1 /* 2 * Intel 5000(P/V/X) class Memory Controllers kernel module 3 * 4 * This file may be distributed under the terms of the 5 * GNU General Public License. 6 * 7 * Written by Douglas Thompson Linux Networx (http://lnxi.com) 8 * norsk5@xmission.com 9 * 10 * This module is based on the following document: 11 * 12 * Intel 5000X Chipset Memory Controller Hub (MCH) - Datasheet 13 * http://developer.intel.com/design/chipsets/datashts/313070.htm 14 * 15 */ 16 17 #include <linux/module.h> 18 #include <linux/init.h> 19 #include <linux/pci.h> 20 #include <linux/pci_ids.h> 21 #include <linux/slab.h> 22 #include <linux/edac.h> 23 #include <asm/mmzone.h> 24 25 #include "edac_core.h" 26 27 /* 28 * Alter this version for the I5000 module when modifications are made 29 */ 30 #define I5000_REVISION " Ver: 2.0.12" 31 #define EDAC_MOD_STR "i5000_edac" 32 33 #define i5000_printk(level, fmt, arg...) \ 34 edac_printk(level, "i5000", fmt, ##arg) 35 36 #define i5000_mc_printk(mci, level, fmt, arg...) \ 37 edac_mc_chipset_printk(mci, level, "i5000", fmt, ##arg) 38 39 #ifndef PCI_DEVICE_ID_INTEL_FBD_0 40 #define PCI_DEVICE_ID_INTEL_FBD_0 0x25F5 41 #endif 42 #ifndef PCI_DEVICE_ID_INTEL_FBD_1 43 #define PCI_DEVICE_ID_INTEL_FBD_1 0x25F6 44 #endif 45 46 /* Device 16, 47 * Function 0: System Address 48 * Function 1: Memory Branch Map, Control, Errors Register 49 * Function 2: FSB Error Registers 50 * 51 * All 3 functions of Device 16 (0,1,2) share the SAME DID 52 */ 53 #define PCI_DEVICE_ID_INTEL_I5000_DEV16 0x25F0 54 55 /* OFFSETS for Function 0 */ 56 57 /* OFFSETS for Function 1 */ 58 #define AMBASE 0x48 59 #define MAXCH 0x56 60 #define MAXDIMMPERCH 0x57 61 #define TOLM 0x6C 62 #define REDMEMB 0x7C 63 #define RED_ECC_LOCATOR(x) ((x) & 0x3FFFF) 64 #define REC_ECC_LOCATOR_EVEN(x) ((x) & 0x001FF) 65 #define REC_ECC_LOCATOR_ODD(x) ((x) & 0x3FE00) 66 #define MIR0 0x80 67 #define MIR1 0x84 68 #define MIR2 0x88 69 #define AMIR0 0x8C 70 #define AMIR1 0x90 71 #define AMIR2 0x94 72 73 #define FERR_FAT_FBD 0x98 74 #define NERR_FAT_FBD 0x9C 75 #define EXTRACT_FBDCHAN_INDX(x) (((x)>>28) & 0x3) 76 #define FERR_FAT_FBDCHAN 0x30000000 77 #define FERR_FAT_M3ERR 0x00000004 78 #define FERR_FAT_M2ERR 0x00000002 79 #define FERR_FAT_M1ERR 0x00000001 80 #define FERR_FAT_MASK (FERR_FAT_M1ERR | \ 81 FERR_FAT_M2ERR | \ 82 FERR_FAT_M3ERR) 83 84 #define FERR_NF_FBD 0xA0 85 86 /* Thermal and SPD or BFD errors */ 87 #define FERR_NF_M28ERR 0x01000000 88 #define FERR_NF_M27ERR 0x00800000 89 #define FERR_NF_M26ERR 0x00400000 90 #define FERR_NF_M25ERR 0x00200000 91 #define FERR_NF_M24ERR 0x00100000 92 #define FERR_NF_M23ERR 0x00080000 93 #define FERR_NF_M22ERR 0x00040000 94 #define FERR_NF_M21ERR 0x00020000 95 96 /* Correctable errors */ 97 #define FERR_NF_M20ERR 0x00010000 98 #define FERR_NF_M19ERR 0x00008000 99 #define FERR_NF_M18ERR 0x00004000 100 #define FERR_NF_M17ERR 0x00002000 101 102 /* Non-Retry or redundant Retry errors */ 103 #define FERR_NF_M16ERR 0x00001000 104 #define FERR_NF_M15ERR 0x00000800 105 #define FERR_NF_M14ERR 0x00000400 106 #define FERR_NF_M13ERR 0x00000200 107 108 /* Uncorrectable errors */ 109 #define FERR_NF_M12ERR 0x00000100 110 #define FERR_NF_M11ERR 0x00000080 111 #define FERR_NF_M10ERR 0x00000040 112 #define FERR_NF_M9ERR 0x00000020 113 #define FERR_NF_M8ERR 0x00000010 114 #define FERR_NF_M7ERR 0x00000008 115 #define FERR_NF_M6ERR 0x00000004 116 #define FERR_NF_M5ERR 0x00000002 117 #define FERR_NF_M4ERR 0x00000001 118 119 #define FERR_NF_UNCORRECTABLE (FERR_NF_M12ERR | \ 120 FERR_NF_M11ERR | \ 121 FERR_NF_M10ERR | \ 122 FERR_NF_M9ERR | \ 123 FERR_NF_M8ERR | \ 124 FERR_NF_M7ERR | \ 125 FERR_NF_M6ERR | \ 126 FERR_NF_M5ERR | \ 127 FERR_NF_M4ERR) 128 #define FERR_NF_CORRECTABLE (FERR_NF_M20ERR | \ 129 FERR_NF_M19ERR | \ 130 FERR_NF_M18ERR | \ 131 FERR_NF_M17ERR) 132 #define FERR_NF_DIMM_SPARE (FERR_NF_M27ERR | \ 133 FERR_NF_M28ERR) 134 #define FERR_NF_THERMAL (FERR_NF_M26ERR | \ 135 FERR_NF_M25ERR | \ 136 FERR_NF_M24ERR | \ 137 FERR_NF_M23ERR) 138 #define FERR_NF_SPD_PROTOCOL (FERR_NF_M22ERR) 139 #define FERR_NF_NORTH_CRC (FERR_NF_M21ERR) 140 #define FERR_NF_NON_RETRY (FERR_NF_M13ERR | \ 141 FERR_NF_M14ERR | \ 142 FERR_NF_M15ERR) 143 144 #define NERR_NF_FBD 0xA4 145 #define FERR_NF_MASK (FERR_NF_UNCORRECTABLE | \ 146 FERR_NF_CORRECTABLE | \ 147 FERR_NF_DIMM_SPARE | \ 148 FERR_NF_THERMAL | \ 149 FERR_NF_SPD_PROTOCOL | \ 150 FERR_NF_NORTH_CRC | \ 151 FERR_NF_NON_RETRY) 152 153 #define EMASK_FBD 0xA8 154 #define EMASK_FBD_M28ERR 0x08000000 155 #define EMASK_FBD_M27ERR 0x04000000 156 #define EMASK_FBD_M26ERR 0x02000000 157 #define EMASK_FBD_M25ERR 0x01000000 158 #define EMASK_FBD_M24ERR 0x00800000 159 #define EMASK_FBD_M23ERR 0x00400000 160 #define EMASK_FBD_M22ERR 0x00200000 161 #define EMASK_FBD_M21ERR 0x00100000 162 #define EMASK_FBD_M20ERR 0x00080000 163 #define EMASK_FBD_M19ERR 0x00040000 164 #define EMASK_FBD_M18ERR 0x00020000 165 #define EMASK_FBD_M17ERR 0x00010000 166 167 #define EMASK_FBD_M15ERR 0x00004000 168 #define EMASK_FBD_M14ERR 0x00002000 169 #define EMASK_FBD_M13ERR 0x00001000 170 #define EMASK_FBD_M12ERR 0x00000800 171 #define EMASK_FBD_M11ERR 0x00000400 172 #define EMASK_FBD_M10ERR 0x00000200 173 #define EMASK_FBD_M9ERR 0x00000100 174 #define EMASK_FBD_M8ERR 0x00000080 175 #define EMASK_FBD_M7ERR 0x00000040 176 #define EMASK_FBD_M6ERR 0x00000020 177 #define EMASK_FBD_M5ERR 0x00000010 178 #define EMASK_FBD_M4ERR 0x00000008 179 #define EMASK_FBD_M3ERR 0x00000004 180 #define EMASK_FBD_M2ERR 0x00000002 181 #define EMASK_FBD_M1ERR 0x00000001 182 183 #define ENABLE_EMASK_FBD_FATAL_ERRORS (EMASK_FBD_M1ERR | \ 184 EMASK_FBD_M2ERR | \ 185 EMASK_FBD_M3ERR) 186 187 #define ENABLE_EMASK_FBD_UNCORRECTABLE (EMASK_FBD_M4ERR | \ 188 EMASK_FBD_M5ERR | \ 189 EMASK_FBD_M6ERR | \ 190 EMASK_FBD_M7ERR | \ 191 EMASK_FBD_M8ERR | \ 192 EMASK_FBD_M9ERR | \ 193 EMASK_FBD_M10ERR | \ 194 EMASK_FBD_M11ERR | \ 195 EMASK_FBD_M12ERR) 196 #define ENABLE_EMASK_FBD_CORRECTABLE (EMASK_FBD_M17ERR | \ 197 EMASK_FBD_M18ERR | \ 198 EMASK_FBD_M19ERR | \ 199 EMASK_FBD_M20ERR) 200 #define ENABLE_EMASK_FBD_DIMM_SPARE (EMASK_FBD_M27ERR | \ 201 EMASK_FBD_M28ERR) 202 #define ENABLE_EMASK_FBD_THERMALS (EMASK_FBD_M26ERR | \ 203 EMASK_FBD_M25ERR | \ 204 EMASK_FBD_M24ERR | \ 205 EMASK_FBD_M23ERR) 206 #define ENABLE_EMASK_FBD_SPD_PROTOCOL (EMASK_FBD_M22ERR) 207 #define ENABLE_EMASK_FBD_NORTH_CRC (EMASK_FBD_M21ERR) 208 #define ENABLE_EMASK_FBD_NON_RETRY (EMASK_FBD_M15ERR | \ 209 EMASK_FBD_M14ERR | \ 210 EMASK_FBD_M13ERR) 211 212 #define ENABLE_EMASK_ALL (ENABLE_EMASK_FBD_NON_RETRY | \ 213 ENABLE_EMASK_FBD_NORTH_CRC | \ 214 ENABLE_EMASK_FBD_SPD_PROTOCOL | \ 215 ENABLE_EMASK_FBD_THERMALS | \ 216 ENABLE_EMASK_FBD_DIMM_SPARE | \ 217 ENABLE_EMASK_FBD_FATAL_ERRORS | \ 218 ENABLE_EMASK_FBD_CORRECTABLE | \ 219 ENABLE_EMASK_FBD_UNCORRECTABLE) 220 221 #define ERR0_FBD 0xAC 222 #define ERR1_FBD 0xB0 223 #define ERR2_FBD 0xB4 224 #define MCERR_FBD 0xB8 225 #define NRECMEMA 0xBE 226 #define NREC_BANK(x) (((x)>>12) & 0x7) 227 #define NREC_RDWR(x) (((x)>>11) & 1) 228 #define NREC_RANK(x) (((x)>>8) & 0x7) 229 #define NRECMEMB 0xC0 230 #define NREC_CAS(x) (((x)>>16) & 0xFFFFFF) 231 #define NREC_RAS(x) ((x) & 0x7FFF) 232 #define NRECFGLOG 0xC4 233 #define NREEECFBDA 0xC8 234 #define NREEECFBDB 0xCC 235 #define NREEECFBDC 0xD0 236 #define NREEECFBDD 0xD4 237 #define NREEECFBDE 0xD8 238 #define REDMEMA 0xDC 239 #define RECMEMA 0xE2 240 #define REC_BANK(x) (((x)>>12) & 0x7) 241 #define REC_RDWR(x) (((x)>>11) & 1) 242 #define REC_RANK(x) (((x)>>8) & 0x7) 243 #define RECMEMB 0xE4 244 #define REC_CAS(x) (((x)>>16) & 0xFFFFFF) 245 #define REC_RAS(x) ((x) & 0x7FFF) 246 #define RECFGLOG 0xE8 247 #define RECFBDA 0xEC 248 #define RECFBDB 0xF0 249 #define RECFBDC 0xF4 250 #define RECFBDD 0xF8 251 #define RECFBDE 0xFC 252 253 /* OFFSETS for Function 2 */ 254 255 /* 256 * Device 21, 257 * Function 0: Memory Map Branch 0 258 * 259 * Device 22, 260 * Function 0: Memory Map Branch 1 261 */ 262 #define PCI_DEVICE_ID_I5000_BRANCH_0 0x25F5 263 #define PCI_DEVICE_ID_I5000_BRANCH_1 0x25F6 264 265 #define AMB_PRESENT_0 0x64 266 #define AMB_PRESENT_1 0x66 267 #define MTR0 0x80 268 #define MTR1 0x84 269 #define MTR2 0x88 270 #define MTR3 0x8C 271 272 #define NUM_MTRS 4 273 #define CHANNELS_PER_BRANCH 2 274 #define MAX_BRANCHES 2 275 276 /* Defines to extract the various fields from the 277 * MTRx - Memory Technology Registers 278 */ 279 #define MTR_DIMMS_PRESENT(mtr) ((mtr) & (0x1 << 8)) 280 #define MTR_DRAM_WIDTH(mtr) ((((mtr) >> 6) & 0x1) ? 8 : 4) 281 #define MTR_DRAM_BANKS(mtr) ((((mtr) >> 5) & 0x1) ? 8 : 4) 282 #define MTR_DRAM_BANKS_ADDR_BITS(mtr) ((MTR_DRAM_BANKS(mtr) == 8) ? 3 : 2) 283 #define MTR_DIMM_RANK(mtr) (((mtr) >> 4) & 0x1) 284 #define MTR_DIMM_RANK_ADDR_BITS(mtr) (MTR_DIMM_RANK(mtr) ? 2 : 1) 285 #define MTR_DIMM_ROWS(mtr) (((mtr) >> 2) & 0x3) 286 #define MTR_DIMM_ROWS_ADDR_BITS(mtr) (MTR_DIMM_ROWS(mtr) + 13) 287 #define MTR_DIMM_COLS(mtr) ((mtr) & 0x3) 288 #define MTR_DIMM_COLS_ADDR_BITS(mtr) (MTR_DIMM_COLS(mtr) + 10) 289 290 /* enables the report of miscellaneous messages as CE errors - default off */ 291 static int misc_messages; 292 293 /* Enumeration of supported devices */ 294 enum i5000_chips { 295 I5000P = 0, 296 I5000V = 1, /* future */ 297 I5000X = 2 /* future */ 298 }; 299 300 /* Device name and register DID (Device ID) */ 301 struct i5000_dev_info { 302 const char *ctl_name; /* name for this device */ 303 u16 fsb_mapping_errors; /* DID for the branchmap,control */ 304 }; 305 306 /* Table of devices attributes supported by this driver */ 307 static const struct i5000_dev_info i5000_devs[] = { 308 [I5000P] = { 309 .ctl_name = "I5000", 310 .fsb_mapping_errors = PCI_DEVICE_ID_INTEL_I5000_DEV16, 311 }, 312 }; 313 314 struct i5000_dimm_info { 315 int megabytes; /* size, 0 means not present */ 316 int dual_rank; 317 }; 318 319 #define MAX_CHANNELS 6 /* max possible channels */ 320 #define MAX_CSROWS (8*2) /* max possible csrows per channel */ 321 322 /* driver private data structure */ 323 struct i5000_pvt { 324 struct pci_dev *system_address; /* 16.0 */ 325 struct pci_dev *branchmap_werrors; /* 16.1 */ 326 struct pci_dev *fsb_error_regs; /* 16.2 */ 327 struct pci_dev *branch_0; /* 21.0 */ 328 struct pci_dev *branch_1; /* 22.0 */ 329 330 u16 tolm; /* top of low memory */ 331 union { 332 u64 ambase; /* AMB BAR */ 333 struct { 334 u32 ambase_bottom; 335 u32 ambase_top; 336 } u __packed; 337 }; 338 339 u16 mir0, mir1, mir2; 340 341 u16 b0_mtr[NUM_MTRS]; /* Memory Technlogy Reg */ 342 u16 b0_ambpresent0; /* Branch 0, Channel 0 */ 343 u16 b0_ambpresent1; /* Brnach 0, Channel 1 */ 344 345 u16 b1_mtr[NUM_MTRS]; /* Memory Technlogy Reg */ 346 u16 b1_ambpresent0; /* Branch 1, Channel 8 */ 347 u16 b1_ambpresent1; /* Branch 1, Channel 1 */ 348 349 /* DIMM information matrix, allocating architecture maximums */ 350 struct i5000_dimm_info dimm_info[MAX_CSROWS][MAX_CHANNELS]; 351 352 /* Actual values for this controller */ 353 int maxch; /* Max channels */ 354 int maxdimmperch; /* Max DIMMs per channel */ 355 }; 356 357 /* I5000 MCH error information retrieved from Hardware */ 358 struct i5000_error_info { 359 360 /* These registers are always read from the MC */ 361 u32 ferr_fat_fbd; /* First Errors Fatal */ 362 u32 nerr_fat_fbd; /* Next Errors Fatal */ 363 u32 ferr_nf_fbd; /* First Errors Non-Fatal */ 364 u32 nerr_nf_fbd; /* Next Errors Non-Fatal */ 365 366 /* These registers are input ONLY if there was a Recoverable Error */ 367 u32 redmemb; /* Recoverable Mem Data Error log B */ 368 u16 recmema; /* Recoverable Mem Error log A */ 369 u32 recmemb; /* Recoverable Mem Error log B */ 370 371 /* These registers are input ONLY if there was a 372 * Non-Recoverable Error */ 373 u16 nrecmema; /* Non-Recoverable Mem log A */ 374 u16 nrecmemb; /* Non-Recoverable Mem log B */ 375 376 }; 377 378 static struct edac_pci_ctl_info *i5000_pci; 379 380 /* 381 * i5000_get_error_info Retrieve the hardware error information from 382 * the hardware and cache it in the 'info' 383 * structure 384 */ 385 static void i5000_get_error_info(struct mem_ctl_info *mci, 386 struct i5000_error_info *info) 387 { 388 struct i5000_pvt *pvt; 389 u32 value; 390 391 pvt = mci->pvt_info; 392 393 /* read in the 1st FATAL error register */ 394 pci_read_config_dword(pvt->branchmap_werrors, FERR_FAT_FBD, &value); 395 396 /* Mask only the bits that the doc says are valid 397 */ 398 value &= (FERR_FAT_FBDCHAN | FERR_FAT_MASK); 399 400 /* If there is an error, then read in the */ 401 /* NEXT FATAL error register and the Memory Error Log Register A */ 402 if (value & FERR_FAT_MASK) { 403 info->ferr_fat_fbd = value; 404 405 /* harvest the various error data we need */ 406 pci_read_config_dword(pvt->branchmap_werrors, 407 NERR_FAT_FBD, &info->nerr_fat_fbd); 408 pci_read_config_word(pvt->branchmap_werrors, 409 NRECMEMA, &info->nrecmema); 410 pci_read_config_word(pvt->branchmap_werrors, 411 NRECMEMB, &info->nrecmemb); 412 413 /* Clear the error bits, by writing them back */ 414 pci_write_config_dword(pvt->branchmap_werrors, 415 FERR_FAT_FBD, value); 416 } else { 417 info->ferr_fat_fbd = 0; 418 info->nerr_fat_fbd = 0; 419 info->nrecmema = 0; 420 info->nrecmemb = 0; 421 } 422 423 /* read in the 1st NON-FATAL error register */ 424 pci_read_config_dword(pvt->branchmap_werrors, FERR_NF_FBD, &value); 425 426 /* If there is an error, then read in the 1st NON-FATAL error 427 * register as well */ 428 if (value & FERR_NF_MASK) { 429 info->ferr_nf_fbd = value; 430 431 /* harvest the various error data we need */ 432 pci_read_config_dword(pvt->branchmap_werrors, 433 NERR_NF_FBD, &info->nerr_nf_fbd); 434 pci_read_config_word(pvt->branchmap_werrors, 435 RECMEMA, &info->recmema); 436 pci_read_config_dword(pvt->branchmap_werrors, 437 RECMEMB, &info->recmemb); 438 pci_read_config_dword(pvt->branchmap_werrors, 439 REDMEMB, &info->redmemb); 440 441 /* Clear the error bits, by writing them back */ 442 pci_write_config_dword(pvt->branchmap_werrors, 443 FERR_NF_FBD, value); 444 } else { 445 info->ferr_nf_fbd = 0; 446 info->nerr_nf_fbd = 0; 447 info->recmema = 0; 448 info->recmemb = 0; 449 info->redmemb = 0; 450 } 451 } 452 453 /* 454 * i5000_process_fatal_error_info(struct mem_ctl_info *mci, 455 * struct i5000_error_info *info, 456 * int handle_errors); 457 * 458 * handle the Intel FATAL errors, if any 459 */ 460 static void i5000_process_fatal_error_info(struct mem_ctl_info *mci, 461 struct i5000_error_info *info, 462 int handle_errors) 463 { 464 char msg[EDAC_MC_LABEL_LEN + 1 + 160]; 465 char *specific = NULL; 466 u32 allErrors; 467 int channel; 468 int bank; 469 int rank; 470 int rdwr; 471 int ras, cas; 472 473 /* mask off the Error bits that are possible */ 474 allErrors = (info->ferr_fat_fbd & FERR_FAT_MASK); 475 if (!allErrors) 476 return; /* if no error, return now */ 477 478 channel = EXTRACT_FBDCHAN_INDX(info->ferr_fat_fbd); 479 480 /* Use the NON-Recoverable macros to extract data */ 481 bank = NREC_BANK(info->nrecmema); 482 rank = NREC_RANK(info->nrecmema); 483 rdwr = NREC_RDWR(info->nrecmema); 484 ras = NREC_RAS(info->nrecmemb); 485 cas = NREC_CAS(info->nrecmemb); 486 487 edac_dbg(0, "\t\tCSROW= %d Channel= %d (DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n", 488 rank, channel, bank, 489 rdwr ? "Write" : "Read", ras, cas); 490 491 /* Only 1 bit will be on */ 492 switch (allErrors) { 493 case FERR_FAT_M1ERR: 494 specific = "Alert on non-redundant retry or fast " 495 "reset timeout"; 496 break; 497 case FERR_FAT_M2ERR: 498 specific = "Northbound CRC error on non-redundant " 499 "retry"; 500 break; 501 case FERR_FAT_M3ERR: 502 { 503 static int done; 504 505 /* 506 * This error is generated to inform that the intelligent 507 * throttling is disabled and the temperature passed the 508 * specified middle point. Since this is something the BIOS 509 * should take care of, we'll warn only once to avoid 510 * worthlessly flooding the log. 511 */ 512 if (done) 513 return; 514 done++; 515 516 specific = ">Tmid Thermal event with intelligent " 517 "throttling disabled"; 518 } 519 break; 520 } 521 522 /* Form out message */ 523 snprintf(msg, sizeof(msg), 524 "Bank=%d RAS=%d CAS=%d FATAL Err=0x%x (%s)", 525 bank, ras, cas, allErrors, specific); 526 527 /* Call the helper to output message */ 528 edac_mc_handle_error(HW_EVENT_ERR_FATAL, mci, 1, 0, 0, 0, 529 channel >> 1, channel & 1, rank, 530 rdwr ? "Write error" : "Read error", 531 msg); 532 } 533 534 /* 535 * i5000_process_fatal_error_info(struct mem_ctl_info *mci, 536 * struct i5000_error_info *info, 537 * int handle_errors); 538 * 539 * handle the Intel NON-FATAL errors, if any 540 */ 541 static void i5000_process_nonfatal_error_info(struct mem_ctl_info *mci, 542 struct i5000_error_info *info, 543 int handle_errors) 544 { 545 char msg[EDAC_MC_LABEL_LEN + 1 + 170]; 546 char *specific = NULL; 547 u32 allErrors; 548 u32 ue_errors; 549 u32 ce_errors; 550 u32 misc_errors; 551 int branch; 552 int channel; 553 int bank; 554 int rank; 555 int rdwr; 556 int ras, cas; 557 558 /* mask off the Error bits that are possible */ 559 allErrors = (info->ferr_nf_fbd & FERR_NF_MASK); 560 if (!allErrors) 561 return; /* if no error, return now */ 562 563 /* ONLY ONE of the possible error bits will be set, as per the docs */ 564 ue_errors = allErrors & FERR_NF_UNCORRECTABLE; 565 if (ue_errors) { 566 edac_dbg(0, "\tUncorrected bits= 0x%x\n", ue_errors); 567 568 branch = EXTRACT_FBDCHAN_INDX(info->ferr_nf_fbd); 569 570 /* 571 * According with i5000 datasheet, bit 28 has no significance 572 * for errors M4Err-M12Err and M17Err-M21Err, on FERR_NF_FBD 573 */ 574 channel = branch & 2; 575 576 bank = NREC_BANK(info->nrecmema); 577 rank = NREC_RANK(info->nrecmema); 578 rdwr = NREC_RDWR(info->nrecmema); 579 ras = NREC_RAS(info->nrecmemb); 580 cas = NREC_CAS(info->nrecmemb); 581 582 edac_dbg(0, "\t\tCSROW= %d Channels= %d,%d (Branch= %d DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n", 583 rank, channel, channel + 1, branch >> 1, bank, 584 rdwr ? "Write" : "Read", ras, cas); 585 586 switch (ue_errors) { 587 case FERR_NF_M12ERR: 588 specific = "Non-Aliased Uncorrectable Patrol Data ECC"; 589 break; 590 case FERR_NF_M11ERR: 591 specific = "Non-Aliased Uncorrectable Spare-Copy " 592 "Data ECC"; 593 break; 594 case FERR_NF_M10ERR: 595 specific = "Non-Aliased Uncorrectable Mirrored Demand " 596 "Data ECC"; 597 break; 598 case FERR_NF_M9ERR: 599 specific = "Non-Aliased Uncorrectable Non-Mirrored " 600 "Demand Data ECC"; 601 break; 602 case FERR_NF_M8ERR: 603 specific = "Aliased Uncorrectable Patrol Data ECC"; 604 break; 605 case FERR_NF_M7ERR: 606 specific = "Aliased Uncorrectable Spare-Copy Data ECC"; 607 break; 608 case FERR_NF_M6ERR: 609 specific = "Aliased Uncorrectable Mirrored Demand " 610 "Data ECC"; 611 break; 612 case FERR_NF_M5ERR: 613 specific = "Aliased Uncorrectable Non-Mirrored Demand " 614 "Data ECC"; 615 break; 616 case FERR_NF_M4ERR: 617 specific = "Uncorrectable Data ECC on Replay"; 618 break; 619 } 620 621 /* Form out message */ 622 snprintf(msg, sizeof(msg), 623 "Rank=%d Bank=%d RAS=%d CAS=%d, UE Err=0x%x (%s)", 624 rank, bank, ras, cas, ue_errors, specific); 625 626 /* Call the helper to output message */ 627 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0, 628 channel >> 1, -1, rank, 629 rdwr ? "Write error" : "Read error", 630 msg); 631 } 632 633 /* Check correctable errors */ 634 ce_errors = allErrors & FERR_NF_CORRECTABLE; 635 if (ce_errors) { 636 edac_dbg(0, "\tCorrected bits= 0x%x\n", ce_errors); 637 638 branch = EXTRACT_FBDCHAN_INDX(info->ferr_nf_fbd); 639 640 channel = 0; 641 if (REC_ECC_LOCATOR_ODD(info->redmemb)) 642 channel = 1; 643 644 /* Convert channel to be based from zero, instead of 645 * from branch base of 0 */ 646 channel += branch; 647 648 bank = REC_BANK(info->recmema); 649 rank = REC_RANK(info->recmema); 650 rdwr = REC_RDWR(info->recmema); 651 ras = REC_RAS(info->recmemb); 652 cas = REC_CAS(info->recmemb); 653 654 edac_dbg(0, "\t\tCSROW= %d Channel= %d (Branch %d DRAM Bank= %d rdwr= %s ras= %d cas= %d)\n", 655 rank, channel, branch >> 1, bank, 656 rdwr ? "Write" : "Read", ras, cas); 657 658 switch (ce_errors) { 659 case FERR_NF_M17ERR: 660 specific = "Correctable Non-Mirrored Demand Data ECC"; 661 break; 662 case FERR_NF_M18ERR: 663 specific = "Correctable Mirrored Demand Data ECC"; 664 break; 665 case FERR_NF_M19ERR: 666 specific = "Correctable Spare-Copy Data ECC"; 667 break; 668 case FERR_NF_M20ERR: 669 specific = "Correctable Patrol Data ECC"; 670 break; 671 } 672 673 /* Form out message */ 674 snprintf(msg, sizeof(msg), 675 "Rank=%d Bank=%d RDWR=%s RAS=%d " 676 "CAS=%d, CE Err=0x%x (%s))", branch >> 1, bank, 677 rdwr ? "Write" : "Read", ras, cas, ce_errors, 678 specific); 679 680 /* Call the helper to output message */ 681 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 0, 0, 0, 682 channel >> 1, channel % 2, rank, 683 rdwr ? "Write error" : "Read error", 684 msg); 685 } 686 687 if (!misc_messages) 688 return; 689 690 misc_errors = allErrors & (FERR_NF_NON_RETRY | FERR_NF_NORTH_CRC | 691 FERR_NF_SPD_PROTOCOL | FERR_NF_DIMM_SPARE); 692 if (misc_errors) { 693 switch (misc_errors) { 694 case FERR_NF_M13ERR: 695 specific = "Non-Retry or Redundant Retry FBD Memory " 696 "Alert or Redundant Fast Reset Timeout"; 697 break; 698 case FERR_NF_M14ERR: 699 specific = "Non-Retry or Redundant Retry FBD " 700 "Configuration Alert"; 701 break; 702 case FERR_NF_M15ERR: 703 specific = "Non-Retry or Redundant Retry FBD " 704 "Northbound CRC error on read data"; 705 break; 706 case FERR_NF_M21ERR: 707 specific = "FBD Northbound CRC error on " 708 "FBD Sync Status"; 709 break; 710 case FERR_NF_M22ERR: 711 specific = "SPD protocol error"; 712 break; 713 case FERR_NF_M27ERR: 714 specific = "DIMM-spare copy started"; 715 break; 716 case FERR_NF_M28ERR: 717 specific = "DIMM-spare copy completed"; 718 break; 719 } 720 branch = EXTRACT_FBDCHAN_INDX(info->ferr_nf_fbd); 721 722 /* Form out message */ 723 snprintf(msg, sizeof(msg), 724 "Err=%#x (%s)", misc_errors, specific); 725 726 /* Call the helper to output message */ 727 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 0, 0, 0, 728 branch >> 1, -1, -1, 729 "Misc error", msg); 730 } 731 } 732 733 /* 734 * i5000_process_error_info Process the error info that is 735 * in the 'info' structure, previously retrieved from hardware 736 */ 737 static void i5000_process_error_info(struct mem_ctl_info *mci, 738 struct i5000_error_info *info, 739 int handle_errors) 740 { 741 /* First handle any fatal errors that occurred */ 742 i5000_process_fatal_error_info(mci, info, handle_errors); 743 744 /* now handle any non-fatal errors that occurred */ 745 i5000_process_nonfatal_error_info(mci, info, handle_errors); 746 } 747 748 /* 749 * i5000_clear_error Retrieve any error from the hardware 750 * but do NOT process that error. 751 * Used for 'clearing' out of previous errors 752 * Called by the Core module. 753 */ 754 static void i5000_clear_error(struct mem_ctl_info *mci) 755 { 756 struct i5000_error_info info; 757 758 i5000_get_error_info(mci, &info); 759 } 760 761 /* 762 * i5000_check_error Retrieve and process errors reported by the 763 * hardware. Called by the Core module. 764 */ 765 static void i5000_check_error(struct mem_ctl_info *mci) 766 { 767 struct i5000_error_info info; 768 edac_dbg(4, "MC%d\n", mci->mc_idx); 769 i5000_get_error_info(mci, &info); 770 i5000_process_error_info(mci, &info, 1); 771 } 772 773 /* 774 * i5000_get_devices Find and perform 'get' operation on the MCH's 775 * device/functions we want to reference for this driver 776 * 777 * Need to 'get' device 16 func 1 and func 2 778 */ 779 static int i5000_get_devices(struct mem_ctl_info *mci, int dev_idx) 780 { 781 //const struct i5000_dev_info *i5000_dev = &i5000_devs[dev_idx]; 782 struct i5000_pvt *pvt; 783 struct pci_dev *pdev; 784 785 pvt = mci->pvt_info; 786 787 /* Attempt to 'get' the MCH register we want */ 788 pdev = NULL; 789 while (1) { 790 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 791 PCI_DEVICE_ID_INTEL_I5000_DEV16, pdev); 792 793 /* End of list, leave */ 794 if (pdev == NULL) { 795 i5000_printk(KERN_ERR, 796 "'system address,Process Bus' " 797 "device not found:" 798 "vendor 0x%x device 0x%x FUNC 1 " 799 "(broken BIOS?)\n", 800 PCI_VENDOR_ID_INTEL, 801 PCI_DEVICE_ID_INTEL_I5000_DEV16); 802 803 return 1; 804 } 805 806 /* Scan for device 16 func 1 */ 807 if (PCI_FUNC(pdev->devfn) == 1) 808 break; 809 } 810 811 pvt->branchmap_werrors = pdev; 812 813 /* Attempt to 'get' the MCH register we want */ 814 pdev = NULL; 815 while (1) { 816 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 817 PCI_DEVICE_ID_INTEL_I5000_DEV16, pdev); 818 819 if (pdev == NULL) { 820 i5000_printk(KERN_ERR, 821 "MC: 'branchmap,control,errors' " 822 "device not found:" 823 "vendor 0x%x device 0x%x Func 2 " 824 "(broken BIOS?)\n", 825 PCI_VENDOR_ID_INTEL, 826 PCI_DEVICE_ID_INTEL_I5000_DEV16); 827 828 pci_dev_put(pvt->branchmap_werrors); 829 return 1; 830 } 831 832 /* Scan for device 16 func 1 */ 833 if (PCI_FUNC(pdev->devfn) == 2) 834 break; 835 } 836 837 pvt->fsb_error_regs = pdev; 838 839 edac_dbg(1, "System Address, processor bus- PCI Bus ID: %s %x:%x\n", 840 pci_name(pvt->system_address), 841 pvt->system_address->vendor, pvt->system_address->device); 842 edac_dbg(1, "Branchmap, control and errors - PCI Bus ID: %s %x:%x\n", 843 pci_name(pvt->branchmap_werrors), 844 pvt->branchmap_werrors->vendor, 845 pvt->branchmap_werrors->device); 846 edac_dbg(1, "FSB Error Regs - PCI Bus ID: %s %x:%x\n", 847 pci_name(pvt->fsb_error_regs), 848 pvt->fsb_error_regs->vendor, pvt->fsb_error_regs->device); 849 850 pdev = NULL; 851 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 852 PCI_DEVICE_ID_I5000_BRANCH_0, pdev); 853 854 if (pdev == NULL) { 855 i5000_printk(KERN_ERR, 856 "MC: 'BRANCH 0' device not found:" 857 "vendor 0x%x device 0x%x Func 0 (broken BIOS?)\n", 858 PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_I5000_BRANCH_0); 859 860 pci_dev_put(pvt->branchmap_werrors); 861 pci_dev_put(pvt->fsb_error_regs); 862 return 1; 863 } 864 865 pvt->branch_0 = pdev; 866 867 /* If this device claims to have more than 2 channels then 868 * fetch Branch 1's information 869 */ 870 if (pvt->maxch >= CHANNELS_PER_BRANCH) { 871 pdev = NULL; 872 pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 873 PCI_DEVICE_ID_I5000_BRANCH_1, pdev); 874 875 if (pdev == NULL) { 876 i5000_printk(KERN_ERR, 877 "MC: 'BRANCH 1' device not found:" 878 "vendor 0x%x device 0x%x Func 0 " 879 "(broken BIOS?)\n", 880 PCI_VENDOR_ID_INTEL, 881 PCI_DEVICE_ID_I5000_BRANCH_1); 882 883 pci_dev_put(pvt->branchmap_werrors); 884 pci_dev_put(pvt->fsb_error_regs); 885 pci_dev_put(pvt->branch_0); 886 return 1; 887 } 888 889 pvt->branch_1 = pdev; 890 } 891 892 return 0; 893 } 894 895 /* 896 * i5000_put_devices 'put' all the devices that we have 897 * reserved via 'get' 898 */ 899 static void i5000_put_devices(struct mem_ctl_info *mci) 900 { 901 struct i5000_pvt *pvt; 902 903 pvt = mci->pvt_info; 904 905 pci_dev_put(pvt->branchmap_werrors); /* FUNC 1 */ 906 pci_dev_put(pvt->fsb_error_regs); /* FUNC 2 */ 907 pci_dev_put(pvt->branch_0); /* DEV 21 */ 908 909 /* Only if more than 2 channels do we release the second branch */ 910 if (pvt->maxch >= CHANNELS_PER_BRANCH) 911 pci_dev_put(pvt->branch_1); /* DEV 22 */ 912 } 913 914 /* 915 * determine_amb_resent 916 * 917 * the information is contained in NUM_MTRS different registers 918 * determineing which of the NUM_MTRS requires knowing 919 * which channel is in question 920 * 921 * 2 branches, each with 2 channels 922 * b0_ambpresent0 for channel '0' 923 * b0_ambpresent1 for channel '1' 924 * b1_ambpresent0 for channel '2' 925 * b1_ambpresent1 for channel '3' 926 */ 927 static int determine_amb_present_reg(struct i5000_pvt *pvt, int channel) 928 { 929 int amb_present; 930 931 if (channel < CHANNELS_PER_BRANCH) { 932 if (channel & 0x1) 933 amb_present = pvt->b0_ambpresent1; 934 else 935 amb_present = pvt->b0_ambpresent0; 936 } else { 937 if (channel & 0x1) 938 amb_present = pvt->b1_ambpresent1; 939 else 940 amb_present = pvt->b1_ambpresent0; 941 } 942 943 return amb_present; 944 } 945 946 /* 947 * determine_mtr(pvt, csrow, channel) 948 * 949 * return the proper MTR register as determine by the csrow and channel desired 950 */ 951 static int determine_mtr(struct i5000_pvt *pvt, int slot, int channel) 952 { 953 int mtr; 954 955 if (channel < CHANNELS_PER_BRANCH) 956 mtr = pvt->b0_mtr[slot]; 957 else 958 mtr = pvt->b1_mtr[slot]; 959 960 return mtr; 961 } 962 963 /* 964 */ 965 static void decode_mtr(int slot_row, u16 mtr) 966 { 967 int ans; 968 969 ans = MTR_DIMMS_PRESENT(mtr); 970 971 edac_dbg(2, "\tMTR%d=0x%x: DIMMs are %sPresent\n", 972 slot_row, mtr, ans ? "" : "NOT "); 973 if (!ans) 974 return; 975 976 edac_dbg(2, "\t\tWIDTH: x%d\n", MTR_DRAM_WIDTH(mtr)); 977 edac_dbg(2, "\t\tNUMBANK: %d bank(s)\n", MTR_DRAM_BANKS(mtr)); 978 edac_dbg(2, "\t\tNUMRANK: %s\n", 979 MTR_DIMM_RANK(mtr) ? "double" : "single"); 980 edac_dbg(2, "\t\tNUMROW: %s\n", 981 MTR_DIMM_ROWS(mtr) == 0 ? "8,192 - 13 rows" : 982 MTR_DIMM_ROWS(mtr) == 1 ? "16,384 - 14 rows" : 983 MTR_DIMM_ROWS(mtr) == 2 ? "32,768 - 15 rows" : 984 "reserved"); 985 edac_dbg(2, "\t\tNUMCOL: %s\n", 986 MTR_DIMM_COLS(mtr) == 0 ? "1,024 - 10 columns" : 987 MTR_DIMM_COLS(mtr) == 1 ? "2,048 - 11 columns" : 988 MTR_DIMM_COLS(mtr) == 2 ? "4,096 - 12 columns" : 989 "reserved"); 990 } 991 992 static void handle_channel(struct i5000_pvt *pvt, int slot, int channel, 993 struct i5000_dimm_info *dinfo) 994 { 995 int mtr; 996 int amb_present_reg; 997 int addrBits; 998 999 mtr = determine_mtr(pvt, slot, channel); 1000 if (MTR_DIMMS_PRESENT(mtr)) { 1001 amb_present_reg = determine_amb_present_reg(pvt, channel); 1002 1003 /* Determine if there is a DIMM present in this DIMM slot */ 1004 if (amb_present_reg) { 1005 dinfo->dual_rank = MTR_DIMM_RANK(mtr); 1006 1007 /* Start with the number of bits for a Bank 1008 * on the DRAM */ 1009 addrBits = MTR_DRAM_BANKS_ADDR_BITS(mtr); 1010 /* Add the number of ROW bits */ 1011 addrBits += MTR_DIMM_ROWS_ADDR_BITS(mtr); 1012 /* add the number of COLUMN bits */ 1013 addrBits += MTR_DIMM_COLS_ADDR_BITS(mtr); 1014 1015 /* Dual-rank memories have twice the size */ 1016 if (dinfo->dual_rank) 1017 addrBits++; 1018 1019 addrBits += 6; /* add 64 bits per DIMM */ 1020 addrBits -= 20; /* divide by 2^^20 */ 1021 addrBits -= 3; /* 8 bits per bytes */ 1022 1023 dinfo->megabytes = 1 << addrBits; 1024 } 1025 } 1026 } 1027 1028 /* 1029 * calculate_dimm_size 1030 * 1031 * also will output a DIMM matrix map, if debug is enabled, for viewing 1032 * how the DIMMs are populated 1033 */ 1034 static void calculate_dimm_size(struct i5000_pvt *pvt) 1035 { 1036 struct i5000_dimm_info *dinfo; 1037 int slot, channel, branch; 1038 char *p, *mem_buffer; 1039 int space, n; 1040 1041 /* ================= Generate some debug output ================= */ 1042 space = PAGE_SIZE; 1043 mem_buffer = p = kmalloc(space, GFP_KERNEL); 1044 if (p == NULL) { 1045 i5000_printk(KERN_ERR, "MC: %s:%s() kmalloc() failed\n", 1046 __FILE__, __func__); 1047 return; 1048 } 1049 1050 /* Scan all the actual slots 1051 * and calculate the information for each DIMM 1052 * Start with the highest slot first, to display it first 1053 * and work toward the 0th slot 1054 */ 1055 for (slot = pvt->maxdimmperch - 1; slot >= 0; slot--) { 1056 1057 /* on an odd slot, first output a 'boundary' marker, 1058 * then reset the message buffer */ 1059 if (slot & 0x1) { 1060 n = snprintf(p, space, "--------------------------" 1061 "--------------------------------"); 1062 p += n; 1063 space -= n; 1064 edac_dbg(2, "%s\n", mem_buffer); 1065 p = mem_buffer; 1066 space = PAGE_SIZE; 1067 } 1068 n = snprintf(p, space, "slot %2d ", slot); 1069 p += n; 1070 space -= n; 1071 1072 for (channel = 0; channel < pvt->maxch; channel++) { 1073 dinfo = &pvt->dimm_info[slot][channel]; 1074 handle_channel(pvt, slot, channel, dinfo); 1075 if (dinfo->megabytes) 1076 n = snprintf(p, space, "%4d MB %dR| ", 1077 dinfo->megabytes, dinfo->dual_rank + 1); 1078 else 1079 n = snprintf(p, space, "%4d MB | ", 0); 1080 p += n; 1081 space -= n; 1082 } 1083 p += n; 1084 space -= n; 1085 edac_dbg(2, "%s\n", mem_buffer); 1086 p = mem_buffer; 1087 space = PAGE_SIZE; 1088 } 1089 1090 /* Output the last bottom 'boundary' marker */ 1091 n = snprintf(p, space, "--------------------------" 1092 "--------------------------------"); 1093 p += n; 1094 space -= n; 1095 edac_dbg(2, "%s\n", mem_buffer); 1096 p = mem_buffer; 1097 space = PAGE_SIZE; 1098 1099 /* now output the 'channel' labels */ 1100 n = snprintf(p, space, " "); 1101 p += n; 1102 space -= n; 1103 for (channel = 0; channel < pvt->maxch; channel++) { 1104 n = snprintf(p, space, "channel %d | ", channel); 1105 p += n; 1106 space -= n; 1107 } 1108 edac_dbg(2, "%s\n", mem_buffer); 1109 p = mem_buffer; 1110 space = PAGE_SIZE; 1111 1112 n = snprintf(p, space, " "); 1113 p += n; 1114 for (branch = 0; branch < MAX_BRANCHES; branch++) { 1115 n = snprintf(p, space, " branch %d | ", branch); 1116 p += n; 1117 space -= n; 1118 } 1119 1120 /* output the last message and free buffer */ 1121 edac_dbg(2, "%s\n", mem_buffer); 1122 kfree(mem_buffer); 1123 } 1124 1125 /* 1126 * i5000_get_mc_regs read in the necessary registers and 1127 * cache locally 1128 * 1129 * Fills in the private data members 1130 */ 1131 static void i5000_get_mc_regs(struct mem_ctl_info *mci) 1132 { 1133 struct i5000_pvt *pvt; 1134 u32 actual_tolm; 1135 u16 limit; 1136 int slot_row; 1137 int maxch; 1138 int maxdimmperch; 1139 int way0, way1; 1140 1141 pvt = mci->pvt_info; 1142 1143 pci_read_config_dword(pvt->system_address, AMBASE, 1144 &pvt->u.ambase_bottom); 1145 pci_read_config_dword(pvt->system_address, AMBASE + sizeof(u32), 1146 &pvt->u.ambase_top); 1147 1148 maxdimmperch = pvt->maxdimmperch; 1149 maxch = pvt->maxch; 1150 1151 edac_dbg(2, "AMBASE= 0x%lx MAXCH= %d MAX-DIMM-Per-CH= %d\n", 1152 (long unsigned int)pvt->ambase, pvt->maxch, pvt->maxdimmperch); 1153 1154 /* Get the Branch Map regs */ 1155 pci_read_config_word(pvt->branchmap_werrors, TOLM, &pvt->tolm); 1156 pvt->tolm >>= 12; 1157 edac_dbg(2, "TOLM (number of 256M regions) =%u (0x%x)\n", 1158 pvt->tolm, pvt->tolm); 1159 1160 actual_tolm = pvt->tolm << 28; 1161 edac_dbg(2, "Actual TOLM byte addr=%u (0x%x)\n", 1162 actual_tolm, actual_tolm); 1163 1164 pci_read_config_word(pvt->branchmap_werrors, MIR0, &pvt->mir0); 1165 pci_read_config_word(pvt->branchmap_werrors, MIR1, &pvt->mir1); 1166 pci_read_config_word(pvt->branchmap_werrors, MIR2, &pvt->mir2); 1167 1168 /* Get the MIR[0-2] regs */ 1169 limit = (pvt->mir0 >> 4) & 0x0FFF; 1170 way0 = pvt->mir0 & 0x1; 1171 way1 = pvt->mir0 & 0x2; 1172 edac_dbg(2, "MIR0: limit= 0x%x WAY1= %u WAY0= %x\n", 1173 limit, way1, way0); 1174 limit = (pvt->mir1 >> 4) & 0x0FFF; 1175 way0 = pvt->mir1 & 0x1; 1176 way1 = pvt->mir1 & 0x2; 1177 edac_dbg(2, "MIR1: limit= 0x%x WAY1= %u WAY0= %x\n", 1178 limit, way1, way0); 1179 limit = (pvt->mir2 >> 4) & 0x0FFF; 1180 way0 = pvt->mir2 & 0x1; 1181 way1 = pvt->mir2 & 0x2; 1182 edac_dbg(2, "MIR2: limit= 0x%x WAY1= %u WAY0= %x\n", 1183 limit, way1, way0); 1184 1185 /* Get the MTR[0-3] regs */ 1186 for (slot_row = 0; slot_row < NUM_MTRS; slot_row++) { 1187 int where = MTR0 + (slot_row * sizeof(u32)); 1188 1189 pci_read_config_word(pvt->branch_0, where, 1190 &pvt->b0_mtr[slot_row]); 1191 1192 edac_dbg(2, "MTR%d where=0x%x B0 value=0x%x\n", 1193 slot_row, where, pvt->b0_mtr[slot_row]); 1194 1195 if (pvt->maxch >= CHANNELS_PER_BRANCH) { 1196 pci_read_config_word(pvt->branch_1, where, 1197 &pvt->b1_mtr[slot_row]); 1198 edac_dbg(2, "MTR%d where=0x%x B1 value=0x%x\n", 1199 slot_row, where, pvt->b1_mtr[slot_row]); 1200 } else { 1201 pvt->b1_mtr[slot_row] = 0; 1202 } 1203 } 1204 1205 /* Read and dump branch 0's MTRs */ 1206 edac_dbg(2, "Memory Technology Registers:\n"); 1207 edac_dbg(2, " Branch 0:\n"); 1208 for (slot_row = 0; slot_row < NUM_MTRS; slot_row++) { 1209 decode_mtr(slot_row, pvt->b0_mtr[slot_row]); 1210 } 1211 pci_read_config_word(pvt->branch_0, AMB_PRESENT_0, 1212 &pvt->b0_ambpresent0); 1213 edac_dbg(2, "\t\tAMB-Branch 0-present0 0x%x:\n", pvt->b0_ambpresent0); 1214 pci_read_config_word(pvt->branch_0, AMB_PRESENT_1, 1215 &pvt->b0_ambpresent1); 1216 edac_dbg(2, "\t\tAMB-Branch 0-present1 0x%x:\n", pvt->b0_ambpresent1); 1217 1218 /* Only if we have 2 branchs (4 channels) */ 1219 if (pvt->maxch < CHANNELS_PER_BRANCH) { 1220 pvt->b1_ambpresent0 = 0; 1221 pvt->b1_ambpresent1 = 0; 1222 } else { 1223 /* Read and dump branch 1's MTRs */ 1224 edac_dbg(2, " Branch 1:\n"); 1225 for (slot_row = 0; slot_row < NUM_MTRS; slot_row++) { 1226 decode_mtr(slot_row, pvt->b1_mtr[slot_row]); 1227 } 1228 pci_read_config_word(pvt->branch_1, AMB_PRESENT_0, 1229 &pvt->b1_ambpresent0); 1230 edac_dbg(2, "\t\tAMB-Branch 1-present0 0x%x:\n", 1231 pvt->b1_ambpresent0); 1232 pci_read_config_word(pvt->branch_1, AMB_PRESENT_1, 1233 &pvt->b1_ambpresent1); 1234 edac_dbg(2, "\t\tAMB-Branch 1-present1 0x%x:\n", 1235 pvt->b1_ambpresent1); 1236 } 1237 1238 /* Go and determine the size of each DIMM and place in an 1239 * orderly matrix */ 1240 calculate_dimm_size(pvt); 1241 } 1242 1243 /* 1244 * i5000_init_csrows Initialize the 'csrows' table within 1245 * the mci control structure with the 1246 * addressing of memory. 1247 * 1248 * return: 1249 * 0 success 1250 * 1 no actual memory found on this MC 1251 */ 1252 static int i5000_init_csrows(struct mem_ctl_info *mci) 1253 { 1254 struct i5000_pvt *pvt; 1255 struct dimm_info *dimm; 1256 int empty, channel_count; 1257 int max_csrows; 1258 int mtr; 1259 int csrow_megs; 1260 int channel; 1261 int slot; 1262 1263 pvt = mci->pvt_info; 1264 1265 channel_count = pvt->maxch; 1266 max_csrows = pvt->maxdimmperch * 2; 1267 1268 empty = 1; /* Assume NO memory */ 1269 1270 /* 1271 * FIXME: The memory layout used to map slot/channel into the 1272 * real memory architecture is weird: branch+slot are "csrows" 1273 * and channel is channel. That required an extra array (dimm_info) 1274 * to map the dimms. A good cleanup would be to remove this array, 1275 * and do a loop here with branch, channel, slot 1276 */ 1277 for (slot = 0; slot < max_csrows; slot++) { 1278 for (channel = 0; channel < pvt->maxch; channel++) { 1279 1280 mtr = determine_mtr(pvt, slot, channel); 1281 1282 if (!MTR_DIMMS_PRESENT(mtr)) 1283 continue; 1284 1285 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, 1286 channel / MAX_BRANCHES, 1287 channel % MAX_BRANCHES, slot); 1288 1289 csrow_megs = pvt->dimm_info[slot][channel].megabytes; 1290 dimm->grain = 8; 1291 1292 /* Assume DDR2 for now */ 1293 dimm->mtype = MEM_FB_DDR2; 1294 1295 /* ask what device type on this row */ 1296 if (MTR_DRAM_WIDTH(mtr)) 1297 dimm->dtype = DEV_X8; 1298 else 1299 dimm->dtype = DEV_X4; 1300 1301 dimm->edac_mode = EDAC_S8ECD8ED; 1302 dimm->nr_pages = csrow_megs << 8; 1303 } 1304 1305 empty = 0; 1306 } 1307 1308 return empty; 1309 } 1310 1311 /* 1312 * i5000_enable_error_reporting 1313 * Turn on the memory reporting features of the hardware 1314 */ 1315 static void i5000_enable_error_reporting(struct mem_ctl_info *mci) 1316 { 1317 struct i5000_pvt *pvt; 1318 u32 fbd_error_mask; 1319 1320 pvt = mci->pvt_info; 1321 1322 /* Read the FBD Error Mask Register */ 1323 pci_read_config_dword(pvt->branchmap_werrors, EMASK_FBD, 1324 &fbd_error_mask); 1325 1326 /* Enable with a '0' */ 1327 fbd_error_mask &= ~(ENABLE_EMASK_ALL); 1328 1329 pci_write_config_dword(pvt->branchmap_werrors, EMASK_FBD, 1330 fbd_error_mask); 1331 } 1332 1333 /* 1334 * i5000_get_dimm_and_channel_counts(pdev, &nr_csrows, &num_channels) 1335 * 1336 * ask the device how many channels are present and how many CSROWS 1337 * as well 1338 */ 1339 static void i5000_get_dimm_and_channel_counts(struct pci_dev *pdev, 1340 int *num_dimms_per_channel, 1341 int *num_channels) 1342 { 1343 u8 value; 1344 1345 /* Need to retrieve just how many channels and dimms per channel are 1346 * supported on this memory controller 1347 */ 1348 pci_read_config_byte(pdev, MAXDIMMPERCH, &value); 1349 *num_dimms_per_channel = (int)value; 1350 1351 pci_read_config_byte(pdev, MAXCH, &value); 1352 *num_channels = (int)value; 1353 } 1354 1355 /* 1356 * i5000_probe1 Probe for ONE instance of device to see if it is 1357 * present. 1358 * return: 1359 * 0 for FOUND a device 1360 * < 0 for error code 1361 */ 1362 static int i5000_probe1(struct pci_dev *pdev, int dev_idx) 1363 { 1364 struct mem_ctl_info *mci; 1365 struct edac_mc_layer layers[3]; 1366 struct i5000_pvt *pvt; 1367 int num_channels; 1368 int num_dimms_per_channel; 1369 1370 edac_dbg(0, "MC: pdev bus %u dev=0x%x fn=0x%x\n", 1371 pdev->bus->number, 1372 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 1373 1374 /* We only are looking for func 0 of the set */ 1375 if (PCI_FUNC(pdev->devfn) != 0) 1376 return -ENODEV; 1377 1378 /* Ask the devices for the number of CSROWS and CHANNELS so 1379 * that we can calculate the memory resources, etc 1380 * 1381 * The Chipset will report what it can handle which will be greater 1382 * or equal to what the motherboard manufacturer will implement. 1383 * 1384 * As we don't have a motherboard identification routine to determine 1385 * actual number of slots/dimms per channel, we thus utilize the 1386 * resource as specified by the chipset. Thus, we might have 1387 * have more DIMMs per channel than actually on the mobo, but this 1388 * allows the driver to support up to the chipset max, without 1389 * some fancy mobo determination. 1390 */ 1391 i5000_get_dimm_and_channel_counts(pdev, &num_dimms_per_channel, 1392 &num_channels); 1393 1394 edac_dbg(0, "MC: Number of Branches=2 Channels= %d DIMMS= %d\n", 1395 num_channels, num_dimms_per_channel); 1396 1397 /* allocate a new MC control structure */ 1398 1399 layers[0].type = EDAC_MC_LAYER_BRANCH; 1400 layers[0].size = MAX_BRANCHES; 1401 layers[0].is_virt_csrow = false; 1402 layers[1].type = EDAC_MC_LAYER_CHANNEL; 1403 layers[1].size = num_channels / MAX_BRANCHES; 1404 layers[1].is_virt_csrow = false; 1405 layers[2].type = EDAC_MC_LAYER_SLOT; 1406 layers[2].size = num_dimms_per_channel; 1407 layers[2].is_virt_csrow = true; 1408 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(*pvt)); 1409 if (mci == NULL) 1410 return -ENOMEM; 1411 1412 edac_dbg(0, "MC: mci = %p\n", mci); 1413 1414 mci->pdev = &pdev->dev; /* record ptr to the generic device */ 1415 1416 pvt = mci->pvt_info; 1417 pvt->system_address = pdev; /* Record this device in our private */ 1418 pvt->maxch = num_channels; 1419 pvt->maxdimmperch = num_dimms_per_channel; 1420 1421 /* 'get' the pci devices we want to reserve for our use */ 1422 if (i5000_get_devices(mci, dev_idx)) 1423 goto fail0; 1424 1425 /* Time to get serious */ 1426 i5000_get_mc_regs(mci); /* retrieve the hardware registers */ 1427 1428 mci->mc_idx = 0; 1429 mci->mtype_cap = MEM_FLAG_FB_DDR2; 1430 mci->edac_ctl_cap = EDAC_FLAG_NONE; 1431 mci->edac_cap = EDAC_FLAG_NONE; 1432 mci->mod_name = "i5000_edac.c"; 1433 mci->mod_ver = I5000_REVISION; 1434 mci->ctl_name = i5000_devs[dev_idx].ctl_name; 1435 mci->dev_name = pci_name(pdev); 1436 mci->ctl_page_to_phys = NULL; 1437 1438 /* Set the function pointer to an actual operation function */ 1439 mci->edac_check = i5000_check_error; 1440 1441 /* initialize the MC control structure 'csrows' table 1442 * with the mapping and control information */ 1443 if (i5000_init_csrows(mci)) { 1444 edac_dbg(0, "MC: Setting mci->edac_cap to EDAC_FLAG_NONE because i5000_init_csrows() returned nonzero value\n"); 1445 mci->edac_cap = EDAC_FLAG_NONE; /* no csrows found */ 1446 } else { 1447 edac_dbg(1, "MC: Enable error reporting now\n"); 1448 i5000_enable_error_reporting(mci); 1449 } 1450 1451 /* add this new MC control structure to EDAC's list of MCs */ 1452 if (edac_mc_add_mc(mci)) { 1453 edac_dbg(0, "MC: failed edac_mc_add_mc()\n"); 1454 /* FIXME: perhaps some code should go here that disables error 1455 * reporting if we just enabled it 1456 */ 1457 goto fail1; 1458 } 1459 1460 i5000_clear_error(mci); 1461 1462 /* allocating generic PCI control info */ 1463 i5000_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR); 1464 if (!i5000_pci) { 1465 printk(KERN_WARNING 1466 "%s(): Unable to create PCI control\n", 1467 __func__); 1468 printk(KERN_WARNING 1469 "%s(): PCI error report via EDAC not setup\n", 1470 __func__); 1471 } 1472 1473 return 0; 1474 1475 /* Error exit unwinding stack */ 1476 fail1: 1477 1478 i5000_put_devices(mci); 1479 1480 fail0: 1481 edac_mc_free(mci); 1482 return -ENODEV; 1483 } 1484 1485 /* 1486 * i5000_init_one constructor for one instance of device 1487 * 1488 * returns: 1489 * negative on error 1490 * count (>= 0) 1491 */ 1492 static int i5000_init_one(struct pci_dev *pdev, const struct pci_device_id *id) 1493 { 1494 int rc; 1495 1496 edac_dbg(0, "MC:\n"); 1497 1498 /* wake up device */ 1499 rc = pci_enable_device(pdev); 1500 if (rc) 1501 return rc; 1502 1503 /* now probe and enable the device */ 1504 return i5000_probe1(pdev, id->driver_data); 1505 } 1506 1507 /* 1508 * i5000_remove_one destructor for one instance of device 1509 * 1510 */ 1511 static void i5000_remove_one(struct pci_dev *pdev) 1512 { 1513 struct mem_ctl_info *mci; 1514 1515 edac_dbg(0, "\n"); 1516 1517 if (i5000_pci) 1518 edac_pci_release_generic_ctl(i5000_pci); 1519 1520 if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL) 1521 return; 1522 1523 /* retrieve references to resources, and free those resources */ 1524 i5000_put_devices(mci); 1525 edac_mc_free(mci); 1526 } 1527 1528 /* 1529 * pci_device_id table for which devices we are looking for 1530 * 1531 * The "E500P" device is the first device supported. 1532 */ 1533 static const struct pci_device_id i5000_pci_tbl[] = { 1534 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_I5000_DEV16), 1535 .driver_data = I5000P}, 1536 1537 {0,} /* 0 terminated list. */ 1538 }; 1539 1540 MODULE_DEVICE_TABLE(pci, i5000_pci_tbl); 1541 1542 /* 1543 * i5000_driver pci_driver structure for this module 1544 * 1545 */ 1546 static struct pci_driver i5000_driver = { 1547 .name = KBUILD_BASENAME, 1548 .probe = i5000_init_one, 1549 .remove = i5000_remove_one, 1550 .id_table = i5000_pci_tbl, 1551 }; 1552 1553 /* 1554 * i5000_init Module entry function 1555 * Try to initialize this module for its devices 1556 */ 1557 static int __init i5000_init(void) 1558 { 1559 int pci_rc; 1560 1561 edac_dbg(2, "MC:\n"); 1562 1563 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 1564 opstate_init(); 1565 1566 pci_rc = pci_register_driver(&i5000_driver); 1567 1568 return (pci_rc < 0) ? pci_rc : 0; 1569 } 1570 1571 /* 1572 * i5000_exit() Module exit function 1573 * Unregister the driver 1574 */ 1575 static void __exit i5000_exit(void) 1576 { 1577 edac_dbg(2, "MC:\n"); 1578 pci_unregister_driver(&i5000_driver); 1579 } 1580 1581 module_init(i5000_init); 1582 module_exit(i5000_exit); 1583 1584 MODULE_LICENSE("GPL"); 1585 MODULE_AUTHOR 1586 ("Linux Networx (http://lnxi.com) Doug Thompson <norsk5@xmission.com>"); 1587 MODULE_DESCRIPTION("MC Driver for Intel I5000 memory controllers - " 1588 I5000_REVISION); 1589 1590 module_param(edac_op_state, int, 0444); 1591 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 1592 module_param(misc_messages, int, 0444); 1593 MODULE_PARM_DESC(misc_messages, "Log miscellaneous non fatal messages"); 1594 1595