1 /* 2 * Intel 5100 Memory Controllers kernel module 3 * 4 * This file may be distributed under the terms of the 5 * GNU General Public License. 6 * 7 * This module is based on the following document: 8 * 9 * Intel 5100X Chipset Memory Controller Hub (MCH) - Datasheet 10 * http://download.intel.com/design/chipsets/datashts/318378.pdf 11 * 12 * The intel 5100 has two independent channels. EDAC core currently 13 * can not reflect this configuration so instead the chip-select 14 * rows for each respective channel are laid out one after another, 15 * the first half belonging to channel 0, the second half belonging 16 * to channel 1. 17 * 18 * This driver is for DDR2 DIMMs, and it uses chip select to select among the 19 * several ranks. However, instead of showing memories as ranks, it outputs 20 * them as DIMM's. An internal table creates the association between ranks 21 * and DIMM's. 22 */ 23 #include <linux/module.h> 24 #include <linux/init.h> 25 #include <linux/pci.h> 26 #include <linux/pci_ids.h> 27 #include <linux/edac.h> 28 #include <linux/delay.h> 29 #include <linux/mmzone.h> 30 #include <linux/debugfs.h> 31 32 #include "edac_module.h" 33 34 /* register addresses */ 35 36 /* device 16, func 1 */ 37 #define I5100_MC 0x40 /* Memory Control Register */ 38 #define I5100_MC_SCRBEN_MASK (1 << 7) 39 #define I5100_MC_SCRBDONE_MASK (1 << 4) 40 #define I5100_MS 0x44 /* Memory Status Register */ 41 #define I5100_SPDDATA 0x48 /* Serial Presence Detect Status Reg */ 42 #define I5100_SPDCMD 0x4c /* Serial Presence Detect Command Reg */ 43 #define I5100_TOLM 0x6c /* Top of Low Memory */ 44 #define I5100_MIR0 0x80 /* Memory Interleave Range 0 */ 45 #define I5100_MIR1 0x84 /* Memory Interleave Range 1 */ 46 #define I5100_AMIR_0 0x8c /* Adjusted Memory Interleave Range 0 */ 47 #define I5100_AMIR_1 0x90 /* Adjusted Memory Interleave Range 1 */ 48 #define I5100_FERR_NF_MEM 0xa0 /* MC First Non Fatal Errors */ 49 #define I5100_FERR_NF_MEM_M16ERR_MASK (1 << 16) 50 #define I5100_FERR_NF_MEM_M15ERR_MASK (1 << 15) 51 #define I5100_FERR_NF_MEM_M14ERR_MASK (1 << 14) 52 #define I5100_FERR_NF_MEM_M12ERR_MASK (1 << 12) 53 #define I5100_FERR_NF_MEM_M11ERR_MASK (1 << 11) 54 #define I5100_FERR_NF_MEM_M10ERR_MASK (1 << 10) 55 #define I5100_FERR_NF_MEM_M6ERR_MASK (1 << 6) 56 #define I5100_FERR_NF_MEM_M5ERR_MASK (1 << 5) 57 #define I5100_FERR_NF_MEM_M4ERR_MASK (1 << 4) 58 #define I5100_FERR_NF_MEM_M1ERR_MASK (1 << 1) 59 #define I5100_FERR_NF_MEM_ANY_MASK \ 60 (I5100_FERR_NF_MEM_M16ERR_MASK | \ 61 I5100_FERR_NF_MEM_M15ERR_MASK | \ 62 I5100_FERR_NF_MEM_M14ERR_MASK | \ 63 I5100_FERR_NF_MEM_M12ERR_MASK | \ 64 I5100_FERR_NF_MEM_M11ERR_MASK | \ 65 I5100_FERR_NF_MEM_M10ERR_MASK | \ 66 I5100_FERR_NF_MEM_M6ERR_MASK | \ 67 I5100_FERR_NF_MEM_M5ERR_MASK | \ 68 I5100_FERR_NF_MEM_M4ERR_MASK | \ 69 I5100_FERR_NF_MEM_M1ERR_MASK) 70 #define I5100_NERR_NF_MEM 0xa4 /* MC Next Non-Fatal Errors */ 71 #define I5100_EMASK_MEM 0xa8 /* MC Error Mask Register */ 72 #define I5100_MEM0EINJMSK0 0x200 /* Injection Mask0 Register Channel 0 */ 73 #define I5100_MEM1EINJMSK0 0x208 /* Injection Mask0 Register Channel 1 */ 74 #define I5100_MEMXEINJMSK0_EINJEN (1 << 27) 75 #define I5100_MEM0EINJMSK1 0x204 /* Injection Mask1 Register Channel 0 */ 76 #define I5100_MEM1EINJMSK1 0x206 /* Injection Mask1 Register Channel 1 */ 77 78 /* Device 19, Function 0 */ 79 #define I5100_DINJ0 0x9a 80 81 /* device 21 and 22, func 0 */ 82 #define I5100_MTR_0 0x154 /* Memory Technology Registers 0-3 */ 83 #define I5100_DMIR 0x15c /* DIMM Interleave Range */ 84 #define I5100_VALIDLOG 0x18c /* Valid Log Markers */ 85 #define I5100_NRECMEMA 0x190 /* Non-Recoverable Memory Error Log Reg A */ 86 #define I5100_NRECMEMB 0x194 /* Non-Recoverable Memory Error Log Reg B */ 87 #define I5100_REDMEMA 0x198 /* Recoverable Memory Data Error Log Reg A */ 88 #define I5100_REDMEMB 0x19c /* Recoverable Memory Data Error Log Reg B */ 89 #define I5100_RECMEMA 0x1a0 /* Recoverable Memory Error Log Reg A */ 90 #define I5100_RECMEMB 0x1a4 /* Recoverable Memory Error Log Reg B */ 91 #define I5100_MTR_4 0x1b0 /* Memory Technology Registers 4,5 */ 92 93 /* bit field accessors */ 94 95 static inline u32 i5100_mc_scrben(u32 mc) 96 { 97 return mc >> 7 & 1; 98 } 99 100 static inline u32 i5100_mc_errdeten(u32 mc) 101 { 102 return mc >> 5 & 1; 103 } 104 105 static inline u32 i5100_mc_scrbdone(u32 mc) 106 { 107 return mc >> 4 & 1; 108 } 109 110 static inline u16 i5100_spddata_rdo(u16 a) 111 { 112 return a >> 15 & 1; 113 } 114 115 static inline u16 i5100_spddata_sbe(u16 a) 116 { 117 return a >> 13 & 1; 118 } 119 120 static inline u16 i5100_spddata_busy(u16 a) 121 { 122 return a >> 12 & 1; 123 } 124 125 static inline u16 i5100_spddata_data(u16 a) 126 { 127 return a & ((1 << 8) - 1); 128 } 129 130 static inline u32 i5100_spdcmd_create(u32 dti, u32 ckovrd, u32 sa, u32 ba, 131 u32 data, u32 cmd) 132 { 133 return ((dti & ((1 << 4) - 1)) << 28) | 134 ((ckovrd & 1) << 27) | 135 ((sa & ((1 << 3) - 1)) << 24) | 136 ((ba & ((1 << 8) - 1)) << 16) | 137 ((data & ((1 << 8) - 1)) << 8) | 138 (cmd & 1); 139 } 140 141 static inline u16 i5100_tolm_tolm(u16 a) 142 { 143 return a >> 12 & ((1 << 4) - 1); 144 } 145 146 static inline u16 i5100_mir_limit(u16 a) 147 { 148 return a >> 4 & ((1 << 12) - 1); 149 } 150 151 static inline u16 i5100_mir_way1(u16 a) 152 { 153 return a >> 1 & 1; 154 } 155 156 static inline u16 i5100_mir_way0(u16 a) 157 { 158 return a & 1; 159 } 160 161 static inline u32 i5100_ferr_nf_mem_chan_indx(u32 a) 162 { 163 return a >> 28 & 1; 164 } 165 166 static inline u32 i5100_ferr_nf_mem_any(u32 a) 167 { 168 return a & I5100_FERR_NF_MEM_ANY_MASK; 169 } 170 171 static inline u32 i5100_nerr_nf_mem_any(u32 a) 172 { 173 return i5100_ferr_nf_mem_any(a); 174 } 175 176 static inline u32 i5100_dmir_limit(u32 a) 177 { 178 return a >> 16 & ((1 << 11) - 1); 179 } 180 181 static inline u32 i5100_dmir_rank(u32 a, u32 i) 182 { 183 return a >> (4 * i) & ((1 << 2) - 1); 184 } 185 186 static inline u16 i5100_mtr_present(u16 a) 187 { 188 return a >> 10 & 1; 189 } 190 191 static inline u16 i5100_mtr_ethrottle(u16 a) 192 { 193 return a >> 9 & 1; 194 } 195 196 static inline u16 i5100_mtr_width(u16 a) 197 { 198 return a >> 8 & 1; 199 } 200 201 static inline u16 i5100_mtr_numbank(u16 a) 202 { 203 return a >> 6 & 1; 204 } 205 206 static inline u16 i5100_mtr_numrow(u16 a) 207 { 208 return a >> 2 & ((1 << 2) - 1); 209 } 210 211 static inline u16 i5100_mtr_numcol(u16 a) 212 { 213 return a & ((1 << 2) - 1); 214 } 215 216 217 static inline u32 i5100_validlog_redmemvalid(u32 a) 218 { 219 return a >> 2 & 1; 220 } 221 222 static inline u32 i5100_validlog_recmemvalid(u32 a) 223 { 224 return a >> 1 & 1; 225 } 226 227 static inline u32 i5100_validlog_nrecmemvalid(u32 a) 228 { 229 return a & 1; 230 } 231 232 static inline u32 i5100_nrecmema_merr(u32 a) 233 { 234 return a >> 15 & ((1 << 5) - 1); 235 } 236 237 static inline u32 i5100_nrecmema_bank(u32 a) 238 { 239 return a >> 12 & ((1 << 3) - 1); 240 } 241 242 static inline u32 i5100_nrecmema_rank(u32 a) 243 { 244 return a >> 8 & ((1 << 3) - 1); 245 } 246 247 static inline u32 i5100_nrecmema_dm_buf_id(u32 a) 248 { 249 return a & ((1 << 8) - 1); 250 } 251 252 static inline u32 i5100_nrecmemb_cas(u32 a) 253 { 254 return a >> 16 & ((1 << 13) - 1); 255 } 256 257 static inline u32 i5100_nrecmemb_ras(u32 a) 258 { 259 return a & ((1 << 16) - 1); 260 } 261 262 static inline u32 i5100_redmemb_ecc_locator(u32 a) 263 { 264 return a & ((1 << 18) - 1); 265 } 266 267 static inline u32 i5100_recmema_merr(u32 a) 268 { 269 return i5100_nrecmema_merr(a); 270 } 271 272 static inline u32 i5100_recmema_bank(u32 a) 273 { 274 return i5100_nrecmema_bank(a); 275 } 276 277 static inline u32 i5100_recmema_rank(u32 a) 278 { 279 return i5100_nrecmema_rank(a); 280 } 281 282 static inline u32 i5100_recmemb_cas(u32 a) 283 { 284 return i5100_nrecmemb_cas(a); 285 } 286 287 static inline u32 i5100_recmemb_ras(u32 a) 288 { 289 return i5100_nrecmemb_ras(a); 290 } 291 292 /* some generic limits */ 293 #define I5100_MAX_RANKS_PER_CHAN 6 294 #define I5100_CHANNELS 2 295 #define I5100_MAX_RANKS_PER_DIMM 4 296 #define I5100_DIMM_ADDR_LINES (6 - 3) /* 64 bits / 8 bits per byte */ 297 #define I5100_MAX_DIMM_SLOTS_PER_CHAN 4 298 #define I5100_MAX_RANK_INTERLEAVE 4 299 #define I5100_MAX_DMIRS 5 300 #define I5100_SCRUB_REFRESH_RATE (5 * 60 * HZ) 301 302 struct i5100_priv { 303 /* ranks on each dimm -- 0 maps to not present -- obtained via SPD */ 304 int dimm_numrank[I5100_CHANNELS][I5100_MAX_DIMM_SLOTS_PER_CHAN]; 305 306 /* 307 * mainboard chip select map -- maps i5100 chip selects to 308 * DIMM slot chip selects. In the case of only 4 ranks per 309 * channel, the mapping is fairly obvious but not unique. 310 * we map -1 -> NC and assume both channels use the same 311 * map... 312 * 313 */ 314 int dimm_csmap[I5100_MAX_DIMM_SLOTS_PER_CHAN][I5100_MAX_RANKS_PER_DIMM]; 315 316 /* memory interleave range */ 317 struct { 318 u64 limit; 319 unsigned way[2]; 320 } mir[I5100_CHANNELS]; 321 322 /* adjusted memory interleave range register */ 323 unsigned amir[I5100_CHANNELS]; 324 325 /* dimm interleave range */ 326 struct { 327 unsigned rank[I5100_MAX_RANK_INTERLEAVE]; 328 u64 limit; 329 } dmir[I5100_CHANNELS][I5100_MAX_DMIRS]; 330 331 /* memory technology registers... */ 332 struct { 333 unsigned present; /* 0 or 1 */ 334 unsigned ethrottle; /* 0 or 1 */ 335 unsigned width; /* 4 or 8 bits */ 336 unsigned numbank; /* 2 or 3 lines */ 337 unsigned numrow; /* 13 .. 16 lines */ 338 unsigned numcol; /* 11 .. 12 lines */ 339 } mtr[I5100_CHANNELS][I5100_MAX_RANKS_PER_CHAN]; 340 341 u64 tolm; /* top of low memory in bytes */ 342 unsigned ranksperchan; /* number of ranks per channel */ 343 344 struct pci_dev *mc; /* device 16 func 1 */ 345 struct pci_dev *einj; /* device 19 func 0 */ 346 struct pci_dev *ch0mm; /* device 21 func 0 */ 347 struct pci_dev *ch1mm; /* device 22 func 0 */ 348 349 struct delayed_work i5100_scrubbing; 350 int scrub_enable; 351 352 /* Error injection */ 353 u8 inject_channel; 354 u8 inject_hlinesel; 355 u8 inject_deviceptr1; 356 u8 inject_deviceptr2; 357 u16 inject_eccmask1; 358 u16 inject_eccmask2; 359 360 struct dentry *debugfs; 361 }; 362 363 static struct dentry *i5100_debugfs; 364 365 /* map a rank/chan to a slot number on the mainboard */ 366 static int i5100_rank_to_slot(const struct mem_ctl_info *mci, 367 int chan, int rank) 368 { 369 const struct i5100_priv *priv = mci->pvt_info; 370 int i; 371 372 for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) { 373 int j; 374 const int numrank = priv->dimm_numrank[chan][i]; 375 376 for (j = 0; j < numrank; j++) 377 if (priv->dimm_csmap[i][j] == rank) 378 return i * 2 + chan; 379 } 380 381 return -1; 382 } 383 384 static const char *i5100_err_msg(unsigned err) 385 { 386 static const char *merrs[] = { 387 "unknown", /* 0 */ 388 "uncorrectable data ECC on replay", /* 1 */ 389 "unknown", /* 2 */ 390 "unknown", /* 3 */ 391 "aliased uncorrectable demand data ECC", /* 4 */ 392 "aliased uncorrectable spare-copy data ECC", /* 5 */ 393 "aliased uncorrectable patrol data ECC", /* 6 */ 394 "unknown", /* 7 */ 395 "unknown", /* 8 */ 396 "unknown", /* 9 */ 397 "non-aliased uncorrectable demand data ECC", /* 10 */ 398 "non-aliased uncorrectable spare-copy data ECC", /* 11 */ 399 "non-aliased uncorrectable patrol data ECC", /* 12 */ 400 "unknown", /* 13 */ 401 "correctable demand data ECC", /* 14 */ 402 "correctable spare-copy data ECC", /* 15 */ 403 "correctable patrol data ECC", /* 16 */ 404 "unknown", /* 17 */ 405 "SPD protocol error", /* 18 */ 406 "unknown", /* 19 */ 407 "spare copy initiated", /* 20 */ 408 "spare copy completed", /* 21 */ 409 }; 410 unsigned i; 411 412 for (i = 0; i < ARRAY_SIZE(merrs); i++) 413 if (1 << i & err) 414 return merrs[i]; 415 416 return "none"; 417 } 418 419 /* convert csrow index into a rank (per channel -- 0..5) */ 420 static unsigned int i5100_csrow_to_rank(const struct mem_ctl_info *mci, 421 unsigned int csrow) 422 { 423 const struct i5100_priv *priv = mci->pvt_info; 424 425 return csrow % priv->ranksperchan; 426 } 427 428 /* convert csrow index into a channel (0..1) */ 429 static unsigned int i5100_csrow_to_chan(const struct mem_ctl_info *mci, 430 unsigned int csrow) 431 { 432 const struct i5100_priv *priv = mci->pvt_info; 433 434 return csrow / priv->ranksperchan; 435 } 436 437 static void i5100_handle_ce(struct mem_ctl_info *mci, 438 int chan, 439 unsigned bank, 440 unsigned rank, 441 unsigned long syndrome, 442 unsigned cas, 443 unsigned ras, 444 const char *msg) 445 { 446 char detail[80]; 447 448 /* Form out message */ 449 snprintf(detail, sizeof(detail), 450 "bank %u, cas %u, ras %u\n", 451 bank, cas, ras); 452 453 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 454 0, 0, syndrome, 455 chan, rank, -1, 456 msg, detail); 457 } 458 459 static void i5100_handle_ue(struct mem_ctl_info *mci, 460 int chan, 461 unsigned bank, 462 unsigned rank, 463 unsigned long syndrome, 464 unsigned cas, 465 unsigned ras, 466 const char *msg) 467 { 468 char detail[80]; 469 470 /* Form out message */ 471 snprintf(detail, sizeof(detail), 472 "bank %u, cas %u, ras %u\n", 473 bank, cas, ras); 474 475 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 476 0, 0, syndrome, 477 chan, rank, -1, 478 msg, detail); 479 } 480 481 static void i5100_read_log(struct mem_ctl_info *mci, int chan, 482 u32 ferr, u32 nerr) 483 { 484 struct i5100_priv *priv = mci->pvt_info; 485 struct pci_dev *pdev = (chan) ? priv->ch1mm : priv->ch0mm; 486 u32 dw; 487 u32 dw2; 488 unsigned syndrome = 0; 489 unsigned ecc_loc = 0; 490 unsigned merr; 491 unsigned bank; 492 unsigned rank; 493 unsigned cas; 494 unsigned ras; 495 496 pci_read_config_dword(pdev, I5100_VALIDLOG, &dw); 497 498 if (i5100_validlog_redmemvalid(dw)) { 499 pci_read_config_dword(pdev, I5100_REDMEMA, &dw2); 500 syndrome = dw2; 501 pci_read_config_dword(pdev, I5100_REDMEMB, &dw2); 502 ecc_loc = i5100_redmemb_ecc_locator(dw2); 503 } 504 505 if (i5100_validlog_recmemvalid(dw)) { 506 const char *msg; 507 508 pci_read_config_dword(pdev, I5100_RECMEMA, &dw2); 509 merr = i5100_recmema_merr(dw2); 510 bank = i5100_recmema_bank(dw2); 511 rank = i5100_recmema_rank(dw2); 512 513 pci_read_config_dword(pdev, I5100_RECMEMB, &dw2); 514 cas = i5100_recmemb_cas(dw2); 515 ras = i5100_recmemb_ras(dw2); 516 517 /* FIXME: not really sure if this is what merr is... 518 */ 519 if (!merr) 520 msg = i5100_err_msg(ferr); 521 else 522 msg = i5100_err_msg(nerr); 523 524 i5100_handle_ce(mci, chan, bank, rank, syndrome, cas, ras, msg); 525 } 526 527 if (i5100_validlog_nrecmemvalid(dw)) { 528 const char *msg; 529 530 pci_read_config_dword(pdev, I5100_NRECMEMA, &dw2); 531 merr = i5100_nrecmema_merr(dw2); 532 bank = i5100_nrecmema_bank(dw2); 533 rank = i5100_nrecmema_rank(dw2); 534 535 pci_read_config_dword(pdev, I5100_NRECMEMB, &dw2); 536 cas = i5100_nrecmemb_cas(dw2); 537 ras = i5100_nrecmemb_ras(dw2); 538 539 /* FIXME: not really sure if this is what merr is... 540 */ 541 if (!merr) 542 msg = i5100_err_msg(ferr); 543 else 544 msg = i5100_err_msg(nerr); 545 546 i5100_handle_ue(mci, chan, bank, rank, syndrome, cas, ras, msg); 547 } 548 549 pci_write_config_dword(pdev, I5100_VALIDLOG, dw); 550 } 551 552 static void i5100_check_error(struct mem_ctl_info *mci) 553 { 554 struct i5100_priv *priv = mci->pvt_info; 555 u32 dw, dw2; 556 557 pci_read_config_dword(priv->mc, I5100_FERR_NF_MEM, &dw); 558 if (i5100_ferr_nf_mem_any(dw)) { 559 560 pci_read_config_dword(priv->mc, I5100_NERR_NF_MEM, &dw2); 561 562 i5100_read_log(mci, i5100_ferr_nf_mem_chan_indx(dw), 563 i5100_ferr_nf_mem_any(dw), 564 i5100_nerr_nf_mem_any(dw2)); 565 566 pci_write_config_dword(priv->mc, I5100_NERR_NF_MEM, dw2); 567 } 568 pci_write_config_dword(priv->mc, I5100_FERR_NF_MEM, dw); 569 } 570 571 /* The i5100 chipset will scrub the entire memory once, then 572 * set a done bit. Continuous scrubbing is achieved by enqueing 573 * delayed work to a workqueue, checking every few minutes if 574 * the scrubbing has completed and if so reinitiating it. 575 */ 576 577 static void i5100_refresh_scrubbing(struct work_struct *work) 578 { 579 struct delayed_work *i5100_scrubbing = to_delayed_work(work); 580 struct i5100_priv *priv = container_of(i5100_scrubbing, 581 struct i5100_priv, 582 i5100_scrubbing); 583 u32 dw; 584 585 pci_read_config_dword(priv->mc, I5100_MC, &dw); 586 587 if (priv->scrub_enable) { 588 589 pci_read_config_dword(priv->mc, I5100_MC, &dw); 590 591 if (i5100_mc_scrbdone(dw)) { 592 dw |= I5100_MC_SCRBEN_MASK; 593 pci_write_config_dword(priv->mc, I5100_MC, dw); 594 pci_read_config_dword(priv->mc, I5100_MC, &dw); 595 } 596 597 schedule_delayed_work(&(priv->i5100_scrubbing), 598 I5100_SCRUB_REFRESH_RATE); 599 } 600 } 601 /* 602 * The bandwidth is based on experimentation, feel free to refine it. 603 */ 604 static int i5100_set_scrub_rate(struct mem_ctl_info *mci, u32 bandwidth) 605 { 606 struct i5100_priv *priv = mci->pvt_info; 607 u32 dw; 608 609 pci_read_config_dword(priv->mc, I5100_MC, &dw); 610 if (bandwidth) { 611 priv->scrub_enable = 1; 612 dw |= I5100_MC_SCRBEN_MASK; 613 schedule_delayed_work(&(priv->i5100_scrubbing), 614 I5100_SCRUB_REFRESH_RATE); 615 } else { 616 priv->scrub_enable = 0; 617 dw &= ~I5100_MC_SCRBEN_MASK; 618 cancel_delayed_work(&(priv->i5100_scrubbing)); 619 } 620 pci_write_config_dword(priv->mc, I5100_MC, dw); 621 622 pci_read_config_dword(priv->mc, I5100_MC, &dw); 623 624 bandwidth = 5900000 * i5100_mc_scrben(dw); 625 626 return bandwidth; 627 } 628 629 static int i5100_get_scrub_rate(struct mem_ctl_info *mci) 630 { 631 struct i5100_priv *priv = mci->pvt_info; 632 u32 dw; 633 634 pci_read_config_dword(priv->mc, I5100_MC, &dw); 635 636 return 5900000 * i5100_mc_scrben(dw); 637 } 638 639 static struct pci_dev *pci_get_device_func(unsigned vendor, 640 unsigned device, 641 unsigned func) 642 { 643 struct pci_dev *ret = NULL; 644 645 while (1) { 646 ret = pci_get_device(vendor, device, ret); 647 648 if (!ret) 649 break; 650 651 if (PCI_FUNC(ret->devfn) == func) 652 break; 653 } 654 655 return ret; 656 } 657 658 static unsigned long i5100_npages(struct mem_ctl_info *mci, unsigned int csrow) 659 { 660 struct i5100_priv *priv = mci->pvt_info; 661 const unsigned int chan_rank = i5100_csrow_to_rank(mci, csrow); 662 const unsigned int chan = i5100_csrow_to_chan(mci, csrow); 663 unsigned addr_lines; 664 665 /* dimm present? */ 666 if (!priv->mtr[chan][chan_rank].present) 667 return 0ULL; 668 669 addr_lines = 670 I5100_DIMM_ADDR_LINES + 671 priv->mtr[chan][chan_rank].numcol + 672 priv->mtr[chan][chan_rank].numrow + 673 priv->mtr[chan][chan_rank].numbank; 674 675 return (unsigned long) 676 ((unsigned long long) (1ULL << addr_lines) / PAGE_SIZE); 677 } 678 679 static void i5100_init_mtr(struct mem_ctl_info *mci) 680 { 681 struct i5100_priv *priv = mci->pvt_info; 682 struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm }; 683 int i; 684 685 for (i = 0; i < I5100_CHANNELS; i++) { 686 int j; 687 struct pci_dev *pdev = mms[i]; 688 689 for (j = 0; j < I5100_MAX_RANKS_PER_CHAN; j++) { 690 const unsigned addr = 691 (j < 4) ? I5100_MTR_0 + j * 2 : 692 I5100_MTR_4 + (j - 4) * 2; 693 u16 w; 694 695 pci_read_config_word(pdev, addr, &w); 696 697 priv->mtr[i][j].present = i5100_mtr_present(w); 698 priv->mtr[i][j].ethrottle = i5100_mtr_ethrottle(w); 699 priv->mtr[i][j].width = 4 + 4 * i5100_mtr_width(w); 700 priv->mtr[i][j].numbank = 2 + i5100_mtr_numbank(w); 701 priv->mtr[i][j].numrow = 13 + i5100_mtr_numrow(w); 702 priv->mtr[i][j].numcol = 10 + i5100_mtr_numcol(w); 703 } 704 } 705 } 706 707 /* 708 * FIXME: make this into a real i2c adapter (so that dimm-decode 709 * will work)? 710 */ 711 static int i5100_read_spd_byte(const struct mem_ctl_info *mci, 712 u8 ch, u8 slot, u8 addr, u8 *byte) 713 { 714 struct i5100_priv *priv = mci->pvt_info; 715 u16 w; 716 unsigned long et; 717 718 pci_read_config_word(priv->mc, I5100_SPDDATA, &w); 719 if (i5100_spddata_busy(w)) 720 return -1; 721 722 pci_write_config_dword(priv->mc, I5100_SPDCMD, 723 i5100_spdcmd_create(0xa, 1, ch * 4 + slot, addr, 724 0, 0)); 725 726 /* wait up to 100ms */ 727 et = jiffies + HZ / 10; 728 udelay(100); 729 while (1) { 730 pci_read_config_word(priv->mc, I5100_SPDDATA, &w); 731 if (!i5100_spddata_busy(w)) 732 break; 733 udelay(100); 734 } 735 736 if (!i5100_spddata_rdo(w) || i5100_spddata_sbe(w)) 737 return -1; 738 739 *byte = i5100_spddata_data(w); 740 741 return 0; 742 } 743 744 /* 745 * fill dimm chip select map 746 * 747 * FIXME: 748 * o not the only way to may chip selects to dimm slots 749 * o investigate if there is some way to obtain this map from the bios 750 */ 751 static void i5100_init_dimm_csmap(struct mem_ctl_info *mci) 752 { 753 struct i5100_priv *priv = mci->pvt_info; 754 int i; 755 756 for (i = 0; i < I5100_MAX_DIMM_SLOTS_PER_CHAN; i++) { 757 int j; 758 759 for (j = 0; j < I5100_MAX_RANKS_PER_DIMM; j++) 760 priv->dimm_csmap[i][j] = -1; /* default NC */ 761 } 762 763 /* only 2 chip selects per slot... */ 764 if (priv->ranksperchan == 4) { 765 priv->dimm_csmap[0][0] = 0; 766 priv->dimm_csmap[0][1] = 3; 767 priv->dimm_csmap[1][0] = 1; 768 priv->dimm_csmap[1][1] = 2; 769 priv->dimm_csmap[2][0] = 2; 770 priv->dimm_csmap[3][0] = 3; 771 } else { 772 priv->dimm_csmap[0][0] = 0; 773 priv->dimm_csmap[0][1] = 1; 774 priv->dimm_csmap[1][0] = 2; 775 priv->dimm_csmap[1][1] = 3; 776 priv->dimm_csmap[2][0] = 4; 777 priv->dimm_csmap[2][1] = 5; 778 } 779 } 780 781 static void i5100_init_dimm_layout(struct pci_dev *pdev, 782 struct mem_ctl_info *mci) 783 { 784 struct i5100_priv *priv = mci->pvt_info; 785 int i; 786 787 for (i = 0; i < I5100_CHANNELS; i++) { 788 int j; 789 790 for (j = 0; j < I5100_MAX_DIMM_SLOTS_PER_CHAN; j++) { 791 u8 rank; 792 793 if (i5100_read_spd_byte(mci, i, j, 5, &rank) < 0) 794 priv->dimm_numrank[i][j] = 0; 795 else 796 priv->dimm_numrank[i][j] = (rank & 3) + 1; 797 } 798 } 799 800 i5100_init_dimm_csmap(mci); 801 } 802 803 static void i5100_init_interleaving(struct pci_dev *pdev, 804 struct mem_ctl_info *mci) 805 { 806 u16 w; 807 u32 dw; 808 struct i5100_priv *priv = mci->pvt_info; 809 struct pci_dev *mms[2] = { priv->ch0mm, priv->ch1mm }; 810 int i; 811 812 pci_read_config_word(pdev, I5100_TOLM, &w); 813 priv->tolm = (u64) i5100_tolm_tolm(w) * 256 * 1024 * 1024; 814 815 pci_read_config_word(pdev, I5100_MIR0, &w); 816 priv->mir[0].limit = (u64) i5100_mir_limit(w) << 28; 817 priv->mir[0].way[1] = i5100_mir_way1(w); 818 priv->mir[0].way[0] = i5100_mir_way0(w); 819 820 pci_read_config_word(pdev, I5100_MIR1, &w); 821 priv->mir[1].limit = (u64) i5100_mir_limit(w) << 28; 822 priv->mir[1].way[1] = i5100_mir_way1(w); 823 priv->mir[1].way[0] = i5100_mir_way0(w); 824 825 pci_read_config_word(pdev, I5100_AMIR_0, &w); 826 priv->amir[0] = w; 827 pci_read_config_word(pdev, I5100_AMIR_1, &w); 828 priv->amir[1] = w; 829 830 for (i = 0; i < I5100_CHANNELS; i++) { 831 int j; 832 833 for (j = 0; j < 5; j++) { 834 int k; 835 836 pci_read_config_dword(mms[i], I5100_DMIR + j * 4, &dw); 837 838 priv->dmir[i][j].limit = 839 (u64) i5100_dmir_limit(dw) << 28; 840 for (k = 0; k < I5100_MAX_RANKS_PER_DIMM; k++) 841 priv->dmir[i][j].rank[k] = 842 i5100_dmir_rank(dw, k); 843 } 844 } 845 846 i5100_init_mtr(mci); 847 } 848 849 static void i5100_init_csrows(struct mem_ctl_info *mci) 850 { 851 int i; 852 struct i5100_priv *priv = mci->pvt_info; 853 854 for (i = 0; i < mci->tot_dimms; i++) { 855 struct dimm_info *dimm; 856 const unsigned long npages = i5100_npages(mci, i); 857 const unsigned int chan = i5100_csrow_to_chan(mci, i); 858 const unsigned int rank = i5100_csrow_to_rank(mci, i); 859 860 if (!npages) 861 continue; 862 863 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, 864 chan, rank, 0); 865 866 dimm->nr_pages = npages; 867 dimm->grain = 32; 868 dimm->dtype = (priv->mtr[chan][rank].width == 4) ? 869 DEV_X4 : DEV_X8; 870 dimm->mtype = MEM_RDDR2; 871 dimm->edac_mode = EDAC_SECDED; 872 snprintf(dimm->label, sizeof(dimm->label), "DIMM%u", 873 i5100_rank_to_slot(mci, chan, rank)); 874 875 edac_dbg(2, "dimm channel %d, rank %d, size %ld\n", 876 chan, rank, (long)PAGES_TO_MiB(npages)); 877 } 878 } 879 880 /**************************************************************************** 881 * Error injection routines 882 ****************************************************************************/ 883 884 static void i5100_do_inject(struct mem_ctl_info *mci) 885 { 886 struct i5100_priv *priv = mci->pvt_info; 887 u32 mask0; 888 u16 mask1; 889 890 /* MEM[1:0]EINJMSK0 891 * 31 - ADDRMATCHEN 892 * 29:28 - HLINESEL 893 * 00 Reserved 894 * 01 Lower half of cache line 895 * 10 Upper half of cache line 896 * 11 Both upper and lower parts of cache line 897 * 27 - EINJEN 898 * 25:19 - XORMASK1 for deviceptr1 899 * 9:5 - SEC2RAM for deviceptr2 900 * 4:0 - FIR2RAM for deviceptr1 901 */ 902 mask0 = ((priv->inject_hlinesel & 0x3) << 28) | 903 I5100_MEMXEINJMSK0_EINJEN | 904 ((priv->inject_eccmask1 & 0xffff) << 10) | 905 ((priv->inject_deviceptr2 & 0x1f) << 5) | 906 (priv->inject_deviceptr1 & 0x1f); 907 908 /* MEM[1:0]EINJMSK1 909 * 15:0 - XORMASK2 for deviceptr2 910 */ 911 mask1 = priv->inject_eccmask2; 912 913 if (priv->inject_channel == 0) { 914 pci_write_config_dword(priv->mc, I5100_MEM0EINJMSK0, mask0); 915 pci_write_config_word(priv->mc, I5100_MEM0EINJMSK1, mask1); 916 } else { 917 pci_write_config_dword(priv->mc, I5100_MEM1EINJMSK0, mask0); 918 pci_write_config_word(priv->mc, I5100_MEM1EINJMSK1, mask1); 919 } 920 921 /* Error Injection Response Function 922 * Intel 5100 Memory Controller Hub Chipset (318378) datasheet 923 * hints about this register but carry no data about them. All 924 * data regarding device 19 is based on experimentation and the 925 * Intel 7300 Chipset Memory Controller Hub (318082) datasheet 926 * which appears to be accurate for the i5100 in this area. 927 * 928 * The injection code don't work without setting this register. 929 * The register needs to be flipped off then on else the hardware 930 * will only preform the first injection. 931 * 932 * Stop condition bits 7:4 933 * 1010 - Stop after one injection 934 * 1011 - Never stop injecting faults 935 * 936 * Start condition bits 3:0 937 * 1010 - Never start 938 * 1011 - Start immediately 939 */ 940 pci_write_config_byte(priv->einj, I5100_DINJ0, 0xaa); 941 pci_write_config_byte(priv->einj, I5100_DINJ0, 0xab); 942 } 943 944 #define to_mci(k) container_of(k, struct mem_ctl_info, dev) 945 static ssize_t inject_enable_write(struct file *file, const char __user *data, 946 size_t count, loff_t *ppos) 947 { 948 struct device *dev = file->private_data; 949 struct mem_ctl_info *mci = to_mci(dev); 950 951 i5100_do_inject(mci); 952 953 return count; 954 } 955 956 static const struct file_operations i5100_inject_enable_fops = { 957 .open = simple_open, 958 .write = inject_enable_write, 959 .llseek = generic_file_llseek, 960 }; 961 962 static int i5100_setup_debugfs(struct mem_ctl_info *mci) 963 { 964 struct i5100_priv *priv = mci->pvt_info; 965 966 if (!i5100_debugfs) 967 return -ENODEV; 968 969 priv->debugfs = edac_debugfs_create_dir_at(mci->bus->name, i5100_debugfs); 970 971 if (!priv->debugfs) 972 return -ENOMEM; 973 974 edac_debugfs_create_x8("inject_channel", S_IRUGO | S_IWUSR, priv->debugfs, 975 &priv->inject_channel); 976 edac_debugfs_create_x8("inject_hlinesel", S_IRUGO | S_IWUSR, priv->debugfs, 977 &priv->inject_hlinesel); 978 edac_debugfs_create_x8("inject_deviceptr1", S_IRUGO | S_IWUSR, priv->debugfs, 979 &priv->inject_deviceptr1); 980 edac_debugfs_create_x8("inject_deviceptr2", S_IRUGO | S_IWUSR, priv->debugfs, 981 &priv->inject_deviceptr2); 982 edac_debugfs_create_x16("inject_eccmask1", S_IRUGO | S_IWUSR, priv->debugfs, 983 &priv->inject_eccmask1); 984 edac_debugfs_create_x16("inject_eccmask2", S_IRUGO | S_IWUSR, priv->debugfs, 985 &priv->inject_eccmask2); 986 edac_debugfs_create_file("inject_enable", S_IWUSR, priv->debugfs, 987 &mci->dev, &i5100_inject_enable_fops); 988 989 return 0; 990 991 } 992 993 static int i5100_init_one(struct pci_dev *pdev, const struct pci_device_id *id) 994 { 995 int rc; 996 struct mem_ctl_info *mci; 997 struct edac_mc_layer layers[2]; 998 struct i5100_priv *priv; 999 struct pci_dev *ch0mm, *ch1mm, *einj; 1000 int ret = 0; 1001 u32 dw; 1002 int ranksperch; 1003 1004 if (PCI_FUNC(pdev->devfn) != 1) 1005 return -ENODEV; 1006 1007 rc = pci_enable_device(pdev); 1008 if (rc < 0) { 1009 ret = rc; 1010 goto bail; 1011 } 1012 1013 /* ECC enabled? */ 1014 pci_read_config_dword(pdev, I5100_MC, &dw); 1015 if (!i5100_mc_errdeten(dw)) { 1016 printk(KERN_INFO "i5100_edac: ECC not enabled.\n"); 1017 ret = -ENODEV; 1018 goto bail_pdev; 1019 } 1020 1021 /* figure out how many ranks, from strapped state of 48GB_Mode input */ 1022 pci_read_config_dword(pdev, I5100_MS, &dw); 1023 ranksperch = !!(dw & (1 << 8)) * 2 + 4; 1024 1025 /* enable error reporting... */ 1026 pci_read_config_dword(pdev, I5100_EMASK_MEM, &dw); 1027 dw &= ~I5100_FERR_NF_MEM_ANY_MASK; 1028 pci_write_config_dword(pdev, I5100_EMASK_MEM, dw); 1029 1030 /* device 21, func 0, Channel 0 Memory Map, Error Flag/Mask, etc... */ 1031 ch0mm = pci_get_device_func(PCI_VENDOR_ID_INTEL, 1032 PCI_DEVICE_ID_INTEL_5100_21, 0); 1033 if (!ch0mm) { 1034 ret = -ENODEV; 1035 goto bail_pdev; 1036 } 1037 1038 rc = pci_enable_device(ch0mm); 1039 if (rc < 0) { 1040 ret = rc; 1041 goto bail_ch0; 1042 } 1043 1044 /* device 22, func 0, Channel 1 Memory Map, Error Flag/Mask, etc... */ 1045 ch1mm = pci_get_device_func(PCI_VENDOR_ID_INTEL, 1046 PCI_DEVICE_ID_INTEL_5100_22, 0); 1047 if (!ch1mm) { 1048 ret = -ENODEV; 1049 goto bail_disable_ch0; 1050 } 1051 1052 rc = pci_enable_device(ch1mm); 1053 if (rc < 0) { 1054 ret = rc; 1055 goto bail_ch1; 1056 } 1057 1058 layers[0].type = EDAC_MC_LAYER_CHANNEL; 1059 layers[0].size = 2; 1060 layers[0].is_virt_csrow = false; 1061 layers[1].type = EDAC_MC_LAYER_SLOT; 1062 layers[1].size = ranksperch; 1063 layers[1].is_virt_csrow = true; 1064 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 1065 sizeof(*priv)); 1066 if (!mci) { 1067 ret = -ENOMEM; 1068 goto bail_disable_ch1; 1069 } 1070 1071 1072 /* device 19, func 0, Error injection */ 1073 einj = pci_get_device_func(PCI_VENDOR_ID_INTEL, 1074 PCI_DEVICE_ID_INTEL_5100_19, 0); 1075 if (!einj) { 1076 ret = -ENODEV; 1077 goto bail_einj; 1078 } 1079 1080 rc = pci_enable_device(einj); 1081 if (rc < 0) { 1082 ret = rc; 1083 goto bail_disable_einj; 1084 } 1085 1086 1087 mci->pdev = &pdev->dev; 1088 1089 priv = mci->pvt_info; 1090 priv->ranksperchan = ranksperch; 1091 priv->mc = pdev; 1092 priv->ch0mm = ch0mm; 1093 priv->ch1mm = ch1mm; 1094 priv->einj = einj; 1095 1096 INIT_DELAYED_WORK(&(priv->i5100_scrubbing), i5100_refresh_scrubbing); 1097 1098 /* If scrubbing was already enabled by the bios, start maintaining it */ 1099 pci_read_config_dword(pdev, I5100_MC, &dw); 1100 if (i5100_mc_scrben(dw)) { 1101 priv->scrub_enable = 1; 1102 schedule_delayed_work(&(priv->i5100_scrubbing), 1103 I5100_SCRUB_REFRESH_RATE); 1104 } 1105 1106 i5100_init_dimm_layout(pdev, mci); 1107 i5100_init_interleaving(pdev, mci); 1108 1109 mci->mtype_cap = MEM_FLAG_FB_DDR2; 1110 mci->edac_ctl_cap = EDAC_FLAG_SECDED; 1111 mci->edac_cap = EDAC_FLAG_SECDED; 1112 mci->mod_name = "i5100_edac.c"; 1113 mci->ctl_name = "i5100"; 1114 mci->dev_name = pci_name(pdev); 1115 mci->ctl_page_to_phys = NULL; 1116 1117 mci->edac_check = i5100_check_error; 1118 mci->set_sdram_scrub_rate = i5100_set_scrub_rate; 1119 mci->get_sdram_scrub_rate = i5100_get_scrub_rate; 1120 1121 priv->inject_channel = 0; 1122 priv->inject_hlinesel = 0; 1123 priv->inject_deviceptr1 = 0; 1124 priv->inject_deviceptr2 = 0; 1125 priv->inject_eccmask1 = 0; 1126 priv->inject_eccmask2 = 0; 1127 1128 i5100_init_csrows(mci); 1129 1130 /* this strange construction seems to be in every driver, dunno why */ 1131 switch (edac_op_state) { 1132 case EDAC_OPSTATE_POLL: 1133 case EDAC_OPSTATE_NMI: 1134 break; 1135 default: 1136 edac_op_state = EDAC_OPSTATE_POLL; 1137 break; 1138 } 1139 1140 if (edac_mc_add_mc(mci)) { 1141 ret = -ENODEV; 1142 goto bail_scrub; 1143 } 1144 1145 i5100_setup_debugfs(mci); 1146 1147 return ret; 1148 1149 bail_scrub: 1150 priv->scrub_enable = 0; 1151 cancel_delayed_work_sync(&(priv->i5100_scrubbing)); 1152 edac_mc_free(mci); 1153 1154 bail_disable_einj: 1155 pci_disable_device(einj); 1156 1157 bail_einj: 1158 pci_dev_put(einj); 1159 1160 bail_disable_ch1: 1161 pci_disable_device(ch1mm); 1162 1163 bail_ch1: 1164 pci_dev_put(ch1mm); 1165 1166 bail_disable_ch0: 1167 pci_disable_device(ch0mm); 1168 1169 bail_ch0: 1170 pci_dev_put(ch0mm); 1171 1172 bail_pdev: 1173 pci_disable_device(pdev); 1174 1175 bail: 1176 return ret; 1177 } 1178 1179 static void i5100_remove_one(struct pci_dev *pdev) 1180 { 1181 struct mem_ctl_info *mci; 1182 struct i5100_priv *priv; 1183 1184 mci = edac_mc_del_mc(&pdev->dev); 1185 1186 if (!mci) 1187 return; 1188 1189 priv = mci->pvt_info; 1190 1191 edac_debugfs_remove_recursive(priv->debugfs); 1192 1193 priv->scrub_enable = 0; 1194 cancel_delayed_work_sync(&(priv->i5100_scrubbing)); 1195 1196 pci_disable_device(pdev); 1197 pci_disable_device(priv->ch0mm); 1198 pci_disable_device(priv->ch1mm); 1199 pci_disable_device(priv->einj); 1200 pci_dev_put(priv->ch0mm); 1201 pci_dev_put(priv->ch1mm); 1202 pci_dev_put(priv->einj); 1203 1204 edac_mc_free(mci); 1205 } 1206 1207 static const struct pci_device_id i5100_pci_tbl[] = { 1208 /* Device 16, Function 0, Channel 0 Memory Map, Error Flag/Mask, ... */ 1209 { PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_5100_16) }, 1210 { 0, } 1211 }; 1212 MODULE_DEVICE_TABLE(pci, i5100_pci_tbl); 1213 1214 static struct pci_driver i5100_driver = { 1215 .name = KBUILD_BASENAME, 1216 .probe = i5100_init_one, 1217 .remove = i5100_remove_one, 1218 .id_table = i5100_pci_tbl, 1219 }; 1220 1221 static int __init i5100_init(void) 1222 { 1223 int pci_rc; 1224 1225 i5100_debugfs = edac_debugfs_create_dir_at("i5100_edac", NULL); 1226 1227 pci_rc = pci_register_driver(&i5100_driver); 1228 return (pci_rc < 0) ? pci_rc : 0; 1229 } 1230 1231 static void __exit i5100_exit(void) 1232 { 1233 edac_debugfs_remove(i5100_debugfs); 1234 1235 pci_unregister_driver(&i5100_driver); 1236 } 1237 1238 module_init(i5100_init); 1239 module_exit(i5100_exit); 1240 1241 MODULE_LICENSE("GPL"); 1242 MODULE_AUTHOR 1243 ("Arthur Jones <ajones@riverbed.com>"); 1244 MODULE_DESCRIPTION("MC Driver for Intel I5100 memory controllers"); 1245