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_module.h"
  17
  18 #include <linux/io-64-nonatomic-lo-hi.h>
  19
  20 #define EDAC_MOD_STR        "i3200_edac"
  21
  22 #define PCI_DEVICE_ID_INTEL_3200_HB    0x29f0
  23
  24 #define I3200_DIMMS             4
  25 #define I3200_RANKS             8
  26 #define I3200_RANKS_PER_CHANNEL 4
  27 #define I3200_CHANNELS          2
  28
  29 /* Intel 3200 register addresses - device 0 function 0 - DRAM Controller */
  30
  31 #define I3200_MCHBAR_LOW        0x48    /* MCH Memory Mapped Register BAR */
  32 #define I3200_MCHBAR_HIGH       0x4c
  33 #define I3200_MCHBAR_MASK       0xfffffc000ULL  /* bits 35:14 */
  34 #define I3200_MMR_WINDOW_SIZE   16384
  35
  36 #define I3200_TOM               0xa0    /* Top of Memory (16b)
  37                  *
  38                  * 15:10 reserved
  39                  *  9:0  total populated physical memory
  40                  */
  41 #define I3200_TOM_MASK          0x3ff   /* bits 9:0 */
  42 #define I3200_TOM_SHIFT         26      /* 64MiB grain */
  43
  44 #define I3200_ERRSTS            0xc8    /* Error Status Register (16b)
  45                  *
  46                  * 15    reserved
  47                  * 14    Isochronous TBWRR Run Behind FIFO Full
  48                  *       (ITCV)
  49                  * 13    Isochronous TBWRR Run Behind FIFO Put
  50                  *       (ITSTV)
  51                  * 12    reserved
  52                  * 11    MCH Thermal Sensor Event
  53                  *       for SMI/SCI/SERR (GTSE)
  54                  * 10    reserved
  55                  *  9    LOCK to non-DRAM Memory Flag (LCKF)
  56                  *  8    reserved
  57                  *  7    DRAM Throttle Flag (DTF)
  58                  *  6:2  reserved
  59                  *  1    Multi-bit DRAM ECC Error Flag (DMERR)
  60                  *  0    Single-bit DRAM ECC Error Flag (DSERR)
  61                  */
  62 #define I3200_ERRSTS_UE         0x0002
  63 #define I3200_ERRSTS_CE         0x0001
  64 #define I3200_ERRSTS_BITS       (I3200_ERRSTS_UE | I3200_ERRSTS_CE)
  65
  66
  67 /* Intel  MMIO register space - device 0 function 0 - MMR space */
  68
  69 #define I3200_C0DRB     0x200   /* Channel 0 DRAM Rank Boundary (16b x 4)
  70                  *
  71                  * 15:10 reserved
  72                  *  9:0  Channel 0 DRAM Rank Boundary Address
  73                  */
  74 #define I3200_C1DRB     0x600   /* Channel 1 DRAM Rank Boundary (16b x 4) */
  75 #define I3200_DRB_MASK  0x3ff   /* bits 9:0 */
  76 #define I3200_DRB_SHIFT 26      /* 64MiB grain */
  77
  78 #define I3200_C0ECCERRLOG       0x280   /* Channel 0 ECC Error Log (64b)
  79                  *
  80                  * 63:48 Error Column Address (ERRCOL)
  81                  * 47:32 Error Row Address (ERRROW)
  82                  * 31:29 Error Bank Address (ERRBANK)
  83                  * 28:27 Error Rank Address (ERRRANK)
  84                  * 26:24 reserved
  85                  * 23:16 Error Syndrome (ERRSYND)
  86                  * 15: 2 reserved
  87                  *    1  Multiple Bit Error Status (MERRSTS)
  88                  *    0  Correctable Error Status (CERRSTS)
  89                  */
  90 #define I3200_C1ECCERRLOG               0x680   /* Chan 1 ECC Error Log (64b) */
  91 #define I3200_ECCERRLOG_CE              0x1
  92 #define I3200_ECCERRLOG_UE              0x2
  93 #define I3200_ECCERRLOG_RANK_BITS       0x18000000
  94 #define I3200_ECCERRLOG_RANK_SHIFT      27
  95 #define I3200_ECCERRLOG_SYNDROME_BITS   0xff0000
  96 #define I3200_ECCERRLOG_SYNDROME_SHIFT  16
  97 #define I3200_CAPID0                    0xe0    /* P.95 of spec for details */
  98
  99 struct i3200_priv {
 100         void __iomem *window;
 101 };
 102
 103 static int nr_channels;
 104
 105 static int how_many_channels(struct pci_dev *pdev)
 106 {
 107         int n_channels;
 108
 109         unsigned char capid0_8b; /* 8th byte of CAPID0 */
 110
 111         pci_read_config_byte(pdev, I3200_CAPID0 + 8, &capid0_8b);
 112
 113         if (capid0_8b & 0x20) { /* check DCD: Dual Channel Disable */
 114                 edac_dbg(0, "In single channel mode\n");
 115                 n_channels = 1;
 116         } else {
 117                 edac_dbg(0, "In dual channel mode\n");
 118                 n_channels = 2;
 119         }
 120
 121         if (capid0_8b & 0x10) /* check if both channels are filled */
 122                 edac_dbg(0, "2 DIMMS per channel disabled\n");
 123         else
 124                 edac_dbg(0, "2 DIMMS per channel enabled\n");
 125
 126         return n_channels;
 127 }
 128
 129 static unsigned long eccerrlog_syndrome(u64 log)
 130 {
 131         return (log & I3200_ECCERRLOG_SYNDROME_BITS) >>
 132                 I3200_ECCERRLOG_SYNDROME_SHIFT;
 133 }
 134
 135 static int eccerrlog_row(int channel, u64 log)
 136 {
 137         u64 rank = ((log & I3200_ECCERRLOG_RANK_BITS) >>
 138                 I3200_ECCERRLOG_RANK_SHIFT);
 139         return rank | (channel * I3200_RANKS_PER_CHANNEL);
 140 }
 141
 142 enum i3200_chips {
 143         I3200 = 0,
 144 };
 145
 146 struct i3200_dev_info {
 147         const char *ctl_name;
 148 };
 149
 150 struct i3200_error_info {
 151         u16 errsts;
 152         u16 errsts2;
 153         u64 eccerrlog[I3200_CHANNELS];
 154 };
 155
 156 static const struct i3200_dev_info i3200_devs[] = {
 157         [I3200] = {
 158                 .ctl_name = "i3200"
 159         },
 160 };
 161
 162 static struct pci_dev *mci_pdev;
 163 static int i3200_registered = 1;
 164
 165
 166 static void i3200_clear_error_info(struct mem_ctl_info *mci)
 167 {
 168         struct pci_dev *pdev;
 169
 170         pdev = to_pci_dev(mci->pdev);
 171
 172         /*
 173          * Clear any error bits.
 174          * (Yes, we really clear bits by writing 1 to them.)
 175          */
 176         pci_write_bits16(pdev, I3200_ERRSTS, I3200_ERRSTS_BITS,
 177                 I3200_ERRSTS_BITS);
 178 }
 179
 180 static void i3200_get_and_clear_error_info(struct mem_ctl_info *mci,
 181                 struct i3200_error_info *info)
 182 {
 183         struct pci_dev *pdev;
 184         struct i3200_priv *priv = mci->pvt_info;
 185         void __iomem *window = priv->window;
 186
 187         pdev = to_pci_dev(mci->pdev);
 188
 189         /*
 190          * This is a mess because there is no atomic way to read all the
 191          * registers at once and the registers can transition from CE being
 192          * overwritten by UE.
 193          */
 194         pci_read_config_word(pdev, I3200_ERRSTS, &info->errsts);
 195         if (!(info->errsts & I3200_ERRSTS_BITS))
 196                 return;
 197
 198         info->eccerrlog[0] = readq(window + I3200_C0ECCERRLOG);
 199         if (nr_channels == 2)
 200                 info->eccerrlog[1] = readq(window + I3200_C1ECCERRLOG);
 201
 202         pci_read_config_word(pdev, I3200_ERRSTS, &info->errsts2);
 203
 204         /*
 205          * If the error is the same for both reads then the first set
 206          * of reads is valid.  If there is a change then there is a CE
 207          * with no info and the second set of reads is valid and
 208          * should be UE info.
 209          */
 210         if ((info->errsts ^ info->errsts2) & I3200_ERRSTS_BITS) {
 211                 info->eccerrlog[0] = readq(window + I3200_C0ECCERRLOG);
 212                 if (nr_channels == 2)
 213                         info->eccerrlog[1] = readq(window + I3200_C1ECCERRLOG);
 214         }
 215
 216         i3200_clear_error_info(mci);
 217 }
 218
 219 static void i3200_process_error_info(struct mem_ctl_info *mci,
 220                 struct i3200_error_info *info)
 221 {
 222         int channel;
 223         u64 log;
 224
 225         if (!(info->errsts & I3200_ERRSTS_BITS))
 226                 return;
 227
 228         if ((info->errsts ^ info->errsts2) & I3200_ERRSTS_BITS) {
 229                 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
 230                                      -1, -1, -1, "UE overwrote CE", "");
 231                 info->errsts = info->errsts2;
 232         }
 233
 234         for (channel = 0; channel < nr_channels; channel++) {
 235                 log = info->eccerrlog[channel];
 236                 if (log & I3200_ECCERRLOG_UE) {
 237                         edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 238                                              0, 0, 0,
 239                                              eccerrlog_row(channel, log),
 240                                              -1, -1,
 241                                              "i3000 UE", "");
 242                 } else if (log & I3200_ECCERRLOG_CE) {
 243                         edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
 244                                              0, 0, eccerrlog_syndrome(log),
 245                                              eccerrlog_row(channel, log),
 246                                              -1, -1,
 247                                              "i3000 CE", "");
 248                 }
 249         }
 250 }
 251
 252 static void i3200_check(struct mem_ctl_info *mci)
 253 {
 254         struct i3200_error_info info;
 255
 256         edac_dbg(1, "MC%d\n", mci->mc_idx);
 257         i3200_get_and_clear_error_info(mci, &info);
 258         i3200_process_error_info(mci, &info);
 259 }
 260
 261 static void __iomem *i3200_map_mchbar(struct pci_dev *pdev)
 262 {
 263         union {
 264                 u64 mchbar;
 265                 struct {
 266                         u32 mchbar_low;
 267                         u32 mchbar_high;
 268                 };
 269         } u;
 270         void __iomem *window;
 271
 272         pci_read_config_dword(pdev, I3200_MCHBAR_LOW, &u.mchbar_low);
 273         pci_read_config_dword(pdev, I3200_MCHBAR_HIGH, &u.mchbar_high);
 274         u.mchbar &= I3200_MCHBAR_MASK;
 275
 276         if (u.mchbar != (resource_size_t)u.mchbar) {
 277                 printk(KERN_ERR
 278                         "i3200: mmio space beyond accessible range (0x%llx)\n",
 279                         (unsigned long long)u.mchbar);
 280                 return NULL;
 281         }
 282
 283         window = ioremap_nocache(u.mchbar, I3200_MMR_WINDOW_SIZE);
 284         if (!window)
 285                 printk(KERN_ERR "i3200: cannot map mmio space at 0x%llx\n",
 286                         (unsigned long long)u.mchbar);
 287
 288         return window;
 289 }
 290
 291
 292 static void i3200_get_drbs(void __iomem *window,
 293         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL])
 294 {
 295         int i;
 296
 297         for (i = 0; i < I3200_RANKS_PER_CHANNEL; i++) {
 298                 drbs[0][i] = readw(window + I3200_C0DRB + 2*i) & I3200_DRB_MASK;
 299                 drbs[1][i] = readw(window + I3200_C1DRB + 2*i) & I3200_DRB_MASK;
 300
 301                 edac_dbg(0, "drb[0][%d] = %d, drb[1][%d] = %d\n", i, drbs[0][i], i, drbs[1][i]);
 302         }
 303 }
 304
 305 static bool i3200_is_stacked(struct pci_dev *pdev,
 306         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL])
 307 {
 308         u16 tom;
 309
 310         pci_read_config_word(pdev, I3200_TOM, &tom);
 311         tom &= I3200_TOM_MASK;
 312
 313         return drbs[I3200_CHANNELS - 1][I3200_RANKS_PER_CHANNEL - 1] == tom;
 314 }
 315
 316 static unsigned long drb_to_nr_pages(
 317         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL], bool stacked,
 318         int channel, int rank)
 319 {
 320         int n;
 321
 322         n = drbs[channel][rank];
 323         if (!n)
 324                 return 0;
 325
 326         if (rank > 0)
 327                 n -= drbs[channel][rank - 1];
 328         if (stacked && (channel == 1) &&
 329         drbs[channel][rank] == drbs[channel][I3200_RANKS_PER_CHANNEL - 1])
 330                 n -= drbs[0][I3200_RANKS_PER_CHANNEL - 1];
 331
 332         n <<= (I3200_DRB_SHIFT - PAGE_SHIFT);
 333         return n;
 334 }
 335
 336 static int i3200_probe1(struct pci_dev *pdev, int dev_idx)
 337 {
 338         int rc;
 339         int i, j;
 340         struct mem_ctl_info *mci = NULL;
 341         struct edac_mc_layer layers[2];
 342         u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL];
 343         bool stacked;
 344         void __iomem *window;
 345         struct i3200_priv *priv;
 346
 347         edac_dbg(0, "MC:\n");
 348
 349         window = i3200_map_mchbar(pdev);
 350         if (!window)
 351                 return -ENODEV;
 352
 353         i3200_get_drbs(window, drbs);
 354         nr_channels = how_many_channels(pdev);
 355
 356         layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 357         layers[0].size = I3200_DIMMS;
 358         layers[0].is_virt_csrow = true;
 359         layers[1].type = EDAC_MC_LAYER_CHANNEL;
 360         layers[1].size = nr_channels;
 361         layers[1].is_virt_csrow = false;
 362         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
 363                             sizeof(struct i3200_priv));
 364         if (!mci)
 365                 return -ENOMEM;
 366
 367         edac_dbg(3, "MC: init mci\n");
 368
 369         mci->pdev = &pdev->dev;
 370         mci->mtype_cap = MEM_FLAG_DDR2;
 371
 372         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 373         mci->edac_cap = EDAC_FLAG_SECDED;
 374
 375         mci->mod_name = EDAC_MOD_STR;
 376         mci->ctl_name = i3200_devs[dev_idx].ctl_name;
 377         mci->dev_name = pci_name(pdev);
 378         mci->edac_check = i3200_check;
 379         mci->ctl_page_to_phys = NULL;
 380         priv = mci->pvt_info;
 381         priv->window = window;
 382
 383         stacked = i3200_is_stacked(pdev, drbs);
 384
 385         /*
 386          * The dram rank boundary (DRB) reg values are boundary addresses
 387          * for each DRAM rank with a granularity of 64MB.  DRB regs are
 388          * cumulative; the last one will contain the total memory
 389          * contained in all ranks.
 390          */
 391         for (i = 0; i < I3200_DIMMS; i++) {
 392                 unsigned long nr_pages;
 393
 394                 for (j = 0; j < nr_channels; j++) {
 395                         struct dimm_info *dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
 396                                                                mci->n_layers, i, j, 0);
 397
 398                         nr_pages = drb_to_nr_pages(drbs, stacked, j, i);
 399                         if (nr_pages == 0)
 400                                 continue;
 401
 402                         edac_dbg(0, "csrow %d, channel %d%s, size = %ld Mb\n", i, j,
 403                                  stacked ? " (stacked)" : "", PAGES_TO_MiB(nr_pages));
 404
 405                         dimm->nr_pages = nr_pages;
 406                         dimm->grain = nr_pages << PAGE_SHIFT;
 407                         dimm->mtype = MEM_DDR2;
 408                         dimm->dtype = DEV_UNKNOWN;
 409                         dimm->edac_mode = EDAC_UNKNOWN;
 410                 }
 411         }
 412
 413         i3200_clear_error_info(mci);
 414
 415         rc = -ENODEV;
 416         if (edac_mc_add_mc(mci)) {
 417                 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
 418                 goto fail;
 419         }
 420
 421         /* get this far and it's successful */
 422         edac_dbg(3, "MC: success\n");
 423         return 0;
 424
 425 fail:
 426         iounmap(window);
 427         if (mci)
 428                 edac_mc_free(mci);
 429
 430         return rc;
 431 }
 432
 433 static int i3200_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 434 {
 435         int rc;
 436
 437         edac_dbg(0, "MC:\n");
 438
 439         if (pci_enable_device(pdev) < 0)
 440                 return -EIO;
 441
 442         rc = i3200_probe1(pdev, ent->driver_data);
 443         if (!mci_pdev)
 444                 mci_pdev = pci_dev_get(pdev);
 445
 446         return rc;
 447 }
 448
 449 static void i3200_remove_one(struct pci_dev *pdev)
 450 {
 451         struct mem_ctl_info *mci;
 452         struct i3200_priv *priv;
 453
 454         edac_dbg(0, "\n");
 455
 456         mci = edac_mc_del_mc(&pdev->dev);
 457         if (!mci)
 458                 return;
 459
 460         priv = mci->pvt_info;
 461         iounmap(priv->window);
 462
 463         edac_mc_free(mci);
 464
 465         pci_disable_device(pdev);
 466 }
 467
 468 static const struct pci_device_id i3200_pci_tbl[] = {
 469         {
 470                 PCI_VEND_DEV(INTEL, 3200_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 471                 I3200},
 472         {
 473                 0,
 474         }            /* 0 terminated list. */
 475 };
 476
 477 MODULE_DEVICE_TABLE(pci, i3200_pci_tbl);
 478
 479 static struct pci_driver i3200_driver = {
 480         .name = EDAC_MOD_STR,
 481         .probe = i3200_init_one,
 482         .remove = i3200_remove_one,
 483         .id_table = i3200_pci_tbl,
 484 };
 485
 486 static int __init i3200_init(void)
 487 {
 488         int pci_rc;
 489
 490         edac_dbg(3, "MC:\n");
 491
 492         /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 493         opstate_init();
 494
 495         pci_rc = pci_register_driver(&i3200_driver);
 496         if (pci_rc < 0)
 497                 goto fail0;
 498
 499         if (!mci_pdev) {
 500                 i3200_registered = 0;
 501                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 502                                 PCI_DEVICE_ID_INTEL_3200_HB, NULL);
 503                 if (!mci_pdev) {
 504                         edac_dbg(0, "i3200 pci_get_device fail\n");
 505                         pci_rc = -ENODEV;
 506                         goto fail1;
 507                 }
 508
 509                 pci_rc = i3200_init_one(mci_pdev, i3200_pci_tbl);
 510                 if (pci_rc < 0) {
 511                         edac_dbg(0, "i3200 init fail\n");
 512                         pci_rc = -ENODEV;
 513                         goto fail1;
 514                 }
 515         }
 516
 517         return 0;
 518
 519 fail1:
 520         pci_unregister_driver(&i3200_driver);
 521
 522 fail0:
 523         pci_dev_put(mci_pdev);
 524
 525         return pci_rc;
 526 }
 527
 528 static void __exit i3200_exit(void)
 529 {
 530         edac_dbg(3, "MC:\n");
 531
 532         pci_unregister_driver(&i3200_driver);
 533         if (!i3200_registered) {
 534                 i3200_remove_one(mci_pdev);
 535                 pci_dev_put(mci_pdev);
 536         }
 537 }
 538
 539 module_init(i3200_init);
 540 module_exit(i3200_exit);
 541
 542 MODULE_LICENSE("GPL");
 543 MODULE_AUTHOR("Akamai Technologies, Inc.");
 544 MODULE_DESCRIPTION("MC support for Intel 3200 memory hub controllers");
 545
 546 module_param(edac_op_state, int, 0444);
 547 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");