity check microcode file. |
| 177 | * @mc: Pointer to the microcode file contents. |
| 178 | * @print_err: Display failure reason if true, silent if false. |
| 179 | * @hdr_type: Type of file, i.e. normal microcode file or In Field Scan file. |
| 180 | * Validate if the microcode header type matches with the type |
| 181 | * specified here. |
| 182 | * |
| 183 | * Validate certain header fields and verify if computed checksum matches |
| 184 | * with the one specified in the header. |
| 185 | * |
| 186 | * Return: 0 if the file passes all the checks, -EINVAL if any of the checks |
| 187 | * fail. |
| 188 | */ |
| 189 | int intel_microcode_sanity_check(void *mc, bool print_err, int hdr_type) |
| 190 | { |
| 191 | unsigned long total_size, data_size, ext_table_size; |
| 192 | struct microcode_header_intel * = mc; |
| 193 | struct extended_sigtable * = NULL; |
| 194 | u32 sum, orig_sum, ext_sigcount = 0, i; |
| 195 | struct extended_signature *ext_sig; |
| 196 | |
| 197 | total_size = get_totalsize(hdr: mc_header); |
| 198 | data_size = intel_microcode_get_datasize(hdr: mc_header); |
| 199 | |
| 200 | if (data_size + MC_HEADER_SIZE > total_size) { |
| 201 | if (print_err) |
| 202 | pr_err("Error: bad microcode data file size.\n" ); |
| 203 | return -EINVAL; |
| 204 | } |
| 205 | |
| 206 | if (mc_header->ldrver != 1 || mc_header->hdrver != hdr_type) { |
| 207 | if (print_err) |
| 208 | pr_err("Error: invalid/unknown microcode update format. Header type %d\n" , |
| 209 | mc_header->hdrver); |
| 210 | return -EINVAL; |
| 211 | } |
| 212 | |
| 213 | ext_table_size = total_size - (MC_HEADER_SIZE + data_size); |
| 214 | if (ext_table_size) { |
| 215 | u32 ext_table_sum = 0; |
| 216 | u32 *ext_tablep; |
| 217 | |
| 218 | if (ext_table_size < EXT_HEADER_SIZE || |
| 219 | ((ext_table_size - EXT_HEADER_SIZE) % EXT_SIGNATURE_SIZE)) { |
| 220 | if (print_err) |
| 221 | pr_err("Error: truncated extended signature table.\n" ); |
| 222 | return -EINVAL; |
| 223 | } |
| 224 | |
| 225 | ext_header = mc + MC_HEADER_SIZE + data_size; |
| 226 | if (ext_table_size != exttable_size(et: ext_header)) { |
| 227 | if (print_err) |
| 228 | pr_err("Error: extended signature table size mismatch.\n" ); |
| 229 | return -EFAULT; |
| 230 | } |
| 231 | |
| 232 | ext_sigcount = ext_header->count; |
| 233 | |
| 234 | /* |
| 235 | * Check extended table checksum: the sum of all dwords that |
| 236 | * comprise a valid table must be 0. |
| 237 | */ |
| 238 | ext_tablep = (u32 *)ext_header; |
| 239 | |
| 240 | i = ext_table_size / sizeof(u32); |
| 241 | while (i--) |
| 242 | ext_table_sum += ext_tablep[i]; |
| 243 | |
| 244 | if (ext_table_sum) { |
| 245 | if (print_err) |
| 246 | pr_warn("Bad extended signature table checksum, aborting.\n" ); |
| 247 | return -EINVAL; |
| 248 | } |
| 249 | } |
| 250 | |
| 251 | /* |
| 252 | * Calculate the checksum of update data and header. The checksum of |
| 253 | * valid update data and header including the extended signature table |
| 254 | * must be 0. |
| 255 | */ |
| 256 | orig_sum = 0; |
| 257 | i = (MC_HEADER_SIZE + data_size) / sizeof(u32); |
| 258 | while (i--) |
| 259 | orig_sum += ((u32 *)mc)[i]; |
| 260 | |
| 261 | if (orig_sum) { |
| 262 | if (print_err) |
| 263 | pr_err("Bad microcode data checksum, aborting.\n" ); |
| 264 | return -EINVAL; |
| 265 | } |
| 266 | |
| 267 | if (!ext_table_size) |
| 268 | return 0; |
| 269 | |
| 270 | /* |
| 271 | * Check extended signature checksum: 0 => valid. |
| 272 | */ |
| 273 | for (i = 0; i < ext_sigcount; i++) { |
| 274 | ext_sig = (void *)ext_header + EXT_HEADER_SIZE + |
| 275 | EXT_SIGNATURE_SIZE * i; |
| 276 | |
| 277 | sum = (mc_header->sig + mc_header->pf + mc_header->cksum) - |
| 278 | (ext_sig->sig + ext_sig->pf + ext_sig->cksum); |
| 279 | if (sum) { |
| 280 | if (print_err) |
| 281 | pr_err("Bad extended signature checksum, aborting.\n" ); |
| 282 | return -EINVAL; |
| 283 | } |
| 284 | } |
| 285 | return 0; |
| 286 | } |
| 287 | EXPORT_SYMBOL_GPL(intel_microcode_sanity_check); |
| 288 | |
| 289 | static void update_ucode_pointer(struct microcode_intel *mc) |
| 290 | { |
| 291 | kvfree(addr: ucode_patch_va); |
| 292 | |
| 293 | /* |
| 294 | * Save the virtual address for early loading and for eventual free |
| 295 | * on late loading. |
| 296 | */ |
| 297 | ucode_patch_va = mc; |
| 298 | } |
| 299 | |
| 300 | static void save_microcode_patch(struct microcode_intel *patch) |
| 301 | { |
| 302 | unsigned int size = get_totalsize(hdr: &patch->hdr); |
| 303 | struct microcode_intel *mc; |
| 304 | |
| 305 | mc = kvmemdup(src: patch, len: size, GFP_KERNEL); |
| 306 | if (mc) |
| 307 | update_ucode_pointer(mc); |
| 308 | else |
| 309 | pr_err("Unable to allocate microcode memory size: %u\n" , size); |
| 310 | } |
| 311 | |
| 312 | /* Scan blob for microcode matching the boot CPUs family, model, stepping */ |
| 313 | static __init struct microcode_intel *scan_microcode(void *data, size_t size, |
| 314 | struct ucode_cpu_info *uci, |
| 315 | bool save) |
| 316 | { |
| 317 | struct microcode_header_intel *; |
| 318 | struct microcode_intel *patch = NULL; |
| 319 | u32 cur_rev = uci->cpu_sig.rev; |
| 320 | unsigned int mc_size; |
| 321 | |
| 322 | for (; size >= sizeof(struct microcode_header_intel); size -= mc_size, data += mc_size) { |
| 323 | mc_header = (struct microcode_header_intel *)data; |
| 324 | |
| 325 | mc_size = get_totalsize(hdr: mc_header); |
| 326 | if (!mc_size || mc_size > size || |
| 327 | intel_microcode_sanity_check(data, false, MC_HEADER_TYPE_MICROCODE) < 0) |
| 328 | break; |
| 329 | |
| 330 | if (!intel_find_matching_signature(data, &uci->cpu_sig)) |
| 331 | continue; |
| 332 | |
| 333 | /* |
| 334 | * For saving the early microcode, find the matching revision which |
| 335 | * was loaded on the BSP. |
| 336 | * |
| 337 | * On the BSP during early boot, find a newer revision than |
| 338 | * actually loaded in the CPU. |
| 339 | */ |
| 340 | if (save) { |
| 341 | if (cur_rev != mc_header->rev) |
| 342 | continue; |
| 343 | } else if (cur_rev >= mc_header->rev) { |
| 344 | continue; |
| 345 | } |
| 346 | |
| 347 | patch = data; |
| 348 | cur_rev = mc_header->rev; |
| 349 | } |
| 350 | |
| 351 | return size ? NULL : patch; |
| 352 | } |
| 353 | |
| 354 | static inline u32 read_mbox_dword(void __iomem *mmio_base) |
| 355 | { |
| 356 | u32 dword = readl(addr: mmio_base + MBOX_RDDATA_OFFSET); |
| 357 | |
| 358 | /* Acknowledge read completion to the staging hardware */ |
| 359 | writel(val: 0, addr: mmio_base + MBOX_RDDATA_OFFSET); |
| 360 | return dword; |
| 361 | } |
| 362 | |
| 363 | static inline void write_mbox_dword(void __iomem *mmio_base, u32 dword) |
| 364 | { |
| 365 | writel(val: dword, addr: mmio_base + MBOX_WRDATA_OFFSET); |
| 366 | } |
| 367 | |
| 368 | static inline u64 (void __iomem *mmio_base) |
| 369 | { |
| 370 | u32 high, low; |
| 371 | |
| 372 | low = read_mbox_dword(mmio_base); |
| 373 | high = read_mbox_dword(mmio_base); |
| 374 | |
| 375 | return ((u64)high << 32) | low; |
| 376 | } |
| 377 | |
| 378 | static inline void (void __iomem *mmio_base, u64 value) |
| 379 | { |
| 380 | write_mbox_dword(mmio_base, dword: value); |
| 381 | write_mbox_dword(mmio_base, dword: value >> 32); |
| 382 | } |
| 383 | |
| 384 | static void write_mbox_data(void __iomem *mmio_base, u32 *chunk, unsigned int chunk_bytes) |
| 385 | { |
| 386 | int i; |
| 387 | |
| 388 | /* |
| 389 | * The MMIO space is mapped as Uncached (UC). Each write arrives |
| 390 | * at the device as an individual transaction in program order. |
| 391 | * The device can then reassemble the sequence accordingly. |
| 392 | */ |
| 393 | for (i = 0; i < chunk_bytes / sizeof(u32); i++) |
| 394 | write_mbox_dword(mmio_base, dword: chunk[i]); |
| 395 | } |
| 396 | |
| 397 | /* |
| 398 | * Prepare for a new microcode transfer: reset hardware and record the |
| 399 | * image size. |
| 400 | */ |
| 401 | static void init_stage(struct staging_state *ss) |
| 402 | { |
| 403 | ss->ucode_len = get_totalsize(hdr: &ucode_patch_late->hdr); |
| 404 | |
| 405 | /* |
| 406 | * Abort any ongoing process, effectively resetting the device. |
| 407 | * Unlike regular mailbox data processing requests, this |
| 408 | * operation does not require a status check. |
| 409 | */ |
| 410 | writel(MASK_MBOX_CTRL_ABORT, addr: ss->mmio_base + MBOX_CONTROL_OFFSET); |
| 411 | } |
| 412 | |
| 413 | /* |
| 414 | * Update the chunk size and decide whether another chunk can be sent. |
| 415 | * This accounts for remaining data and retry limits. |
| 416 | */ |
| 417 | static bool can_send_next_chunk(struct staging_state *ss, int *err) |
| 418 | { |
| 419 | /* A page size or remaining bytes if this is the final chunk */ |
| 420 | ss->chunk_size = min(PAGE_SIZE, ss->ucode_len - ss->offset); |
| 421 | |
| 422 | /* |
| 423 | * Each microcode image is divided into chunks, each at most |
| 424 | * one page size. A 10-chunk image would typically require 10 |
| 425 | * transactions. |
| 426 | * |
| 427 | * However, the hardware managing the mailbox has limited |
| 428 | * resources and may not cache the entire image, potentially |
| 429 | * requesting the same chunk multiple times. |
| 430 | * |
| 431 | * To tolerate this behavior, allow up to twice the expected |
| 432 | * number of transactions (i.e., a 10-chunk image can take up to |
| 433 | * 20 attempts). |
| 434 | * |
| 435 | * If the number of attempts exceeds this limit, treat it as |
| 436 | * exceeding the maximum allowed transfer size. |
| 437 | */ |
| 438 | if (ss->bytes_sent + ss->chunk_size > ss->ucode_len * 2) { |
| 439 | *err = -EMSGSIZE; |
| 440 | return false; |
| 441 | } |
| 442 | |
| 443 | *err = 0; |
| 444 | return true; |
| 445 | } |
| 446 | |
| 447 | /* |
| 448 | * The hardware indicates completion by returning a sentinel end offset. |
| 449 | */ |
| 450 | static inline bool is_end_offset(u32 offset) |
| 451 | { |
| 452 | return offset == UINT_MAX; |
| 453 | } |
| 454 | |
| 455 | /* |
| 456 | * Determine whether staging is complete: either the hardware signaled |
| 457 | * the end offset, or no more transactions are permitted (retry limit |
| 458 | * reached). |
| 459 | */ |
| 460 | static inline bool staging_is_complete(struct staging_state *ss, int *err) |
| 461 | { |
| 462 | return is_end_offset(offset: ss->offset) || !can_send_next_chunk(ss, err); |
| 463 | } |
| 464 | |
| 465 | /* |
| 466 | * Wait for the hardware to complete a transaction. |
| 467 | * Return 0 on success, or an error code on failure. |
| 468 | */ |
| 469 | static int wait_for_transaction(struct staging_state *ss) |
| 470 | { |
| 471 | u32 timeout, status; |
| 472 | |
| 473 | /* Allow time for hardware to complete the operation: */ |
| 474 | for (timeout = 0; timeout < MBOX_XACTION_TIMEOUT_MS; timeout++) { |
| 475 | msleep(msecs: 1); |
| 476 | |
| 477 | status = readl(addr: ss->mmio_base + MBOX_STATUS_OFFSET); |
| 478 | /* Break out early if the hardware is ready: */ |
| 479 | if (status & MASK_MBOX_STATUS_READY) |
| 480 | break; |
| 481 | } |
| 482 | |
| 483 | /* Check for explicit error response */ |
| 484 | if (status & MASK_MBOX_STATUS_ERROR) |
| 485 | return -EIO; |
| 486 | |
| 487 | /* |
| 488 | * Hardware has neither responded to the action nor signaled any |
| 489 | * error. Treat this as a timeout. |
| 490 | */ |
| 491 | if (!(status & MASK_MBOX_STATUS_READY)) |
| 492 | return -ETIMEDOUT; |
| 493 | |
| 494 | return 0; |
| 495 | } |
| 496 | |
| 497 | /* |
| 498 | * Transmit a chunk of the microcode image to the hardware. |
| 499 | * Return 0 on success, or an error code on failure. |
| 500 | */ |
| 501 | static int send_data_chunk(struct staging_state *ss, void *ucode_ptr) |
| 502 | { |
| 503 | u32 *src_chunk = ucode_ptr + ss->offset; |
| 504 | u16 mbox_size; |
| 505 | |
| 506 | /* |
| 507 | * Write a 'request' mailbox object in this order: |
| 508 | * 1. Mailbox header includes total size |
| 509 | * 2. Command header specifies the load operation |
| 510 | * 3. Data section contains a microcode chunk |
| 511 | * |
| 512 | * Thus, the mailbox size is two headers plus the chunk size. |
| 513 | */ |
| 514 | mbox_size = MBOX_HEADER_SIZE * 2 + ss->chunk_size; |
| 515 | write_mbox_header(mmio_base: ss->mmio_base, MBOX_HEADER(mbox_size)); |
| 516 | write_mbox_header(mmio_base: ss->mmio_base, MBOX_CMD_LOAD); |
| 517 | write_mbox_data(mmio_base: ss->mmio_base, chunk: src_chunk, chunk_bytes: ss->chunk_size); |
| 518 | ss->bytes_sent += ss->chunk_size; |
| 519 | |
| 520 | /* Notify the hardware that the mailbox is ready for processing. */ |
| 521 | writel(MASK_MBOX_CTRL_GO, addr: ss->mmio_base + MBOX_CONTROL_OFFSET); |
| 522 | |
| 523 | return wait_for_transaction(ss); |
| 524 | } |
| 525 | |
| 526 | /* |
| 527 | * Retrieve the next offset from the hardware response. |
| 528 | * Return 0 on success, or an error code on failure. |
| 529 | */ |
| 530 | static int fetch_next_offset(struct staging_state *ss) |
| 531 | { |
| 532 | const u64 = MBOX_HEADER(MBOX_HEADER_SIZE + MBOX_RESPONSE_SIZE); |
| 533 | u32 offset, status; |
| 534 | u64 ; |
| 535 | |
| 536 | /* |
| 537 | * The 'response' mailbox returns three fields, in order: |
| 538 | * 1. Header |
| 539 | * 2. Next offset in the microcode image |
| 540 | * 3. Status flags |
| 541 | */ |
| 542 | header = read_mbox_header(mmio_base: ss->mmio_base); |
| 543 | offset = read_mbox_dword(mmio_base: ss->mmio_base); |
| 544 | status = read_mbox_dword(mmio_base: ss->mmio_base); |
| 545 | |
| 546 | /* All valid responses must start with the expected header. */ |
| 547 | if (header != expected_header) { |
| 548 | pr_err_once("staging: invalid response header (0x%llx)\n" , header); |
| 549 | return -EBADR; |
| 550 | } |
| 551 | |
| 552 | /* |
| 553 | * Verify the offset: If not at the end marker, it must not |
| 554 | * exceed the microcode image length. |
| 555 | */ |
| 556 | if (!is_end_offset(offset) && offset > ss->ucode_len) { |
| 557 | pr_err_once("staging: invalid offset (%u) past the image end (%u)\n" , |
| 558 | offset, ss->ucode_len); |
| 559 | return -EINVAL; |
| 560 | } |
| 561 | |
| 562 | /* Hardware may report errors explicitly in the status field */ |
| 563 | if (status & MASK_MBOX_RESP_ERROR) |
| 564 | return -EPROTO; |
| 565 | |
| 566 | ss->offset = offset; |
| 567 | return 0; |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * Handle the staging process using the mailbox MMIO interface. The |
| 572 | * microcode image is transferred in chunks until completion. |
| 573 | * Return 0 on success or an error code on failure. |
| 574 | */ |
| 575 | static int do_stage(u64 mmio_pa) |
| 576 | { |
| 577 | struct staging_state ss = {}; |
| 578 | int err; |
| 579 | |
| 580 | ss.mmio_base = ioremap(offset: mmio_pa, MBOX_REG_NUM * MBOX_REG_SIZE); |
| 581 | if (WARN_ON_ONCE(!ss.mmio_base)) |
| 582 | return -EADDRNOTAVAIL; |
| 583 | |
| 584 | init_stage(ss: &ss); |
| 585 | |
| 586 | /* Perform the staging process while within the retry limit */ |
| 587 | while (!staging_is_complete(ss: &ss, err: &err)) { |
| 588 | /* Send a chunk of microcode each time: */ |
| 589 | err = send_data_chunk(ss: &ss, ucode_ptr: ucode_patch_late); |
| 590 | if (err) |
| 591 | break; |
| 592 | /* |
| 593 | * Then, ask the hardware which piece of the image it |
| 594 | * needs next. The same piece may be sent more than once. |
| 595 | */ |
| 596 | err = fetch_next_offset(ss: &ss); |
| 597 | if (err) |
| 598 | break; |
| 599 | } |
| 600 | |
| 601 | iounmap(addr: ss.mmio_base); |
| 602 | |
| 603 | return err; |
| 604 | } |
| 605 | |
| 606 | static void stage_microcode(void) |
| 607 | { |
| 608 | unsigned int pkg_id = UINT_MAX; |
| 609 | int cpu, err; |
| 610 | u64 mmio_pa; |
| 611 | |
| 612 | if (!IS_ALIGNED(get_totalsize(&ucode_patch_late->hdr), sizeof(u32))) { |
| 613 | pr_err("Microcode image 32-bit misaligned (0x%x), staging failed.\n" , |
| 614 | get_totalsize(&ucode_patch_late->hdr)); |
| 615 | return; |
| 616 | } |
| 617 | |
| 618 | lockdep_assert_cpus_held(); |
| 619 | |
| 620 | /* |
| 621 | * The MMIO address is unique per package, and all the SMT |
| 622 | * primary threads are online here. Find each MMIO space by |
| 623 | * their package IDs to avoid duplicate staging. |
| 624 | */ |
| 625 | for_each_cpu(cpu, cpu_primary_thread_mask) { |
| 626 | if (topology_logical_package_id(cpu) == pkg_id) |
| 627 | continue; |
| 628 | |
| 629 | pkg_id = topology_logical_package_id(cpu); |
| 630 | |
| 631 | err = rdmsrq_on_cpu(cpu, MSR_IA32_MCU_STAGING_MBOX_ADDR, q: &mmio_pa); |
| 632 | if (WARN_ON_ONCE(err)) |
| 633 | return; |
| 634 | |
| 635 | err = do_stage(mmio_pa); |
| 636 | if (err) { |
| 637 | pr_err("Error: staging failed (%d) for CPU%d at package %u.\n" , |
| 638 | err, cpu, pkg_id); |
| 639 | return; |
| 640 | } |
| 641 | } |
| 642 | |
| 643 | pr_info("Staging of patch revision 0x%x succeeded.\n" , ucode_patch_late->hdr.rev); |
| 644 | } |
| 645 | |
| 646 | static enum ucode_state __apply_microcode(struct ucode_cpu_info *uci, |
| 647 | struct microcode_intel *mc, |
| 648 | u32 *cur_rev) |
| 649 | { |
| 650 | u32 rev; |
| 651 | |
| 652 | if (!mc) |
| 653 | return UCODE_NFOUND; |
| 654 | |
| 655 | /* |
| 656 | * Save us the MSR write below - which is a particular expensive |
| 657 | * operation - when the other hyperthread has updated the microcode |
| 658 | * already. |
| 659 | */ |
| 660 | *cur_rev = intel_get_microcode_revision(); |
| 661 | if (*cur_rev >= mc->hdr.rev) { |
| 662 | uci->cpu_sig.rev = *cur_rev; |
| 663 | return UCODE_OK; |
| 664 | } |
| 665 | |
| 666 | /* write microcode via MSR 0x79 */ |
| 667 | native_wrmsrq(MSR_IA32_UCODE_WRITE, (unsigned long)mc->bits); |
| 668 | |
| 669 | rev = intel_get_microcode_revision(); |
| 670 | if (rev != mc->hdr.rev) |
| 671 | return UCODE_ERROR; |
| 672 | |
| 673 | uci->cpu_sig.rev = rev; |
| 674 | return UCODE_UPDATED; |
| 675 | } |
| 676 | |
| 677 | static enum ucode_state apply_microcode_early(struct ucode_cpu_info *uci) |
| 678 | { |
| 679 | struct microcode_intel *mc = uci->mc; |
| 680 | u32 cur_rev; |
| 681 | |
| 682 | return __apply_microcode(uci, mc, cur_rev: &cur_rev); |
| 683 | } |
| 684 | |
| 685 | static __init bool load_builtin_intel_microcode(struct cpio_data *cp) |
| 686 | { |
| 687 | unsigned int eax = 1, ebx, ecx = 0, edx; |
| 688 | struct firmware fw; |
| 689 | char name[30]; |
| 690 | |
| 691 | if (IS_ENABLED(CONFIG_X86_32)) |
| 692 | return false; |
| 693 | |
| 694 | native_cpuid(eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx); |
| 695 | |
| 696 | sprintf(buf: name, fmt: "intel-ucode/%02x-%02x-%02x" , |
| 697 | x86_family(sig: eax), x86_model(sig: eax), x86_stepping(sig: eax)); |
| 698 | |
| 699 | if (firmware_request_builtin(fw: &fw, name)) { |
| 700 | cp->size = fw.size; |
| 701 | cp->data = (void *)fw.data; |
| 702 | return true; |
| 703 | } |
| 704 | return false; |
| 705 | } |
| 706 | |
| 707 | static __init struct microcode_intel *get_microcode_blob(struct ucode_cpu_info *uci, bool save) |
| 708 | { |
| 709 | struct cpio_data cp; |
| 710 | |
| 711 | intel_collect_cpu_info(&uci->cpu_sig); |
| 712 | |
| 713 | if (!load_builtin_intel_microcode(cp: &cp)) |
| 714 | cp = find_microcode_in_initrd(path: ucode_path); |
| 715 | |
| 716 | if (!(cp.data && cp.size)) |
| 717 | return NULL; |
| 718 | |
| 719 | return scan_microcode(data: cp.data, size: cp.size, uci, save); |
| 720 | } |
| 721 | |
| 722 | /* |
| 723 | * Invoked from an early init call to save the microcode blob which was |
| 724 | * selected during early boot when mm was not usable. The microcode must be |
| 725 | * saved because initrd is going away. It's an early init call so the APs |
| 726 | * just can use the pointer and do not have to scan initrd/builtin firmware |
| 727 | * again. |
| 728 | */ |
| 729 | static int __init save_builtin_microcode(void) |
| 730 | { |
| 731 | struct ucode_cpu_info uci; |
| 732 | |
| 733 | if (xchg(&ucode_patch_va, NULL) != UCODE_BSP_LOADED) |
| 734 | return 0; |
| 735 | |
| 736 | if (microcode_loader_disabled() || boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) |
| 737 | return 0; |
| 738 | |
| 739 | uci.mc = get_microcode_blob(uci: &uci, save: true); |
| 740 | if (uci.mc) |
| 741 | save_microcode_patch(patch: uci.mc); |
| 742 | return 0; |
| 743 | } |
| 744 | early_initcall(save_builtin_microcode); |
| 745 | |
| 746 | /* Load microcode on BSP from initrd or builtin blobs */ |
| 747 | void __init load_ucode_intel_bsp(struct early_load_data *ed) |
| 748 | { |
| 749 | struct ucode_cpu_info uci; |
| 750 | |
| 751 | uci.mc = get_microcode_blob(uci: &uci, save: false); |
| 752 | ed->old_rev = uci.cpu_sig.rev; |
| 753 | |
| 754 | if (uci.mc && apply_microcode_early(uci: &uci) == UCODE_UPDATED) { |
| 755 | ucode_patch_va = UCODE_BSP_LOADED; |
| 756 | ed->new_rev = uci.cpu_sig.rev; |
| 757 | } |
| 758 | } |
| 759 | |
| 760 | void load_ucode_intel_ap(void) |
| 761 | { |
| 762 | struct ucode_cpu_info uci; |
| 763 | |
| 764 | uci.mc = ucode_patch_va; |
| 765 | if (uci.mc) |
| 766 | apply_microcode_early(uci: &uci); |
| 767 | } |
| 768 | |
| 769 | /* Reload microcode on resume */ |
| 770 | void reload_ucode_intel(void) |
| 771 | { |
| 772 | struct ucode_cpu_info uci = { .mc = ucode_patch_va, }; |
| 773 | |
| 774 | if (uci.mc) |
| 775 | apply_microcode_early(uci: &uci); |
| 776 | } |
| 777 | |
| 778 | static int collect_cpu_info(int cpu_num, struct cpu_signature *csig) |
| 779 | { |
| 780 | intel_collect_cpu_info(csig); |
| 781 | return 0; |
| 782 | } |
| 783 | |
| 784 | static enum ucode_state apply_microcode_late(int cpu) |
| 785 | { |
| 786 | struct ucode_cpu_info *uci = ucode_cpu_info + cpu; |
| 787 | struct microcode_intel *mc = ucode_patch_late; |
| 788 | enum ucode_state ret; |
| 789 | u32 cur_rev; |
| 790 | |
| 791 | if (WARN_ON_ONCE(smp_processor_id() != cpu)) |
| 792 | return UCODE_ERROR; |
| 793 | |
| 794 | ret = __apply_microcode(uci, mc, cur_rev: &cur_rev); |
| 795 | if (ret != UCODE_UPDATED && ret != UCODE_OK) |
| 796 | return ret; |
| 797 | |
| 798 | cpu_data(cpu).microcode = uci->cpu_sig.rev; |
| 799 | if (!cpu) |
| 800 | boot_cpu_data.microcode = uci->cpu_sig.rev; |
| 801 | |
| 802 | return ret; |
| 803 | } |
| 804 | |
| 805 | static bool ucode_validate_minrev(struct microcode_header_intel *) |
| 806 | { |
| 807 | int cur_rev = boot_cpu_data.microcode; |
| 808 | |
| 809 | /* |
| 810 | * When late-loading, ensure the header declares a minimum revision |
| 811 | * required to perform a late-load. The previously reserved field |
| 812 | * is 0 in older microcode blobs. |
| 813 | */ |
| 814 | if (!mc_header->min_req_ver) { |
| 815 | pr_info("Unsafe microcode update: Microcode header does not specify a required min version\n" ); |
| 816 | return false; |
| 817 | } |
| 818 | |
| 819 | /* |
| 820 | * Check whether the current revision is either greater or equal to |
| 821 | * to the minimum revision specified in the header. |
| 822 | */ |
| 823 | if (cur_rev < mc_header->min_req_ver) { |
| 824 | pr_info("Unsafe microcode update: Current revision 0x%x too old\n" , cur_rev); |
| 825 | pr_info("Current should be at 0x%x or higher. Use early loading instead\n" , mc_header->min_req_ver); |
| 826 | return false; |
| 827 | } |
| 828 | return true; |
| 829 | } |
| 830 | |
| 831 | static enum ucode_state parse_microcode_blobs(int cpu, struct iov_iter *iter) |
| 832 | { |
| 833 | struct ucode_cpu_info *uci = ucode_cpu_info + cpu; |
| 834 | bool is_safe, new_is_safe = false; |
| 835 | int cur_rev = uci->cpu_sig.rev; |
| 836 | unsigned int curr_mc_size = 0; |
| 837 | u8 *new_mc = NULL, *mc = NULL; |
| 838 | |
| 839 | while (iov_iter_count(i: iter)) { |
| 840 | struct microcode_header_intel ; |
| 841 | unsigned int mc_size, data_size; |
| 842 | u8 *data; |
| 843 | |
| 844 | if (!copy_from_iter_full(addr: &mc_header, bytes: sizeof(mc_header), i: iter)) { |
| 845 | pr_err("error! Truncated or inaccessible header in microcode data file\n" ); |
| 846 | goto fail; |
| 847 | } |
| 848 | |
| 849 | mc_size = get_totalsize(hdr: &mc_header); |
| 850 | if (mc_size < sizeof(mc_header)) { |
| 851 | pr_err("error! Bad data in microcode data file (totalsize too small)\n" ); |
| 852 | goto fail; |
| 853 | } |
| 854 | data_size = mc_size - sizeof(mc_header); |
| 855 | if (data_size > iov_iter_count(i: iter)) { |
| 856 | pr_err("error! Bad data in microcode data file (truncated file?)\n" ); |
| 857 | goto fail; |
| 858 | } |
| 859 | |
| 860 | /* For performance reasons, reuse mc area when possible */ |
| 861 | if (!mc || mc_size > curr_mc_size) { |
| 862 | kvfree(addr: mc); |
| 863 | mc = kvmalloc(mc_size, GFP_KERNEL); |
| 864 | if (!mc) |
| 865 | goto fail; |
| 866 | curr_mc_size = mc_size; |
| 867 | } |
| 868 | |
| 869 | memcpy(mc, &mc_header, sizeof(mc_header)); |
| 870 | data = mc + sizeof(mc_header); |
| 871 | if (!copy_from_iter_full(addr: data, bytes: data_size, i: iter) || |
| 872 | intel_microcode_sanity_check(mc, true, MC_HEADER_TYPE_MICROCODE) < 0) |
| 873 | goto fail; |
| 874 | |
| 875 | if (cur_rev >= mc_header.rev) |
| 876 | continue; |
| 877 | |
| 878 | if (!intel_find_matching_signature(mc, &uci->cpu_sig)) |
| 879 | continue; |
| 880 | |
| 881 | is_safe = ucode_validate_minrev(mc_header: &mc_header); |
| 882 | if (force_minrev && !is_safe) |
| 883 | continue; |
| 884 | |
| 885 | kvfree(addr: new_mc); |
| 886 | cur_rev = mc_header.rev; |
| 887 | new_mc = mc; |
| 888 | new_is_safe = is_safe; |
| 889 | mc = NULL; |
| 890 | } |
| 891 | |
| 892 | if (iov_iter_count(i: iter)) |
| 893 | goto fail; |
| 894 | |
| 895 | kvfree(addr: mc); |
| 896 | if (!new_mc) |
| 897 | return UCODE_NFOUND; |
| 898 | |
| 899 | ucode_patch_late = (struct microcode_intel *)new_mc; |
| 900 | return new_is_safe ? UCODE_NEW_SAFE : UCODE_NEW; |
| 901 | |
| 902 | fail: |
| 903 | kvfree(addr: mc); |
| 904 | kvfree(addr: new_mc); |
| 905 | return UCODE_ERROR; |
| 906 | } |
| 907 | |
| 908 | static bool is_blacklisted(unsigned int cpu) |
| 909 | { |
| 910 | struct cpuinfo_x86 *c = &cpu_data(cpu); |
| 911 | |
| 912 | /* |
| 913 | * Late loading on model 79 with microcode revision less than 0x0b000021 |
| 914 | * and LLC size per core bigger than 2.5MB may result in a system hang. |
| 915 | * This behavior is documented in item BDX90, #334165 (Intel Xeon |
| 916 | * Processor E7-8800/4800 v4 Product Family). |
| 917 | */ |
| 918 | if (c->x86_vfm == INTEL_BROADWELL_X && |
| 919 | c->x86_stepping == 0x01 && |
| 920 | llc_size_per_core > 2621440 && |
| 921 | c->microcode < 0x0b000021) { |
| 922 | pr_err_once("Erratum BDX90: late loading with revision < 0x0b000021 (0x%x) disabled.\n" , c->microcode); |
| 923 | pr_err_once("Please consider either early loading through initrd/built-in or a potential BIOS update.\n" ); |
| 924 | return true; |
| 925 | } |
| 926 | |
| 927 | return false; |
| 928 | } |
| 929 | |
| 930 | static enum ucode_state request_microcode_fw(int cpu, struct device *device) |
| 931 | { |
| 932 | struct cpuinfo_x86 *c = &cpu_data(cpu); |
| 933 | const struct firmware *firmware; |
| 934 | struct iov_iter iter; |
| 935 | enum ucode_state ret; |
| 936 | struct kvec kvec; |
| 937 | char name[30]; |
| 938 | |
| 939 | if (is_blacklisted(cpu)) |
| 940 | return UCODE_NFOUND; |
| 941 | |
| 942 | sprintf(buf: name, fmt: "intel-ucode/%02x-%02x-%02x" , |
| 943 | c->x86, c->x86_model, c->x86_stepping); |
| 944 | |
| 945 | if (request_firmware_direct(fw: &firmware, name, device)) { |
| 946 | pr_debug("data file %s load failed\n" , name); |
| 947 | return UCODE_NFOUND; |
| 948 | } |
| 949 | |
| 950 | kvec.iov_base = (void *)firmware->data; |
| 951 | kvec.iov_len = firmware->size; |
| 952 | iov_iter_kvec(i: &iter, ITER_SOURCE, kvec: &kvec, nr_segs: 1, count: firmware->size); |
| 953 | ret = parse_microcode_blobs(cpu, iter: &iter); |
| 954 | |
| 955 | release_firmware(fw: firmware); |
| 956 | |
| 957 | return ret; |
| 958 | } |
| 959 | |
| 960 | static void finalize_late_load(int result) |
| 961 | { |
| 962 | if (!result) |
| 963 | update_ucode_pointer(mc: ucode_patch_late); |
| 964 | else |
| 965 | kvfree(addr: ucode_patch_late); |
| 966 | ucode_patch_late = NULL; |
| 967 | } |
| 968 | |
| 969 | static struct microcode_ops microcode_intel_ops = { |
| 970 | .request_microcode_fw = request_microcode_fw, |
| 971 | .collect_cpu_info = collect_cpu_info, |
| 972 | .apply_microcode = apply_microcode_late, |
| 973 | .finalize_late_load = finalize_late_load, |
| 974 | .stage_microcode = stage_microcode, |
| 975 | .use_nmi = IS_ENABLED(CONFIG_X86_64), |
| 976 | }; |
| 977 | |
| 978 | static __init void calc_llc_size_per_core(struct cpuinfo_x86 *c) |
| 979 | { |
| 980 | u64 llc_size = c->x86_cache_size * 1024ULL; |
| 981 | |
| 982 | do_div(llc_size, topology_num_cores_per_package()); |
| 983 | llc_size_per_core = (unsigned int)llc_size; |
| 984 | } |
| 985 | |
| 986 | static __init bool staging_available(void) |
| 987 | { |
| 988 | u64 val; |
| 989 | |
| 990 | val = x86_read_arch_cap_msr(); |
| 991 | if (!(val & ARCH_CAP_MCU_ENUM)) |
| 992 | return false; |
| 993 | |
| 994 | rdmsrq(MSR_IA32_MCU_ENUMERATION, val); |
| 995 | return !!(val & MCU_STAGING); |
| 996 | } |
| 997 | |
| 998 | struct microcode_ops * __init init_intel_microcode(void) |
| 999 | { |
| 1000 | struct cpuinfo_x86 *c = &boot_cpu_data; |
| 1001 | |
| 1002 | if (c->x86_vendor != X86_VENDOR_INTEL || c->x86 < 6 || |
| 1003 | cpu_has(c, X86_FEATURE_IA64)) { |
| 1004 | pr_err("Intel CPU family 0x%x not supported\n" , c->x86); |
| 1005 | return NULL; |
| 1006 | } |
| 1007 | |
| 1008 | if (staging_available()) { |
| 1009 | microcode_intel_ops.use_staging = true; |
| 1010 | pr_info("Enabled staging feature.\n" ); |
| 1011 | } |
| 1012 | |
| 1013 | calc_llc_size_per_core(c); |
| 1014 | |
| 1015 | return µcode_intel_ops; |
| 1016 | } |
| 1017 | |