e752x_edac.c (41680B)
1/* 2 * Intel e752x Memory Controller kernel module 3 * (C) 2004 Linux Networx (http://lnxi.com) 4 * This file may be distributed under the terms of the 5 * GNU General Public License. 6 * 7 * Implement support for the e7520, E7525, e7320 and i3100 memory controllers. 8 * 9 * Datasheets: 10 * https://www.intel.in/content/www/in/en/chipsets/e7525-memory-controller-hub-datasheet.html 11 * ftp://download.intel.com/design/intarch/datashts/31345803.pdf 12 * 13 * Written by Tom Zimmerman 14 * 15 * Contributors: 16 * Thayne Harbaugh at realmsys.com (?) 17 * Wang Zhenyu at intel.com 18 * Dave Jiang at mvista.com 19 * 20 */ 21 22#include <linux/module.h> 23#include <linux/init.h> 24#include <linux/pci.h> 25#include <linux/pci_ids.h> 26#include <linux/edac.h> 27#include "edac_module.h" 28 29#define EDAC_MOD_STR "e752x_edac" 30 31static int report_non_memory_errors; 32static int force_function_unhide; 33static int sysbus_parity = -1; 34 35static struct edac_pci_ctl_info *e752x_pci; 36 37#define e752x_printk(level, fmt, arg...) \ 38 edac_printk(level, "e752x", fmt, ##arg) 39 40#define e752x_mc_printk(mci, level, fmt, arg...) \ 41 edac_mc_chipset_printk(mci, level, "e752x", fmt, ##arg) 42 43#ifndef PCI_DEVICE_ID_INTEL_7520_0 44#define PCI_DEVICE_ID_INTEL_7520_0 0x3590 45#endif /* PCI_DEVICE_ID_INTEL_7520_0 */ 46 47#ifndef PCI_DEVICE_ID_INTEL_7520_1_ERR 48#define PCI_DEVICE_ID_INTEL_7520_1_ERR 0x3591 49#endif /* PCI_DEVICE_ID_INTEL_7520_1_ERR */ 50 51#ifndef PCI_DEVICE_ID_INTEL_7525_0 52#define PCI_DEVICE_ID_INTEL_7525_0 0x359E 53#endif /* PCI_DEVICE_ID_INTEL_7525_0 */ 54 55#ifndef PCI_DEVICE_ID_INTEL_7525_1_ERR 56#define PCI_DEVICE_ID_INTEL_7525_1_ERR 0x3593 57#endif /* PCI_DEVICE_ID_INTEL_7525_1_ERR */ 58 59#ifndef PCI_DEVICE_ID_INTEL_7320_0 60#define PCI_DEVICE_ID_INTEL_7320_0 0x3592 61#endif /* PCI_DEVICE_ID_INTEL_7320_0 */ 62 63#ifndef PCI_DEVICE_ID_INTEL_7320_1_ERR 64#define PCI_DEVICE_ID_INTEL_7320_1_ERR 0x3593 65#endif /* PCI_DEVICE_ID_INTEL_7320_1_ERR */ 66 67#ifndef PCI_DEVICE_ID_INTEL_3100_0 68#define PCI_DEVICE_ID_INTEL_3100_0 0x35B0 69#endif /* PCI_DEVICE_ID_INTEL_3100_0 */ 70 71#ifndef PCI_DEVICE_ID_INTEL_3100_1_ERR 72#define PCI_DEVICE_ID_INTEL_3100_1_ERR 0x35B1 73#endif /* PCI_DEVICE_ID_INTEL_3100_1_ERR */ 74 75#define E752X_NR_CSROWS 8 /* number of csrows */ 76 77/* E752X register addresses - device 0 function 0 */ 78#define E752X_MCHSCRB 0x52 /* Memory Scrub register (16b) */ 79 /* 80 * 6:5 Scrub Completion Count 81 * 3:2 Scrub Rate (i3100 only) 82 * 01=fast 10=normal 83 * 1:0 Scrub Mode enable 84 * 00=off 10=on 85 */ 86#define E752X_DRB 0x60 /* DRAM row boundary register (8b) */ 87#define E752X_DRA 0x70 /* DRAM row attribute register (8b) */ 88 /* 89 * 31:30 Device width row 7 90 * 01=x8 10=x4 11=x8 DDR2 91 * 27:26 Device width row 6 92 * 23:22 Device width row 5 93 * 19:20 Device width row 4 94 * 15:14 Device width row 3 95 * 11:10 Device width row 2 96 * 7:6 Device width row 1 97 * 3:2 Device width row 0 98 */ 99#define E752X_DRC 0x7C /* DRAM controller mode reg (32b) */ 100 /* FIXME:IS THIS RIGHT? */ 101 /* 102 * 22 Number channels 0=1,1=2 103 * 19:18 DRB Granularity 32/64MB 104 */ 105#define E752X_DRM 0x80 /* Dimm mapping register */ 106#define E752X_DDRCSR 0x9A /* DDR control and status reg (16b) */ 107 /* 108 * 14:12 1 single A, 2 single B, 3 dual 109 */ 110#define E752X_TOLM 0xC4 /* DRAM top of low memory reg (16b) */ 111#define E752X_REMAPBASE 0xC6 /* DRAM remap base address reg (16b) */ 112#define E752X_REMAPLIMIT 0xC8 /* DRAM remap limit address reg (16b) */ 113#define E752X_REMAPOFFSET 0xCA /* DRAM remap limit offset reg (16b) */ 114 115/* E752X register addresses - device 0 function 1 */ 116#define E752X_FERR_GLOBAL 0x40 /* Global first error register (32b) */ 117#define E752X_NERR_GLOBAL 0x44 /* Global next error register (32b) */ 118#define E752X_HI_FERR 0x50 /* Hub interface first error reg (8b) */ 119#define E752X_HI_NERR 0x52 /* Hub interface next error reg (8b) */ 120#define E752X_HI_ERRMASK 0x54 /* Hub interface error mask reg (8b) */ 121#define E752X_HI_SMICMD 0x5A /* Hub interface SMI command reg (8b) */ 122#define E752X_SYSBUS_FERR 0x60 /* System buss first error reg (16b) */ 123#define E752X_SYSBUS_NERR 0x62 /* System buss next error reg (16b) */ 124#define E752X_SYSBUS_ERRMASK 0x64 /* System buss error mask reg (16b) */ 125#define E752X_SYSBUS_SMICMD 0x6A /* System buss SMI command reg (16b) */ 126#define E752X_BUF_FERR 0x70 /* Memory buffer first error reg (8b) */ 127#define E752X_BUF_NERR 0x72 /* Memory buffer next error reg (8b) */ 128#define E752X_BUF_ERRMASK 0x74 /* Memory buffer error mask reg (8b) */ 129#define E752X_BUF_SMICMD 0x7A /* Memory buffer SMI cmd reg (8b) */ 130#define E752X_DRAM_FERR 0x80 /* DRAM first error register (16b) */ 131#define E752X_DRAM_NERR 0x82 /* DRAM next error register (16b) */ 132#define E752X_DRAM_ERRMASK 0x84 /* DRAM error mask register (8b) */ 133#define E752X_DRAM_SMICMD 0x8A /* DRAM SMI command register (8b) */ 134#define E752X_DRAM_RETR_ADD 0xAC /* DRAM Retry address register (32b) */ 135#define E752X_DRAM_SEC1_ADD 0xA0 /* DRAM first correctable memory */ 136 /* error address register (32b) */ 137 /* 138 * 31 Reserved 139 * 30:2 CE address (64 byte block 34:6 140 * 1 Reserved 141 * 0 HiLoCS 142 */ 143#define E752X_DRAM_SEC2_ADD 0xC8 /* DRAM first correctable memory */ 144 /* error address register (32b) */ 145 /* 146 * 31 Reserved 147 * 30:2 CE address (64 byte block 34:6) 148 * 1 Reserved 149 * 0 HiLoCS 150 */ 151#define E752X_DRAM_DED_ADD 0xA4 /* DRAM first uncorrectable memory */ 152 /* error address register (32b) */ 153 /* 154 * 31 Reserved 155 * 30:2 CE address (64 byte block 34:6) 156 * 1 Reserved 157 * 0 HiLoCS 158 */ 159#define E752X_DRAM_SCRB_ADD 0xA8 /* DRAM 1st uncorrectable scrub mem */ 160 /* error address register (32b) */ 161 /* 162 * 31 Reserved 163 * 30:2 CE address (64 byte block 34:6 164 * 1 Reserved 165 * 0 HiLoCS 166 */ 167#define E752X_DRAM_SEC1_SYNDROME 0xC4 /* DRAM first correctable memory */ 168 /* error syndrome register (16b) */ 169#define E752X_DRAM_SEC2_SYNDROME 0xC6 /* DRAM second correctable memory */ 170 /* error syndrome register (16b) */ 171#define E752X_DEVPRES1 0xF4 /* Device Present 1 register (8b) */ 172 173/* 3100 IMCH specific register addresses - device 0 function 1 */ 174#define I3100_NSI_FERR 0x48 /* NSI first error reg (32b) */ 175#define I3100_NSI_NERR 0x4C /* NSI next error reg (32b) */ 176#define I3100_NSI_SMICMD 0x54 /* NSI SMI command register (32b) */ 177#define I3100_NSI_EMASK 0x90 /* NSI error mask register (32b) */ 178 179/* ICH5R register addresses - device 30 function 0 */ 180#define ICH5R_PCI_STAT 0x06 /* PCI status register (16b) */ 181#define ICH5R_PCI_2ND_STAT 0x1E /* PCI status secondary reg (16b) */ 182#define ICH5R_PCI_BRIDGE_CTL 0x3E /* PCI bridge control register (16b) */ 183 184enum e752x_chips { 185 E7520 = 0, 186 E7525 = 1, 187 E7320 = 2, 188 I3100 = 3 189}; 190 191/* 192 * Those chips Support single-rank and dual-rank memories only. 193 * 194 * On e752x chips, the odd rows are present only on dual-rank memories. 195 * Dividing the rank by two will provide the dimm# 196 * 197 * i3100 MC has a different mapping: it supports only 4 ranks. 198 * 199 * The mapping is (from 1 to n): 200 * slot single-ranked double-ranked 201 * dimm #1 -> rank #4 NA 202 * dimm #2 -> rank #3 NA 203 * dimm #3 -> rank #2 Ranks 2 and 3 204 * dimm #4 -> rank $1 Ranks 1 and 4 205 * 206 * FIXME: The current mapping for i3100 considers that it supports up to 8 207 * ranks/chanel, but datasheet says that the MC supports only 4 ranks. 208 */ 209 210struct e752x_pvt { 211 struct pci_dev *dev_d0f0; 212 struct pci_dev *dev_d0f1; 213 u32 tolm; 214 u32 remapbase; 215 u32 remaplimit; 216 int mc_symmetric; 217 u8 map[8]; 218 int map_type; 219 const struct e752x_dev_info *dev_info; 220}; 221 222struct e752x_dev_info { 223 u16 err_dev; 224 u16 ctl_dev; 225 const char *ctl_name; 226}; 227 228struct e752x_error_info { 229 u32 ferr_global; 230 u32 nerr_global; 231 u32 nsi_ferr; /* 3100 only */ 232 u32 nsi_nerr; /* 3100 only */ 233 u8 hi_ferr; /* all but 3100 */ 234 u8 hi_nerr; /* all but 3100 */ 235 u16 sysbus_ferr; 236 u16 sysbus_nerr; 237 u8 buf_ferr; 238 u8 buf_nerr; 239 u16 dram_ferr; 240 u16 dram_nerr; 241 u32 dram_sec1_add; 242 u32 dram_sec2_add; 243 u16 dram_sec1_syndrome; 244 u16 dram_sec2_syndrome; 245 u32 dram_ded_add; 246 u32 dram_scrb_add; 247 u32 dram_retr_add; 248}; 249 250static const struct e752x_dev_info e752x_devs[] = { 251 [E7520] = { 252 .err_dev = PCI_DEVICE_ID_INTEL_7520_1_ERR, 253 .ctl_dev = PCI_DEVICE_ID_INTEL_7520_0, 254 .ctl_name = "E7520"}, 255 [E7525] = { 256 .err_dev = PCI_DEVICE_ID_INTEL_7525_1_ERR, 257 .ctl_dev = PCI_DEVICE_ID_INTEL_7525_0, 258 .ctl_name = "E7525"}, 259 [E7320] = { 260 .err_dev = PCI_DEVICE_ID_INTEL_7320_1_ERR, 261 .ctl_dev = PCI_DEVICE_ID_INTEL_7320_0, 262 .ctl_name = "E7320"}, 263 [I3100] = { 264 .err_dev = PCI_DEVICE_ID_INTEL_3100_1_ERR, 265 .ctl_dev = PCI_DEVICE_ID_INTEL_3100_0, 266 .ctl_name = "3100"}, 267}; 268 269/* Valid scrub rates for the e752x/3100 hardware memory scrubber. We 270 * map the scrubbing bandwidth to a hardware register value. The 'set' 271 * operation finds the 'matching or higher value'. Note that scrubbing 272 * on the e752x can only be enabled/disabled. The 3100 supports 273 * a normal and fast mode. 274 */ 275 276#define SDRATE_EOT 0xFFFFFFFF 277 278struct scrubrate { 279 u32 bandwidth; /* bandwidth consumed by scrubbing in bytes/sec */ 280 u16 scrubval; /* register value for scrub rate */ 281}; 282 283/* Rate below assumes same performance as i3100 using PC3200 DDR2 in 284 * normal mode. e752x bridges don't support choosing normal or fast mode, 285 * so the scrubbing bandwidth value isn't all that important - scrubbing is 286 * either on or off. 287 */ 288static const struct scrubrate scrubrates_e752x[] = { 289 {0, 0x00}, /* Scrubbing Off */ 290 {500000, 0x02}, /* Scrubbing On */ 291 {SDRATE_EOT, 0x00} /* End of Table */ 292}; 293 294/* Fast mode: 2 GByte PC3200 DDR2 scrubbed in 33s = 63161283 bytes/s 295 * Normal mode: 125 (32000 / 256) times slower than fast mode. 296 */ 297static const struct scrubrate scrubrates_i3100[] = { 298 {0, 0x00}, /* Scrubbing Off */ 299 {500000, 0x0a}, /* Normal mode - 32k clocks */ 300 {62500000, 0x06}, /* Fast mode - 256 clocks */ 301 {SDRATE_EOT, 0x00} /* End of Table */ 302}; 303 304static unsigned long ctl_page_to_phys(struct mem_ctl_info *mci, 305 unsigned long page) 306{ 307 u32 remap; 308 struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info; 309 310 edac_dbg(3, "\n"); 311 312 if (page < pvt->tolm) 313 return page; 314 315 if ((page >= 0x100000) && (page < pvt->remapbase)) 316 return page; 317 318 remap = (page - pvt->tolm) + pvt->remapbase; 319 320 if (remap < pvt->remaplimit) 321 return remap; 322 323 e752x_printk(KERN_ERR, "Invalid page %lx - out of range\n", page); 324 return pvt->tolm - 1; 325} 326 327static void do_process_ce(struct mem_ctl_info *mci, u16 error_one, 328 u32 sec1_add, u16 sec1_syndrome) 329{ 330 u32 page; 331 int row; 332 int channel; 333 int i; 334 struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info; 335 336 edac_dbg(3, "\n"); 337 338 /* convert the addr to 4k page */ 339 page = sec1_add >> (PAGE_SHIFT - 4); 340 341 /* FIXME - check for -1 */ 342 if (pvt->mc_symmetric) { 343 /* chip select are bits 14 & 13 */ 344 row = ((page >> 1) & 3); 345 e752x_printk(KERN_WARNING, 346 "Test row %d Table %d %d %d %d %d %d %d %d\n", row, 347 pvt->map[0], pvt->map[1], pvt->map[2], pvt->map[3], 348 pvt->map[4], pvt->map[5], pvt->map[6], 349 pvt->map[7]); 350 351 /* test for channel remapping */ 352 for (i = 0; i < 8; i++) { 353 if (pvt->map[i] == row) 354 break; 355 } 356 357 e752x_printk(KERN_WARNING, "Test computed row %d\n", i); 358 359 if (i < 8) 360 row = i; 361 else 362 e752x_mc_printk(mci, KERN_WARNING, 363 "row %d not found in remap table\n", 364 row); 365 } else 366 row = edac_mc_find_csrow_by_page(mci, page); 367 368 /* 0 = channel A, 1 = channel B */ 369 channel = !(error_one & 1); 370 371 /* e752x mc reads 34:6 of the DRAM linear address */ 372 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 373 page, offset_in_page(sec1_add << 4), sec1_syndrome, 374 row, channel, -1, 375 "e752x CE", ""); 376} 377 378static inline void process_ce(struct mem_ctl_info *mci, u16 error_one, 379 u32 sec1_add, u16 sec1_syndrome, int *error_found, 380 int handle_error) 381{ 382 *error_found = 1; 383 384 if (handle_error) 385 do_process_ce(mci, error_one, sec1_add, sec1_syndrome); 386} 387 388static void do_process_ue(struct mem_ctl_info *mci, u16 error_one, 389 u32 ded_add, u32 scrb_add) 390{ 391 u32 error_2b, block_page; 392 int row; 393 struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info; 394 395 edac_dbg(3, "\n"); 396 397 if (error_one & 0x0202) { 398 error_2b = ded_add; 399 400 /* convert to 4k address */ 401 block_page = error_2b >> (PAGE_SHIFT - 4); 402 403 row = pvt->mc_symmetric ? 404 /* chip select are bits 14 & 13 */ 405 ((block_page >> 1) & 3) : 406 edac_mc_find_csrow_by_page(mci, block_page); 407 408 /* e752x mc reads 34:6 of the DRAM linear address */ 409 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 410 block_page, 411 offset_in_page(error_2b << 4), 0, 412 row, -1, -1, 413 "e752x UE from Read", ""); 414 415 } 416 if (error_one & 0x0404) { 417 error_2b = scrb_add; 418 419 /* convert to 4k address */ 420 block_page = error_2b >> (PAGE_SHIFT - 4); 421 422 row = pvt->mc_symmetric ? 423 /* chip select are bits 14 & 13 */ 424 ((block_page >> 1) & 3) : 425 edac_mc_find_csrow_by_page(mci, block_page); 426 427 /* e752x mc reads 34:6 of the DRAM linear address */ 428 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 429 block_page, 430 offset_in_page(error_2b << 4), 0, 431 row, -1, -1, 432 "e752x UE from Scruber", ""); 433 } 434} 435 436static inline void process_ue(struct mem_ctl_info *mci, u16 error_one, 437 u32 ded_add, u32 scrb_add, int *error_found, 438 int handle_error) 439{ 440 *error_found = 1; 441 442 if (handle_error) 443 do_process_ue(mci, error_one, ded_add, scrb_add); 444} 445 446static inline void process_ue_no_info_wr(struct mem_ctl_info *mci, 447 int *error_found, int handle_error) 448{ 449 *error_found = 1; 450 451 if (!handle_error) 452 return; 453 454 edac_dbg(3, "\n"); 455 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0, 456 -1, -1, -1, 457 "e752x UE log memory write", ""); 458} 459 460static void do_process_ded_retry(struct mem_ctl_info *mci, u16 error, 461 u32 retry_add) 462{ 463 u32 error_1b, page; 464 int row; 465 struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info; 466 467 error_1b = retry_add; 468 page = error_1b >> (PAGE_SHIFT - 4); /* convert the addr to 4k page */ 469 470 /* chip select are bits 14 & 13 */ 471 row = pvt->mc_symmetric ? ((page >> 1) & 3) : 472 edac_mc_find_csrow_by_page(mci, page); 473 474 e752x_mc_printk(mci, KERN_WARNING, 475 "CE page 0x%lx, row %d : Memory read retry\n", 476 (long unsigned int)page, row); 477} 478 479static inline void process_ded_retry(struct mem_ctl_info *mci, u16 error, 480 u32 retry_add, int *error_found, 481 int handle_error) 482{ 483 *error_found = 1; 484 485 if (handle_error) 486 do_process_ded_retry(mci, error, retry_add); 487} 488 489static inline void process_threshold_ce(struct mem_ctl_info *mci, u16 error, 490 int *error_found, int handle_error) 491{ 492 *error_found = 1; 493 494 if (handle_error) 495 e752x_mc_printk(mci, KERN_WARNING, "Memory threshold CE\n"); 496} 497 498static char *global_message[11] = { 499 "PCI Express C1", 500 "PCI Express C", 501 "PCI Express B1", 502 "PCI Express B", 503 "PCI Express A1", 504 "PCI Express A", 505 "DMA Controller", 506 "HUB or NS Interface", 507 "System Bus", 508 "DRAM Controller", /* 9th entry */ 509 "Internal Buffer" 510}; 511 512#define DRAM_ENTRY 9 513 514static char *fatal_message[2] = { "Non-Fatal ", "Fatal " }; 515 516static void do_global_error(int fatal, u32 errors) 517{ 518 int i; 519 520 for (i = 0; i < 11; i++) { 521 if (errors & (1 << i)) { 522 /* If the error is from DRAM Controller OR 523 * we are to report ALL errors, then 524 * report the error 525 */ 526 if ((i == DRAM_ENTRY) || report_non_memory_errors) 527 e752x_printk(KERN_WARNING, "%sError %s\n", 528 fatal_message[fatal], 529 global_message[i]); 530 } 531 } 532} 533 534static inline void global_error(int fatal, u32 errors, int *error_found, 535 int handle_error) 536{ 537 *error_found = 1; 538 539 if (handle_error) 540 do_global_error(fatal, errors); 541} 542 543static char *hub_message[7] = { 544 "HI Address or Command Parity", "HI Illegal Access", 545 "HI Internal Parity", "Out of Range Access", 546 "HI Data Parity", "Enhanced Config Access", 547 "Hub Interface Target Abort" 548}; 549 550static void do_hub_error(int fatal, u8 errors) 551{ 552 int i; 553 554 for (i = 0; i < 7; i++) { 555 if (errors & (1 << i)) 556 e752x_printk(KERN_WARNING, "%sError %s\n", 557 fatal_message[fatal], hub_message[i]); 558 } 559} 560 561static inline void hub_error(int fatal, u8 errors, int *error_found, 562 int handle_error) 563{ 564 *error_found = 1; 565 566 if (handle_error) 567 do_hub_error(fatal, errors); 568} 569 570#define NSI_FATAL_MASK 0x0c080081 571#define NSI_NON_FATAL_MASK 0x23a0ba64 572#define NSI_ERR_MASK (NSI_FATAL_MASK | NSI_NON_FATAL_MASK) 573 574static char *nsi_message[30] = { 575 "NSI Link Down", /* NSI_FERR/NSI_NERR bit 0, fatal error */ 576 "", /* reserved */ 577 "NSI Parity Error", /* bit 2, non-fatal */ 578 "", /* reserved */ 579 "", /* reserved */ 580 "Correctable Error Message", /* bit 5, non-fatal */ 581 "Non-Fatal Error Message", /* bit 6, non-fatal */ 582 "Fatal Error Message", /* bit 7, fatal */ 583 "", /* reserved */ 584 "Receiver Error", /* bit 9, non-fatal */ 585 "", /* reserved */ 586 "Bad TLP", /* bit 11, non-fatal */ 587 "Bad DLLP", /* bit 12, non-fatal */ 588 "REPLAY_NUM Rollover", /* bit 13, non-fatal */ 589 "", /* reserved */ 590 "Replay Timer Timeout", /* bit 15, non-fatal */ 591 "", /* reserved */ 592 "", /* reserved */ 593 "", /* reserved */ 594 "Data Link Protocol Error", /* bit 19, fatal */ 595 "", /* reserved */ 596 "Poisoned TLP", /* bit 21, non-fatal */ 597 "", /* reserved */ 598 "Completion Timeout", /* bit 23, non-fatal */ 599 "Completer Abort", /* bit 24, non-fatal */ 600 "Unexpected Completion", /* bit 25, non-fatal */ 601 "Receiver Overflow", /* bit 26, fatal */ 602 "Malformed TLP", /* bit 27, fatal */ 603 "", /* reserved */ 604 "Unsupported Request" /* bit 29, non-fatal */ 605}; 606 607static void do_nsi_error(int fatal, u32 errors) 608{ 609 int i; 610 611 for (i = 0; i < 30; i++) { 612 if (errors & (1 << i)) 613 printk(KERN_WARNING "%sError %s\n", 614 fatal_message[fatal], nsi_message[i]); 615 } 616} 617 618static inline void nsi_error(int fatal, u32 errors, int *error_found, 619 int handle_error) 620{ 621 *error_found = 1; 622 623 if (handle_error) 624 do_nsi_error(fatal, errors); 625} 626 627static char *membuf_message[4] = { 628 "Internal PMWB to DRAM parity", 629 "Internal PMWB to System Bus Parity", 630 "Internal System Bus or IO to PMWB Parity", 631 "Internal DRAM to PMWB Parity" 632}; 633 634static void do_membuf_error(u8 errors) 635{ 636 int i; 637 638 for (i = 0; i < 4; i++) { 639 if (errors & (1 << i)) 640 e752x_printk(KERN_WARNING, "Non-Fatal Error %s\n", 641 membuf_message[i]); 642 } 643} 644 645static inline void membuf_error(u8 errors, int *error_found, int handle_error) 646{ 647 *error_found = 1; 648 649 if (handle_error) 650 do_membuf_error(errors); 651} 652 653static char *sysbus_message[10] = { 654 "Addr or Request Parity", 655 "Data Strobe Glitch", 656 "Addr Strobe Glitch", 657 "Data Parity", 658 "Addr Above TOM", 659 "Non DRAM Lock Error", 660 "MCERR", "BINIT", 661 "Memory Parity", 662 "IO Subsystem Parity" 663}; 664 665static void do_sysbus_error(int fatal, u32 errors) 666{ 667 int i; 668 669 for (i = 0; i < 10; i++) { 670 if (errors & (1 << i)) 671 e752x_printk(KERN_WARNING, "%sError System Bus %s\n", 672 fatal_message[fatal], sysbus_message[i]); 673 } 674} 675 676static inline void sysbus_error(int fatal, u32 errors, int *error_found, 677 int handle_error) 678{ 679 *error_found = 1; 680 681 if (handle_error) 682 do_sysbus_error(fatal, errors); 683} 684 685static void e752x_check_hub_interface(struct e752x_error_info *info, 686 int *error_found, int handle_error) 687{ 688 u8 stat8; 689 690 //pci_read_config_byte(dev,E752X_HI_FERR,&stat8); 691 692 stat8 = info->hi_ferr; 693 694 if (stat8 & 0x7f) { /* Error, so process */ 695 stat8 &= 0x7f; 696 697 if (stat8 & 0x2b) 698 hub_error(1, stat8 & 0x2b, error_found, handle_error); 699 700 if (stat8 & 0x54) 701 hub_error(0, stat8 & 0x54, error_found, handle_error); 702 } 703 //pci_read_config_byte(dev,E752X_HI_NERR,&stat8); 704 705 stat8 = info->hi_nerr; 706 707 if (stat8 & 0x7f) { /* Error, so process */ 708 stat8 &= 0x7f; 709 710 if (stat8 & 0x2b) 711 hub_error(1, stat8 & 0x2b, error_found, handle_error); 712 713 if (stat8 & 0x54) 714 hub_error(0, stat8 & 0x54, error_found, handle_error); 715 } 716} 717 718static void e752x_check_ns_interface(struct e752x_error_info *info, 719 int *error_found, int handle_error) 720{ 721 u32 stat32; 722 723 stat32 = info->nsi_ferr; 724 if (stat32 & NSI_ERR_MASK) { /* Error, so process */ 725 if (stat32 & NSI_FATAL_MASK) /* check for fatal errors */ 726 nsi_error(1, stat32 & NSI_FATAL_MASK, error_found, 727 handle_error); 728 if (stat32 & NSI_NON_FATAL_MASK) /* check for non-fatal ones */ 729 nsi_error(0, stat32 & NSI_NON_FATAL_MASK, error_found, 730 handle_error); 731 } 732 stat32 = info->nsi_nerr; 733 if (stat32 & NSI_ERR_MASK) { 734 if (stat32 & NSI_FATAL_MASK) 735 nsi_error(1, stat32 & NSI_FATAL_MASK, error_found, 736 handle_error); 737 if (stat32 & NSI_NON_FATAL_MASK) 738 nsi_error(0, stat32 & NSI_NON_FATAL_MASK, error_found, 739 handle_error); 740 } 741} 742 743static void e752x_check_sysbus(struct e752x_error_info *info, 744 int *error_found, int handle_error) 745{ 746 u32 stat32, error32; 747 748 //pci_read_config_dword(dev,E752X_SYSBUS_FERR,&stat32); 749 stat32 = info->sysbus_ferr + (info->sysbus_nerr << 16); 750 751 if (stat32 == 0) 752 return; /* no errors */ 753 754 error32 = (stat32 >> 16) & 0x3ff; 755 stat32 = stat32 & 0x3ff; 756 757 if (stat32 & 0x087) 758 sysbus_error(1, stat32 & 0x087, error_found, handle_error); 759 760 if (stat32 & 0x378) 761 sysbus_error(0, stat32 & 0x378, error_found, handle_error); 762 763 if (error32 & 0x087) 764 sysbus_error(1, error32 & 0x087, error_found, handle_error); 765 766 if (error32 & 0x378) 767 sysbus_error(0, error32 & 0x378, error_found, handle_error); 768} 769 770static void e752x_check_membuf(struct e752x_error_info *info, 771 int *error_found, int handle_error) 772{ 773 u8 stat8; 774 775 stat8 = info->buf_ferr; 776 777 if (stat8 & 0x0f) { /* Error, so process */ 778 stat8 &= 0x0f; 779 membuf_error(stat8, error_found, handle_error); 780 } 781 782 stat8 = info->buf_nerr; 783 784 if (stat8 & 0x0f) { /* Error, so process */ 785 stat8 &= 0x0f; 786 membuf_error(stat8, error_found, handle_error); 787 } 788} 789 790static void e752x_check_dram(struct mem_ctl_info *mci, 791 struct e752x_error_info *info, int *error_found, 792 int handle_error) 793{ 794 u16 error_one, error_next; 795 796 error_one = info->dram_ferr; 797 error_next = info->dram_nerr; 798 799 /* decode and report errors */ 800 if (error_one & 0x0101) /* check first error correctable */ 801 process_ce(mci, error_one, info->dram_sec1_add, 802 info->dram_sec1_syndrome, error_found, handle_error); 803 804 if (error_next & 0x0101) /* check next error correctable */ 805 process_ce(mci, error_next, info->dram_sec2_add, 806 info->dram_sec2_syndrome, error_found, handle_error); 807 808 if (error_one & 0x4040) 809 process_ue_no_info_wr(mci, error_found, handle_error); 810 811 if (error_next & 0x4040) 812 process_ue_no_info_wr(mci, error_found, handle_error); 813 814 if (error_one & 0x2020) 815 process_ded_retry(mci, error_one, info->dram_retr_add, 816 error_found, handle_error); 817 818 if (error_next & 0x2020) 819 process_ded_retry(mci, error_next, info->dram_retr_add, 820 error_found, handle_error); 821 822 if (error_one & 0x0808) 823 process_threshold_ce(mci, error_one, error_found, handle_error); 824 825 if (error_next & 0x0808) 826 process_threshold_ce(mci, error_next, error_found, 827 handle_error); 828 829 if (error_one & 0x0606) 830 process_ue(mci, error_one, info->dram_ded_add, 831 info->dram_scrb_add, error_found, handle_error); 832 833 if (error_next & 0x0606) 834 process_ue(mci, error_next, info->dram_ded_add, 835 info->dram_scrb_add, error_found, handle_error); 836} 837 838static void e752x_get_error_info(struct mem_ctl_info *mci, 839 struct e752x_error_info *info) 840{ 841 struct pci_dev *dev; 842 struct e752x_pvt *pvt; 843 844 memset(info, 0, sizeof(*info)); 845 pvt = (struct e752x_pvt *)mci->pvt_info; 846 dev = pvt->dev_d0f1; 847 pci_read_config_dword(dev, E752X_FERR_GLOBAL, &info->ferr_global); 848 849 if (info->ferr_global) { 850 if (pvt->dev_info->err_dev == PCI_DEVICE_ID_INTEL_3100_1_ERR) { 851 pci_read_config_dword(dev, I3100_NSI_FERR, 852 &info->nsi_ferr); 853 info->hi_ferr = 0; 854 } else { 855 pci_read_config_byte(dev, E752X_HI_FERR, 856 &info->hi_ferr); 857 info->nsi_ferr = 0; 858 } 859 pci_read_config_word(dev, E752X_SYSBUS_FERR, 860 &info->sysbus_ferr); 861 pci_read_config_byte(dev, E752X_BUF_FERR, &info->buf_ferr); 862 pci_read_config_word(dev, E752X_DRAM_FERR, &info->dram_ferr); 863 pci_read_config_dword(dev, E752X_DRAM_SEC1_ADD, 864 &info->dram_sec1_add); 865 pci_read_config_word(dev, E752X_DRAM_SEC1_SYNDROME, 866 &info->dram_sec1_syndrome); 867 pci_read_config_dword(dev, E752X_DRAM_DED_ADD, 868 &info->dram_ded_add); 869 pci_read_config_dword(dev, E752X_DRAM_SCRB_ADD, 870 &info->dram_scrb_add); 871 pci_read_config_dword(dev, E752X_DRAM_RETR_ADD, 872 &info->dram_retr_add); 873 874 /* ignore the reserved bits just in case */ 875 if (info->hi_ferr & 0x7f) 876 pci_write_config_byte(dev, E752X_HI_FERR, 877 info->hi_ferr); 878 879 if (info->nsi_ferr & NSI_ERR_MASK) 880 pci_write_config_dword(dev, I3100_NSI_FERR, 881 info->nsi_ferr); 882 883 if (info->sysbus_ferr) 884 pci_write_config_word(dev, E752X_SYSBUS_FERR, 885 info->sysbus_ferr); 886 887 if (info->buf_ferr & 0x0f) 888 pci_write_config_byte(dev, E752X_BUF_FERR, 889 info->buf_ferr); 890 891 if (info->dram_ferr) 892 pci_write_bits16(pvt->dev_d0f1, E752X_DRAM_FERR, 893 info->dram_ferr, info->dram_ferr); 894 895 pci_write_config_dword(dev, E752X_FERR_GLOBAL, 896 info->ferr_global); 897 } 898 899 pci_read_config_dword(dev, E752X_NERR_GLOBAL, &info->nerr_global); 900 901 if (info->nerr_global) { 902 if (pvt->dev_info->err_dev == PCI_DEVICE_ID_INTEL_3100_1_ERR) { 903 pci_read_config_dword(dev, I3100_NSI_NERR, 904 &info->nsi_nerr); 905 info->hi_nerr = 0; 906 } else { 907 pci_read_config_byte(dev, E752X_HI_NERR, 908 &info->hi_nerr); 909 info->nsi_nerr = 0; 910 } 911 pci_read_config_word(dev, E752X_SYSBUS_NERR, 912 &info->sysbus_nerr); 913 pci_read_config_byte(dev, E752X_BUF_NERR, &info->buf_nerr); 914 pci_read_config_word(dev, E752X_DRAM_NERR, &info->dram_nerr); 915 pci_read_config_dword(dev, E752X_DRAM_SEC2_ADD, 916 &info->dram_sec2_add); 917 pci_read_config_word(dev, E752X_DRAM_SEC2_SYNDROME, 918 &info->dram_sec2_syndrome); 919 920 if (info->hi_nerr & 0x7f) 921 pci_write_config_byte(dev, E752X_HI_NERR, 922 info->hi_nerr); 923 924 if (info->nsi_nerr & NSI_ERR_MASK) 925 pci_write_config_dword(dev, I3100_NSI_NERR, 926 info->nsi_nerr); 927 928 if (info->sysbus_nerr) 929 pci_write_config_word(dev, E752X_SYSBUS_NERR, 930 info->sysbus_nerr); 931 932 if (info->buf_nerr & 0x0f) 933 pci_write_config_byte(dev, E752X_BUF_NERR, 934 info->buf_nerr); 935 936 if (info->dram_nerr) 937 pci_write_bits16(pvt->dev_d0f1, E752X_DRAM_NERR, 938 info->dram_nerr, info->dram_nerr); 939 940 pci_write_config_dword(dev, E752X_NERR_GLOBAL, 941 info->nerr_global); 942 } 943} 944 945static int e752x_process_error_info(struct mem_ctl_info *mci, 946 struct e752x_error_info *info, 947 int handle_errors) 948{ 949 u32 error32, stat32; 950 int error_found; 951 952 error_found = 0; 953 error32 = (info->ferr_global >> 18) & 0x3ff; 954 stat32 = (info->ferr_global >> 4) & 0x7ff; 955 956 if (error32) 957 global_error(1, error32, &error_found, handle_errors); 958 959 if (stat32) 960 global_error(0, stat32, &error_found, handle_errors); 961 962 error32 = (info->nerr_global >> 18) & 0x3ff; 963 stat32 = (info->nerr_global >> 4) & 0x7ff; 964 965 if (error32) 966 global_error(1, error32, &error_found, handle_errors); 967 968 if (stat32) 969 global_error(0, stat32, &error_found, handle_errors); 970 971 e752x_check_hub_interface(info, &error_found, handle_errors); 972 e752x_check_ns_interface(info, &error_found, handle_errors); 973 e752x_check_sysbus(info, &error_found, handle_errors); 974 e752x_check_membuf(info, &error_found, handle_errors); 975 e752x_check_dram(mci, info, &error_found, handle_errors); 976 return error_found; 977} 978 979static void e752x_check(struct mem_ctl_info *mci) 980{ 981 struct e752x_error_info info; 982 983 e752x_get_error_info(mci, &info); 984 e752x_process_error_info(mci, &info, 1); 985} 986 987/* Program byte/sec bandwidth scrub rate to hardware */ 988static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw) 989{ 990 const struct scrubrate *scrubrates; 991 struct e752x_pvt *pvt = (struct e752x_pvt *) mci->pvt_info; 992 struct pci_dev *pdev = pvt->dev_d0f0; 993 int i; 994 995 if (pvt->dev_info->ctl_dev == PCI_DEVICE_ID_INTEL_3100_0) 996 scrubrates = scrubrates_i3100; 997 else 998 scrubrates = scrubrates_e752x; 999 1000 /* Translate the desired scrub rate to a e752x/3100 register value. 1001 * Search for the bandwidth that is equal or greater than the 1002 * desired rate and program the cooresponding register value. 1003 */ 1004 for (i = 0; scrubrates[i].bandwidth != SDRATE_EOT; i++) 1005 if (scrubrates[i].bandwidth >= new_bw) 1006 break; 1007 1008 if (scrubrates[i].bandwidth == SDRATE_EOT) 1009 return -1; 1010 1011 pci_write_config_word(pdev, E752X_MCHSCRB, scrubrates[i].scrubval); 1012 1013 return scrubrates[i].bandwidth; 1014} 1015 1016/* Convert current scrub rate value into byte/sec bandwidth */ 1017static int get_sdram_scrub_rate(struct mem_ctl_info *mci) 1018{ 1019 const struct scrubrate *scrubrates; 1020 struct e752x_pvt *pvt = (struct e752x_pvt *) mci->pvt_info; 1021 struct pci_dev *pdev = pvt->dev_d0f0; 1022 u16 scrubval; 1023 int i; 1024 1025 if (pvt->dev_info->ctl_dev == PCI_DEVICE_ID_INTEL_3100_0) 1026 scrubrates = scrubrates_i3100; 1027 else 1028 scrubrates = scrubrates_e752x; 1029 1030 /* Find the bandwidth matching the memory scrubber configuration */ 1031 pci_read_config_word(pdev, E752X_MCHSCRB, &scrubval); 1032 scrubval = scrubval & 0x0f; 1033 1034 for (i = 0; scrubrates[i].bandwidth != SDRATE_EOT; i++) 1035 if (scrubrates[i].scrubval == scrubval) 1036 break; 1037 1038 if (scrubrates[i].bandwidth == SDRATE_EOT) { 1039 e752x_printk(KERN_WARNING, 1040 "Invalid sdram scrub control value: 0x%x\n", scrubval); 1041 return -1; 1042 } 1043 return scrubrates[i].bandwidth; 1044 1045} 1046 1047/* Return 1 if dual channel mode is active. Else return 0. */ 1048static inline int dual_channel_active(u16 ddrcsr) 1049{ 1050 return (((ddrcsr >> 12) & 3) == 3); 1051} 1052 1053/* Remap csrow index numbers if map_type is "reverse" 1054 */ 1055static inline int remap_csrow_index(struct mem_ctl_info *mci, int index) 1056{ 1057 struct e752x_pvt *pvt = mci->pvt_info; 1058 1059 if (!pvt->map_type) 1060 return (7 - index); 1061 1062 return (index); 1063} 1064 1065static void e752x_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev, 1066 u16 ddrcsr) 1067{ 1068 struct csrow_info *csrow; 1069 enum edac_type edac_mode; 1070 unsigned long last_cumul_size; 1071 int index, mem_dev, drc_chan; 1072 int drc_drbg; /* DRB granularity 0=64mb, 1=128mb */ 1073 int drc_ddim; /* DRAM Data Integrity Mode 0=none, 2=edac */ 1074 u8 value; 1075 u32 dra, drc, cumul_size, i, nr_pages; 1076 1077 dra = 0; 1078 for (index = 0; index < 4; index++) { 1079 u8 dra_reg; 1080 pci_read_config_byte(pdev, E752X_DRA + index, &dra_reg); 1081 dra |= dra_reg << (index * 8); 1082 } 1083 pci_read_config_dword(pdev, E752X_DRC, &drc); 1084 drc_chan = dual_channel_active(ddrcsr) ? 1 : 0; 1085 drc_drbg = drc_chan + 1; /* 128 in dual mode, 64 in single */ 1086 drc_ddim = (drc >> 20) & 0x3; 1087 1088 /* The dram row boundary (DRB) reg values are boundary address for 1089 * each DRAM row with a granularity of 64 or 128MB (single/dual 1090 * channel operation). DRB regs are cumulative; therefore DRB7 will 1091 * contain the total memory contained in all eight rows. 1092 */ 1093 for (last_cumul_size = index = 0; index < mci->nr_csrows; index++) { 1094 /* mem_dev 0=x8, 1=x4 */ 1095 mem_dev = (dra >> (index * 4 + 2)) & 0x3; 1096 csrow = mci->csrows[remap_csrow_index(mci, index)]; 1097 1098 mem_dev = (mem_dev == 2); 1099 pci_read_config_byte(pdev, E752X_DRB + index, &value); 1100 /* convert a 128 or 64 MiB DRB to a page size. */ 1101 cumul_size = value << (25 + drc_drbg - PAGE_SHIFT); 1102 edac_dbg(3, "(%d) cumul_size 0x%x\n", index, cumul_size); 1103 if (cumul_size == last_cumul_size) 1104 continue; /* not populated */ 1105 1106 csrow->first_page = last_cumul_size; 1107 csrow->last_page = cumul_size - 1; 1108 nr_pages = cumul_size - last_cumul_size; 1109 last_cumul_size = cumul_size; 1110 1111 /* 1112 * if single channel or x8 devices then SECDED 1113 * if dual channel and x4 then S4ECD4ED 1114 */ 1115 if (drc_ddim) { 1116 if (drc_chan && mem_dev) { 1117 edac_mode = EDAC_S4ECD4ED; 1118 mci->edac_cap |= EDAC_FLAG_S4ECD4ED; 1119 } else { 1120 edac_mode = EDAC_SECDED; 1121 mci->edac_cap |= EDAC_FLAG_SECDED; 1122 } 1123 } else 1124 edac_mode = EDAC_NONE; 1125 for (i = 0; i < csrow->nr_channels; i++) { 1126 struct dimm_info *dimm = csrow->channels[i]->dimm; 1127 1128 edac_dbg(3, "Initializing rank at (%i,%i)\n", index, i); 1129 dimm->nr_pages = nr_pages / csrow->nr_channels; 1130 dimm->grain = 1 << 12; /* 4KiB - resolution of CELOG */ 1131 dimm->mtype = MEM_RDDR; /* only one type supported */ 1132 dimm->dtype = mem_dev ? DEV_X4 : DEV_X8; 1133 dimm->edac_mode = edac_mode; 1134 } 1135 } 1136} 1137 1138static void e752x_init_mem_map_table(struct pci_dev *pdev, 1139 struct e752x_pvt *pvt) 1140{ 1141 int index; 1142 u8 value, last, row; 1143 1144 last = 0; 1145 row = 0; 1146 1147 for (index = 0; index < 8; index += 2) { 1148 pci_read_config_byte(pdev, E752X_DRB + index, &value); 1149 /* test if there is a dimm in this slot */ 1150 if (value == last) { 1151 /* no dimm in the slot, so flag it as empty */ 1152 pvt->map[index] = 0xff; 1153 pvt->map[index + 1] = 0xff; 1154 } else { /* there is a dimm in the slot */ 1155 pvt->map[index] = row; 1156 row++; 1157 last = value; 1158 /* test the next value to see if the dimm is double 1159 * sided 1160 */ 1161 pci_read_config_byte(pdev, E752X_DRB + index + 1, 1162 &value); 1163 1164 /* the dimm is single sided, so flag as empty */ 1165 /* this is a double sided dimm to save the next row #*/ 1166 pvt->map[index + 1] = (value == last) ? 0xff : row; 1167 row++; 1168 last = value; 1169 } 1170 } 1171} 1172 1173/* Return 0 on success or 1 on failure. */ 1174static int e752x_get_devs(struct pci_dev *pdev, int dev_idx, 1175 struct e752x_pvt *pvt) 1176{ 1177 pvt->dev_d0f1 = pci_get_device(PCI_VENDOR_ID_INTEL, 1178 pvt->dev_info->err_dev, NULL); 1179 1180 if (pvt->dev_d0f1 == NULL) { 1181 pvt->dev_d0f1 = pci_scan_single_device(pdev->bus, 1182 PCI_DEVFN(0, 1)); 1183 pci_dev_get(pvt->dev_d0f1); 1184 } 1185 1186 if (pvt->dev_d0f1 == NULL) { 1187 e752x_printk(KERN_ERR, "error reporting device not found:" 1188 "vendor %x device 0x%x (broken BIOS?)\n", 1189 PCI_VENDOR_ID_INTEL, e752x_devs[dev_idx].err_dev); 1190 return 1; 1191 } 1192 1193 pvt->dev_d0f0 = pci_get_device(PCI_VENDOR_ID_INTEL, 1194 e752x_devs[dev_idx].ctl_dev, 1195 NULL); 1196 1197 if (pvt->dev_d0f0 == NULL) 1198 goto fail; 1199 1200 return 0; 1201 1202fail: 1203 pci_dev_put(pvt->dev_d0f1); 1204 return 1; 1205} 1206 1207/* Setup system bus parity mask register. 1208 * Sysbus parity supported on: 1209 * e7320/e7520/e7525 + Xeon 1210 */ 1211static void e752x_init_sysbus_parity_mask(struct e752x_pvt *pvt) 1212{ 1213 char *cpu_id = cpu_data(0).x86_model_id; 1214 struct pci_dev *dev = pvt->dev_d0f1; 1215 int enable = 1; 1216 1217 /* Allow module parameter override, else see if CPU supports parity */ 1218 if (sysbus_parity != -1) { 1219 enable = sysbus_parity; 1220 } else if (cpu_id[0] && !strstr(cpu_id, "Xeon")) { 1221 e752x_printk(KERN_INFO, "System Bus Parity not " 1222 "supported by CPU, disabling\n"); 1223 enable = 0; 1224 } 1225 1226 if (enable) 1227 pci_write_config_word(dev, E752X_SYSBUS_ERRMASK, 0x0000); 1228 else 1229 pci_write_config_word(dev, E752X_SYSBUS_ERRMASK, 0x0309); 1230} 1231 1232static void e752x_init_error_reporting_regs(struct e752x_pvt *pvt) 1233{ 1234 struct pci_dev *dev; 1235 1236 dev = pvt->dev_d0f1; 1237 /* Turn off error disable & SMI in case the BIOS turned it on */ 1238 if (pvt->dev_info->err_dev == PCI_DEVICE_ID_INTEL_3100_1_ERR) { 1239 pci_write_config_dword(dev, I3100_NSI_EMASK, 0); 1240 pci_write_config_dword(dev, I3100_NSI_SMICMD, 0); 1241 } else { 1242 pci_write_config_byte(dev, E752X_HI_ERRMASK, 0x00); 1243 pci_write_config_byte(dev, E752X_HI_SMICMD, 0x00); 1244 } 1245 1246 e752x_init_sysbus_parity_mask(pvt); 1247 1248 pci_write_config_word(dev, E752X_SYSBUS_SMICMD, 0x00); 1249 pci_write_config_byte(dev, E752X_BUF_ERRMASK, 0x00); 1250 pci_write_config_byte(dev, E752X_BUF_SMICMD, 0x00); 1251 pci_write_config_byte(dev, E752X_DRAM_ERRMASK, 0x00); 1252 pci_write_config_byte(dev, E752X_DRAM_SMICMD, 0x00); 1253} 1254 1255static int e752x_probe1(struct pci_dev *pdev, int dev_idx) 1256{ 1257 u16 pci_data; 1258 u8 stat8; 1259 struct mem_ctl_info *mci; 1260 struct edac_mc_layer layers[2]; 1261 struct e752x_pvt *pvt; 1262 u16 ddrcsr; 1263 int drc_chan; /* Number of channels 0=1chan,1=2chan */ 1264 struct e752x_error_info discard; 1265 1266 edac_dbg(0, "mci\n"); 1267 edac_dbg(0, "Starting Probe1\n"); 1268 1269 /* check to see if device 0 function 1 is enabled; if it isn't, we 1270 * assume the BIOS has reserved it for a reason and is expecting 1271 * exclusive access, we take care not to violate that assumption and 1272 * fail the probe. */ 1273 pci_read_config_byte(pdev, E752X_DEVPRES1, &stat8); 1274 if (!force_function_unhide && !(stat8 & (1 << 5))) { 1275 printk(KERN_INFO "Contact your BIOS vendor to see if the " 1276 "E752x error registers can be safely un-hidden\n"); 1277 return -ENODEV; 1278 } 1279 stat8 |= (1 << 5); 1280 pci_write_config_byte(pdev, E752X_DEVPRES1, stat8); 1281 1282 pci_read_config_word(pdev, E752X_DDRCSR, &ddrcsr); 1283 /* FIXME: should check >>12 or 0xf, true for all? */ 1284 /* Dual channel = 1, Single channel = 0 */ 1285 drc_chan = dual_channel_active(ddrcsr); 1286 1287 layers[0].type = EDAC_MC_LAYER_CHIP_SELECT; 1288 layers[0].size = E752X_NR_CSROWS; 1289 layers[0].is_virt_csrow = true; 1290 layers[1].type = EDAC_MC_LAYER_CHANNEL; 1291 layers[1].size = drc_chan + 1; 1292 layers[1].is_virt_csrow = false; 1293 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(*pvt)); 1294 if (mci == NULL) 1295 return -ENOMEM; 1296 1297 edac_dbg(3, "init mci\n"); 1298 mci->mtype_cap = MEM_FLAG_RDDR; 1299 /* 3100 IMCH supports SECDEC only */ 1300 mci->edac_ctl_cap = (dev_idx == I3100) ? EDAC_FLAG_SECDED : 1301 (EDAC_FLAG_NONE | EDAC_FLAG_SECDED | EDAC_FLAG_S4ECD4ED); 1302 /* FIXME - what if different memory types are in different csrows? */ 1303 mci->mod_name = EDAC_MOD_STR; 1304 mci->pdev = &pdev->dev; 1305 1306 edac_dbg(3, "init pvt\n"); 1307 pvt = (struct e752x_pvt *)mci->pvt_info; 1308 pvt->dev_info = &e752x_devs[dev_idx]; 1309 pvt->mc_symmetric = ((ddrcsr & 0x10) != 0); 1310 1311 if (e752x_get_devs(pdev, dev_idx, pvt)) { 1312 edac_mc_free(mci); 1313 return -ENODEV; 1314 } 1315 1316 edac_dbg(3, "more mci init\n"); 1317 mci->ctl_name = pvt->dev_info->ctl_name; 1318 mci->dev_name = pci_name(pdev); 1319 mci->edac_check = e752x_check; 1320 mci->ctl_page_to_phys = ctl_page_to_phys; 1321 mci->set_sdram_scrub_rate = set_sdram_scrub_rate; 1322 mci->get_sdram_scrub_rate = get_sdram_scrub_rate; 1323 1324 /* set the map type. 1 = normal, 0 = reversed 1325 * Must be set before e752x_init_csrows in case csrow mapping 1326 * is reversed. 1327 */ 1328 pci_read_config_byte(pdev, E752X_DRM, &stat8); 1329 pvt->map_type = ((stat8 & 0x0f) > ((stat8 >> 4) & 0x0f)); 1330 1331 e752x_init_csrows(mci, pdev, ddrcsr); 1332 e752x_init_mem_map_table(pdev, pvt); 1333 1334 if (dev_idx == I3100) 1335 mci->edac_cap = EDAC_FLAG_SECDED; /* the only mode supported */ 1336 else 1337 mci->edac_cap |= EDAC_FLAG_NONE; 1338 edac_dbg(3, "tolm, remapbase, remaplimit\n"); 1339 1340 /* load the top of low memory, remap base, and remap limit vars */ 1341 pci_read_config_word(pdev, E752X_TOLM, &pci_data); 1342 pvt->tolm = ((u32) pci_data) << 4; 1343 pci_read_config_word(pdev, E752X_REMAPBASE, &pci_data); 1344 pvt->remapbase = ((u32) pci_data) << 14; 1345 pci_read_config_word(pdev, E752X_REMAPLIMIT, &pci_data); 1346 pvt->remaplimit = ((u32) pci_data) << 14; 1347 e752x_printk(KERN_INFO, 1348 "tolm = %x, remapbase = %x, remaplimit = %x\n", 1349 pvt->tolm, pvt->remapbase, pvt->remaplimit); 1350 1351 /* Here we assume that we will never see multiple instances of this 1352 * type of memory controller. The ID is therefore hardcoded to 0. 1353 */ 1354 if (edac_mc_add_mc(mci)) { 1355 edac_dbg(3, "failed edac_mc_add_mc()\n"); 1356 goto fail; 1357 } 1358 1359 e752x_init_error_reporting_regs(pvt); 1360 e752x_get_error_info(mci, &discard); /* clear other MCH errors */ 1361 1362 /* allocating generic PCI control info */ 1363 e752x_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR); 1364 if (!e752x_pci) { 1365 printk(KERN_WARNING 1366 "%s(): Unable to create PCI control\n", __func__); 1367 printk(KERN_WARNING 1368 "%s(): PCI error report via EDAC not setup\n", 1369 __func__); 1370 } 1371 1372 /* get this far and it's successful */ 1373 edac_dbg(3, "success\n"); 1374 return 0; 1375 1376fail: 1377 pci_dev_put(pvt->dev_d0f0); 1378 pci_dev_put(pvt->dev_d0f1); 1379 edac_mc_free(mci); 1380 1381 return -ENODEV; 1382} 1383 1384/* returns count (>= 0), or negative on error */ 1385static int e752x_init_one(struct pci_dev *pdev, const struct pci_device_id *ent) 1386{ 1387 edac_dbg(0, "\n"); 1388 1389 /* wake up and enable device */ 1390 if (pci_enable_device(pdev) < 0) 1391 return -EIO; 1392 1393 return e752x_probe1(pdev, ent->driver_data); 1394} 1395 1396static void e752x_remove_one(struct pci_dev *pdev) 1397{ 1398 struct mem_ctl_info *mci; 1399 struct e752x_pvt *pvt; 1400 1401 edac_dbg(0, "\n"); 1402 1403 if (e752x_pci) 1404 edac_pci_release_generic_ctl(e752x_pci); 1405 1406 if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL) 1407 return; 1408 1409 pvt = (struct e752x_pvt *)mci->pvt_info; 1410 pci_dev_put(pvt->dev_d0f0); 1411 pci_dev_put(pvt->dev_d0f1); 1412 edac_mc_free(mci); 1413} 1414 1415static const struct pci_device_id e752x_pci_tbl[] = { 1416 { 1417 PCI_VEND_DEV(INTEL, 7520_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1418 E7520}, 1419 { 1420 PCI_VEND_DEV(INTEL, 7525_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1421 E7525}, 1422 { 1423 PCI_VEND_DEV(INTEL, 7320_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1424 E7320}, 1425 { 1426 PCI_VEND_DEV(INTEL, 3100_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0, 1427 I3100}, 1428 { 1429 0, 1430 } /* 0 terminated list. */ 1431}; 1432 1433MODULE_DEVICE_TABLE(pci, e752x_pci_tbl); 1434 1435static struct pci_driver e752x_driver = { 1436 .name = EDAC_MOD_STR, 1437 .probe = e752x_init_one, 1438 .remove = e752x_remove_one, 1439 .id_table = e752x_pci_tbl, 1440}; 1441 1442static int __init e752x_init(void) 1443{ 1444 int pci_rc; 1445 1446 edac_dbg(3, "\n"); 1447 1448 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 1449 opstate_init(); 1450 1451 pci_rc = pci_register_driver(&e752x_driver); 1452 return (pci_rc < 0) ? pci_rc : 0; 1453} 1454 1455static void __exit e752x_exit(void) 1456{ 1457 edac_dbg(3, "\n"); 1458 pci_unregister_driver(&e752x_driver); 1459} 1460 1461module_init(e752x_init); 1462module_exit(e752x_exit); 1463 1464MODULE_LICENSE("GPL"); 1465MODULE_AUTHOR("Linux Networx (http://lnxi.com) Tom Zimmerman\n"); 1466MODULE_DESCRIPTION("MC support for Intel e752x/3100 memory controllers"); 1467 1468module_param(force_function_unhide, int, 0444); 1469MODULE_PARM_DESC(force_function_unhide, "if BIOS sets Dev0:Fun1 up as hidden:" 1470 " 1=force unhide and hope BIOS doesn't fight driver for " 1471 "Dev0:Fun1 access"); 1472 1473module_param(edac_op_state, int, 0444); 1474MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 1475 1476module_param(sysbus_parity, int, 0444); 1477MODULE_PARM_DESC(sysbus_parity, "0=disable system bus parity checking," 1478 " 1=enable system bus parity checking, default=auto-detect"); 1479module_param(report_non_memory_errors, int, 0644); 1480MODULE_PARM_DESC(report_non_memory_errors, "0=disable non-memory error " 1481 "reporting, 1=enable non-memory error reporting");