1 /* Intel i7 core/Nehalem Memory Controller kernel module 2 * 3 * This driver supports the memory controllers found on the Intel 4 * processor families i7core, i7core 7xx/8xx, i5core, Xeon 35xx, 5 * Xeon 55xx and Xeon 56xx also known as Nehalem, Nehalem-EP, Lynnfield 6 * and Westmere-EP. 7 * 8 * This file may be distributed under the terms of the 9 * GNU General Public License version 2 only. 10 * 11 * Copyright (c) 2009-2010 by: 12 * Mauro Carvalho Chehab <mchehab@redhat.com> 13 * 14 * Red Hat Inc. http://www.redhat.com 15 * 16 * Forked and adapted from the i5400_edac driver 17 * 18 * Based on the following public Intel datasheets: 19 * Intel Core i7 Processor Extreme Edition and Intel Core i7 Processor 20 * Datasheet, Volume 2: 21 * http://download.intel.com/design/processor/datashts/320835.pdf 22 * Intel Xeon Processor 5500 Series Datasheet Volume 2 23 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf 24 * also available at: 25 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf 26 */ 27 28 #include <linux/module.h> 29 #include <linux/init.h> 30 #include <linux/pci.h> 31 #include <linux/pci_ids.h> 32 #include <linux/slab.h> 33 #include <linux/delay.h> 34 #include <linux/dmi.h> 35 #include <linux/edac.h> 36 #include <linux/mmzone.h> 37 #include <linux/smp.h> 38 #include <asm/mce.h> 39 #include <asm/processor.h> 40 #include <asm/div64.h> 41 42 #include "edac_core.h" 43 44 /* Static vars */ 45 static LIST_HEAD(i7core_edac_list); 46 static DEFINE_MUTEX(i7core_edac_lock); 47 static int probed; 48 49 static int use_pci_fixup; 50 module_param(use_pci_fixup, int, 0444); 51 MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices"); 52 /* 53 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core 54 * registers start at bus 255, and are not reported by BIOS. 55 * We currently find devices with only 2 sockets. In order to support more QPI 56 * Quick Path Interconnect, just increment this number. 57 */ 58 #define MAX_SOCKET_BUSES 2 59 60 61 /* 62 * Alter this version for the module when modifications are made 63 */ 64 #define I7CORE_REVISION " Ver: 1.0.0" 65 #define EDAC_MOD_STR "i7core_edac" 66 67 /* 68 * Debug macros 69 */ 70 #define i7core_printk(level, fmt, arg...) \ 71 edac_printk(level, "i7core", fmt, ##arg) 72 73 #define i7core_mc_printk(mci, level, fmt, arg...) \ 74 edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg) 75 76 /* 77 * i7core Memory Controller Registers 78 */ 79 80 /* OFFSETS for Device 0 Function 0 */ 81 82 #define MC_CFG_CONTROL 0x90 83 #define MC_CFG_UNLOCK 0x02 84 #define MC_CFG_LOCK 0x00 85 86 /* OFFSETS for Device 3 Function 0 */ 87 88 #define MC_CONTROL 0x48 89 #define MC_STATUS 0x4c 90 #define MC_MAX_DOD 0x64 91 92 /* 93 * OFFSETS for Device 3 Function 4, as indicated on Xeon 5500 datasheet: 94 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf 95 */ 96 97 #define MC_TEST_ERR_RCV1 0x60 98 #define DIMM2_COR_ERR(r) ((r) & 0x7fff) 99 100 #define MC_TEST_ERR_RCV0 0x64 101 #define DIMM1_COR_ERR(r) (((r) >> 16) & 0x7fff) 102 #define DIMM0_COR_ERR(r) ((r) & 0x7fff) 103 104 /* OFFSETS for Device 3 Function 2, as indicated on Xeon 5500 datasheet */ 105 #define MC_SSRCONTROL 0x48 106 #define SSR_MODE_DISABLE 0x00 107 #define SSR_MODE_ENABLE 0x01 108 #define SSR_MODE_MASK 0x03 109 110 #define MC_SCRUB_CONTROL 0x4c 111 #define STARTSCRUB (1 << 24) 112 #define SCRUBINTERVAL_MASK 0xffffff 113 114 #define MC_COR_ECC_CNT_0 0x80 115 #define MC_COR_ECC_CNT_1 0x84 116 #define MC_COR_ECC_CNT_2 0x88 117 #define MC_COR_ECC_CNT_3 0x8c 118 #define MC_COR_ECC_CNT_4 0x90 119 #define MC_COR_ECC_CNT_5 0x94 120 121 #define DIMM_TOP_COR_ERR(r) (((r) >> 16) & 0x7fff) 122 #define DIMM_BOT_COR_ERR(r) ((r) & 0x7fff) 123 124 125 /* OFFSETS for Devices 4,5 and 6 Function 0 */ 126 127 #define MC_CHANNEL_DIMM_INIT_PARAMS 0x58 128 #define THREE_DIMMS_PRESENT (1 << 24) 129 #define SINGLE_QUAD_RANK_PRESENT (1 << 23) 130 #define QUAD_RANK_PRESENT (1 << 22) 131 #define REGISTERED_DIMM (1 << 15) 132 133 #define MC_CHANNEL_MAPPER 0x60 134 #define RDLCH(r, ch) ((((r) >> (3 + (ch * 6))) & 0x07) - 1) 135 #define WRLCH(r, ch) ((((r) >> (ch * 6)) & 0x07) - 1) 136 137 #define MC_CHANNEL_RANK_PRESENT 0x7c 138 #define RANK_PRESENT_MASK 0xffff 139 140 #define MC_CHANNEL_ADDR_MATCH 0xf0 141 #define MC_CHANNEL_ERROR_MASK 0xf8 142 #define MC_CHANNEL_ERROR_INJECT 0xfc 143 #define INJECT_ADDR_PARITY 0x10 144 #define INJECT_ECC 0x08 145 #define MASK_CACHELINE 0x06 146 #define MASK_FULL_CACHELINE 0x06 147 #define MASK_MSB32_CACHELINE 0x04 148 #define MASK_LSB32_CACHELINE 0x02 149 #define NO_MASK_CACHELINE 0x00 150 #define REPEAT_EN 0x01 151 152 /* OFFSETS for Devices 4,5 and 6 Function 1 */ 153 154 #define MC_DOD_CH_DIMM0 0x48 155 #define MC_DOD_CH_DIMM1 0x4c 156 #define MC_DOD_CH_DIMM2 0x50 157 #define RANKOFFSET_MASK ((1 << 12) | (1 << 11) | (1 << 10)) 158 #define RANKOFFSET(x) ((x & RANKOFFSET_MASK) >> 10) 159 #define DIMM_PRESENT_MASK (1 << 9) 160 #define DIMM_PRESENT(x) (((x) & DIMM_PRESENT_MASK) >> 9) 161 #define MC_DOD_NUMBANK_MASK ((1 << 8) | (1 << 7)) 162 #define MC_DOD_NUMBANK(x) (((x) & MC_DOD_NUMBANK_MASK) >> 7) 163 #define MC_DOD_NUMRANK_MASK ((1 << 6) | (1 << 5)) 164 #define MC_DOD_NUMRANK(x) (((x) & MC_DOD_NUMRANK_MASK) >> 5) 165 #define MC_DOD_NUMROW_MASK ((1 << 4) | (1 << 3) | (1 << 2)) 166 #define MC_DOD_NUMROW(x) (((x) & MC_DOD_NUMROW_MASK) >> 2) 167 #define MC_DOD_NUMCOL_MASK 3 168 #define MC_DOD_NUMCOL(x) ((x) & MC_DOD_NUMCOL_MASK) 169 170 #define MC_RANK_PRESENT 0x7c 171 172 #define MC_SAG_CH_0 0x80 173 #define MC_SAG_CH_1 0x84 174 #define MC_SAG_CH_2 0x88 175 #define MC_SAG_CH_3 0x8c 176 #define MC_SAG_CH_4 0x90 177 #define MC_SAG_CH_5 0x94 178 #define MC_SAG_CH_6 0x98 179 #define MC_SAG_CH_7 0x9c 180 181 #define MC_RIR_LIMIT_CH_0 0x40 182 #define MC_RIR_LIMIT_CH_1 0x44 183 #define MC_RIR_LIMIT_CH_2 0x48 184 #define MC_RIR_LIMIT_CH_3 0x4C 185 #define MC_RIR_LIMIT_CH_4 0x50 186 #define MC_RIR_LIMIT_CH_5 0x54 187 #define MC_RIR_LIMIT_CH_6 0x58 188 #define MC_RIR_LIMIT_CH_7 0x5C 189 #define MC_RIR_LIMIT_MASK ((1 << 10) - 1) 190 191 #define MC_RIR_WAY_CH 0x80 192 #define MC_RIR_WAY_OFFSET_MASK (((1 << 14) - 1) & ~0x7) 193 #define MC_RIR_WAY_RANK_MASK 0x7 194 195 /* 196 * i7core structs 197 */ 198 199 #define NUM_CHANS 3 200 #define MAX_DIMMS 3 /* Max DIMMS per channel */ 201 #define MAX_MCR_FUNC 4 202 #define MAX_CHAN_FUNC 3 203 204 struct i7core_info { 205 u32 mc_control; 206 u32 mc_status; 207 u32 max_dod; 208 u32 ch_map; 209 }; 210 211 212 struct i7core_inject { 213 int enable; 214 215 u32 section; 216 u32 type; 217 u32 eccmask; 218 219 /* Error address mask */ 220 int channel, dimm, rank, bank, page, col; 221 }; 222 223 struct i7core_channel { 224 bool is_3dimms_present; 225 bool is_single_4rank; 226 bool has_4rank; 227 u32 dimms; 228 }; 229 230 struct pci_id_descr { 231 int dev; 232 int func; 233 int dev_id; 234 int optional; 235 }; 236 237 struct pci_id_table { 238 const struct pci_id_descr *descr; 239 int n_devs; 240 }; 241 242 struct i7core_dev { 243 struct list_head list; 244 u8 socket; 245 struct pci_dev **pdev; 246 int n_devs; 247 struct mem_ctl_info *mci; 248 }; 249 250 struct i7core_pvt { 251 struct device *addrmatch_dev, *chancounts_dev; 252 253 struct pci_dev *pci_noncore; 254 struct pci_dev *pci_mcr[MAX_MCR_FUNC + 1]; 255 struct pci_dev *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1]; 256 257 struct i7core_dev *i7core_dev; 258 259 struct i7core_info info; 260 struct i7core_inject inject; 261 struct i7core_channel channel[NUM_CHANS]; 262 263 int ce_count_available; 264 265 /* ECC corrected errors counts per udimm */ 266 unsigned long udimm_ce_count[MAX_DIMMS]; 267 int udimm_last_ce_count[MAX_DIMMS]; 268 /* ECC corrected errors counts per rdimm */ 269 unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS]; 270 int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS]; 271 272 bool is_registered, enable_scrub; 273 274 /* Fifo double buffers */ 275 struct mce mce_entry[MCE_LOG_LEN]; 276 struct mce mce_outentry[MCE_LOG_LEN]; 277 278 /* Fifo in/out counters */ 279 unsigned mce_in, mce_out; 280 281 /* Count indicator to show errors not got */ 282 unsigned mce_overrun; 283 284 /* DCLK Frequency used for computing scrub rate */ 285 int dclk_freq; 286 287 /* Struct to control EDAC polling */ 288 struct edac_pci_ctl_info *i7core_pci; 289 }; 290 291 #define PCI_DESCR(device, function, device_id) \ 292 .dev = (device), \ 293 .func = (function), \ 294 .dev_id = (device_id) 295 296 static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = { 297 /* Memory controller */ 298 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) }, 299 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) }, 300 /* Exists only for RDIMM */ 301 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 }, 302 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) }, 303 304 /* Channel 0 */ 305 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) }, 306 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) }, 307 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) }, 308 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC) }, 309 310 /* Channel 1 */ 311 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) }, 312 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) }, 313 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) }, 314 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC) }, 315 316 /* Channel 2 */ 317 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) }, 318 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) }, 319 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) }, 320 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) }, 321 322 /* Generic Non-core registers */ 323 /* 324 * This is the PCI device on i7core and on Xeon 35xx (8086:2c41) 325 * On Xeon 55xx, however, it has a different id (8086:2c40). So, 326 * the probing code needs to test for the other address in case of 327 * failure of this one 328 */ 329 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) }, 330 331 }; 332 333 static const struct pci_id_descr pci_dev_descr_lynnfield[] = { 334 { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR) }, 335 { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD) }, 336 { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST) }, 337 338 { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) }, 339 { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) }, 340 { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) }, 341 { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC) }, 342 343 { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) }, 344 { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) }, 345 { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) }, 346 { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) }, 347 348 /* 349 * This is the PCI device has an alternate address on some 350 * processors like Core i7 860 351 */ 352 { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) }, 353 }; 354 355 static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = { 356 /* Memory controller */ 357 { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2) }, 358 { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2) }, 359 /* Exists only for RDIMM */ 360 { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1 }, 361 { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) }, 362 363 /* Channel 0 */ 364 { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) }, 365 { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) }, 366 { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) }, 367 { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2) }, 368 369 /* Channel 1 */ 370 { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) }, 371 { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) }, 372 { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) }, 373 { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2) }, 374 375 /* Channel 2 */ 376 { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) }, 377 { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) }, 378 { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) }, 379 { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) }, 380 381 /* Generic Non-core registers */ 382 { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) }, 383 384 }; 385 386 #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) } 387 static const struct pci_id_table pci_dev_table[] = { 388 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem), 389 PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield), 390 PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere), 391 {0,} /* 0 terminated list. */ 392 }; 393 394 /* 395 * pci_device_id table for which devices we are looking for 396 */ 397 static const struct pci_device_id i7core_pci_tbl[] = { 398 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)}, 399 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)}, 400 {0,} /* 0 terminated list. */ 401 }; 402 403 /**************************************************************************** 404 Ancillary status routines 405 ****************************************************************************/ 406 407 /* MC_CONTROL bits */ 408 #define CH_ACTIVE(pvt, ch) ((pvt)->info.mc_control & (1 << (8 + ch))) 409 #define ECCx8(pvt) ((pvt)->info.mc_control & (1 << 1)) 410 411 /* MC_STATUS bits */ 412 #define ECC_ENABLED(pvt) ((pvt)->info.mc_status & (1 << 4)) 413 #define CH_DISABLED(pvt, ch) ((pvt)->info.mc_status & (1 << ch)) 414 415 /* MC_MAX_DOD read functions */ 416 static inline int numdimms(u32 dimms) 417 { 418 return (dimms & 0x3) + 1; 419 } 420 421 static inline int numrank(u32 rank) 422 { 423 static const int ranks[] = { 1, 2, 4, -EINVAL }; 424 425 return ranks[rank & 0x3]; 426 } 427 428 static inline int numbank(u32 bank) 429 { 430 static const int banks[] = { 4, 8, 16, -EINVAL }; 431 432 return banks[bank & 0x3]; 433 } 434 435 static inline int numrow(u32 row) 436 { 437 static const int rows[] = { 438 1 << 12, 1 << 13, 1 << 14, 1 << 15, 439 1 << 16, -EINVAL, -EINVAL, -EINVAL, 440 }; 441 442 return rows[row & 0x7]; 443 } 444 445 static inline int numcol(u32 col) 446 { 447 static const int cols[] = { 448 1 << 10, 1 << 11, 1 << 12, -EINVAL, 449 }; 450 return cols[col & 0x3]; 451 } 452 453 static struct i7core_dev *get_i7core_dev(u8 socket) 454 { 455 struct i7core_dev *i7core_dev; 456 457 list_for_each_entry(i7core_dev, &i7core_edac_list, list) { 458 if (i7core_dev->socket == socket) 459 return i7core_dev; 460 } 461 462 return NULL; 463 } 464 465 static struct i7core_dev *alloc_i7core_dev(u8 socket, 466 const struct pci_id_table *table) 467 { 468 struct i7core_dev *i7core_dev; 469 470 i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL); 471 if (!i7core_dev) 472 return NULL; 473 474 i7core_dev->pdev = kzalloc(sizeof(*i7core_dev->pdev) * table->n_devs, 475 GFP_KERNEL); 476 if (!i7core_dev->pdev) { 477 kfree(i7core_dev); 478 return NULL; 479 } 480 481 i7core_dev->socket = socket; 482 i7core_dev->n_devs = table->n_devs; 483 list_add_tail(&i7core_dev->list, &i7core_edac_list); 484 485 return i7core_dev; 486 } 487 488 static void free_i7core_dev(struct i7core_dev *i7core_dev) 489 { 490 list_del(&i7core_dev->list); 491 kfree(i7core_dev->pdev); 492 kfree(i7core_dev); 493 } 494 495 /**************************************************************************** 496 Memory check routines 497 ****************************************************************************/ 498 499 static int get_dimm_config(struct mem_ctl_info *mci) 500 { 501 struct i7core_pvt *pvt = mci->pvt_info; 502 struct pci_dev *pdev; 503 int i, j; 504 enum edac_type mode; 505 enum mem_type mtype; 506 struct dimm_info *dimm; 507 508 /* Get data from the MC register, function 0 */ 509 pdev = pvt->pci_mcr[0]; 510 if (!pdev) 511 return -ENODEV; 512 513 /* Device 3 function 0 reads */ 514 pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control); 515 pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status); 516 pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod); 517 pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map); 518 519 edac_dbg(0, "QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n", 520 pvt->i7core_dev->socket, pvt->info.mc_control, 521 pvt->info.mc_status, pvt->info.max_dod, pvt->info.ch_map); 522 523 if (ECC_ENABLED(pvt)) { 524 edac_dbg(0, "ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4); 525 if (ECCx8(pvt)) 526 mode = EDAC_S8ECD8ED; 527 else 528 mode = EDAC_S4ECD4ED; 529 } else { 530 edac_dbg(0, "ECC disabled\n"); 531 mode = EDAC_NONE; 532 } 533 534 /* FIXME: need to handle the error codes */ 535 edac_dbg(0, "DOD Max limits: DIMMS: %d, %d-ranked, %d-banked x%x x 0x%x\n", 536 numdimms(pvt->info.max_dod), 537 numrank(pvt->info.max_dod >> 2), 538 numbank(pvt->info.max_dod >> 4), 539 numrow(pvt->info.max_dod >> 6), 540 numcol(pvt->info.max_dod >> 9)); 541 542 for (i = 0; i < NUM_CHANS; i++) { 543 u32 data, dimm_dod[3], value[8]; 544 545 if (!pvt->pci_ch[i][0]) 546 continue; 547 548 if (!CH_ACTIVE(pvt, i)) { 549 edac_dbg(0, "Channel %i is not active\n", i); 550 continue; 551 } 552 if (CH_DISABLED(pvt, i)) { 553 edac_dbg(0, "Channel %i is disabled\n", i); 554 continue; 555 } 556 557 /* Devices 4-6 function 0 */ 558 pci_read_config_dword(pvt->pci_ch[i][0], 559 MC_CHANNEL_DIMM_INIT_PARAMS, &data); 560 561 562 if (data & THREE_DIMMS_PRESENT) 563 pvt->channel[i].is_3dimms_present = true; 564 565 if (data & SINGLE_QUAD_RANK_PRESENT) 566 pvt->channel[i].is_single_4rank = true; 567 568 if (data & QUAD_RANK_PRESENT) 569 pvt->channel[i].has_4rank = true; 570 571 if (data & REGISTERED_DIMM) 572 mtype = MEM_RDDR3; 573 else 574 mtype = MEM_DDR3; 575 576 /* Devices 4-6 function 1 */ 577 pci_read_config_dword(pvt->pci_ch[i][1], 578 MC_DOD_CH_DIMM0, &dimm_dod[0]); 579 pci_read_config_dword(pvt->pci_ch[i][1], 580 MC_DOD_CH_DIMM1, &dimm_dod[1]); 581 pci_read_config_dword(pvt->pci_ch[i][1], 582 MC_DOD_CH_DIMM2, &dimm_dod[2]); 583 584 edac_dbg(0, "Ch%d phy rd%d, wr%d (0x%08x): %s%s%s%cDIMMs\n", 585 i, 586 RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i), 587 data, 588 pvt->channel[i].is_3dimms_present ? "3DIMMS " : "", 589 pvt->channel[i].is_3dimms_present ? "SINGLE_4R " : "", 590 pvt->channel[i].has_4rank ? "HAS_4R " : "", 591 (data & REGISTERED_DIMM) ? 'R' : 'U'); 592 593 for (j = 0; j < 3; j++) { 594 u32 banks, ranks, rows, cols; 595 u32 size, npages; 596 597 if (!DIMM_PRESENT(dimm_dod[j])) 598 continue; 599 600 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, 601 i, j, 0); 602 banks = numbank(MC_DOD_NUMBANK(dimm_dod[j])); 603 ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j])); 604 rows = numrow(MC_DOD_NUMROW(dimm_dod[j])); 605 cols = numcol(MC_DOD_NUMCOL(dimm_dod[j])); 606 607 /* DDR3 has 8 I/O banks */ 608 size = (rows * cols * banks * ranks) >> (20 - 3); 609 610 edac_dbg(0, "\tdimm %d %d Mb offset: %x, bank: %d, rank: %d, row: %#x, col: %#x\n", 611 j, size, 612 RANKOFFSET(dimm_dod[j]), 613 banks, ranks, rows, cols); 614 615 npages = MiB_TO_PAGES(size); 616 617 dimm->nr_pages = npages; 618 619 switch (banks) { 620 case 4: 621 dimm->dtype = DEV_X4; 622 break; 623 case 8: 624 dimm->dtype = DEV_X8; 625 break; 626 case 16: 627 dimm->dtype = DEV_X16; 628 break; 629 default: 630 dimm->dtype = DEV_UNKNOWN; 631 } 632 633 snprintf(dimm->label, sizeof(dimm->label), 634 "CPU#%uChannel#%u_DIMM#%u", 635 pvt->i7core_dev->socket, i, j); 636 dimm->grain = 8; 637 dimm->edac_mode = mode; 638 dimm->mtype = mtype; 639 } 640 641 pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]); 642 pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]); 643 pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]); 644 pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]); 645 pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]); 646 pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]); 647 pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]); 648 pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]); 649 edac_dbg(1, "\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i); 650 for (j = 0; j < 8; j++) 651 edac_dbg(1, "\t\t%#x\t%#x\t%#x\n", 652 (value[j] >> 27) & 0x1, 653 (value[j] >> 24) & 0x7, 654 (value[j] & ((1 << 24) - 1))); 655 } 656 657 return 0; 658 } 659 660 /**************************************************************************** 661 Error insertion routines 662 ****************************************************************************/ 663 664 #define to_mci(k) container_of(k, struct mem_ctl_info, dev) 665 666 /* The i7core has independent error injection features per channel. 667 However, to have a simpler code, we don't allow enabling error injection 668 on more than one channel. 669 Also, since a change at an inject parameter will be applied only at enable, 670 we're disabling error injection on all write calls to the sysfs nodes that 671 controls the error code injection. 672 */ 673 static int disable_inject(const struct mem_ctl_info *mci) 674 { 675 struct i7core_pvt *pvt = mci->pvt_info; 676 677 pvt->inject.enable = 0; 678 679 if (!pvt->pci_ch[pvt->inject.channel][0]) 680 return -ENODEV; 681 682 pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0], 683 MC_CHANNEL_ERROR_INJECT, 0); 684 685 return 0; 686 } 687 688 /* 689 * i7core inject inject.section 690 * 691 * accept and store error injection inject.section value 692 * bit 0 - refers to the lower 32-byte half cacheline 693 * bit 1 - refers to the upper 32-byte half cacheline 694 */ 695 static ssize_t i7core_inject_section_store(struct device *dev, 696 struct device_attribute *mattr, 697 const char *data, size_t count) 698 { 699 struct mem_ctl_info *mci = to_mci(dev); 700 struct i7core_pvt *pvt = mci->pvt_info; 701 unsigned long value; 702 int rc; 703 704 if (pvt->inject.enable) 705 disable_inject(mci); 706 707 rc = kstrtoul(data, 10, &value); 708 if ((rc < 0) || (value > 3)) 709 return -EIO; 710 711 pvt->inject.section = (u32) value; 712 return count; 713 } 714 715 static ssize_t i7core_inject_section_show(struct device *dev, 716 struct device_attribute *mattr, 717 char *data) 718 { 719 struct mem_ctl_info *mci = to_mci(dev); 720 struct i7core_pvt *pvt = mci->pvt_info; 721 return sprintf(data, "0x%08x\n", pvt->inject.section); 722 } 723 724 /* 725 * i7core inject.type 726 * 727 * accept and store error injection inject.section value 728 * bit 0 - repeat enable - Enable error repetition 729 * bit 1 - inject ECC error 730 * bit 2 - inject parity error 731 */ 732 static ssize_t i7core_inject_type_store(struct device *dev, 733 struct device_attribute *mattr, 734 const char *data, size_t count) 735 { 736 struct mem_ctl_info *mci = to_mci(dev); 737 struct i7core_pvt *pvt = mci->pvt_info; 738 unsigned long value; 739 int rc; 740 741 if (pvt->inject.enable) 742 disable_inject(mci); 743 744 rc = kstrtoul(data, 10, &value); 745 if ((rc < 0) || (value > 7)) 746 return -EIO; 747 748 pvt->inject.type = (u32) value; 749 return count; 750 } 751 752 static ssize_t i7core_inject_type_show(struct device *dev, 753 struct device_attribute *mattr, 754 char *data) 755 { 756 struct mem_ctl_info *mci = to_mci(dev); 757 struct i7core_pvt *pvt = mci->pvt_info; 758 759 return sprintf(data, "0x%08x\n", pvt->inject.type); 760 } 761 762 /* 763 * i7core_inject_inject.eccmask_store 764 * 765 * The type of error (UE/CE) will depend on the inject.eccmask value: 766 * Any bits set to a 1 will flip the corresponding ECC bit 767 * Correctable errors can be injected by flipping 1 bit or the bits within 768 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or 769 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an 770 * uncorrectable error to be injected. 771 */ 772 static ssize_t i7core_inject_eccmask_store(struct device *dev, 773 struct device_attribute *mattr, 774 const char *data, size_t count) 775 { 776 struct mem_ctl_info *mci = to_mci(dev); 777 struct i7core_pvt *pvt = mci->pvt_info; 778 unsigned long value; 779 int rc; 780 781 if (pvt->inject.enable) 782 disable_inject(mci); 783 784 rc = kstrtoul(data, 10, &value); 785 if (rc < 0) 786 return -EIO; 787 788 pvt->inject.eccmask = (u32) value; 789 return count; 790 } 791 792 static ssize_t i7core_inject_eccmask_show(struct device *dev, 793 struct device_attribute *mattr, 794 char *data) 795 { 796 struct mem_ctl_info *mci = to_mci(dev); 797 struct i7core_pvt *pvt = mci->pvt_info; 798 799 return sprintf(data, "0x%08x\n", pvt->inject.eccmask); 800 } 801 802 /* 803 * i7core_addrmatch 804 * 805 * The type of error (UE/CE) will depend on the inject.eccmask value: 806 * Any bits set to a 1 will flip the corresponding ECC bit 807 * Correctable errors can be injected by flipping 1 bit or the bits within 808 * a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or 809 * 23:16 and 31:24). Flipping bits in two symbol pairs will cause an 810 * uncorrectable error to be injected. 811 */ 812 813 #define DECLARE_ADDR_MATCH(param, limit) \ 814 static ssize_t i7core_inject_store_##param( \ 815 struct device *dev, \ 816 struct device_attribute *mattr, \ 817 const char *data, size_t count) \ 818 { \ 819 struct mem_ctl_info *mci = dev_get_drvdata(dev); \ 820 struct i7core_pvt *pvt; \ 821 long value; \ 822 int rc; \ 823 \ 824 edac_dbg(1, "\n"); \ 825 pvt = mci->pvt_info; \ 826 \ 827 if (pvt->inject.enable) \ 828 disable_inject(mci); \ 829 \ 830 if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\ 831 value = -1; \ 832 else { \ 833 rc = kstrtoul(data, 10, &value); \ 834 if ((rc < 0) || (value >= limit)) \ 835 return -EIO; \ 836 } \ 837 \ 838 pvt->inject.param = value; \ 839 \ 840 return count; \ 841 } \ 842 \ 843 static ssize_t i7core_inject_show_##param( \ 844 struct device *dev, \ 845 struct device_attribute *mattr, \ 846 char *data) \ 847 { \ 848 struct mem_ctl_info *mci = dev_get_drvdata(dev); \ 849 struct i7core_pvt *pvt; \ 850 \ 851 pvt = mci->pvt_info; \ 852 edac_dbg(1, "pvt=%p\n", pvt); \ 853 if (pvt->inject.param < 0) \ 854 return sprintf(data, "any\n"); \ 855 else \ 856 return sprintf(data, "%d\n", pvt->inject.param);\ 857 } 858 859 #define ATTR_ADDR_MATCH(param) \ 860 static DEVICE_ATTR(param, S_IRUGO | S_IWUSR, \ 861 i7core_inject_show_##param, \ 862 i7core_inject_store_##param) 863 864 DECLARE_ADDR_MATCH(channel, 3); 865 DECLARE_ADDR_MATCH(dimm, 3); 866 DECLARE_ADDR_MATCH(rank, 4); 867 DECLARE_ADDR_MATCH(bank, 32); 868 DECLARE_ADDR_MATCH(page, 0x10000); 869 DECLARE_ADDR_MATCH(col, 0x4000); 870 871 ATTR_ADDR_MATCH(channel); 872 ATTR_ADDR_MATCH(dimm); 873 ATTR_ADDR_MATCH(rank); 874 ATTR_ADDR_MATCH(bank); 875 ATTR_ADDR_MATCH(page); 876 ATTR_ADDR_MATCH(col); 877 878 static int write_and_test(struct pci_dev *dev, const int where, const u32 val) 879 { 880 u32 read; 881 int count; 882 883 edac_dbg(0, "setting pci %02x:%02x.%x reg=%02x value=%08x\n", 884 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn), 885 where, val); 886 887 for (count = 0; count < 10; count++) { 888 if (count) 889 msleep(100); 890 pci_write_config_dword(dev, where, val); 891 pci_read_config_dword(dev, where, &read); 892 893 if (read == val) 894 return 0; 895 } 896 897 i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x " 898 "write=%08x. Read=%08x\n", 899 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn), 900 where, val, read); 901 902 return -EINVAL; 903 } 904 905 /* 906 * This routine prepares the Memory Controller for error injection. 907 * The error will be injected when some process tries to write to the 908 * memory that matches the given criteria. 909 * The criteria can be set in terms of a mask where dimm, rank, bank, page 910 * and col can be specified. 911 * A -1 value for any of the mask items will make the MCU to ignore 912 * that matching criteria for error injection. 913 * 914 * It should be noticed that the error will only happen after a write operation 915 * on a memory that matches the condition. if REPEAT_EN is not enabled at 916 * inject mask, then it will produce just one error. Otherwise, it will repeat 917 * until the injectmask would be cleaned. 918 * 919 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD 920 * is reliable enough to check if the MC is using the 921 * three channels. However, this is not clear at the datasheet. 922 */ 923 static ssize_t i7core_inject_enable_store(struct device *dev, 924 struct device_attribute *mattr, 925 const char *data, size_t count) 926 { 927 struct mem_ctl_info *mci = to_mci(dev); 928 struct i7core_pvt *pvt = mci->pvt_info; 929 u32 injectmask; 930 u64 mask = 0; 931 int rc; 932 long enable; 933 934 if (!pvt->pci_ch[pvt->inject.channel][0]) 935 return 0; 936 937 rc = kstrtoul(data, 10, &enable); 938 if ((rc < 0)) 939 return 0; 940 941 if (enable) { 942 pvt->inject.enable = 1; 943 } else { 944 disable_inject(mci); 945 return count; 946 } 947 948 /* Sets pvt->inject.dimm mask */ 949 if (pvt->inject.dimm < 0) 950 mask |= 1LL << 41; 951 else { 952 if (pvt->channel[pvt->inject.channel].dimms > 2) 953 mask |= (pvt->inject.dimm & 0x3LL) << 35; 954 else 955 mask |= (pvt->inject.dimm & 0x1LL) << 36; 956 } 957 958 /* Sets pvt->inject.rank mask */ 959 if (pvt->inject.rank < 0) 960 mask |= 1LL << 40; 961 else { 962 if (pvt->channel[pvt->inject.channel].dimms > 2) 963 mask |= (pvt->inject.rank & 0x1LL) << 34; 964 else 965 mask |= (pvt->inject.rank & 0x3LL) << 34; 966 } 967 968 /* Sets pvt->inject.bank mask */ 969 if (pvt->inject.bank < 0) 970 mask |= 1LL << 39; 971 else 972 mask |= (pvt->inject.bank & 0x15LL) << 30; 973 974 /* Sets pvt->inject.page mask */ 975 if (pvt->inject.page < 0) 976 mask |= 1LL << 38; 977 else 978 mask |= (pvt->inject.page & 0xffff) << 14; 979 980 /* Sets pvt->inject.column mask */ 981 if (pvt->inject.col < 0) 982 mask |= 1LL << 37; 983 else 984 mask |= (pvt->inject.col & 0x3fff); 985 986 /* 987 * bit 0: REPEAT_EN 988 * bits 1-2: MASK_HALF_CACHELINE 989 * bit 3: INJECT_ECC 990 * bit 4: INJECT_ADDR_PARITY 991 */ 992 993 injectmask = (pvt->inject.type & 1) | 994 (pvt->inject.section & 0x3) << 1 | 995 (pvt->inject.type & 0x6) << (3 - 1); 996 997 /* Unlock writes to registers - this register is write only */ 998 pci_write_config_dword(pvt->pci_noncore, 999 MC_CFG_CONTROL, 0x2); 1000 1001 write_and_test(pvt->pci_ch[pvt->inject.channel][0], 1002 MC_CHANNEL_ADDR_MATCH, mask); 1003 write_and_test(pvt->pci_ch[pvt->inject.channel][0], 1004 MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L); 1005 1006 write_and_test(pvt->pci_ch[pvt->inject.channel][0], 1007 MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask); 1008 1009 write_and_test(pvt->pci_ch[pvt->inject.channel][0], 1010 MC_CHANNEL_ERROR_INJECT, injectmask); 1011 1012 /* 1013 * This is something undocumented, based on my tests 1014 * Without writing 8 to this register, errors aren't injected. Not sure 1015 * why. 1016 */ 1017 pci_write_config_dword(pvt->pci_noncore, 1018 MC_CFG_CONTROL, 8); 1019 1020 edac_dbg(0, "Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n", 1021 mask, pvt->inject.eccmask, injectmask); 1022 1023 1024 return count; 1025 } 1026 1027 static ssize_t i7core_inject_enable_show(struct device *dev, 1028 struct device_attribute *mattr, 1029 char *data) 1030 { 1031 struct mem_ctl_info *mci = to_mci(dev); 1032 struct i7core_pvt *pvt = mci->pvt_info; 1033 u32 injectmask; 1034 1035 if (!pvt->pci_ch[pvt->inject.channel][0]) 1036 return 0; 1037 1038 pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0], 1039 MC_CHANNEL_ERROR_INJECT, &injectmask); 1040 1041 edac_dbg(0, "Inject error read: 0x%018x\n", injectmask); 1042 1043 if (injectmask & 0x0c) 1044 pvt->inject.enable = 1; 1045 1046 return sprintf(data, "%d\n", pvt->inject.enable); 1047 } 1048 1049 #define DECLARE_COUNTER(param) \ 1050 static ssize_t i7core_show_counter_##param( \ 1051 struct device *dev, \ 1052 struct device_attribute *mattr, \ 1053 char *data) \ 1054 { \ 1055 struct mem_ctl_info *mci = dev_get_drvdata(dev); \ 1056 struct i7core_pvt *pvt = mci->pvt_info; \ 1057 \ 1058 edac_dbg(1, "\n"); \ 1059 if (!pvt->ce_count_available || (pvt->is_registered)) \ 1060 return sprintf(data, "data unavailable\n"); \ 1061 return sprintf(data, "%lu\n", \ 1062 pvt->udimm_ce_count[param]); \ 1063 } 1064 1065 #define ATTR_COUNTER(param) \ 1066 static DEVICE_ATTR(udimm##param, S_IRUGO | S_IWUSR, \ 1067 i7core_show_counter_##param, \ 1068 NULL) 1069 1070 DECLARE_COUNTER(0); 1071 DECLARE_COUNTER(1); 1072 DECLARE_COUNTER(2); 1073 1074 ATTR_COUNTER(0); 1075 ATTR_COUNTER(1); 1076 ATTR_COUNTER(2); 1077 1078 /* 1079 * inject_addrmatch device sysfs struct 1080 */ 1081 1082 static struct attribute *i7core_addrmatch_attrs[] = { 1083 &dev_attr_channel.attr, 1084 &dev_attr_dimm.attr, 1085 &dev_attr_rank.attr, 1086 &dev_attr_bank.attr, 1087 &dev_attr_page.attr, 1088 &dev_attr_col.attr, 1089 NULL 1090 }; 1091 1092 static struct attribute_group addrmatch_grp = { 1093 .attrs = i7core_addrmatch_attrs, 1094 }; 1095 1096 static const struct attribute_group *addrmatch_groups[] = { 1097 &addrmatch_grp, 1098 NULL 1099 }; 1100 1101 static void addrmatch_release(struct device *device) 1102 { 1103 edac_dbg(1, "Releasing device %s\n", dev_name(device)); 1104 kfree(device); 1105 } 1106 1107 static struct device_type addrmatch_type = { 1108 .groups = addrmatch_groups, 1109 .release = addrmatch_release, 1110 }; 1111 1112 /* 1113 * all_channel_counts sysfs struct 1114 */ 1115 1116 static struct attribute *i7core_udimm_counters_attrs[] = { 1117 &dev_attr_udimm0.attr, 1118 &dev_attr_udimm1.attr, 1119 &dev_attr_udimm2.attr, 1120 NULL 1121 }; 1122 1123 static struct attribute_group all_channel_counts_grp = { 1124 .attrs = i7core_udimm_counters_attrs, 1125 }; 1126 1127 static const struct attribute_group *all_channel_counts_groups[] = { 1128 &all_channel_counts_grp, 1129 NULL 1130 }; 1131 1132 static void all_channel_counts_release(struct device *device) 1133 { 1134 edac_dbg(1, "Releasing device %s\n", dev_name(device)); 1135 kfree(device); 1136 } 1137 1138 static struct device_type all_channel_counts_type = { 1139 .groups = all_channel_counts_groups, 1140 .release = all_channel_counts_release, 1141 }; 1142 1143 /* 1144 * inject sysfs attributes 1145 */ 1146 1147 static DEVICE_ATTR(inject_section, S_IRUGO | S_IWUSR, 1148 i7core_inject_section_show, i7core_inject_section_store); 1149 1150 static DEVICE_ATTR(inject_type, S_IRUGO | S_IWUSR, 1151 i7core_inject_type_show, i7core_inject_type_store); 1152 1153 1154 static DEVICE_ATTR(inject_eccmask, S_IRUGO | S_IWUSR, 1155 i7core_inject_eccmask_show, i7core_inject_eccmask_store); 1156 1157 static DEVICE_ATTR(inject_enable, S_IRUGO | S_IWUSR, 1158 i7core_inject_enable_show, i7core_inject_enable_store); 1159 1160 static int i7core_create_sysfs_devices(struct mem_ctl_info *mci) 1161 { 1162 struct i7core_pvt *pvt = mci->pvt_info; 1163 int rc; 1164 1165 rc = device_create_file(&mci->dev, &dev_attr_inject_section); 1166 if (rc < 0) 1167 return rc; 1168 rc = device_create_file(&mci->dev, &dev_attr_inject_type); 1169 if (rc < 0) 1170 return rc; 1171 rc = device_create_file(&mci->dev, &dev_attr_inject_eccmask); 1172 if (rc < 0) 1173 return rc; 1174 rc = device_create_file(&mci->dev, &dev_attr_inject_enable); 1175 if (rc < 0) 1176 return rc; 1177 1178 pvt->addrmatch_dev = kzalloc(sizeof(*pvt->addrmatch_dev), GFP_KERNEL); 1179 if (!pvt->addrmatch_dev) 1180 return rc; 1181 1182 pvt->addrmatch_dev->type = &addrmatch_type; 1183 pvt->addrmatch_dev->bus = mci->dev.bus; 1184 device_initialize(pvt->addrmatch_dev); 1185 pvt->addrmatch_dev->parent = &mci->dev; 1186 dev_set_name(pvt->addrmatch_dev, "inject_addrmatch"); 1187 dev_set_drvdata(pvt->addrmatch_dev, mci); 1188 1189 edac_dbg(1, "creating %s\n", dev_name(pvt->addrmatch_dev)); 1190 1191 rc = device_add(pvt->addrmatch_dev); 1192 if (rc < 0) 1193 return rc; 1194 1195 if (!pvt->is_registered) { 1196 pvt->chancounts_dev = kzalloc(sizeof(*pvt->chancounts_dev), 1197 GFP_KERNEL); 1198 if (!pvt->chancounts_dev) { 1199 put_device(pvt->addrmatch_dev); 1200 device_del(pvt->addrmatch_dev); 1201 return rc; 1202 } 1203 1204 pvt->chancounts_dev->type = &all_channel_counts_type; 1205 pvt->chancounts_dev->bus = mci->dev.bus; 1206 device_initialize(pvt->chancounts_dev); 1207 pvt->chancounts_dev->parent = &mci->dev; 1208 dev_set_name(pvt->chancounts_dev, "all_channel_counts"); 1209 dev_set_drvdata(pvt->chancounts_dev, mci); 1210 1211 edac_dbg(1, "creating %s\n", dev_name(pvt->chancounts_dev)); 1212 1213 rc = device_add(pvt->chancounts_dev); 1214 if (rc < 0) 1215 return rc; 1216 } 1217 return 0; 1218 } 1219 1220 static void i7core_delete_sysfs_devices(struct mem_ctl_info *mci) 1221 { 1222 struct i7core_pvt *pvt = mci->pvt_info; 1223 1224 edac_dbg(1, "\n"); 1225 1226 device_remove_file(&mci->dev, &dev_attr_inject_section); 1227 device_remove_file(&mci->dev, &dev_attr_inject_type); 1228 device_remove_file(&mci->dev, &dev_attr_inject_eccmask); 1229 device_remove_file(&mci->dev, &dev_attr_inject_enable); 1230 1231 if (!pvt->is_registered) { 1232 put_device(pvt->chancounts_dev); 1233 device_del(pvt->chancounts_dev); 1234 } 1235 put_device(pvt->addrmatch_dev); 1236 device_del(pvt->addrmatch_dev); 1237 } 1238 1239 /**************************************************************************** 1240 Device initialization routines: put/get, init/exit 1241 ****************************************************************************/ 1242 1243 /* 1244 * i7core_put_all_devices 'put' all the devices that we have 1245 * reserved via 'get' 1246 */ 1247 static void i7core_put_devices(struct i7core_dev *i7core_dev) 1248 { 1249 int i; 1250 1251 edac_dbg(0, "\n"); 1252 for (i = 0; i < i7core_dev->n_devs; i++) { 1253 struct pci_dev *pdev = i7core_dev->pdev[i]; 1254 if (!pdev) 1255 continue; 1256 edac_dbg(0, "Removing dev %02x:%02x.%d\n", 1257 pdev->bus->number, 1258 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 1259 pci_dev_put(pdev); 1260 } 1261 } 1262 1263 static void i7core_put_all_devices(void) 1264 { 1265 struct i7core_dev *i7core_dev, *tmp; 1266 1267 list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) { 1268 i7core_put_devices(i7core_dev); 1269 free_i7core_dev(i7core_dev); 1270 } 1271 } 1272 1273 static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table) 1274 { 1275 struct pci_dev *pdev = NULL; 1276 int i; 1277 1278 /* 1279 * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses 1280 * aren't announced by acpi. So, we need to use a legacy scan probing 1281 * to detect them 1282 */ 1283 while (table && table->descr) { 1284 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL); 1285 if (unlikely(!pdev)) { 1286 for (i = 0; i < MAX_SOCKET_BUSES; i++) 1287 pcibios_scan_specific_bus(255-i); 1288 } 1289 pci_dev_put(pdev); 1290 table++; 1291 } 1292 } 1293 1294 static unsigned i7core_pci_lastbus(void) 1295 { 1296 int last_bus = 0, bus; 1297 struct pci_bus *b = NULL; 1298 1299 while ((b = pci_find_next_bus(b)) != NULL) { 1300 bus = b->number; 1301 edac_dbg(0, "Found bus %d\n", bus); 1302 if (bus > last_bus) 1303 last_bus = bus; 1304 } 1305 1306 edac_dbg(0, "Last bus %d\n", last_bus); 1307 1308 return last_bus; 1309 } 1310 1311 /* 1312 * i7core_get_all_devices Find and perform 'get' operation on the MCH's 1313 * device/functions we want to reference for this driver 1314 * 1315 * Need to 'get' device 16 func 1 and func 2 1316 */ 1317 static int i7core_get_onedevice(struct pci_dev **prev, 1318 const struct pci_id_table *table, 1319 const unsigned devno, 1320 const unsigned last_bus) 1321 { 1322 struct i7core_dev *i7core_dev; 1323 const struct pci_id_descr *dev_descr = &table->descr[devno]; 1324 1325 struct pci_dev *pdev = NULL; 1326 u8 bus = 0; 1327 u8 socket = 0; 1328 1329 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 1330 dev_descr->dev_id, *prev); 1331 1332 /* 1333 * On Xeon 55xx, the Intel QuickPath Arch Generic Non-core regs 1334 * is at addr 8086:2c40, instead of 8086:2c41. So, we need 1335 * to probe for the alternate address in case of failure 1336 */ 1337 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev) { 1338 pci_dev_get(*prev); /* pci_get_device will put it */ 1339 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 1340 PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev); 1341 } 1342 1343 if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && 1344 !pdev) { 1345 pci_dev_get(*prev); /* pci_get_device will put it */ 1346 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 1347 PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT, 1348 *prev); 1349 } 1350 1351 if (!pdev) { 1352 if (*prev) { 1353 *prev = pdev; 1354 return 0; 1355 } 1356 1357 if (dev_descr->optional) 1358 return 0; 1359 1360 if (devno == 0) 1361 return -ENODEV; 1362 1363 i7core_printk(KERN_INFO, 1364 "Device not found: dev %02x.%d PCI ID %04x:%04x\n", 1365 dev_descr->dev, dev_descr->func, 1366 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 1367 1368 /* End of list, leave */ 1369 return -ENODEV; 1370 } 1371 bus = pdev->bus->number; 1372 1373 socket = last_bus - bus; 1374 1375 i7core_dev = get_i7core_dev(socket); 1376 if (!i7core_dev) { 1377 i7core_dev = alloc_i7core_dev(socket, table); 1378 if (!i7core_dev) { 1379 pci_dev_put(pdev); 1380 return -ENOMEM; 1381 } 1382 } 1383 1384 if (i7core_dev->pdev[devno]) { 1385 i7core_printk(KERN_ERR, 1386 "Duplicated device for " 1387 "dev %02x:%02x.%d PCI ID %04x:%04x\n", 1388 bus, dev_descr->dev, dev_descr->func, 1389 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 1390 pci_dev_put(pdev); 1391 return -ENODEV; 1392 } 1393 1394 i7core_dev->pdev[devno] = pdev; 1395 1396 /* Sanity check */ 1397 if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev || 1398 PCI_FUNC(pdev->devfn) != dev_descr->func)) { 1399 i7core_printk(KERN_ERR, 1400 "Device PCI ID %04x:%04x " 1401 "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n", 1402 PCI_VENDOR_ID_INTEL, dev_descr->dev_id, 1403 bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 1404 bus, dev_descr->dev, dev_descr->func); 1405 return -ENODEV; 1406 } 1407 1408 /* Be sure that the device is enabled */ 1409 if (unlikely(pci_enable_device(pdev) < 0)) { 1410 i7core_printk(KERN_ERR, 1411 "Couldn't enable " 1412 "dev %02x:%02x.%d PCI ID %04x:%04x\n", 1413 bus, dev_descr->dev, dev_descr->func, 1414 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 1415 return -ENODEV; 1416 } 1417 1418 edac_dbg(0, "Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n", 1419 socket, bus, dev_descr->dev, 1420 dev_descr->func, 1421 PCI_VENDOR_ID_INTEL, dev_descr->dev_id); 1422 1423 /* 1424 * As stated on drivers/pci/search.c, the reference count for 1425 * @from is always decremented if it is not %NULL. So, as we need 1426 * to get all devices up to null, we need to do a get for the device 1427 */ 1428 pci_dev_get(pdev); 1429 1430 *prev = pdev; 1431 1432 return 0; 1433 } 1434 1435 static int i7core_get_all_devices(void) 1436 { 1437 int i, rc, last_bus; 1438 struct pci_dev *pdev = NULL; 1439 const struct pci_id_table *table = pci_dev_table; 1440 1441 last_bus = i7core_pci_lastbus(); 1442 1443 while (table && table->descr) { 1444 for (i = 0; i < table->n_devs; i++) { 1445 pdev = NULL; 1446 do { 1447 rc = i7core_get_onedevice(&pdev, table, i, 1448 last_bus); 1449 if (rc < 0) { 1450 if (i == 0) { 1451 i = table->n_devs; 1452 break; 1453 } 1454 i7core_put_all_devices(); 1455 return -ENODEV; 1456 } 1457 } while (pdev); 1458 } 1459 table++; 1460 } 1461 1462 return 0; 1463 } 1464 1465 static int mci_bind_devs(struct mem_ctl_info *mci, 1466 struct i7core_dev *i7core_dev) 1467 { 1468 struct i7core_pvt *pvt = mci->pvt_info; 1469 struct pci_dev *pdev; 1470 int i, func, slot; 1471 char *family; 1472 1473 pvt->is_registered = false; 1474 pvt->enable_scrub = false; 1475 for (i = 0; i < i7core_dev->n_devs; i++) { 1476 pdev = i7core_dev->pdev[i]; 1477 if (!pdev) 1478 continue; 1479 1480 func = PCI_FUNC(pdev->devfn); 1481 slot = PCI_SLOT(pdev->devfn); 1482 if (slot == 3) { 1483 if (unlikely(func > MAX_MCR_FUNC)) 1484 goto error; 1485 pvt->pci_mcr[func] = pdev; 1486 } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) { 1487 if (unlikely(func > MAX_CHAN_FUNC)) 1488 goto error; 1489 pvt->pci_ch[slot - 4][func] = pdev; 1490 } else if (!slot && !func) { 1491 pvt->pci_noncore = pdev; 1492 1493 /* Detect the processor family */ 1494 switch (pdev->device) { 1495 case PCI_DEVICE_ID_INTEL_I7_NONCORE: 1496 family = "Xeon 35xx/ i7core"; 1497 pvt->enable_scrub = false; 1498 break; 1499 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT: 1500 family = "i7-800/i5-700"; 1501 pvt->enable_scrub = false; 1502 break; 1503 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE: 1504 family = "Xeon 34xx"; 1505 pvt->enable_scrub = false; 1506 break; 1507 case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT: 1508 family = "Xeon 55xx"; 1509 pvt->enable_scrub = true; 1510 break; 1511 case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2: 1512 family = "Xeon 56xx / i7-900"; 1513 pvt->enable_scrub = true; 1514 break; 1515 default: 1516 family = "unknown"; 1517 pvt->enable_scrub = false; 1518 } 1519 edac_dbg(0, "Detected a processor type %s\n", family); 1520 } else 1521 goto error; 1522 1523 edac_dbg(0, "Associated fn %d.%d, dev = %p, socket %d\n", 1524 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), 1525 pdev, i7core_dev->socket); 1526 1527 if (PCI_SLOT(pdev->devfn) == 3 && 1528 PCI_FUNC(pdev->devfn) == 2) 1529 pvt->is_registered = true; 1530 } 1531 1532 return 0; 1533 1534 error: 1535 i7core_printk(KERN_ERR, "Device %d, function %d " 1536 "is out of the expected range\n", 1537 slot, func); 1538 return -EINVAL; 1539 } 1540 1541 /**************************************************************************** 1542 Error check routines 1543 ****************************************************************************/ 1544 1545 static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci, 1546 const int chan, 1547 const int new0, 1548 const int new1, 1549 const int new2) 1550 { 1551 struct i7core_pvt *pvt = mci->pvt_info; 1552 int add0 = 0, add1 = 0, add2 = 0; 1553 /* Updates CE counters if it is not the first time here */ 1554 if (pvt->ce_count_available) { 1555 /* Updates CE counters */ 1556 1557 add2 = new2 - pvt->rdimm_last_ce_count[chan][2]; 1558 add1 = new1 - pvt->rdimm_last_ce_count[chan][1]; 1559 add0 = new0 - pvt->rdimm_last_ce_count[chan][0]; 1560 1561 if (add2 < 0) 1562 add2 += 0x7fff; 1563 pvt->rdimm_ce_count[chan][2] += add2; 1564 1565 if (add1 < 0) 1566 add1 += 0x7fff; 1567 pvt->rdimm_ce_count[chan][1] += add1; 1568 1569 if (add0 < 0) 1570 add0 += 0x7fff; 1571 pvt->rdimm_ce_count[chan][0] += add0; 1572 } else 1573 pvt->ce_count_available = 1; 1574 1575 /* Store the new values */ 1576 pvt->rdimm_last_ce_count[chan][2] = new2; 1577 pvt->rdimm_last_ce_count[chan][1] = new1; 1578 pvt->rdimm_last_ce_count[chan][0] = new0; 1579 1580 /*updated the edac core */ 1581 if (add0 != 0) 1582 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add0, 1583 0, 0, 0, 1584 chan, 0, -1, "error", ""); 1585 if (add1 != 0) 1586 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add1, 1587 0, 0, 0, 1588 chan, 1, -1, "error", ""); 1589 if (add2 != 0) 1590 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add2, 1591 0, 0, 0, 1592 chan, 2, -1, "error", ""); 1593 } 1594 1595 static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci) 1596 { 1597 struct i7core_pvt *pvt = mci->pvt_info; 1598 u32 rcv[3][2]; 1599 int i, new0, new1, new2; 1600 1601 /*Read DEV 3: FUN 2: MC_COR_ECC_CNT regs directly*/ 1602 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0, 1603 &rcv[0][0]); 1604 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1, 1605 &rcv[0][1]); 1606 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2, 1607 &rcv[1][0]); 1608 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3, 1609 &rcv[1][1]); 1610 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4, 1611 &rcv[2][0]); 1612 pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5, 1613 &rcv[2][1]); 1614 for (i = 0 ; i < 3; i++) { 1615 edac_dbg(3, "MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n", 1616 (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]); 1617 /*if the channel has 3 dimms*/ 1618 if (pvt->channel[i].dimms > 2) { 1619 new0 = DIMM_BOT_COR_ERR(rcv[i][0]); 1620 new1 = DIMM_TOP_COR_ERR(rcv[i][0]); 1621 new2 = DIMM_BOT_COR_ERR(rcv[i][1]); 1622 } else { 1623 new0 = DIMM_TOP_COR_ERR(rcv[i][0]) + 1624 DIMM_BOT_COR_ERR(rcv[i][0]); 1625 new1 = DIMM_TOP_COR_ERR(rcv[i][1]) + 1626 DIMM_BOT_COR_ERR(rcv[i][1]); 1627 new2 = 0; 1628 } 1629 1630 i7core_rdimm_update_ce_count(mci, i, new0, new1, new2); 1631 } 1632 } 1633 1634 /* This function is based on the device 3 function 4 registers as described on: 1635 * Intel Xeon Processor 5500 Series Datasheet Volume 2 1636 * http://www.intel.com/Assets/PDF/datasheet/321322.pdf 1637 * also available at: 1638 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf 1639 */ 1640 static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci) 1641 { 1642 struct i7core_pvt *pvt = mci->pvt_info; 1643 u32 rcv1, rcv0; 1644 int new0, new1, new2; 1645 1646 if (!pvt->pci_mcr[4]) { 1647 edac_dbg(0, "MCR registers not found\n"); 1648 return; 1649 } 1650 1651 /* Corrected test errors */ 1652 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1); 1653 pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0); 1654 1655 /* Store the new values */ 1656 new2 = DIMM2_COR_ERR(rcv1); 1657 new1 = DIMM1_COR_ERR(rcv0); 1658 new0 = DIMM0_COR_ERR(rcv0); 1659 1660 /* Updates CE counters if it is not the first time here */ 1661 if (pvt->ce_count_available) { 1662 /* Updates CE counters */ 1663 int add0, add1, add2; 1664 1665 add2 = new2 - pvt->udimm_last_ce_count[2]; 1666 add1 = new1 - pvt->udimm_last_ce_count[1]; 1667 add0 = new0 - pvt->udimm_last_ce_count[0]; 1668 1669 if (add2 < 0) 1670 add2 += 0x7fff; 1671 pvt->udimm_ce_count[2] += add2; 1672 1673 if (add1 < 0) 1674 add1 += 0x7fff; 1675 pvt->udimm_ce_count[1] += add1; 1676 1677 if (add0 < 0) 1678 add0 += 0x7fff; 1679 pvt->udimm_ce_count[0] += add0; 1680 1681 if (add0 | add1 | add2) 1682 i7core_printk(KERN_ERR, "New Corrected error(s): " 1683 "dimm0: +%d, dimm1: +%d, dimm2 +%d\n", 1684 add0, add1, add2); 1685 } else 1686 pvt->ce_count_available = 1; 1687 1688 /* Store the new values */ 1689 pvt->udimm_last_ce_count[2] = new2; 1690 pvt->udimm_last_ce_count[1] = new1; 1691 pvt->udimm_last_ce_count[0] = new0; 1692 } 1693 1694 /* 1695 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32 1696 * Architectures Software Developer’s Manual Volume 3B. 1697 * Nehalem are defined as family 0x06, model 0x1a 1698 * 1699 * The MCA registers used here are the following ones: 1700 * struct mce field MCA Register 1701 * m->status MSR_IA32_MC8_STATUS 1702 * m->addr MSR_IA32_MC8_ADDR 1703 * m->misc MSR_IA32_MC8_MISC 1704 * In the case of Nehalem, the error information is masked at .status and .misc 1705 * fields 1706 */ 1707 static void i7core_mce_output_error(struct mem_ctl_info *mci, 1708 const struct mce *m) 1709 { 1710 struct i7core_pvt *pvt = mci->pvt_info; 1711 char *type, *optype, *err; 1712 enum hw_event_mc_err_type tp_event; 1713 unsigned long error = m->status & 0x1ff0000l; 1714 bool uncorrected_error = m->mcgstatus & 1ll << 61; 1715 bool ripv = m->mcgstatus & 1; 1716 u32 optypenum = (m->status >> 4) & 0x07; 1717 u32 core_err_cnt = (m->status >> 38) & 0x7fff; 1718 u32 dimm = (m->misc >> 16) & 0x3; 1719 u32 channel = (m->misc >> 18) & 0x3; 1720 u32 syndrome = m->misc >> 32; 1721 u32 errnum = find_first_bit(&error, 32); 1722 1723 if (uncorrected_error) { 1724 if (ripv) { 1725 type = "FATAL"; 1726 tp_event = HW_EVENT_ERR_FATAL; 1727 } else { 1728 type = "NON_FATAL"; 1729 tp_event = HW_EVENT_ERR_UNCORRECTED; 1730 } 1731 } else { 1732 type = "CORRECTED"; 1733 tp_event = HW_EVENT_ERR_CORRECTED; 1734 } 1735 1736 switch (optypenum) { 1737 case 0: 1738 optype = "generic undef request"; 1739 break; 1740 case 1: 1741 optype = "read error"; 1742 break; 1743 case 2: 1744 optype = "write error"; 1745 break; 1746 case 3: 1747 optype = "addr/cmd error"; 1748 break; 1749 case 4: 1750 optype = "scrubbing error"; 1751 break; 1752 default: 1753 optype = "reserved"; 1754 break; 1755 } 1756 1757 switch (errnum) { 1758 case 16: 1759 err = "read ECC error"; 1760 break; 1761 case 17: 1762 err = "RAS ECC error"; 1763 break; 1764 case 18: 1765 err = "write parity error"; 1766 break; 1767 case 19: 1768 err = "redundacy loss"; 1769 break; 1770 case 20: 1771 err = "reserved"; 1772 break; 1773 case 21: 1774 err = "memory range error"; 1775 break; 1776 case 22: 1777 err = "RTID out of range"; 1778 break; 1779 case 23: 1780 err = "address parity error"; 1781 break; 1782 case 24: 1783 err = "byte enable parity error"; 1784 break; 1785 default: 1786 err = "unknown"; 1787 } 1788 1789 /* 1790 * Call the helper to output message 1791 * FIXME: what to do if core_err_cnt > 1? Currently, it generates 1792 * only one event 1793 */ 1794 if (uncorrected_error || !pvt->is_registered) 1795 edac_mc_handle_error(tp_event, mci, core_err_cnt, 1796 m->addr >> PAGE_SHIFT, 1797 m->addr & ~PAGE_MASK, 1798 syndrome, 1799 channel, dimm, -1, 1800 err, optype); 1801 } 1802 1803 /* 1804 * i7core_check_error Retrieve and process errors reported by the 1805 * hardware. Called by the Core module. 1806 */ 1807 static void i7core_check_error(struct mem_ctl_info *mci) 1808 { 1809 struct i7core_pvt *pvt = mci->pvt_info; 1810 int i; 1811 unsigned count = 0; 1812 struct mce *m; 1813 1814 /* 1815 * MCE first step: Copy all mce errors into a temporary buffer 1816 * We use a double buffering here, to reduce the risk of 1817 * losing an error. 1818 */ 1819 smp_rmb(); 1820 count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in) 1821 % MCE_LOG_LEN; 1822 if (!count) 1823 goto check_ce_error; 1824 1825 m = pvt->mce_outentry; 1826 if (pvt->mce_in + count > MCE_LOG_LEN) { 1827 unsigned l = MCE_LOG_LEN - pvt->mce_in; 1828 1829 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l); 1830 smp_wmb(); 1831 pvt->mce_in = 0; 1832 count -= l; 1833 m += l; 1834 } 1835 memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count); 1836 smp_wmb(); 1837 pvt->mce_in += count; 1838 1839 smp_rmb(); 1840 if (pvt->mce_overrun) { 1841 i7core_printk(KERN_ERR, "Lost %d memory errors\n", 1842 pvt->mce_overrun); 1843 smp_wmb(); 1844 pvt->mce_overrun = 0; 1845 } 1846 1847 /* 1848 * MCE second step: parse errors and display 1849 */ 1850 for (i = 0; i < count; i++) 1851 i7core_mce_output_error(mci, &pvt->mce_outentry[i]); 1852 1853 /* 1854 * Now, let's increment CE error counts 1855 */ 1856 check_ce_error: 1857 if (!pvt->is_registered) 1858 i7core_udimm_check_mc_ecc_err(mci); 1859 else 1860 i7core_rdimm_check_mc_ecc_err(mci); 1861 } 1862 1863 /* 1864 * i7core_mce_check_error Replicates mcelog routine to get errors 1865 * This routine simply queues mcelog errors, and 1866 * return. The error itself should be handled later 1867 * by i7core_check_error. 1868 * WARNING: As this routine should be called at NMI time, extra care should 1869 * be taken to avoid deadlocks, and to be as fast as possible. 1870 */ 1871 static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val, 1872 void *data) 1873 { 1874 struct mce *mce = (struct mce *)data; 1875 struct i7core_dev *i7_dev; 1876 struct mem_ctl_info *mci; 1877 struct i7core_pvt *pvt; 1878 1879 i7_dev = get_i7core_dev(mce->socketid); 1880 if (!i7_dev) 1881 return NOTIFY_BAD; 1882 1883 mci = i7_dev->mci; 1884 pvt = mci->pvt_info; 1885 1886 /* 1887 * Just let mcelog handle it if the error is 1888 * outside the memory controller 1889 */ 1890 if (((mce->status & 0xffff) >> 7) != 1) 1891 return NOTIFY_DONE; 1892 1893 /* Bank 8 registers are the only ones that we know how to handle */ 1894 if (mce->bank != 8) 1895 return NOTIFY_DONE; 1896 1897 smp_rmb(); 1898 if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) { 1899 smp_wmb(); 1900 pvt->mce_overrun++; 1901 return NOTIFY_DONE; 1902 } 1903 1904 /* Copy memory error at the ringbuffer */ 1905 memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce)); 1906 smp_wmb(); 1907 pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN; 1908 1909 /* Handle fatal errors immediately */ 1910 if (mce->mcgstatus & 1) 1911 i7core_check_error(mci); 1912 1913 /* Advise mcelog that the errors were handled */ 1914 return NOTIFY_STOP; 1915 } 1916 1917 static struct notifier_block i7_mce_dec = { 1918 .notifier_call = i7core_mce_check_error, 1919 }; 1920 1921 struct memdev_dmi_entry { 1922 u8 type; 1923 u8 length; 1924 u16 handle; 1925 u16 phys_mem_array_handle; 1926 u16 mem_err_info_handle; 1927 u16 total_width; 1928 u16 data_width; 1929 u16 size; 1930 u8 form; 1931 u8 device_set; 1932 u8 device_locator; 1933 u8 bank_locator; 1934 u8 memory_type; 1935 u16 type_detail; 1936 u16 speed; 1937 u8 manufacturer; 1938 u8 serial_number; 1939 u8 asset_tag; 1940 u8 part_number; 1941 u8 attributes; 1942 u32 extended_size; 1943 u16 conf_mem_clk_speed; 1944 } __attribute__((__packed__)); 1945 1946 1947 /* 1948 * Decode the DRAM Clock Frequency, be paranoid, make sure that all 1949 * memory devices show the same speed, and if they don't then consider 1950 * all speeds to be invalid. 1951 */ 1952 static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq) 1953 { 1954 int *dclk_freq = _dclk_freq; 1955 u16 dmi_mem_clk_speed; 1956 1957 if (*dclk_freq == -1) 1958 return; 1959 1960 if (dh->type == DMI_ENTRY_MEM_DEVICE) { 1961 struct memdev_dmi_entry *memdev_dmi_entry = 1962 (struct memdev_dmi_entry *)dh; 1963 unsigned long conf_mem_clk_speed_offset = 1964 (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed - 1965 (unsigned long)&memdev_dmi_entry->type; 1966 unsigned long speed_offset = 1967 (unsigned long)&memdev_dmi_entry->speed - 1968 (unsigned long)&memdev_dmi_entry->type; 1969 1970 /* Check that a DIMM is present */ 1971 if (memdev_dmi_entry->size == 0) 1972 return; 1973 1974 /* 1975 * Pick the configured speed if it's available, otherwise 1976 * pick the DIMM speed, or we don't have a speed. 1977 */ 1978 if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) { 1979 dmi_mem_clk_speed = 1980 memdev_dmi_entry->conf_mem_clk_speed; 1981 } else if (memdev_dmi_entry->length > speed_offset) { 1982 dmi_mem_clk_speed = memdev_dmi_entry->speed; 1983 } else { 1984 *dclk_freq = -1; 1985 return; 1986 } 1987 1988 if (*dclk_freq == 0) { 1989 /* First pass, speed was 0 */ 1990 if (dmi_mem_clk_speed > 0) { 1991 /* Set speed if a valid speed is read */ 1992 *dclk_freq = dmi_mem_clk_speed; 1993 } else { 1994 /* Otherwise we don't have a valid speed */ 1995 *dclk_freq = -1; 1996 } 1997 } else if (*dclk_freq > 0 && 1998 *dclk_freq != dmi_mem_clk_speed) { 1999 /* 2000 * If we have a speed, check that all DIMMS are the same 2001 * speed, otherwise set the speed as invalid. 2002 */ 2003 *dclk_freq = -1; 2004 } 2005 } 2006 } 2007 2008 /* 2009 * The default DCLK frequency is used as a fallback if we 2010 * fail to find anything reliable in the DMI. The value 2011 * is taken straight from the datasheet. 2012 */ 2013 #define DEFAULT_DCLK_FREQ 800 2014 2015 static int get_dclk_freq(void) 2016 { 2017 int dclk_freq = 0; 2018 2019 dmi_walk(decode_dclk, (void *)&dclk_freq); 2020 2021 if (dclk_freq < 1) 2022 return DEFAULT_DCLK_FREQ; 2023 2024 return dclk_freq; 2025 } 2026 2027 /* 2028 * set_sdram_scrub_rate This routine sets byte/sec bandwidth scrub rate 2029 * to hardware according to SCRUBINTERVAL formula 2030 * found in datasheet. 2031 */ 2032 static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw) 2033 { 2034 struct i7core_pvt *pvt = mci->pvt_info; 2035 struct pci_dev *pdev; 2036 u32 dw_scrub; 2037 u32 dw_ssr; 2038 2039 /* Get data from the MC register, function 2 */ 2040 pdev = pvt->pci_mcr[2]; 2041 if (!pdev) 2042 return -ENODEV; 2043 2044 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub); 2045 2046 if (new_bw == 0) { 2047 /* Prepare to disable petrol scrub */ 2048 dw_scrub &= ~STARTSCRUB; 2049 /* Stop the patrol scrub engine */ 2050 write_and_test(pdev, MC_SCRUB_CONTROL, 2051 dw_scrub & ~SCRUBINTERVAL_MASK); 2052 2053 /* Get current status of scrub rate and set bit to disable */ 2054 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr); 2055 dw_ssr &= ~SSR_MODE_MASK; 2056 dw_ssr |= SSR_MODE_DISABLE; 2057 } else { 2058 const int cache_line_size = 64; 2059 const u32 freq_dclk_mhz = pvt->dclk_freq; 2060 unsigned long long scrub_interval; 2061 /* 2062 * Translate the desired scrub rate to a register value and 2063 * program the corresponding register value. 2064 */ 2065 scrub_interval = (unsigned long long)freq_dclk_mhz * 2066 cache_line_size * 1000000; 2067 do_div(scrub_interval, new_bw); 2068 2069 if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK) 2070 return -EINVAL; 2071 2072 dw_scrub = SCRUBINTERVAL_MASK & scrub_interval; 2073 2074 /* Start the patrol scrub engine */ 2075 pci_write_config_dword(pdev, MC_SCRUB_CONTROL, 2076 STARTSCRUB | dw_scrub); 2077 2078 /* Get current status of scrub rate and set bit to enable */ 2079 pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr); 2080 dw_ssr &= ~SSR_MODE_MASK; 2081 dw_ssr |= SSR_MODE_ENABLE; 2082 } 2083 /* Disable or enable scrubbing */ 2084 pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr); 2085 2086 return new_bw; 2087 } 2088 2089 /* 2090 * get_sdram_scrub_rate This routine convert current scrub rate value 2091 * into byte/sec bandwidth according to 2092 * SCRUBINTERVAL formula found in datasheet. 2093 */ 2094 static int get_sdram_scrub_rate(struct mem_ctl_info *mci) 2095 { 2096 struct i7core_pvt *pvt = mci->pvt_info; 2097 struct pci_dev *pdev; 2098 const u32 cache_line_size = 64; 2099 const u32 freq_dclk_mhz = pvt->dclk_freq; 2100 unsigned long long scrub_rate; 2101 u32 scrubval; 2102 2103 /* Get data from the MC register, function 2 */ 2104 pdev = pvt->pci_mcr[2]; 2105 if (!pdev) 2106 return -ENODEV; 2107 2108 /* Get current scrub control data */ 2109 pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval); 2110 2111 /* Mask highest 8-bits to 0 */ 2112 scrubval &= SCRUBINTERVAL_MASK; 2113 if (!scrubval) 2114 return 0; 2115 2116 /* Calculate scrub rate value into byte/sec bandwidth */ 2117 scrub_rate = (unsigned long long)freq_dclk_mhz * 2118 1000000 * cache_line_size; 2119 do_div(scrub_rate, scrubval); 2120 return (int)scrub_rate; 2121 } 2122 2123 static void enable_sdram_scrub_setting(struct mem_ctl_info *mci) 2124 { 2125 struct i7core_pvt *pvt = mci->pvt_info; 2126 u32 pci_lock; 2127 2128 /* Unlock writes to pci registers */ 2129 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock); 2130 pci_lock &= ~0x3; 2131 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, 2132 pci_lock | MC_CFG_UNLOCK); 2133 2134 mci->set_sdram_scrub_rate = set_sdram_scrub_rate; 2135 mci->get_sdram_scrub_rate = get_sdram_scrub_rate; 2136 } 2137 2138 static void disable_sdram_scrub_setting(struct mem_ctl_info *mci) 2139 { 2140 struct i7core_pvt *pvt = mci->pvt_info; 2141 u32 pci_lock; 2142 2143 /* Lock writes to pci registers */ 2144 pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock); 2145 pci_lock &= ~0x3; 2146 pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, 2147 pci_lock | MC_CFG_LOCK); 2148 } 2149 2150 static void i7core_pci_ctl_create(struct i7core_pvt *pvt) 2151 { 2152 pvt->i7core_pci = edac_pci_create_generic_ctl( 2153 &pvt->i7core_dev->pdev[0]->dev, 2154 EDAC_MOD_STR); 2155 if (unlikely(!pvt->i7core_pci)) 2156 i7core_printk(KERN_WARNING, 2157 "Unable to setup PCI error report via EDAC\n"); 2158 } 2159 2160 static void i7core_pci_ctl_release(struct i7core_pvt *pvt) 2161 { 2162 if (likely(pvt->i7core_pci)) 2163 edac_pci_release_generic_ctl(pvt->i7core_pci); 2164 else 2165 i7core_printk(KERN_ERR, 2166 "Couldn't find mem_ctl_info for socket %d\n", 2167 pvt->i7core_dev->socket); 2168 pvt->i7core_pci = NULL; 2169 } 2170 2171 static void i7core_unregister_mci(struct i7core_dev *i7core_dev) 2172 { 2173 struct mem_ctl_info *mci = i7core_dev->mci; 2174 struct i7core_pvt *pvt; 2175 2176 if (unlikely(!mci || !mci->pvt_info)) { 2177 edac_dbg(0, "MC: dev = %p\n", &i7core_dev->pdev[0]->dev); 2178 2179 i7core_printk(KERN_ERR, "Couldn't find mci handler\n"); 2180 return; 2181 } 2182 2183 pvt = mci->pvt_info; 2184 2185 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev); 2186 2187 /* Disable scrubrate setting */ 2188 if (pvt->enable_scrub) 2189 disable_sdram_scrub_setting(mci); 2190 2191 /* Disable EDAC polling */ 2192 i7core_pci_ctl_release(pvt); 2193 2194 /* Remove MC sysfs nodes */ 2195 i7core_delete_sysfs_devices(mci); 2196 edac_mc_del_mc(mci->pdev); 2197 2198 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name); 2199 kfree(mci->ctl_name); 2200 edac_mc_free(mci); 2201 i7core_dev->mci = NULL; 2202 } 2203 2204 static int i7core_register_mci(struct i7core_dev *i7core_dev) 2205 { 2206 struct mem_ctl_info *mci; 2207 struct i7core_pvt *pvt; 2208 int rc; 2209 struct edac_mc_layer layers[2]; 2210 2211 /* allocate a new MC control structure */ 2212 2213 layers[0].type = EDAC_MC_LAYER_CHANNEL; 2214 layers[0].size = NUM_CHANS; 2215 layers[0].is_virt_csrow = false; 2216 layers[1].type = EDAC_MC_LAYER_SLOT; 2217 layers[1].size = MAX_DIMMS; 2218 layers[1].is_virt_csrow = true; 2219 mci = edac_mc_alloc(i7core_dev->socket, ARRAY_SIZE(layers), layers, 2220 sizeof(*pvt)); 2221 if (unlikely(!mci)) 2222 return -ENOMEM; 2223 2224 edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev); 2225 2226 pvt = mci->pvt_info; 2227 memset(pvt, 0, sizeof(*pvt)); 2228 2229 /* Associates i7core_dev and mci for future usage */ 2230 pvt->i7core_dev = i7core_dev; 2231 i7core_dev->mci = mci; 2232 2233 /* 2234 * FIXME: how to handle RDDR3 at MCI level? It is possible to have 2235 * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different 2236 * memory channels 2237 */ 2238 mci->mtype_cap = MEM_FLAG_DDR3; 2239 mci->edac_ctl_cap = EDAC_FLAG_NONE; 2240 mci->edac_cap = EDAC_FLAG_NONE; 2241 mci->mod_name = "i7core_edac.c"; 2242 mci->mod_ver = I7CORE_REVISION; 2243 mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d", 2244 i7core_dev->socket); 2245 mci->dev_name = pci_name(i7core_dev->pdev[0]); 2246 mci->ctl_page_to_phys = NULL; 2247 2248 /* Store pci devices at mci for faster access */ 2249 rc = mci_bind_devs(mci, i7core_dev); 2250 if (unlikely(rc < 0)) 2251 goto fail0; 2252 2253 2254 /* Get dimm basic config */ 2255 get_dimm_config(mci); 2256 /* record ptr to the generic device */ 2257 mci->pdev = &i7core_dev->pdev[0]->dev; 2258 /* Set the function pointer to an actual operation function */ 2259 mci->edac_check = i7core_check_error; 2260 2261 /* Enable scrubrate setting */ 2262 if (pvt->enable_scrub) 2263 enable_sdram_scrub_setting(mci); 2264 2265 /* add this new MC control structure to EDAC's list of MCs */ 2266 if (unlikely(edac_mc_add_mc(mci))) { 2267 edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); 2268 /* FIXME: perhaps some code should go here that disables error 2269 * reporting if we just enabled it 2270 */ 2271 2272 rc = -EINVAL; 2273 goto fail0; 2274 } 2275 if (i7core_create_sysfs_devices(mci)) { 2276 edac_dbg(0, "MC: failed to create sysfs nodes\n"); 2277 edac_mc_del_mc(mci->pdev); 2278 rc = -EINVAL; 2279 goto fail0; 2280 } 2281 2282 /* Default error mask is any memory */ 2283 pvt->inject.channel = 0; 2284 pvt->inject.dimm = -1; 2285 pvt->inject.rank = -1; 2286 pvt->inject.bank = -1; 2287 pvt->inject.page = -1; 2288 pvt->inject.col = -1; 2289 2290 /* allocating generic PCI control info */ 2291 i7core_pci_ctl_create(pvt); 2292 2293 /* DCLK for scrub rate setting */ 2294 pvt->dclk_freq = get_dclk_freq(); 2295 2296 return 0; 2297 2298 fail0: 2299 kfree(mci->ctl_name); 2300 edac_mc_free(mci); 2301 i7core_dev->mci = NULL; 2302 return rc; 2303 } 2304 2305 /* 2306 * i7core_probe Probe for ONE instance of device to see if it is 2307 * present. 2308 * return: 2309 * 0 for FOUND a device 2310 * < 0 for error code 2311 */ 2312 2313 static int i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id) 2314 { 2315 int rc, count = 0; 2316 struct i7core_dev *i7core_dev; 2317 2318 /* get the pci devices we want to reserve for our use */ 2319 mutex_lock(&i7core_edac_lock); 2320 2321 /* 2322 * All memory controllers are allocated at the first pass. 2323 */ 2324 if (unlikely(probed >= 1)) { 2325 mutex_unlock(&i7core_edac_lock); 2326 return -ENODEV; 2327 } 2328 probed++; 2329 2330 rc = i7core_get_all_devices(); 2331 if (unlikely(rc < 0)) 2332 goto fail0; 2333 2334 list_for_each_entry(i7core_dev, &i7core_edac_list, list) { 2335 count++; 2336 rc = i7core_register_mci(i7core_dev); 2337 if (unlikely(rc < 0)) 2338 goto fail1; 2339 } 2340 2341 /* 2342 * Nehalem-EX uses a different memory controller. However, as the 2343 * memory controller is not visible on some Nehalem/Nehalem-EP, we 2344 * need to indirectly probe via a X58 PCI device. The same devices 2345 * are found on (some) Nehalem-EX. So, on those machines, the 2346 * probe routine needs to return -ENODEV, as the actual Memory 2347 * Controller registers won't be detected. 2348 */ 2349 if (!count) { 2350 rc = -ENODEV; 2351 goto fail1; 2352 } 2353 2354 i7core_printk(KERN_INFO, 2355 "Driver loaded, %d memory controller(s) found.\n", 2356 count); 2357 2358 mutex_unlock(&i7core_edac_lock); 2359 return 0; 2360 2361 fail1: 2362 list_for_each_entry(i7core_dev, &i7core_edac_list, list) 2363 i7core_unregister_mci(i7core_dev); 2364 2365 i7core_put_all_devices(); 2366 fail0: 2367 mutex_unlock(&i7core_edac_lock); 2368 return rc; 2369 } 2370 2371 /* 2372 * i7core_remove destructor for one instance of device 2373 * 2374 */ 2375 static void i7core_remove(struct pci_dev *pdev) 2376 { 2377 struct i7core_dev *i7core_dev; 2378 2379 edac_dbg(0, "\n"); 2380 2381 /* 2382 * we have a trouble here: pdev value for removal will be wrong, since 2383 * it will point to the X58 register used to detect that the machine 2384 * is a Nehalem or upper design. However, due to the way several PCI 2385 * devices are grouped together to provide MC functionality, we need 2386 * to use a different method for releasing the devices 2387 */ 2388 2389 mutex_lock(&i7core_edac_lock); 2390 2391 if (unlikely(!probed)) { 2392 mutex_unlock(&i7core_edac_lock); 2393 return; 2394 } 2395 2396 list_for_each_entry(i7core_dev, &i7core_edac_list, list) 2397 i7core_unregister_mci(i7core_dev); 2398 2399 /* Release PCI resources */ 2400 i7core_put_all_devices(); 2401 2402 probed--; 2403 2404 mutex_unlock(&i7core_edac_lock); 2405 } 2406 2407 MODULE_DEVICE_TABLE(pci, i7core_pci_tbl); 2408 2409 /* 2410 * i7core_driver pci_driver structure for this module 2411 * 2412 */ 2413 static struct pci_driver i7core_driver = { 2414 .name = "i7core_edac", 2415 .probe = i7core_probe, 2416 .remove = i7core_remove, 2417 .id_table = i7core_pci_tbl, 2418 }; 2419 2420 /* 2421 * i7core_init Module entry function 2422 * Try to initialize this module for its devices 2423 */ 2424 static int __init i7core_init(void) 2425 { 2426 int pci_rc; 2427 2428 edac_dbg(2, "\n"); 2429 2430 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 2431 opstate_init(); 2432 2433 if (use_pci_fixup) 2434 i7core_xeon_pci_fixup(pci_dev_table); 2435 2436 pci_rc = pci_register_driver(&i7core_driver); 2437 2438 if (pci_rc >= 0) { 2439 mce_register_decode_chain(&i7_mce_dec); 2440 return 0; 2441 } 2442 2443 i7core_printk(KERN_ERR, "Failed to register device with error %d.\n", 2444 pci_rc); 2445 2446 return pci_rc; 2447 } 2448 2449 /* 2450 * i7core_exit() Module exit function 2451 * Unregister the driver 2452 */ 2453 static void __exit i7core_exit(void) 2454 { 2455 edac_dbg(2, "\n"); 2456 pci_unregister_driver(&i7core_driver); 2457 mce_unregister_decode_chain(&i7_mce_dec); 2458 } 2459 2460 module_init(i7core_init); 2461 module_exit(i7core_exit); 2462 2463 MODULE_LICENSE("GPL"); 2464 MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>"); 2465 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)"); 2466 MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - " 2467 I7CORE_REVISION); 2468 2469 module_param(edac_op_state, int, 0444); 2470 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 2471