drivers/edac/x38_edac.c

   1 /*
   2  * Intel X38 Memory Controller kernel module
   3  * Copyright (C) 2008 Cluster Computing, Inc.
   4  *
   5  * This file may be distributed under the terms of the
   6  * GNU General Public License.
   7  *
   8  * This file is based on i3200_edac.c
   9  *
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/pci.h>
  15 #include <linux/pci_ids.h>
  16 #include <linux/edac.h>
  17
  18 #include <linux/io-64-nonatomic-lo-hi.h>
  19 #include "edac_core.h"
  20
  21 #define X38_REVISION            "1.1"
  22
  23 #define EDAC_MOD_STR            "x38_edac"
  24
  25 #define PCI_DEVICE_ID_INTEL_X38_HB      0x29e0
  26
  27 #define X38_RANKS               8
  28 #define X38_RANKS_PER_CHANNEL   4
  29 #define X38_CHANNELS            2
  30
  31 /* Intel X38 register addresses - device 0 function 0 - DRAM Controller */
  32
  33 #define X38_MCHBAR_LOW  0x48    /* MCH Memory Mapped Register BAR */
  34 #define X38_MCHBAR_HIGH 0x4c
  35 #define X38_MCHBAR_MASK 0xfffffc000ULL  /* bits 35:14 */
  36 #define X38_MMR_WINDOW_SIZE     16384
  37
  38 #define X38_TOM 0xa0    /* Top of Memory (16b)
  39                                  *
  40                                  * 15:10 reserved
  41                                  *  9:0  total populated physical memory
  42                                  */
  43 #define X38_TOM_MASK    0x3ff   /* bits 9:0 */
  44 #define X38_TOM_SHIFT 26        /* 64MiB grain */
  45
  46 #define X38_ERRSTS      0xc8    /* Error Status Register (16b)
  47                                  *
  48                                  * 15    reserved
  49                                  * 14    Isochronous TBWRR Run Behind FIFO Full
  50                                  *       (ITCV)
  51                                  * 13    Isochronous TBWRR Run Behind FIFO Put
  52                                  *       (ITSTV)
  53                                  * 12    reserved
  54                                  * 11    MCH Thermal Sensor Event
  55                                  *       for SMI/SCI/SERR (GTSE)
  56                                  * 10    reserved
  57                                  *  9    LOCK to non-DRAM Memory Flag (LCKF)
  58                                  *  8    reserved
  59                                  *  7    DRAM Throttle Flag (DTF)
  60                                  *  6:2  reserved
  61                                  *  1    Multi-bit DRAM ECC Error Flag (DMERR)
  62                                  *  0    Single-bit DRAM ECC Error Flag (DSERR)
  63                                  */
  64 #define X38_ERRSTS_UE           0x0002
  65 #define X38_ERRSTS_CE           0x0001
  66 #define X38_ERRSTS_BITS (X38_ERRSTS_UE | X38_ERRSTS_CE)
  67
  68
  69 /* Intel  MMIO register space - device 0 function 0 - MMR space */
  70
  71 #define X38_C0DRB       0x200   /* Channel 0 DRAM Rank Boundary (16b x 4)
  72                                  *
  73                                  * 15:10 reserved
  74                                  *  9:0  Channel 0 DRAM Rank Boundary Address
  75                                  */
  76 #define X38_C1DRB       0x600   /* Channel 1 DRAM Rank Boundary (16b x 4) */
  77 #define X38_DRB_MASK    0x3ff   /* bits 9:0 */
  78 #define X38_DRB_SHIFT 26        /* 64MiB grain */
  79
  80 #define X38_C0ECCERRLOG 0x280   /* Channel 0 ECC Error Log (64b)
  81                                  *
  82                                  * 63:48 Error Column Address (ERRCOL)
  83                                  * 47:32 Error Row Address (ERRROW)
  84                                  * 31:29 Error Bank Address (ERRBANK)
  85                                  * 28:27 Error Rank Address (ERRRANK)
  86                                  * 26:24 reserved
  87                                  * 23:16 Error Syndrome (ERRSYND)
  88                                  * 15: 2 reserved
  89                                  *    1  Multiple Bit Error Status (MERRSTS)
  90                                  *    0  Correctable Error Status (CERRSTS)
  91                                  */
  92 #define X38_C1ECCERRLOG 0x680   /* Channel 1 ECC Error Log (64b) */
  93 #define X38_ECCERRLOG_CE        0x1
  94 #define X38_ECCERRLOG_UE        0x2
  95 #define X38_ECCERRLOG_RANK_BITS 0x18000000
  96 #define X38_ECCERRLOG_SYNDROME_BITS     0xff0000
  97
  98 #define X38_CAPID0 0xe0 /* see P.94 of spec for details */
  99
 100 static int x38_channel_num;
 101
 102 static int how_many_channel(struct pci_dev *pdev)
 103 {
 104         unsigned char capid0_8b; /* 8th byte of CAPID0 */
 105
 106         pci_read_config_byte(pdev, X38_CAPID0 + 8, &capid0_8b);
 107         if (capid0_8b & 0x20) { /* check DCD: Dual Channel Disable */
 108                 edac_dbg(0, "In single channel mode\n");
 109                 x38_channel_num = 1;
 110         } else {
 111                 edac_dbg(0, "In dual channel mode\n");
 112                 x38_channel_num = 2;
 113         }
 114
 115         return x38_channel_num;
 116 }
 117
 118 static unsigned long eccerrlog_syndrome(u64 log)
 119 {
 120         return (log & X38_ECCERRLOG_SYNDROME_BITS) >> 16;
 121 }
 122
 123 static int eccerrlog_row(int channel, u64 log)
 124 {
 125         return ((log & X38_ECCERRLOG_RANK_BITS) >> 27) |
 126                 (channel * X38_RANKS_PER_CHANNEL);
 127 }
 128
 129 enum x38_chips {
 130         X38 = 0,
 131 };
 132
 133 struct x38_dev_info {
 134         const char *ctl_name;
 135 };
 136
 137 struct x38_error_info {
 138         u16 errsts;
 139         u16 errsts2;
 140         u64 eccerrlog[X38_CHANNELS];
 141 };
 142
 143 static const struct x38_dev_info x38_devs[] = {
 144         [X38] = {
 145                 .ctl_name = "x38"},
 146 };
 147
 148 static struct pci_dev *mci_pdev;
 149 static int x38_registered = 1;
 150
 151
 152 static void x38_clear_error_info(struct mem_ctl_info *mci)
 153 {
 154         struct pci_dev *pdev;
 155
 156         pdev = to_pci_dev(mci->pdev);
 157
 158         /*
 159          * Clear any error bits.
 160          * (Yes, we really clear bits by writing 1 to them.)
 161          */
 162         pci_write_bits16(pdev, X38_ERRSTS, X38_ERRSTS_BITS,
 163                          X38_ERRSTS_BITS);
 164 }
 165
 166 static void x38_get_and_clear_error_info(struct mem_ctl_info *mci,
 167                                  struct x38_error_info *info)
 168 {
 169         struct pci_dev *pdev;
 170         void __iomem *window = mci->pvt_info;
 171
 172         pdev = to_pci_dev(mci->pdev);
 173
 174         /*
 175          * This is a mess because there is no atomic way to read all the
 176          * registers at once and the registers can transition from CE being
 177          * overwritten by UE.
 178          */
 179         pci_read_config_word(pdev, X38_ERRSTS, &info->errsts);
 180         if (!(info->errsts & X38_ERRSTS_BITS))
 181                 return;
 182
 183         info->eccerrlog[0] = lo_hi_readq(window + X38_C0ECCERRLOG);
 184         if (x38_channel_num == 2)
 185                 info->eccerrlog[1] = lo_hi_readq(window + X38_C1ECCERRLOG);
 186
 187         pci_read_config_word(pdev, X38_ERRSTS, &info->errsts2);
 188
 189         /*
 190          * If the error is the same for both reads then the first set
 191          * of reads is valid.  If there is a change then there is a CE
 192          * with no info and the second set of reads is valid and
 193          * should be UE info.
 194          */
 195         if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) {
 196                 info->eccerrlog[0] = lo_hi_readq(window + X38_C0ECCERRLOG);
 197                 if (x38_channel_num == 2)
 198                         info->eccerrlog[1] =
 199                                 lo_hi_readq(window + X38_C1ECCERRLOG);
 200         }
 201
 202         x38_clear_error_info(mci);
 203 }
 204
 205 static void x38_process_error_info(struct mem_ctl_info *mci,
 206                                 struct x38_error_info *info)
 207 {
 208         int channel;
 209         u64 log;
 210
 211         if (!(info->errsts & X38_ERRSTS_BITS))
 212                 return;
 213
 214         if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) {
 215                 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
 216                                      -1, -1, -1,
 217                                      "UE overwrote CE", "");
 218                 info->errsts = info->errsts2;
 219         }
 220
 221         for (channel = 0; channel < x38_channel_num; channel++) {
 222                 log = info->eccerrlog[channel];
 223                 if (log & X38_ECCERRLOG_UE) {
 224                         edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 225                                              0, 0, 0,
 226                                              eccerrlog_row(channel, log),
 227                                              -1, -1,
 228                                              "x38 UE", "");
 229                 } else if (log & X38_ECCERRLOG_CE) {
 230                         edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
 231                                              0, 0, eccerrlog_syndrome(log),
 232                                              eccerrlog_row(channel, log),
 233                                              -1, -1,
 234                                              "x38 CE", "");
 235                 }
 236         }
 237 }
 238
 239 static void x38_check(struct mem_ctl_info *mci)
 240 {
 241         struct x38_error_info info;
 242
 243         edac_dbg(1, "MC%d\n", mci->mc_idx);
 244         x38_get_and_clear_error_info(mci, &info);
 245         x38_process_error_info(mci, &info);
 246 }
 247
 248 static void __iomem *x38_map_mchbar(struct pci_dev *pdev)
 249 {
 250         union {
 251                 u64 mchbar;
 252                 struct {
 253                         u32 mchbar_low;
 254                         u32 mchbar_high;
 255                 };
 256         } u;
 257         void __iomem *window;
 258
 259         pci_read_config_dword(pdev, X38_MCHBAR_LOW, &u.mchbar_low);
 260         pci_write_config_dword(pdev, X38_MCHBAR_LOW, u.mchbar_low | 0x1);
 261         pci_read_config_dword(pdev, X38_MCHBAR_HIGH, &u.mchbar_high);
 262         u.mchbar &= X38_MCHBAR_MASK;
 263
 264         if (u.mchbar != (resource_size_t)u.mchbar) {
 265                 printk(KERN_ERR
 266                         "x38: mmio space beyond accessible range (0x%llx)\n",
 267                         (unsigned long long)u.mchbar);
 268                 return NULL;
 269         }
 270
 271         window = ioremap_nocache(u.mchbar, X38_MMR_WINDOW_SIZE);
 272         if (!window)
 273                 printk(KERN_ERR "x38: cannot map mmio space at 0x%llx\n",
 274                         (unsigned long long)u.mchbar);
 275
 276         return window;
 277 }
 278
 279
 280 static void x38_get_drbs(void __iomem *window,
 281                         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])
 282 {
 283         int i;
 284
 285         for (i = 0; i < X38_RANKS_PER_CHANNEL; i++) {
 286                 drbs[0][i] = readw(window + X38_C0DRB + 2*i) & X38_DRB_MASK;
 287                 drbs[1][i] = readw(window + X38_C1DRB + 2*i) & X38_DRB_MASK;
 288         }
 289 }
 290
 291 static bool x38_is_stacked(struct pci_dev *pdev,
 292                         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])
 293 {
 294         u16 tom;
 295
 296         pci_read_config_word(pdev, X38_TOM, &tom);
 297         tom &= X38_TOM_MASK;
 298
 299         return drbs[X38_CHANNELS - 1][X38_RANKS_PER_CHANNEL - 1] == tom;
 300 }
 301
 302 static unsigned long drb_to_nr_pages(
 303                         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL],
 304                         bool stacked, int channel, int rank)
 305 {
 306         int n;
 307
 308         n = drbs[channel][rank];
 309         if (rank > 0)
 310                 n -= drbs[channel][rank - 1];
 311         if (stacked && (channel == 1) && drbs[channel][rank] ==
 312                                 drbs[channel][X38_RANKS_PER_CHANNEL - 1]) {
 313                 n -= drbs[0][X38_RANKS_PER_CHANNEL - 1];
 314         }
 315
 316         n <<= (X38_DRB_SHIFT - PAGE_SHIFT);
 317         return n;
 318 }
 319
 320 static int x38_probe1(struct pci_dev *pdev, int dev_idx)
 321 {
 322         int rc;
 323         int i, j;
 324         struct mem_ctl_info *mci = NULL;
 325         struct edac_mc_layer layers[2];
 326         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL];
 327         bool stacked;
 328         void __iomem *window;
 329
 330         edac_dbg(0, "MC:\n");
 331
 332         window = x38_map_mchbar(pdev);
 333         if (!window)
 334                 return -ENODEV;
 335
 336         x38_get_drbs(window, drbs);
 337
 338         how_many_channel(pdev);
 339
 340         /* FIXME: unconventional pvt_info usage */
 341         layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 342         layers[0].size = X38_RANKS;
 343         layers[0].is_virt_csrow = true;
 344         layers[1].type = EDAC_MC_LAYER_CHANNEL;
 345         layers[1].size = x38_channel_num;
 346         layers[1].is_virt_csrow = false;
 347         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0);
 348         if (!mci)
 349                 return -ENOMEM;
 350
 351         edac_dbg(3, "MC: init mci\n");
 352
 353         mci->pdev = &pdev->dev;
 354         mci->mtype_cap = MEM_FLAG_DDR2;
 355
 356         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 357         mci->edac_cap = EDAC_FLAG_SECDED;
 358
 359         mci->mod_name = EDAC_MOD_STR;
 360         mci->mod_ver = X38_REVISION;
 361         mci->ctl_name = x38_devs[dev_idx].ctl_name;
 362         mci->dev_name = pci_name(pdev);
 363         mci->edac_check = x38_check;
 364         mci->ctl_page_to_phys = NULL;
 365         mci->pvt_info = window;
 366
 367         stacked = x38_is_stacked(pdev, drbs);
 368
 369         /*
 370          * The dram rank boundary (DRB) reg values are boundary addresses
 371          * for each DRAM rank with a granularity of 64MB.  DRB regs are
 372          * cumulative; the last one will contain the total memory
 373          * contained in all ranks.
 374          */
 375         for (i = 0; i < mci->nr_csrows; i++) {
 376                 unsigned long nr_pages;
 377                 struct csrow_info *csrow = mci->csrows[i];
 378
 379                 nr_pages = drb_to_nr_pages(drbs, stacked,
 380                         i / X38_RANKS_PER_CHANNEL,
 381                         i % X38_RANKS_PER_CHANNEL);
 382
 383                 if (nr_pages == 0)
 384                         continue;
 385
 386                 for (j = 0; j < x38_channel_num; j++) {
 387                         struct dimm_info *dimm = csrow->channels[j]->dimm;
 388
 389                         dimm->nr_pages = nr_pages / x38_channel_num;
 390                         dimm->grain = nr_pages << PAGE_SHIFT;
 391                         dimm->mtype = MEM_DDR2;
 392                         dimm->dtype = DEV_UNKNOWN;
 393                         dimm->edac_mode = EDAC_UNKNOWN;
 394                 }
 395         }
 396
 397         x38_clear_error_info(mci);
 398
 399         rc = -ENODEV;
 400         if (edac_mc_add_mc(mci)) {
 401                 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
 402                 goto fail;
 403         }
 404
 405         /* get this far and it's successful */
 406         edac_dbg(3, "MC: success\n");
 407         return 0;
 408
 409 fail:
 410         iounmap(window);
 411         if (mci)
 412                 edac_mc_free(mci);
 413
 414         return rc;
 415 }
 416
 417 static int x38_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 418 {
 419         int rc;
 420
 421         edac_dbg(0, "MC:\n");
 422
 423         if (pci_enable_device(pdev) < 0)
 424                 return -EIO;
 425
 426         rc = x38_probe1(pdev, ent->driver_data);
 427         if (!mci_pdev)
 428                 mci_pdev = pci_dev_get(pdev);
 429
 430         return rc;
 431 }
 432
 433 static void x38_remove_one(struct pci_dev *pdev)
 434 {
 435         struct mem_ctl_info *mci;
 436
 437         edac_dbg(0, "\n");
 438
 439         mci = edac_mc_del_mc(&pdev->dev);
 440         if (!mci)
 441                 return;
 442
 443         iounmap(mci->pvt_info);
 444
 445         edac_mc_free(mci);
 446 }
 447
 448 static const struct pci_device_id x38_pci_tbl[] = {
 449         {
 450          PCI_VEND_DEV(INTEL, X38_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 451          X38},
 452         {
 453          0,
 454          }                      /* 0 terminated list. */
 455 };
 456
 457 MODULE_DEVICE_TABLE(pci, x38_pci_tbl);
 458
 459 static struct pci_driver x38_driver = {
 460         .name = EDAC_MOD_STR,
 461         .probe = x38_init_one,
 462         .remove = x38_remove_one,
 463         .id_table = x38_pci_tbl,
 464 };
 465
 466 static int __init x38_init(void)
 467 {
 468         int pci_rc;
 469
 470         edac_dbg(3, "MC:\n");
 471
 472         /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 473         opstate_init();
 474
 475         pci_rc = pci_register_driver(&x38_driver);
 476         if (pci_rc < 0)
 477                 goto fail0;
 478
 479         if (!mci_pdev) {
 480                 x38_registered = 0;
 481                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 482                                         PCI_DEVICE_ID_INTEL_X38_HB, NULL);
 483                 if (!mci_pdev) {
 484                         edac_dbg(0, "x38 pci_get_device fail\n");
 485                         pci_rc = -ENODEV;
 486                         goto fail1;
 487                 }
 488
 489                 pci_rc = x38_init_one(mci_pdev, x38_pci_tbl);
 490                 if (pci_rc < 0) {
 491                         edac_dbg(0, "x38 init fail\n");
 492                         pci_rc = -ENODEV;
 493                         goto fail1;
 494                 }
 495         }
 496
 497         return 0;
 498
 499 fail1:
 500         pci_unregister_driver(&x38_driver);
 501
 502 fail0:
 503         pci_dev_put(mci_pdev);
 504
 505         return pci_rc;
 506 }
 507
 508 static void __exit x38_exit(void)
 509 {
 510         edac_dbg(3, "MC:\n");
 511
 512         pci_unregister_driver(&x38_driver);
 513         if (!x38_registered) {
 514                 x38_remove_one(mci_pdev);
 515                 pci_dev_put(mci_pdev);
 516         }
 517 }
 518
 519 module_init(x38_init);
 520 module_exit(x38_exit);
 521
 522 MODULE_LICENSE("GPL");
 523 MODULE_AUTHOR("Cluster Computing, Inc. Hitoshi Mitake");
 524 MODULE_DESCRIPTION("MC support for Intel X38 memory hub controllers");
 525
 526 module_param(edac_op_state, int, 0444);
 527 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");