drivers/edac/i3200_edac.c

   1 /*
   2  * Intel 3200/3210 Memory Controller kernel module
   3  * Copyright (C) 2008-2009 Akamai Technologies, Inc.
   4  * Portions by Hitoshi Mitake <h.mitake@gmail.com>.
   5  *
   6  * This file may be distributed under the terms of the
   7  * GNU General Public License.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/init.h>
  12 #include <linux/pci.h>
  13 #include <linux/pci_ids.h>
  14 #include <linux/edac.h>
  15 #include <linux/io.h>
  16 #include "edac_core.h"
  17
  18 #include <asm-generic/io-64-nonatomic-lo-hi.h>
  19
  20 #define I3200_REVISION        "1.1"
  21
  22 #define EDAC_MOD_STR        "i3200_edac"
  23
  24 #define PCI_DEVICE_ID_INTEL_3200_HB    0x29f0
  25
  26 #define I3200_DIMMS             4
  27 #define I3200_RANKS             8
  28 #define I3200_RANKS_PER_CHANNEL 4
  29 #define I3200_CHANNELS          2
  30
  31 /* Intel 3200 register addresses - device 0 function 0 - DRAM Controller */
  32
  33 #define I3200_MCHBAR_LOW        0x48    /* MCH Memory Mapped Register BAR */
  34 #define I3200_MCHBAR_HIGH       0x4c
  35 #define I3200_MCHBAR_MASK       0xfffffc000ULL  /* bits 35:14 */
  36 #define I3200_MMR_WINDOW_SIZE   16384
  37
  38 #define I3200_TOM               0xa0    /* Top of Memory (16b)
  39                  *
  40                  * 15:10 reserved
  41                  *  9:0  total populated physical memory
  42                  */
  43 #define I3200_TOM_MASK          0x3ff   /* bits 9:0 */
  44 #define I3200_TOM_SHIFT         26      /* 64MiB grain */
  45
  46 #define I3200_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 I3200_ERRSTS_UE         0x0002
  65 #define I3200_ERRSTS_CE         0x0001
  66 #define I3200_ERRSTS_BITS       (I3200_ERRSTS_UE | I3200_ERRSTS_CE)
  67
  68
  69 /* Intel  MMIO register space - device 0 function 0 - MMR space */
  70
  71 #define I3200_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 I3200_C1DRB     0x600   /* Channel 1 DRAM Rank Boundary (16b x 4) */
  77 #define I3200_DRB_MASK  0x3ff   /* bits 9:0 */
  78 #define I3200_DRB_SHIFT 26      /* 64MiB grain */
  79
  80 #define I3200_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 I3200_C1ECCERRLOG               0x680   /* Chan 1 ECC Error Log (64b) */
  93 #define I3200_ECCERRLOG_CE              0x1
  94 #define I3200_ECCERRLOG_UE              0x2
  95 #define I3200_ECCERRLOG_RANK_BITS       0x18000000
  96 #define I3200_ECCERRLOG_RANK_SHIFT      27
  97 #define I3200_ECCERRLOG_SYNDROME_BITS   0xff0000
  98 #define I3200_ECCERRLOG_SYNDROME_SHIFT  16
  99 #define I3200_CAPID0                    0xe0    /* P.95 of spec for details */
 100
 101 struct i3200_priv {
 102         void __iomem *window;
 103 };
 104
 105 static int nr_channels;
 106
 107 static int how_many_channels(struct pci_dev *pdev)
 108 {
 109         unsigned char capid0_8b; /* 8th byte of CAPID0 */
 110
 111         pci_read_config_byte(pdev, I3200_CAPID0 + 8, &capid0_8b);
 112         if (capid0_8b & 0x20) { /* check DCD: Dual Channel Disable */
 113                 edac_dbg(0, "In single channel mode\n");
 114                 return 1;
 115         } else {
 116                 edac_dbg(0, "In dual channel mode\n");
 117                 return 2;
 118         }
 119 }
 120
 121 static unsigned long eccerrlog_syndrome(u64 log)
 122 {
 123         return (log & I3200_ECCERRLOG_SYNDROME_BITS) >>
 124                 I3200_ECCERRLOG_SYNDROME_SHIFT;
 125 }
 126
 127 static int eccerrlog_row(int channel, u64 log)
 128 {
 129         u64 rank = ((log & I3200_ECCERRLOG_RANK_BITS) >>
 130                 I3200_ECCERRLOG_RANK_SHIFT);
 131         return rank | (channel * I3200_RANKS_PER_CHANNEL);
 132 }
 133
 134 enum i3200_chips {
 135         I3200 = 0,
 136 };
 137
 138 struct i3200_dev_info {
 139         const char *ctl_name;
 140 };
 141
 142 struct i3200_error_info {
 143         u16 errsts;
 144         u16 errsts2;
 145         u64 eccerrlog[I3200_CHANNELS];
 146 };
 147
 148 static const struct i3200_dev_info i3200_devs[] = {
 149         [I3200] = {
 150                 .ctl_name = "i3200"
 151         },
 152 };
 153
 154 static struct pci_dev *mci_pdev;
 155 static int i3200_registered = 1;
 156
 157
 158 static void i3200_clear_error_info(struct mem_ctl_info *mci)
 159 {
 160         struct pci_dev *pdev;
 161
 162         pdev = to_pci_dev(mci->pdev);
 163
 164         /*
 165          * Clear any error bits.
 166          * (Yes, we really clear bits by writing 1 to them.)
 167          */
 168         pci_write_bits16(pdev, I3200_ERRSTS, I3200_ERRSTS_BITS,
 169                 I3200_ERRSTS_BITS);
 170 }
 171
 172 static void i3200_get_and_clear_error_info(struct mem_ctl_info *mci,
 173                 struct i3200_error_info *info)
 174 {
 175         struct pci_dev *pdev;
 176         struct i3200_priv *priv = mci->pvt_info;
 177         void __iomem *window = priv->window;
 178
 179         pdev = to_pci_dev(mci->pdev);
 180
 181         /*
 182          * This is a mess because there is no atomic way to read all the
 183          * registers at once and the registers can transition from CE being
 184          * overwritten by UE.
 185          */
 186         pci_read_config_word(pdev, I3200_ERRSTS, &info->errsts);
 187         if (!(info->errsts & I3200_ERRSTS_BITS))
 188                 return;
 189
 190         info->eccerrlog[0] = readq(window + I3200_C0ECCERRLOG);
 191         if (nr_channels == 2)
 192                 info->eccerrlog[1] = readq(window + I3200_C1ECCERRLOG);
 193
 194         pci_read_config_word(pdev, I3200_ERRSTS, &info->errsts2);
 195
 196         /*
 197          * If the error is the same for both reads then the first set
 198          * of reads is valid.  If there is a change then there is a CE
 199          * with no info and the second set of reads is valid and
 200          * should be UE info.
 201          */
 202         if ((info->errsts ^ info->errsts2) & I3200_ERRSTS_BITS) {
 203                 info->eccerrlog[0] = readq(window + I3200_C0ECCERRLOG);
 204                 if (nr_channels == 2)
 205                         info->eccerrlog[1] = readq(window + I3200_C1ECCERRLOG);
 206         }
 207
 208         i3200_clear_error_info(mci);
 209 }
 210
 211 static void i3200_process_error_info(struct mem_ctl_info *mci,
 212                 struct i3200_error_info *info)
 213 {
 214         int channel;
 215         u64 log;
 216
 217         if (!(info->errsts & I3200_ERRSTS_BITS))
 218                 return;
 219
 220         if ((info->errsts ^ info->errsts2) & I3200_ERRSTS_BITS) {
 221                 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
 222                                      -1, -1, -1, "UE overwrote CE", "");
 223                 info->errsts = info->errsts2;
 224         }
 225
 226         for (channel = 0; channel < nr_channels; channel++) {
 227                 log = info->eccerrlog[channel];
 228                 if (log & I3200_ECCERRLOG_UE) {
 229                         edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 230                                              0, 0, 0,
 231                                              eccerrlog_row(channel, log),
 232                                              -1, -1,
 233                                              "i3000 UE", "");
 234                 } else if (log & I3200_ECCERRLOG_CE) {
 235                         edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 236                                              0, 0, eccerrlog_syndrome(log),
 237                                              eccerrlog_row(channel, log),
 238                                              -1, -1,
 239                                              "i3000 UE", "");
 240                 }
 241         }
 242 }
 243
 244 static void i3200_check(struct mem_ctl_info *mci)
 245 {
 246         struct i3200_error_info info;
 247
 248         edac_dbg(1, "MC%d\n", mci->mc_idx);
 249         i3200_get_and_clear_error_info(mci, &info);
 250         i3200_process_error_info(mci, &info);
 251 }
 252
 253
 254 void __iomem *i3200_map_mchbar(struct pci_dev *pdev)
 255 {
 256         union {
 257                 u64 mchbar;
 258                 struct {
 259                         u32 mchbar_low;
 260                         u32 mchbar_high;
 261                 };
 262         } u;
 263         void __iomem *window;
 264
 265         pci_read_config_dword(pdev, I3200_MCHBAR_LOW, &u.mchbar_low);
 266         pci_read_config_dword(pdev, I3200_MCHBAR_HIGH, &u.mchbar_high);
 267         u.mchbar &= I3200_MCHBAR_MASK;
 268
 269         if (u.mchbar != (resource_size_t)u.mchbar) {
 270                 printk(KERN_ERR
 271                         "i3200: mmio space beyond accessible range (0x%llx)\n",
 272                         (unsigned long long)u.mchbar);
 273                 return NULL;
 274         }
 275
 276         window = ioremap_nocache(u.mchbar, I3200_MMR_WINDOW_SIZE);
 277         if (!window)
 278                 printk(KERN_ERR "i3200: cannot map mmio space at 0x%llx\n",
 279                         (unsigned long long)u.mchbar);
 280
 281         return window;
 282 }
 283
 284
 285 static void i3200_get_drbs(void __iomem *window,
 286         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL])
 287 {
 288         int i;
 289
 290         for (i = 0; i < I3200_RANKS_PER_CHANNEL; i++) {
 291                 drbs[0][i] = readw(window + I3200_C0DRB + 2*i) & I3200_DRB_MASK;
 292                 drbs[1][i] = readw(window + I3200_C1DRB + 2*i) & I3200_DRB_MASK;
 293         }
 294 }
 295
 296 static bool i3200_is_stacked(struct pci_dev *pdev,
 297         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL])
 298 {
 299         u16 tom;
 300
 301         pci_read_config_word(pdev, I3200_TOM, &tom);
 302         tom &= I3200_TOM_MASK;
 303
 304         return drbs[I3200_CHANNELS - 1][I3200_RANKS_PER_CHANNEL - 1] == tom;
 305 }
 306
 307 static unsigned long drb_to_nr_pages(
 308         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL], bool stacked,
 309         int channel, int rank)
 310 {
 311         int n;
 312
 313         n = drbs[channel][rank];
 314         if (rank > 0)
 315                 n -= drbs[channel][rank - 1];
 316         if (stacked && (channel == 1) &&
 317         drbs[channel][rank] == drbs[channel][I3200_RANKS_PER_CHANNEL - 1])
 318                 n -= drbs[0][I3200_RANKS_PER_CHANNEL - 1];
 319
 320         n <<= (I3200_DRB_SHIFT - PAGE_SHIFT);
 321         return n;
 322 }
 323
 324 static int i3200_probe1(struct pci_dev *pdev, int dev_idx)
 325 {
 326         int rc;
 327         int i, j;
 328         struct mem_ctl_info *mci = NULL;
 329         struct edac_mc_layer layers[2];
 330         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL];
 331         bool stacked;
 332         void __iomem *window;
 333         struct i3200_priv *priv;
 334
 335         edac_dbg(0, "MC:\n");
 336
 337         window = i3200_map_mchbar(pdev);
 338         if (!window)
 339                 return -ENODEV;
 340
 341         i3200_get_drbs(window, drbs);
 342         nr_channels = how_many_channels(pdev);
 343
 344         layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 345         layers[0].size = I3200_DIMMS;
 346         layers[0].is_virt_csrow = true;
 347         layers[1].type = EDAC_MC_LAYER_CHANNEL;
 348         layers[1].size = nr_channels;
 349         layers[1].is_virt_csrow = false;
 350         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
 351                             sizeof(struct i3200_priv));
 352         if (!mci)
 353                 return -ENOMEM;
 354
 355         edac_dbg(3, "MC: init mci\n");
 356
 357         mci->pdev = &pdev->dev;
 358         mci->mtype_cap = MEM_FLAG_DDR2;
 359
 360         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 361         mci->edac_cap = EDAC_FLAG_SECDED;
 362
 363         mci->mod_name = EDAC_MOD_STR;
 364         mci->mod_ver = I3200_REVISION;
 365         mci->ctl_name = i3200_devs[dev_idx].ctl_name;
 366         mci->dev_name = pci_name(pdev);
 367         mci->edac_check = i3200_check;
 368         mci->ctl_page_to_phys = NULL;
 369         priv = mci->pvt_info;
 370         priv->window = window;
 371
 372         stacked = i3200_is_stacked(pdev, drbs);
 373
 374         /*
 375          * The dram rank boundary (DRB) reg values are boundary addresses
 376          * for each DRAM rank with a granularity of 64MB.  DRB regs are
 377          * cumulative; the last one will contain the total memory
 378          * contained in all ranks.
 379          */
 380         for (i = 0; i < mci->nr_csrows; i++) {
 381                 unsigned long nr_pages;
 382                 struct csrow_info *csrow = mci->csrows[i];
 383
 384                 nr_pages = drb_to_nr_pages(drbs, stacked,
 385                         i / I3200_RANKS_PER_CHANNEL,
 386                         i % I3200_RANKS_PER_CHANNEL);
 387
 388                 if (nr_pages == 0)
 389                         continue;
 390
 391                 for (j = 0; j < nr_channels; j++) {
 392                         struct dimm_info *dimm = csrow->channels[j]->dimm;
 393
 394                         dimm->nr_pages = nr_pages;
 395                         dimm->grain = nr_pages << PAGE_SHIFT;
 396                         dimm->mtype = MEM_DDR2;
 397                         dimm->dtype = DEV_UNKNOWN;
 398                         dimm->edac_mode = EDAC_UNKNOWN;
 399                 }
 400         }
 401
 402         i3200_clear_error_info(mci);
 403
 404         rc = -ENODEV;
 405         if (edac_mc_add_mc(mci)) {
 406                 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
 407                 goto fail;
 408         }
 409
 410         /* get this far and it's successful */
 411         edac_dbg(3, "MC: success\n");
 412         return 0;
 413
 414 fail:
 415         iounmap(window);
 416         if (mci)
 417                 edac_mc_free(mci);
 418
 419         return rc;
 420 }
 421
 422 static int __devinit i3200_init_one(struct pci_dev *pdev,
 423                 const struct pci_device_id *ent)
 424 {
 425         int rc;
 426
 427         edac_dbg(0, "MC:\n");
 428
 429         if (pci_enable_device(pdev) < 0)
 430                 return -EIO;
 431
 432         rc = i3200_probe1(pdev, ent->driver_data);
 433         if (!mci_pdev)
 434                 mci_pdev = pci_dev_get(pdev);
 435
 436         return rc;
 437 }
 438
 439 static void __devexit i3200_remove_one(struct pci_dev *pdev)
 440 {
 441         struct mem_ctl_info *mci;
 442         struct i3200_priv *priv;
 443
 444         edac_dbg(0, "\n");
 445
 446         mci = edac_mc_del_mc(&pdev->dev);
 447         if (!mci)
 448                 return;
 449
 450         priv = mci->pvt_info;
 451         iounmap(priv->window);
 452
 453         edac_mc_free(mci);
 454 }
 455
 456 static DEFINE_PCI_DEVICE_TABLE(i3200_pci_tbl) = {
 457         {
 458                 PCI_VEND_DEV(INTEL, 3200_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 459                 I3200},
 460         {
 461                 0,
 462         }            /* 0 terminated list. */
 463 };
 464
 465 MODULE_DEVICE_TABLE(pci, i3200_pci_tbl);
 466
 467 static struct pci_driver i3200_driver = {
 468         .name = EDAC_MOD_STR,
 469         .probe = i3200_init_one,
 470         .remove = __devexit_p(i3200_remove_one),
 471         .id_table = i3200_pci_tbl,
 472 };
 473
 474 static int __init i3200_init(void)
 475 {
 476         int pci_rc;
 477
 478         edac_dbg(3, "MC:\n");
 479
 480         /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 481         opstate_init();
 482
 483         pci_rc = pci_register_driver(&i3200_driver);
 484         if (pci_rc < 0)
 485                 goto fail0;
 486
 487         if (!mci_pdev) {
 488                 i3200_registered = 0;
 489                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 490                                 PCI_DEVICE_ID_INTEL_3200_HB, NULL);
 491                 if (!mci_pdev) {
 492                         edac_dbg(0, "i3200 pci_get_device fail\n");
 493                         pci_rc = -ENODEV;
 494                         goto fail1;
 495                 }
 496
 497                 pci_rc = i3200_init_one(mci_pdev, i3200_pci_tbl);
 498                 if (pci_rc < 0) {
 499                         edac_dbg(0, "i3200 init fail\n");
 500                         pci_rc = -ENODEV;
 501                         goto fail1;
 502                 }
 503         }
 504
 505         return 0;
 506
 507 fail1:
 508         pci_unregister_driver(&i3200_driver);
 509
 510 fail0:
 511         if (mci_pdev)
 512                 pci_dev_put(mci_pdev);
 513
 514         return pci_rc;
 515 }
 516
 517 static void __exit i3200_exit(void)
 518 {
 519         edac_dbg(3, "MC:\n");
 520
 521         pci_unregister_driver(&i3200_driver);
 522         if (!i3200_registered) {
 523                 i3200_remove_one(mci_pdev);
 524                 pci_dev_put(mci_pdev);
 525         }
 526 }
 527
 528 module_init(i3200_init);
 529 module_exit(i3200_exit);
 530
 531 MODULE_LICENSE("GPL");
 532 MODULE_AUTHOR("Akamai Technologies, Inc.");
 533 MODULE_DESCRIPTION("MC support for Intel 3200 memory hub controllers");
 534
 535 module_param(edac_op_state, int, 0444);
 536 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");