i5400_edac.c (40744B)
1/* 2 * Intel 5400 class Memory Controllers kernel module (Seaburg) 3 * 4 * This file may be distributed under the terms of the 5 * GNU General Public License. 6 * 7 * Copyright (c) 2008 by: 8 * Ben Woodard <woodard@redhat.com> 9 * Mauro Carvalho Chehab 10 * 11 * Red Hat Inc. https://www.redhat.com 12 * 13 * Forked and adapted from the i5000_edac driver which was 14 * written by Douglas Thompson Linux Networx <norsk5@xmission.com> 15 * 16 * This module is based on the following document: 17 * 18 * Intel 5400 Chipset Memory Controller Hub (MCH) - Datasheet 19 * http://developer.intel.com/design/chipsets/datashts/313070.htm 20 * 21 * This Memory Controller manages DDR2 FB-DIMMs. It has 2 branches, each with 22 * 2 channels operating in lockstep no-mirror mode. Each channel can have up to 23 * 4 dimm's, each with up to 8GB. 24 * 25 */ 26 27#include <linux/module.h> 28#include <linux/init.h> 29#include <linux/pci.h> 30#include <linux/pci_ids.h> 31#include <linux/slab.h> 32#include <linux/edac.h> 33#include <linux/mmzone.h> 34 35#include "edac_module.h" 36 37/* 38 * Alter this version for the I5400 module when modifications are made 39 */ 40#define I5400_REVISION " Ver: 1.0.0" 41 42#define EDAC_MOD_STR "i5400_edac" 43 44#define i5400_printk(level, fmt, arg...) \ 45 edac_printk(level, "i5400", fmt, ##arg) 46 47#define i5400_mc_printk(mci, level, fmt, arg...) \ 48 edac_mc_chipset_printk(mci, level, "i5400", fmt, ##arg) 49 50/* Limits for i5400 */ 51#define MAX_BRANCHES 2 52#define CHANNELS_PER_BRANCH 2 53#define DIMMS_PER_CHANNEL 4 54#define MAX_CHANNELS (MAX_BRANCHES * CHANNELS_PER_BRANCH) 55 56/* Device 16, 57 * Function 0: System Address 58 * Function 1: Memory Branch Map, Control, Errors Register 59 * Function 2: FSB Error Registers 60 * 61 * All 3 functions of Device 16 (0,1,2) share the SAME DID and 62 * uses PCI_DEVICE_ID_INTEL_5400_ERR for device 16 (0,1,2), 63 * PCI_DEVICE_ID_INTEL_5400_FBD0 and PCI_DEVICE_ID_INTEL_5400_FBD1 64 * for device 21 (0,1). 65 */ 66 67 /* OFFSETS for Function 0 */ 68#define AMBASE 0x48 /* AMB Mem Mapped Reg Region Base */ 69#define MAXCH 0x56 /* Max Channel Number */ 70#define MAXDIMMPERCH 0x57 /* Max DIMM PER Channel Number */ 71 72 /* OFFSETS for Function 1 */ 73#define TOLM 0x6C 74#define REDMEMB 0x7C 75#define REC_ECC_LOCATOR_ODD(x) ((x) & 0x3fe00) /* bits [17:9] indicate ODD, [8:0] indicate EVEN */ 76#define MIR0 0x80 77#define MIR1 0x84 78#define AMIR0 0x8c 79#define AMIR1 0x90 80 81 /* Fatal error registers */ 82#define FERR_FAT_FBD 0x98 /* also called as FERR_FAT_FB_DIMM at datasheet */ 83#define FERR_FAT_FBDCHAN (3<<28) /* channel index where the highest-order error occurred */ 84 85#define NERR_FAT_FBD 0x9c 86#define FERR_NF_FBD 0xa0 /* also called as FERR_NFAT_FB_DIMM at datasheet */ 87 88 /* Non-fatal error register */ 89#define NERR_NF_FBD 0xa4 90 91 /* Enable error mask */ 92#define EMASK_FBD 0xa8 93 94#define ERR0_FBD 0xac 95#define ERR1_FBD 0xb0 96#define ERR2_FBD 0xb4 97#define MCERR_FBD 0xb8 98 99 /* No OFFSETS for Device 16 Function 2 */ 100 101/* 102 * Device 21, 103 * Function 0: Memory Map Branch 0 104 * 105 * Device 22, 106 * Function 0: Memory Map Branch 1 107 */ 108 109 /* OFFSETS for Function 0 */ 110#define AMBPRESENT_0 0x64 111#define AMBPRESENT_1 0x66 112#define MTR0 0x80 113#define MTR1 0x82 114#define MTR2 0x84 115#define MTR3 0x86 116 117 /* OFFSETS for Function 1 */ 118#define NRECFGLOG 0x74 119#define RECFGLOG 0x78 120#define NRECMEMA 0xbe 121#define NRECMEMB 0xc0 122#define NRECFB_DIMMA 0xc4 123#define NRECFB_DIMMB 0xc8 124#define NRECFB_DIMMC 0xcc 125#define NRECFB_DIMMD 0xd0 126#define NRECFB_DIMME 0xd4 127#define NRECFB_DIMMF 0xd8 128#define REDMEMA 0xdC 129#define RECMEMA 0xf0 130#define RECMEMB 0xf4 131#define RECFB_DIMMA 0xf8 132#define RECFB_DIMMB 0xec 133#define RECFB_DIMMC 0xf0 134#define RECFB_DIMMD 0xf4 135#define RECFB_DIMME 0xf8 136#define RECFB_DIMMF 0xfC 137 138/* 139 * Error indicator bits and masks 140 * Error masks are according with Table 5-17 of i5400 datasheet 141 */ 142 143enum error_mask { 144 EMASK_M1 = 1<<0, /* Memory Write error on non-redundant retry */ 145 EMASK_M2 = 1<<1, /* Memory or FB-DIMM configuration CRC read error */ 146 EMASK_M3 = 1<<2, /* Reserved */ 147 EMASK_M4 = 1<<3, /* Uncorrectable Data ECC on Replay */ 148 EMASK_M5 = 1<<4, /* Aliased Uncorrectable Non-Mirrored Demand Data ECC */ 149 EMASK_M6 = 1<<5, /* Unsupported on i5400 */ 150 EMASK_M7 = 1<<6, /* Aliased Uncorrectable Resilver- or Spare-Copy Data ECC */ 151 EMASK_M8 = 1<<7, /* Aliased Uncorrectable Patrol Data ECC */ 152 EMASK_M9 = 1<<8, /* Non-Aliased Uncorrectable Non-Mirrored Demand Data ECC */ 153 EMASK_M10 = 1<<9, /* Unsupported on i5400 */ 154 EMASK_M11 = 1<<10, /* Non-Aliased Uncorrectable Resilver- or Spare-Copy Data ECC */ 155 EMASK_M12 = 1<<11, /* Non-Aliased Uncorrectable Patrol Data ECC */ 156 EMASK_M13 = 1<<12, /* Memory Write error on first attempt */ 157 EMASK_M14 = 1<<13, /* FB-DIMM Configuration Write error on first attempt */ 158 EMASK_M15 = 1<<14, /* Memory or FB-DIMM configuration CRC read error */ 159 EMASK_M16 = 1<<15, /* Channel Failed-Over Occurred */ 160 EMASK_M17 = 1<<16, /* Correctable Non-Mirrored Demand Data ECC */ 161 EMASK_M18 = 1<<17, /* Unsupported on i5400 */ 162 EMASK_M19 = 1<<18, /* Correctable Resilver- or Spare-Copy Data ECC */ 163 EMASK_M20 = 1<<19, /* Correctable Patrol Data ECC */ 164 EMASK_M21 = 1<<20, /* FB-DIMM Northbound parity error on FB-DIMM Sync Status */ 165 EMASK_M22 = 1<<21, /* SPD protocol Error */ 166 EMASK_M23 = 1<<22, /* Non-Redundant Fast Reset Timeout */ 167 EMASK_M24 = 1<<23, /* Refresh error */ 168 EMASK_M25 = 1<<24, /* Memory Write error on redundant retry */ 169 EMASK_M26 = 1<<25, /* Redundant Fast Reset Timeout */ 170 EMASK_M27 = 1<<26, /* Correctable Counter Threshold Exceeded */ 171 EMASK_M28 = 1<<27, /* DIMM-Spare Copy Completed */ 172 EMASK_M29 = 1<<28, /* DIMM-Isolation Completed */ 173}; 174 175/* 176 * Names to translate bit error into something useful 177 */ 178static const char *error_name[] = { 179 [0] = "Memory Write error on non-redundant retry", 180 [1] = "Memory or FB-DIMM configuration CRC read error", 181 /* Reserved */ 182 [3] = "Uncorrectable Data ECC on Replay", 183 [4] = "Aliased Uncorrectable Non-Mirrored Demand Data ECC", 184 /* M6 Unsupported on i5400 */ 185 [6] = "Aliased Uncorrectable Resilver- or Spare-Copy Data ECC", 186 [7] = "Aliased Uncorrectable Patrol Data ECC", 187 [8] = "Non-Aliased Uncorrectable Non-Mirrored Demand Data ECC", 188 /* M10 Unsupported on i5400 */ 189 [10] = "Non-Aliased Uncorrectable Resilver- or Spare-Copy Data ECC", 190 [11] = "Non-Aliased Uncorrectable Patrol Data ECC", 191 [12] = "Memory Write error on first attempt", 192 [13] = "FB-DIMM Configuration Write error on first attempt", 193 [14] = "Memory or FB-DIMM configuration CRC read error", 194 [15] = "Channel Failed-Over Occurred", 195 [16] = "Correctable Non-Mirrored Demand Data ECC", 196 /* M18 Unsupported on i5400 */ 197 [18] = "Correctable Resilver- or Spare-Copy Data ECC", 198 [19] = "Correctable Patrol Data ECC", 199 [20] = "FB-DIMM Northbound parity error on FB-DIMM Sync Status", 200 [21] = "SPD protocol Error", 201 [22] = "Non-Redundant Fast Reset Timeout", 202 [23] = "Refresh error", 203 [24] = "Memory Write error on redundant retry", 204 [25] = "Redundant Fast Reset Timeout", 205 [26] = "Correctable Counter Threshold Exceeded", 206 [27] = "DIMM-Spare Copy Completed", 207 [28] = "DIMM-Isolation Completed", 208}; 209 210/* Fatal errors */ 211#define ERROR_FAT_MASK (EMASK_M1 | \ 212 EMASK_M2 | \ 213 EMASK_M23) 214 215/* Correctable errors */ 216#define ERROR_NF_CORRECTABLE (EMASK_M27 | \ 217 EMASK_M20 | \ 218 EMASK_M19 | \ 219 EMASK_M18 | \ 220 EMASK_M17 | \ 221 EMASK_M16) 222#define ERROR_NF_DIMM_SPARE (EMASK_M29 | \ 223 EMASK_M28) 224#define ERROR_NF_SPD_PROTOCOL (EMASK_M22) 225#define ERROR_NF_NORTH_CRC (EMASK_M21) 226 227/* Recoverable errors */ 228#define ERROR_NF_RECOVERABLE (EMASK_M26 | \ 229 EMASK_M25 | \ 230 EMASK_M24 | \ 231 EMASK_M15 | \ 232 EMASK_M14 | \ 233 EMASK_M13 | \ 234 EMASK_M12 | \ 235 EMASK_M11 | \ 236 EMASK_M9 | \ 237 EMASK_M8 | \ 238 EMASK_M7 | \ 239 EMASK_M5) 240 241/* uncorrectable errors */ 242#define ERROR_NF_UNCORRECTABLE (EMASK_M4) 243 244/* mask to all non-fatal errors */ 245#define ERROR_NF_MASK (ERROR_NF_CORRECTABLE | \ 246 ERROR_NF_UNCORRECTABLE | \ 247 ERROR_NF_RECOVERABLE | \ 248 ERROR_NF_DIMM_SPARE | \ 249 ERROR_NF_SPD_PROTOCOL | \ 250 ERROR_NF_NORTH_CRC) 251 252/* 253 * Define error masks for the several registers 254 */ 255 256/* Enable all fatal and non fatal errors */ 257#define ENABLE_EMASK_ALL (ERROR_FAT_MASK | ERROR_NF_MASK) 258 259/* mask for fatal error registers */ 260#define FERR_FAT_MASK ERROR_FAT_MASK 261 262/* masks for non-fatal error register */ 263static inline int to_nf_mask(unsigned int mask) 264{ 265 return (mask & EMASK_M29) | (mask >> 3); 266}; 267 268static inline int from_nf_ferr(unsigned int mask) 269{ 270 return (mask & EMASK_M29) | /* Bit 28 */ 271 (mask & ((1 << 28) - 1) << 3); /* Bits 0 to 27 */ 272}; 273 274#define FERR_NF_MASK to_nf_mask(ERROR_NF_MASK) 275#define FERR_NF_CORRECTABLE to_nf_mask(ERROR_NF_CORRECTABLE) 276#define FERR_NF_DIMM_SPARE to_nf_mask(ERROR_NF_DIMM_SPARE) 277#define FERR_NF_SPD_PROTOCOL to_nf_mask(ERROR_NF_SPD_PROTOCOL) 278#define FERR_NF_NORTH_CRC to_nf_mask(ERROR_NF_NORTH_CRC) 279#define FERR_NF_RECOVERABLE to_nf_mask(ERROR_NF_RECOVERABLE) 280#define FERR_NF_UNCORRECTABLE to_nf_mask(ERROR_NF_UNCORRECTABLE) 281 282/* Defines to extract the vaious fields from the 283 * MTRx - Memory Technology Registers 284 */ 285#define MTR_DIMMS_PRESENT(mtr) ((mtr) & (1 << 10)) 286#define MTR_DIMMS_ETHROTTLE(mtr) ((mtr) & (1 << 9)) 287#define MTR_DRAM_WIDTH(mtr) (((mtr) & (1 << 8)) ? 8 : 4) 288#define MTR_DRAM_BANKS(mtr) (((mtr) & (1 << 6)) ? 8 : 4) 289#define MTR_DRAM_BANKS_ADDR_BITS(mtr) ((MTR_DRAM_BANKS(mtr) == 8) ? 3 : 2) 290#define MTR_DIMM_RANK(mtr) (((mtr) >> 5) & 0x1) 291#define MTR_DIMM_RANK_ADDR_BITS(mtr) (MTR_DIMM_RANK(mtr) ? 2 : 1) 292#define MTR_DIMM_ROWS(mtr) (((mtr) >> 2) & 0x3) 293#define MTR_DIMM_ROWS_ADDR_BITS(mtr) (MTR_DIMM_ROWS(mtr) + 13) 294#define MTR_DIMM_COLS(mtr) ((mtr) & 0x3) 295#define MTR_DIMM_COLS_ADDR_BITS(mtr) (MTR_DIMM_COLS(mtr) + 10) 296 297/* This applies to FERR_NF_FB-DIMM as well as FERR_FAT_FB-DIMM */ 298static inline int extract_fbdchan_indx(u32 x) 299{ 300 return (x>>28) & 0x3; 301} 302 303/* Device name and register DID (Device ID) */ 304struct i5400_dev_info { 305 const char *ctl_name; /* name for this device */ 306 u16 fsb_mapping_errors; /* DID for the branchmap,control */ 307}; 308 309/* Table of devices attributes supported by this driver */ 310static const struct i5400_dev_info i5400_devs[] = { 311 { 312 .ctl_name = "I5400", 313 .fsb_mapping_errors = PCI_DEVICE_ID_INTEL_5400_ERR, 314 }, 315}; 316 317struct i5400_dimm_info { 318 int megabytes; /* size, 0 means not present */ 319}; 320 321/* driver private data structure */ 322struct i5400_pvt { 323 struct pci_dev *system_address; /* 16.0 */ 324 struct pci_dev *branchmap_werrors; /* 16.1 */ 325 struct pci_dev *fsb_error_regs; /* 16.2 */ 326 struct pci_dev *branch_0; /* 21.0 */ 327 struct pci_dev *branch_1; /* 22.0 */ 328 329 u16 tolm; /* top of low memory */ 330 union { 331 u64 ambase; /* AMB BAR */ 332 struct { 333 u32 ambase_bottom; 334 u32 ambase_top; 335 } u __packed; 336 }; 337 338 u16 mir0, mir1; 339 340 u16 b0_mtr[DIMMS_PER_CHANNEL]; /* Memory Technlogy Reg */ 341 u16 b0_ambpresent0; /* Branch 0, Channel 0 */ 342 u16 b0_ambpresent1; /* Brnach 0, Channel 1 */ 343 344 u16 b1_mtr[DIMMS_PER_CHANNEL]; /* Memory Technlogy Reg */ 345 u16 b1_ambpresent0; /* Branch 1, Channel 8 */ 346 u16 b1_ambpresent1; /* Branch 1, Channel 1 */ 347 348 /* DIMM information matrix, allocating architecture maximums */ 349 struct i5400_dimm_info dimm_info[DIMMS_PER_CHANNEL][MAX_CHANNELS]; 350 351 /* Actual values for this controller */ 352 int maxch; /* Max channels */ 353 int maxdimmperch; /* Max DIMMs per channel */ 354}; 355 356/* I5400 MCH error information retrieved from Hardware */ 357struct i5400_error_info { 358 /* These registers are always read from the MC */ 359 u32 ferr_fat_fbd; /* First Errors Fatal */ 360 u32 nerr_fat_fbd; /* Next Errors Fatal */ 361 u32 ferr_nf_fbd; /* First Errors Non-Fatal */ 362 u32 nerr_nf_fbd; /* Next Errors Non-Fatal */ 363 364 /* These registers are input ONLY if there was a Recoverable Error */ 365 u32 redmemb; /* Recoverable Mem Data Error log B */ 366 u16 recmema; /* Recoverable Mem Error log A */ 367 u32 recmemb; /* Recoverable Mem Error log B */ 368 369 /* These registers are input ONLY if there was a Non-Rec Error */ 370 u16 nrecmema; /* Non-Recoverable Mem log A */ 371 u32 nrecmemb; /* Non-Recoverable Mem log B */ 372 373}; 374 375/* note that nrec_rdwr changed from NRECMEMA to NRECMEMB between the 5000 and 376 5400 better to use an inline function than a macro in this case */ 377static inline int nrec_bank(struct i5400_error_info *info) 378{ 379 return ((info->nrecmema) >> 12) & 0x7; 380} 381static inline int nrec_rank(struct i5400_error_info *info) 382{ 383 return ((info->nrecmema) >> 8) & 0xf; 384} 385static inline int nrec_buf_id(struct i5400_error_info *info) 386{ 387 return ((info->nrecmema)) & 0xff; 388} 389static inline int nrec_rdwr(struct i5400_error_info *info) 390{ 391 return (info->nrecmemb) >> 31; 392} 393/* This applies to both NREC and REC string so it can be used with nrec_rdwr 394 and rec_rdwr */ 395static inline const char *rdwr_str(int rdwr) 396{ 397 return rdwr ? "Write" : "Read"; 398} 399static inline int nrec_cas(struct i5400_error_info *info) 400{ 401 return ((info->nrecmemb) >> 16) & 0x1fff; 402} 403static inline int nrec_ras(struct i5400_error_info *info) 404{ 405 return (info->nrecmemb) & 0xffff; 406} 407static inline int rec_bank(struct i5400_error_info *info) 408{ 409 return ((info->recmema) >> 12) & 0x7; 410} 411static inline int rec_rank(struct i5400_error_info *info) 412{ 413 return ((info->recmema) >> 8) & 0xf; 414} 415static inline int rec_rdwr(struct i5400_error_info *info) 416{ 417 return (info->recmemb) >> 31; 418} 419static inline int rec_cas(struct i5400_error_info *info) 420{ 421 return ((info->recmemb) >> 16) & 0x1fff; 422} 423static inline int rec_ras(struct i5400_error_info *info) 424{ 425 return (info->recmemb) & 0xffff; 426} 427 428static struct edac_pci_ctl_info *i5400_pci; 429 430/* 431 * i5400_get_error_info Retrieve the hardware error information from 432 * the hardware and cache it in the 'info' 433 * structure 434 */ 435static void i5400_get_error_info(struct mem_ctl_info *mci, 436 struct i5400_error_info *info) 437{ 438 struct i5400_pvt *pvt; 439 u32 value; 440 441 pvt = mci->pvt_info; 442 443 /* read in the 1st FATAL error register */ 444 pci_read_config_dword(pvt->branchmap_werrors, FERR_FAT_FBD, &value); 445 446 /* Mask only the bits that the doc says are valid 447 */ 448 value &= (FERR_FAT_FBDCHAN | FERR_FAT_MASK); 449 450 /* If there is an error, then read in the 451 NEXT FATAL error register and the Memory Error Log Register A 452 */ 453 if (value & FERR_FAT_MASK) { 454 info->ferr_fat_fbd = value; 455 456 /* harvest the various error data we need */ 457 pci_read_config_dword(pvt->branchmap_werrors, 458 NERR_FAT_FBD, &info->nerr_fat_fbd); 459 pci_read_config_word(pvt->branchmap_werrors, 460 NRECMEMA, &info->nrecmema); 461 pci_read_config_dword(pvt->branchmap_werrors, 462 NRECMEMB, &info->nrecmemb); 463 464 /* Clear the error bits, by writing them back */ 465 pci_write_config_dword(pvt->branchmap_werrors, 466 FERR_FAT_FBD, value); 467 } else { 468 info->ferr_fat_fbd = 0; 469 info->nerr_fat_fbd = 0; 470 info->nrecmema = 0; 471 info->nrecmemb = 0; 472 } 473 474 /* read in the 1st NON-FATAL error register */ 475 pci_read_config_dword(pvt->branchmap_werrors, FERR_NF_FBD, &value); 476 477 /* If there is an error, then read in the 1st NON-FATAL error 478 * register as well */ 479 if (value & FERR_NF_MASK) { 480 info->ferr_nf_fbd = value; 481 482 /* harvest the various error data we need */ 483 pci_read_config_dword(pvt->branchmap_werrors, 484 NERR_NF_FBD, &info->nerr_nf_fbd); 485 pci_read_config_word(pvt->branchmap_werrors, 486 RECMEMA, &info->recmema); 487 pci_read_config_dword(pvt->branchmap_werrors, 488 RECMEMB, &info->recmemb); 489 pci_read_config_dword(pvt->branchmap_werrors, 490 REDMEMB, &info->redmemb); 491 492 /* Clear the error bits, by writing them back */ 493 pci_write_config_dword(pvt->branchmap_werrors, 494 FERR_NF_FBD, value); 495 } else { 496 info->ferr_nf_fbd = 0; 497 info->nerr_nf_fbd = 0; 498 info->recmema = 0; 499 info->recmemb = 0; 500 info->redmemb = 0; 501 } 502} 503 504/* 505 * i5400_proccess_non_recoverable_info(struct mem_ctl_info *mci, 506 * struct i5400_error_info *info, 507 * int handle_errors); 508 * 509 * handle the Intel FATAL and unrecoverable errors, if any 510 */ 511static void i5400_proccess_non_recoverable_info(struct mem_ctl_info *mci, 512 struct i5400_error_info *info, 513 unsigned long allErrors) 514{ 515 char msg[EDAC_MC_LABEL_LEN + 1 + 90 + 80]; 516 int branch; 517 int channel; 518 int bank; 519 int buf_id; 520 int rank; 521 int rdwr; 522 int ras, cas; 523 int errnum; 524 char *type = NULL; 525 enum hw_event_mc_err_type tp_event = HW_EVENT_ERR_UNCORRECTED; 526 527 if (!allErrors) 528 return; /* if no error, return now */ 529 530 if (allErrors & ERROR_FAT_MASK) { 531 type = "FATAL"; 532 tp_event = HW_EVENT_ERR_FATAL; 533 } else if (allErrors & FERR_NF_UNCORRECTABLE) 534 type = "NON-FATAL uncorrected"; 535 else 536 type = "NON-FATAL recoverable"; 537 538 /* ONLY ONE of the possible error bits will be set, as per the docs */ 539 540 branch = extract_fbdchan_indx(info->ferr_fat_fbd); 541 channel = branch; 542 543 /* Use the NON-Recoverable macros to extract data */ 544 bank = nrec_bank(info); 545 rank = nrec_rank(info); 546 buf_id = nrec_buf_id(info); 547 rdwr = nrec_rdwr(info); 548 ras = nrec_ras(info); 549 cas = nrec_cas(info); 550 551 edac_dbg(0, "\t\t%s DIMM= %d Channels= %d,%d (Branch= %d DRAM Bank= %d Buffer ID = %d rdwr= %s ras= %d cas= %d)\n", 552 type, rank, channel, channel + 1, branch >> 1, bank, 553 buf_id, rdwr_str(rdwr), ras, cas); 554 555 /* Only 1 bit will be on */ 556 errnum = find_first_bit(&allErrors, ARRAY_SIZE(error_name)); 557 558 /* Form out message */ 559 snprintf(msg, sizeof(msg), 560 "Bank=%d Buffer ID = %d RAS=%d CAS=%d Err=0x%lx (%s)", 561 bank, buf_id, ras, cas, allErrors, error_name[errnum]); 562 563 edac_mc_handle_error(tp_event, mci, 1, 0, 0, 0, 564 branch >> 1, -1, rank, 565 rdwr ? "Write error" : "Read error", 566 msg); 567} 568 569/* 570 * i5400_process_fatal_error_info(struct mem_ctl_info *mci, 571 * struct i5400_error_info *info, 572 * int handle_errors); 573 * 574 * handle the Intel NON-FATAL errors, if any 575 */ 576static void i5400_process_nonfatal_error_info(struct mem_ctl_info *mci, 577 struct i5400_error_info *info) 578{ 579 char msg[EDAC_MC_LABEL_LEN + 1 + 90 + 80]; 580 unsigned long allErrors; 581 int branch; 582 int channel; 583 int bank; 584 int rank; 585 int rdwr; 586 int ras, cas; 587 int errnum; 588 589 /* mask off the Error bits that are possible */ 590 allErrors = from_nf_ferr(info->ferr_nf_fbd & FERR_NF_MASK); 591 if (!allErrors) 592 return; /* if no error, return now */ 593 594 /* ONLY ONE of the possible error bits will be set, as per the docs */ 595 596 if (allErrors & (ERROR_NF_UNCORRECTABLE | ERROR_NF_RECOVERABLE)) { 597 i5400_proccess_non_recoverable_info(mci, info, allErrors); 598 return; 599 } 600 601 /* Correctable errors */ 602 if (allErrors & ERROR_NF_CORRECTABLE) { 603 edac_dbg(0, "\tCorrected bits= 0x%lx\n", allErrors); 604 605 branch = extract_fbdchan_indx(info->ferr_nf_fbd); 606 607 channel = 0; 608 if (REC_ECC_LOCATOR_ODD(info->redmemb)) 609 channel = 1; 610 611 /* Convert channel to be based from zero, instead of 612 * from branch base of 0 */ 613 channel += branch; 614 615 bank = rec_bank(info); 616 rank = rec_rank(info); 617 rdwr = rec_rdwr(info); 618 ras = rec_ras(info); 619 cas = rec_cas(info); 620 621 /* Only 1 bit will be on */ 622 errnum = find_first_bit(&allErrors, ARRAY_SIZE(error_name)); 623 624 edac_dbg(0, "\t\tDIMM= %d Channel= %d (Branch %d DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n", 625 rank, channel, branch >> 1, bank, 626 rdwr_str(rdwr), ras, cas); 627 628 /* Form out message */ 629 snprintf(msg, sizeof(msg), 630 "Corrected error (Branch=%d DRAM-Bank=%d RDWR=%s " 631 "RAS=%d CAS=%d, CE Err=0x%lx (%s))", 632 branch >> 1, bank, rdwr_str(rdwr), ras, cas, 633 allErrors, error_name[errnum]); 634 635 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 0, 0, 0, 636 branch >> 1, channel % 2, rank, 637 rdwr ? "Write error" : "Read error", 638 msg); 639 640 return; 641 } 642 643 /* Miscellaneous errors */ 644 errnum = find_first_bit(&allErrors, ARRAY_SIZE(error_name)); 645 646 branch = extract_fbdchan_indx(info->ferr_nf_fbd); 647 648 i5400_mc_printk(mci, KERN_EMERG, 649 "Non-Fatal misc error (Branch=%d Err=%#lx (%s))", 650 branch >> 1, allErrors, error_name[errnum]); 651} 652 653/* 654 * i5400_process_error_info Process the error info that is 655 * in the 'info' structure, previously retrieved from hardware 656 */ 657static void i5400_process_error_info(struct mem_ctl_info *mci, 658 struct i5400_error_info *info) 659{ u32 allErrors; 660 661 /* First handle any fatal errors that occurred */ 662 allErrors = (info->ferr_fat_fbd & FERR_FAT_MASK); 663 i5400_proccess_non_recoverable_info(mci, info, allErrors); 664 665 /* now handle any non-fatal errors that occurred */ 666 i5400_process_nonfatal_error_info(mci, info); 667} 668 669/* 670 * i5400_clear_error Retrieve any error from the hardware 671 * but do NOT process that error. 672 * Used for 'clearing' out of previous errors 673 * Called by the Core module. 674 */ 675static void i5400_clear_error(struct mem_ctl_info *mci) 676{ 677 struct i5400_error_info info; 678 679 i5400_get_error_info(mci, &info); 680} 681 682/* 683 * i5400_check_error Retrieve and process errors reported by the 684 * hardware. Called by the Core module. 685 */ 686static void i5400_check_error(struct mem_ctl_info *mci) 687{ 688 struct i5400_error_info info; 689 690 i5400_get_error_info(mci, &info); 691 i5400_process_error_info(mci, &info); 692} 693 694/* 695 * i5400_put_devices 'put' all the devices that we have 696 * reserved via 'get' 697 */ 698static void i5400_put_devices(struct mem_ctl_info *mci) 699{ 700 struct i5400_pvt *pvt; 701 702 pvt = mci->pvt_info; 703 704 /* Decrement usage count for devices */ 705 pci_dev_put(pvt->branch_1); 706 pci_dev_put(pvt->branch_0); 707 pci_dev_put(pvt->fsb_error_regs); 708 pci_dev_put(pvt->branchmap_werrors); 709} 710 711/* 712 * i5400_get_devices Find and perform 'get' operation on the MCH's 713 * device/functions we want to reference for this driver 714 * 715 * Need to 'get' device 16 func 1 and func 2 716 */ 717static int i5400_get_devices(struct mem_ctl_info *mci, int dev_idx) 718{ 719 struct i5400_pvt *pvt; 720 struct pci_dev *pdev; 721 722 pvt = mci->pvt_info; 723 pvt->branchmap_werrors = NULL; 724 pvt->fsb_error_regs = NULL; 725 pvt->branch_0 = NULL; 726 pvt->branch_1 = NULL; 727 728 /* Attempt to 'get' the MCH register we want */ 729 pdev = NULL; 730 while (1) { 731 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 732 PCI_DEVICE_ID_INTEL_5400_ERR, pdev); 733 if (!pdev) { 734 /* End of list, leave */ 735 i5400_printk(KERN_ERR, 736 "'system address,Process Bus' " 737 "device not found:" 738 "vendor 0x%x device 0x%x ERR func 1 " 739 "(broken BIOS?)\n", 740 PCI_VENDOR_ID_INTEL, 741 PCI_DEVICE_ID_INTEL_5400_ERR); 742 return -ENODEV; 743 } 744 745 /* Store device 16 func 1 */ 746 if (PCI_FUNC(pdev->devfn) == 1) 747 break; 748 } 749 pvt->branchmap_werrors = pdev; 750 751 pdev = NULL; 752 while (1) { 753 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 754 PCI_DEVICE_ID_INTEL_5400_ERR, pdev); 755 if (!pdev) { 756 /* End of list, leave */ 757 i5400_printk(KERN_ERR, 758 "'system address,Process Bus' " 759 "device not found:" 760 "vendor 0x%x device 0x%x ERR func 2 " 761 "(broken BIOS?)\n", 762 PCI_VENDOR_ID_INTEL, 763 PCI_DEVICE_ID_INTEL_5400_ERR); 764 765 pci_dev_put(pvt->branchmap_werrors); 766 return -ENODEV; 767 } 768 769 /* Store device 16 func 2 */ 770 if (PCI_FUNC(pdev->devfn) == 2) 771 break; 772 } 773 pvt->fsb_error_regs = pdev; 774 775 edac_dbg(1, "System Address, processor bus- PCI Bus ID: %s %x:%x\n", 776 pci_name(pvt->system_address), 777 pvt->system_address->vendor, pvt->system_address->device); 778 edac_dbg(1, "Branchmap, control and errors - PCI Bus ID: %s %x:%x\n", 779 pci_name(pvt->branchmap_werrors), 780 pvt->branchmap_werrors->vendor, 781 pvt->branchmap_werrors->device); 782 edac_dbg(1, "FSB Error Regs - PCI Bus ID: %s %x:%x\n", 783 pci_name(pvt->fsb_error_regs), 784 pvt->fsb_error_regs->vendor, pvt->fsb_error_regs->device); 785 786 pvt->branch_0 = pci_get_device(PCI_VENDOR_ID_INTEL, 787 PCI_DEVICE_ID_INTEL_5400_FBD0, NULL); 788 if (!pvt->branch_0) { 789 i5400_printk(KERN_ERR, 790 "MC: 'BRANCH 0' device not found:" 791 "vendor 0x%x device 0x%x Func 0 (broken BIOS?)\n", 792 PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5400_FBD0); 793 794 pci_dev_put(pvt->fsb_error_regs); 795 pci_dev_put(pvt->branchmap_werrors); 796 return -ENODEV; 797 } 798 799 /* If this device claims to have more than 2 channels then 800 * fetch Branch 1's information 801 */ 802 if (pvt->maxch < CHANNELS_PER_BRANCH) 803 return 0; 804 805 pvt->branch_1 = pci_get_device(PCI_VENDOR_ID_INTEL, 806 PCI_DEVICE_ID_INTEL_5400_FBD1, NULL); 807 if (!pvt->branch_1) { 808 i5400_printk(KERN_ERR, 809 "MC: 'BRANCH 1' device not found:" 810 "vendor 0x%x device 0x%x Func 0 " 811 "(broken BIOS?)\n", 812 PCI_VENDOR_ID_INTEL, 813 PCI_DEVICE_ID_INTEL_5400_FBD1); 814 815 pci_dev_put(pvt->branch_0); 816 pci_dev_put(pvt->fsb_error_regs); 817 pci_dev_put(pvt->branchmap_werrors); 818 return -ENODEV; 819 } 820 821 return 0; 822} 823 824/* 825 * determine_amb_present 826 * 827 * the information is contained in DIMMS_PER_CHANNEL different 828 * registers determining which of the DIMMS_PER_CHANNEL requires 829 * knowing which channel is in question 830 * 831 * 2 branches, each with 2 channels 832 * b0_ambpresent0 for channel '0' 833 * b0_ambpresent1 for channel '1' 834 * b1_ambpresent0 for channel '2' 835 * b1_ambpresent1 for channel '3' 836 */ 837static int determine_amb_present_reg(struct i5400_pvt *pvt, int channel) 838{ 839 int amb_present; 840 841 if (channel < CHANNELS_PER_BRANCH) { 842 if (channel & 0x1) 843 amb_present = pvt->b0_ambpresent1; 844 else 845 amb_present = pvt->b0_ambpresent0; 846 } else { 847 if (channel & 0x1) 848 amb_present = pvt->b1_ambpresent1; 849 else 850 amb_present = pvt->b1_ambpresent0; 851 } 852 853 return amb_present; 854} 855 856/* 857 * determine_mtr(pvt, dimm, channel) 858 * 859 * return the proper MTR register as determine by the dimm and desired channel 860 */ 861static int determine_mtr(struct i5400_pvt *pvt, int dimm, int channel) 862{ 863 int mtr; 864 int n; 865 866 /* There is one MTR for each slot pair of FB-DIMMs, 867 Each slot pair may be at branch 0 or branch 1. 868 */ 869 n = dimm; 870 871 if (n >= DIMMS_PER_CHANNEL) { 872 edac_dbg(0, "ERROR: trying to access an invalid dimm: %d\n", 873 dimm); 874 return 0; 875 } 876 877 if (channel < CHANNELS_PER_BRANCH) 878 mtr = pvt->b0_mtr[n]; 879 else 880 mtr = pvt->b1_mtr[n]; 881 882 return mtr; 883} 884 885/* 886 */ 887static void decode_mtr(int slot_row, u16 mtr) 888{ 889 int ans; 890 891 ans = MTR_DIMMS_PRESENT(mtr); 892 893 edac_dbg(2, "\tMTR%d=0x%x: DIMMs are %sPresent\n", 894 slot_row, mtr, ans ? "" : "NOT "); 895 if (!ans) 896 return; 897 898 edac_dbg(2, "\t\tWIDTH: x%d\n", MTR_DRAM_WIDTH(mtr)); 899 900 edac_dbg(2, "\t\tELECTRICAL THROTTLING is %s\n", 901 MTR_DIMMS_ETHROTTLE(mtr) ? "enabled" : "disabled"); 902 903 edac_dbg(2, "\t\tNUMBANK: %d bank(s)\n", MTR_DRAM_BANKS(mtr)); 904 edac_dbg(2, "\t\tNUMRANK: %s\n", 905 MTR_DIMM_RANK(mtr) ? "double" : "single"); 906 edac_dbg(2, "\t\tNUMROW: %s\n", 907 MTR_DIMM_ROWS(mtr) == 0 ? "8,192 - 13 rows" : 908 MTR_DIMM_ROWS(mtr) == 1 ? "16,384 - 14 rows" : 909 MTR_DIMM_ROWS(mtr) == 2 ? "32,768 - 15 rows" : 910 "65,536 - 16 rows"); 911 edac_dbg(2, "\t\tNUMCOL: %s\n", 912 MTR_DIMM_COLS(mtr) == 0 ? "1,024 - 10 columns" : 913 MTR_DIMM_COLS(mtr) == 1 ? "2,048 - 11 columns" : 914 MTR_DIMM_COLS(mtr) == 2 ? "4,096 - 12 columns" : 915 "reserved"); 916} 917 918static void handle_channel(struct i5400_pvt *pvt, int dimm, int channel, 919 struct i5400_dimm_info *dinfo) 920{ 921 int mtr; 922 int amb_present_reg; 923 int addrBits; 924 925 mtr = determine_mtr(pvt, dimm, channel); 926 if (MTR_DIMMS_PRESENT(mtr)) { 927 amb_present_reg = determine_amb_present_reg(pvt, channel); 928 929 /* Determine if there is a DIMM present in this DIMM slot */ 930 if (amb_present_reg & (1 << dimm)) { 931 /* Start with the number of bits for a Bank 932 * on the DRAM */ 933 addrBits = MTR_DRAM_BANKS_ADDR_BITS(mtr); 934 /* Add thenumber of ROW bits */ 935 addrBits += MTR_DIMM_ROWS_ADDR_BITS(mtr); 936 /* add the number of COLUMN bits */ 937 addrBits += MTR_DIMM_COLS_ADDR_BITS(mtr); 938 /* add the number of RANK bits */ 939 addrBits += MTR_DIMM_RANK(mtr); 940 941 addrBits += 6; /* add 64 bits per DIMM */ 942 addrBits -= 20; /* divide by 2^^20 */ 943 addrBits -= 3; /* 8 bits per bytes */ 944 945 dinfo->megabytes = 1 << addrBits; 946 } 947 } 948} 949 950/* 951 * calculate_dimm_size 952 * 953 * also will output a DIMM matrix map, if debug is enabled, for viewing 954 * how the DIMMs are populated 955 */ 956static void calculate_dimm_size(struct i5400_pvt *pvt) 957{ 958 struct i5400_dimm_info *dinfo; 959 int dimm, max_dimms; 960 char *p, *mem_buffer; 961 int space, n; 962 int channel, branch; 963 964 /* ================= Generate some debug output ================= */ 965 space = PAGE_SIZE; 966 mem_buffer = p = kmalloc(space, GFP_KERNEL); 967 if (p == NULL) { 968 i5400_printk(KERN_ERR, "MC: %s:%s() kmalloc() failed\n", 969 __FILE__, __func__); 970 return; 971 } 972 973 /* Scan all the actual DIMMS 974 * and calculate the information for each DIMM 975 * Start with the highest dimm first, to display it first 976 * and work toward the 0th dimm 977 */ 978 max_dimms = pvt->maxdimmperch; 979 for (dimm = max_dimms - 1; dimm >= 0; dimm--) { 980 981 /* on an odd dimm, first output a 'boundary' marker, 982 * then reset the message buffer */ 983 if (dimm & 0x1) { 984 n = snprintf(p, space, "---------------------------" 985 "-------------------------------"); 986 p += n; 987 space -= n; 988 edac_dbg(2, "%s\n", mem_buffer); 989 p = mem_buffer; 990 space = PAGE_SIZE; 991 } 992 n = snprintf(p, space, "dimm %2d ", dimm); 993 p += n; 994 space -= n; 995 996 for (channel = 0; channel < pvt->maxch; channel++) { 997 dinfo = &pvt->dimm_info[dimm][channel]; 998 handle_channel(pvt, dimm, channel, dinfo); 999 n = snprintf(p, space, "%4d MB | ", dinfo->megabytes); 1000 p += n; 1001 space -= n; 1002 } 1003 edac_dbg(2, "%s\n", mem_buffer); 1004 p = mem_buffer; 1005 space = PAGE_SIZE; 1006 } 1007 1008 /* Output the last bottom 'boundary' marker */ 1009 n = snprintf(p, space, "---------------------------" 1010 "-------------------------------"); 1011 p += n; 1012 space -= n; 1013 edac_dbg(2, "%s\n", mem_buffer); 1014 p = mem_buffer; 1015 space = PAGE_SIZE; 1016 1017 /* now output the 'channel' labels */ 1018 n = snprintf(p, space, " "); 1019 p += n; 1020 space -= n; 1021 for (channel = 0; channel < pvt->maxch; channel++) { 1022 n = snprintf(p, space, "channel %d | ", channel); 1023 p += n; 1024 space -= n; 1025 } 1026 1027 space -= n; 1028 edac_dbg(2, "%s\n", mem_buffer); 1029 p = mem_buffer; 1030 space = PAGE_SIZE; 1031 1032 n = snprintf(p, space, " "); 1033 p += n; 1034 for (branch = 0; branch < MAX_BRANCHES; branch++) { 1035 n = snprintf(p, space, " branch %d | ", branch); 1036 p += n; 1037 space -= n; 1038 } 1039 1040 /* output the last message and free buffer */ 1041 edac_dbg(2, "%s\n", mem_buffer); 1042 kfree(mem_buffer); 1043} 1044 1045/* 1046 * i5400_get_mc_regs read in the necessary registers and 1047 * cache locally 1048 * 1049 * Fills in the private data members 1050 */ 1051static void i5400_get_mc_regs(struct mem_ctl_info *mci) 1052{ 1053 struct i5400_pvt *pvt; 1054 u32 actual_tolm; 1055 u16 limit; 1056 int slot_row; 1057 int way0, way1; 1058 1059 pvt = mci->pvt_info; 1060 1061 pci_read_config_dword(pvt->system_address, AMBASE, 1062 &pvt->u.ambase_bottom); 1063 pci_read_config_dword(pvt->system_address, AMBASE + sizeof(u32), 1064 &pvt->u.ambase_top); 1065 1066 edac_dbg(2, "AMBASE= 0x%lx MAXCH= %d MAX-DIMM-Per-CH= %d\n", 1067 (long unsigned int)pvt->ambase, pvt->maxch, pvt->maxdimmperch); 1068 1069 /* Get the Branch Map regs */ 1070 pci_read_config_word(pvt->branchmap_werrors, TOLM, &pvt->tolm); 1071 pvt->tolm >>= 12; 1072 edac_dbg(2, "\nTOLM (number of 256M regions) =%u (0x%x)\n", 1073 pvt->tolm, pvt->tolm); 1074 1075 actual_tolm = (u32) ((1000l * pvt->tolm) >> (30 - 28)); 1076 edac_dbg(2, "Actual TOLM byte addr=%u.%03u GB (0x%x)\n", 1077 actual_tolm/1000, actual_tolm % 1000, pvt->tolm << 28); 1078 1079 pci_read_config_word(pvt->branchmap_werrors, MIR0, &pvt->mir0); 1080 pci_read_config_word(pvt->branchmap_werrors, MIR1, &pvt->mir1); 1081 1082 /* Get the MIR[0-1] regs */ 1083 limit = (pvt->mir0 >> 4) & 0x0fff; 1084 way0 = pvt->mir0 & 0x1; 1085 way1 = pvt->mir0 & 0x2; 1086 edac_dbg(2, "MIR0: limit= 0x%x WAY1= %u WAY0= %x\n", 1087 limit, way1, way0); 1088 limit = (pvt->mir1 >> 4) & 0xfff; 1089 way0 = pvt->mir1 & 0x1; 1090 way1 = pvt->mir1 & 0x2; 1091 edac_dbg(2, "MIR1: limit= 0x%x WAY1= %u WAY0= %x\n", 1092 limit, way1, way0); 1093 1094 /* Get the set of MTR[0-3] regs by each branch */ 1095 for (slot_row = 0; slot_row < DIMMS_PER_CHANNEL; slot_row++) { 1096 int where = MTR0 + (slot_row * sizeof(u16)); 1097 1098 /* Branch 0 set of MTR registers */ 1099 pci_read_config_word(pvt->branch_0, where, 1100 &pvt->b0_mtr[slot_row]); 1101 1102 edac_dbg(2, "MTR%d where=0x%x B0 value=0x%x\n", 1103 slot_row, where, pvt->b0_mtr[slot_row]); 1104 1105 if (pvt->maxch < CHANNELS_PER_BRANCH) { 1106 pvt->b1_mtr[slot_row] = 0; 1107 continue; 1108 } 1109 1110 /* Branch 1 set of MTR registers */ 1111 pci_read_config_word(pvt->branch_1, where, 1112 &pvt->b1_mtr[slot_row]); 1113 edac_dbg(2, "MTR%d where=0x%x B1 value=0x%x\n", 1114 slot_row, where, pvt->b1_mtr[slot_row]); 1115 } 1116 1117 /* Read and dump branch 0's MTRs */ 1118 edac_dbg(2, "Memory Technology Registers:\n"); 1119 edac_dbg(2, " Branch 0:\n"); 1120 for (slot_row = 0; slot_row < DIMMS_PER_CHANNEL; slot_row++) 1121 decode_mtr(slot_row, pvt->b0_mtr[slot_row]); 1122 1123 pci_read_config_word(pvt->branch_0, AMBPRESENT_0, 1124 &pvt->b0_ambpresent0); 1125 edac_dbg(2, "\t\tAMB-Branch 0-present0 0x%x:\n", pvt->b0_ambpresent0); 1126 pci_read_config_word(pvt->branch_0, AMBPRESENT_1, 1127 &pvt->b0_ambpresent1); 1128 edac_dbg(2, "\t\tAMB-Branch 0-present1 0x%x:\n", pvt->b0_ambpresent1); 1129 1130 /* Only if we have 2 branchs (4 channels) */ 1131 if (pvt->maxch < CHANNELS_PER_BRANCH) { 1132 pvt->b1_ambpresent0 = 0; 1133 pvt->b1_ambpresent1 = 0; 1134 } else { 1135 /* Read and dump branch 1's MTRs */ 1136 edac_dbg(2, " Branch 1:\n"); 1137 for (slot_row = 0; slot_row < DIMMS_PER_CHANNEL; slot_row++) 1138 decode_mtr(slot_row, pvt->b1_mtr[slot_row]); 1139 1140 pci_read_config_word(pvt->branch_1, AMBPRESENT_0, 1141 &pvt->b1_ambpresent0); 1142 edac_dbg(2, "\t\tAMB-Branch 1-present0 0x%x:\n", 1143 pvt->b1_ambpresent0); 1144 pci_read_config_word(pvt->branch_1, AMBPRESENT_1, 1145 &pvt->b1_ambpresent1); 1146 edac_dbg(2, "\t\tAMB-Branch 1-present1 0x%x:\n", 1147 pvt->b1_ambpresent1); 1148 } 1149 1150 /* Go and determine the size of each DIMM and place in an 1151 * orderly matrix */ 1152 calculate_dimm_size(pvt); 1153} 1154 1155/* 1156 * i5400_init_dimms Initialize the 'dimms' table within 1157 * the mci control structure with the 1158 * addressing of memory. 1159 * 1160 * return: 1161 * 0 success 1162 * 1 no actual memory found on this MC 1163 */ 1164static int i5400_init_dimms(struct mem_ctl_info *mci) 1165{ 1166 struct i5400_pvt *pvt; 1167 struct dimm_info *dimm; 1168 int ndimms; 1169 int mtr; 1170 int size_mb; 1171 int channel, slot; 1172 1173 pvt = mci->pvt_info; 1174 1175 ndimms = 0; 1176 1177 /* 1178 * FIXME: remove pvt->dimm_info[slot][channel] and use the 3 1179 * layers here. 1180 */ 1181 for (channel = 0; channel < mci->layers[0].size * mci->layers[1].size; 1182 channel++) { 1183 for (slot = 0; slot < mci->layers[2].size; slot++) { 1184 mtr = determine_mtr(pvt, slot, channel); 1185 1186 /* if no DIMMS on this slot, continue */ 1187 if (!MTR_DIMMS_PRESENT(mtr)) 1188 continue; 1189 1190 dimm = edac_get_dimm(mci, channel / 2, channel % 2, slot); 1191 1192 size_mb = pvt->dimm_info[slot][channel].megabytes; 1193 1194 edac_dbg(2, "dimm (branch %d channel %d slot %d): %d.%03d GB\n", 1195 channel / 2, channel % 2, slot, 1196 size_mb / 1000, size_mb % 1000); 1197 1198 dimm->nr_pages = size_mb << 8; 1199 dimm->grain = 8; 1200 dimm->dtype = MTR_DRAM_WIDTH(mtr) == 8 ? 1201 DEV_X8 : DEV_X4; 1202 dimm->mtype = MEM_FB_DDR2; 1203 /* 1204 * The eccc mechanism is SDDC (aka SECC), with 1205 * is similar to Chipkill. 1206 */ 1207 dimm->edac_mode = MTR_DRAM_WIDTH(mtr) == 8 ? 1208 EDAC_S8ECD8ED : EDAC_S4ECD4ED; 1209 ndimms++; 1210 } 1211 } 1212 1213 /* 1214 * When just one memory is provided, it should be at location (0,0,0). 1215 * With such single-DIMM mode, the SDCC algorithm degrades to SECDEC+. 1216 */ 1217 if (ndimms == 1) 1218 mci->dimms[0]->edac_mode = EDAC_SECDED; 1219 1220 return (ndimms == 0); 1221} 1222 1223/* 1224 * i5400_enable_error_reporting 1225 * Turn on the memory reporting features of the hardware 1226 */ 1227static void i5400_enable_error_reporting(struct mem_ctl_info *mci) 1228{ 1229 struct i5400_pvt *pvt; 1230 u32 fbd_error_mask; 1231 1232 pvt = mci->pvt_info; 1233 1234 /* Read the FBD Error Mask Register */ 1235 pci_read_config_dword(pvt->branchmap_werrors, EMASK_FBD, 1236 &fbd_error_mask); 1237 1238 /* Enable with a '0' */ 1239 fbd_error_mask &= ~(ENABLE_EMASK_ALL); 1240 1241 pci_write_config_dword(pvt->branchmap_werrors, EMASK_FBD, 1242 fbd_error_mask); 1243} 1244 1245/* 1246 * i5400_probe1 Probe for ONE instance of device to see if it is 1247 * present. 1248 * return: 1249 * 0 for FOUND a device 1250 * < 0 for error code 1251 */ 1252static int i5400_probe1(struct pci_dev *pdev, int dev_idx) 1253{ 1254 struct mem_ctl_info *mci; 1255 struct i5400_pvt *pvt; 1256 struct edac_mc_layer layers[3]; 1257 1258 if (dev_idx >= ARRAY_SIZE(i5400_devs)) 1259 return -EINVAL; 1260 1261 edac_dbg(0, "MC: pdev bus %u dev=0x%x fn=0x%x\n", 1262 pdev->bus->number, 1263 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 1264 1265 /* We only are looking for func 0 of the set */ 1266 if (PCI_FUNC(pdev->devfn) != 0) 1267 return -ENODEV; 1268 1269 /* 1270 * allocate a new MC control structure 1271 * 1272 * This drivers uses the DIMM slot as "csrow" and the rest as "channel". 1273 */ 1274 layers[0].type = EDAC_MC_LAYER_BRANCH; 1275 layers[0].size = MAX_BRANCHES; 1276 layers[0].is_virt_csrow = false; 1277 layers[1].type = EDAC_MC_LAYER_CHANNEL; 1278 layers[1].size = CHANNELS_PER_BRANCH; 1279 layers[1].is_virt_csrow = false; 1280 layers[2].type = EDAC_MC_LAYER_SLOT; 1281 layers[2].size = DIMMS_PER_CHANNEL; 1282 layers[2].is_virt_csrow = true; 1283 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(*pvt)); 1284 if (mci == NULL) 1285 return -ENOMEM; 1286 1287 edac_dbg(0, "MC: mci = %p\n", mci); 1288 1289 mci->pdev = &pdev->dev; /* record ptr to the generic device */ 1290 1291 pvt = mci->pvt_info; 1292 pvt->system_address = pdev; /* Record this device in our private */ 1293 pvt->maxch = MAX_CHANNELS; 1294 pvt->maxdimmperch = DIMMS_PER_CHANNEL; 1295 1296 /* 'get' the pci devices we want to reserve for our use */ 1297 if (i5400_get_devices(mci, dev_idx)) 1298 goto fail0; 1299 1300 /* Time to get serious */ 1301 i5400_get_mc_regs(mci); /* retrieve the hardware registers */ 1302 1303 mci->mc_idx = 0; 1304 mci->mtype_cap = MEM_FLAG_FB_DDR2; 1305 mci->edac_ctl_cap = EDAC_FLAG_NONE; 1306 mci->edac_cap = EDAC_FLAG_NONE; 1307 mci->mod_name = "i5400_edac.c"; 1308 mci->ctl_name = i5400_devs[dev_idx].ctl_name; 1309 mci->dev_name = pci_name(pdev); 1310 mci->ctl_page_to_phys = NULL; 1311 1312 /* Set the function pointer to an actual operation function */ 1313 mci->edac_check = i5400_check_error; 1314 1315 /* initialize the MC control structure 'dimms' table 1316 * with the mapping and control information */ 1317 if (i5400_init_dimms(mci)) { 1318 edac_dbg(0, "MC: Setting mci->edac_cap to EDAC_FLAG_NONE because i5400_init_dimms() returned nonzero value\n"); 1319 mci->edac_cap = EDAC_FLAG_NONE; /* no dimms found */ 1320 } else { 1321 edac_dbg(1, "MC: Enable error reporting now\n"); 1322 i5400_enable_error_reporting(mci); 1323 } 1324 1325 /* add this new MC control structure to EDAC's list of MCs */ 1326 if (edac_mc_add_mc(mci)) { 1327 edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); 1328 /* FIXME: perhaps some code should go here that disables error 1329 * reporting if we just enabled it 1330 */ 1331 goto fail1; 1332 } 1333 1334 i5400_clear_error(mci); 1335 1336 /* allocating generic PCI control info */ 1337 i5400_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR); 1338 if (!i5400_pci) { 1339 printk(KERN_WARNING 1340 "%s(): Unable to create PCI control\n", 1341 __func__); 1342 printk(KERN_WARNING 1343 "%s(): PCI error report via EDAC not setup\n", 1344 __func__); 1345 } 1346 1347 return 0; 1348 1349 /* Error exit unwinding stack */ 1350fail1: 1351 1352 i5400_put_devices(mci); 1353 1354fail0: 1355 edac_mc_free(mci); 1356 return -ENODEV; 1357} 1358 1359/* 1360 * i5400_init_one constructor for one instance of device 1361 * 1362 * returns: 1363 * negative on error 1364 * count (>= 0) 1365 */ 1366static int i5400_init_one(struct pci_dev *pdev, const struct pci_device_id *id) 1367{ 1368 int rc; 1369 1370 edac_dbg(0, "MC:\n"); 1371 1372 /* wake up device */ 1373 rc = pci_enable_device(pdev); 1374 if (rc) 1375 return rc; 1376 1377 /* now probe and enable the device */ 1378 return i5400_probe1(pdev, id->driver_data); 1379} 1380 1381/* 1382 * i5400_remove_one destructor for one instance of device 1383 * 1384 */ 1385static void i5400_remove_one(struct pci_dev *pdev) 1386{ 1387 struct mem_ctl_info *mci; 1388 1389 edac_dbg(0, "\n"); 1390 1391 if (i5400_pci) 1392 edac_pci_release_generic_ctl(i5400_pci); 1393 1394 mci = edac_mc_del_mc(&pdev->dev); 1395 if (!mci) 1396 return; 1397 1398 /* retrieve references to resources, and free those resources */ 1399 i5400_put_devices(mci); 1400 1401 pci_disable_device(pdev); 1402 1403 edac_mc_free(mci); 1404} 1405 1406/* 1407 * pci_device_id table for which devices we are looking for 1408 * 1409 * The "E500P" device is the first device supported. 1410 */ 1411static const struct pci_device_id i5400_pci_tbl[] = { 1412 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5400_ERR)}, 1413 {0,} /* 0 terminated list. */ 1414}; 1415 1416MODULE_DEVICE_TABLE(pci, i5400_pci_tbl); 1417 1418/* 1419 * i5400_driver pci_driver structure for this module 1420 * 1421 */ 1422static struct pci_driver i5400_driver = { 1423 .name = "i5400_edac", 1424 .probe = i5400_init_one, 1425 .remove = i5400_remove_one, 1426 .id_table = i5400_pci_tbl, 1427}; 1428 1429/* 1430 * i5400_init Module entry function 1431 * Try to initialize this module for its devices 1432 */ 1433static int __init i5400_init(void) 1434{ 1435 int pci_rc; 1436 1437 edac_dbg(2, "MC:\n"); 1438 1439 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 1440 opstate_init(); 1441 1442 pci_rc = pci_register_driver(&i5400_driver); 1443 1444 return (pci_rc < 0) ? pci_rc : 0; 1445} 1446 1447/* 1448 * i5400_exit() Module exit function 1449 * Unregister the driver 1450 */ 1451static void __exit i5400_exit(void) 1452{ 1453 edac_dbg(2, "MC:\n"); 1454 pci_unregister_driver(&i5400_driver); 1455} 1456 1457module_init(i5400_init); 1458module_exit(i5400_exit); 1459 1460MODULE_LICENSE("GPL"); 1461MODULE_AUTHOR("Ben Woodard <woodard@redhat.com>"); 1462MODULE_AUTHOR("Mauro Carvalho Chehab"); 1463MODULE_AUTHOR("Red Hat Inc. (https://www.redhat.com)"); 1464MODULE_DESCRIPTION("MC Driver for Intel I5400 memory controllers - " 1465 I5400_REVISION); 1466 1467module_param(edac_op_state, int, 0444); 1468MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");