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/slab.h>
  17 #include <linux/edac.h>
  18 #include "edac_core.h"
  19
  20 #define X38_REVISION            "1.1"
  21
  22 #define EDAC_MOD_STR            "x38_edac"
  23
  24 #define PCI_DEVICE_ID_INTEL_X38_HB      0x29e0
  25
  26 #define X38_RANKS               8
  27 #define X38_RANKS_PER_CHANNEL   4
  28 #define X38_CHANNELS            2
  29
  30 /* Intel X38 register addresses - device 0 function 0 - DRAM Controller */
  31
  32 #define X38_MCHBAR_LOW  0x48    /* MCH Memory Mapped Register BAR */
  33 #define X38_MCHBAR_HIGH 0x4c
  34 #define X38_MCHBAR_MASK 0xfffffc000ULL  /* bits 35:14 */
  35 #define X38_MMR_WINDOW_SIZE     16384
  36
  37 #define X38_TOM 0xa0    /* Top of Memory (16b)
  38                                  *
  39                                  * 15:10 reserved
  40                                  *  9:0  total populated physical memory
  41                                  */
  42 #define X38_TOM_MASK    0x3ff   /* bits 9:0 */
  43 #define X38_TOM_SHIFT 26        /* 64MiB grain */
  44
  45 #define X38_ERRSTS      0xc8    /* Error Status Register (16b)
  46                                  *
  47                                  * 15    reserved
  48                                  * 14    Isochronous TBWRR Run Behind FIFO Full
  49                                  *       (ITCV)
  50                                  * 13    Isochronous TBWRR Run Behind FIFO Put
  51                                  *       (ITSTV)
  52                                  * 12    reserved
  53                                  * 11    MCH Thermal Sensor Event
  54                                  *       for SMI/SCI/SERR (GTSE)
  55                                  * 10    reserved
  56                                  *  9    LOCK to non-DRAM Memory Flag (LCKF)
  57                                  *  8    reserved
  58                                  *  7    DRAM Throttle Flag (DTF)
  59                                  *  6:2  reserved
  60                                  *  1    Multi-bit DRAM ECC Error Flag (DMERR)
  61                                  *  0    Single-bit DRAM ECC Error Flag (DSERR)
  62                                  */
  63 #define X38_ERRSTS_UE           0x0002
  64 #define X38_ERRSTS_CE           0x0001
  65 #define X38_ERRSTS_BITS (X38_ERRSTS_UE | X38_ERRSTS_CE)
  66
  67
  68 /* Intel  MMIO register space - device 0 function 0 - MMR space */
  69
  70 #define X38_C0DRB       0x200   /* Channel 0 DRAM Rank Boundary (16b x 4)
  71                                  *
  72                                  * 15:10 reserved
  73                                  *  9:0  Channel 0 DRAM Rank Boundary Address
  74                                  */
  75 #define X38_C1DRB       0x600   /* Channel 1 DRAM Rank Boundary (16b x 4) */
  76 #define X38_DRB_MASK    0x3ff   /* bits 9:0 */
  77 #define X38_DRB_SHIFT 26        /* 64MiB grain */
  78
  79 #define X38_C0ECCERRLOG 0x280   /* Channel 0 ECC Error Log (64b)
  80                                  *
  81                                  * 63:48 Error Column Address (ERRCOL)
  82                                  * 47:32 Error Row Address (ERRROW)
  83                                  * 31:29 Error Bank Address (ERRBANK)
  84                                  * 28:27 Error Rank Address (ERRRANK)
  85                                  * 26:24 reserved
  86                                  * 23:16 Error Syndrome (ERRSYND)
  87                                  * 15: 2 reserved
  88                                  *    1  Multiple Bit Error Status (MERRSTS)
  89                                  *    0  Correctable Error Status (CERRSTS)
  90                                  */
  91 #define X38_C1ECCERRLOG 0x680   /* Channel 1 ECC Error Log (64b) */
  92 #define X38_ECCERRLOG_CE        0x1
  93 #define X38_ECCERRLOG_UE        0x2
  94 #define X38_ECCERRLOG_RANK_BITS 0x18000000
  95 #define X38_ECCERRLOG_SYNDROME_BITS     0xff0000
  96
  97 #define X38_CAPID0 0xe0 /* see P.94 of spec for details */
  98
  99 static int x38_channel_num;
 100
 101 static int how_many_channel(struct pci_dev *pdev)
 102 {
 103         unsigned char capid0_8b; /* 8th byte of CAPID0 */
 104
 105         pci_read_config_byte(pdev, X38_CAPID0 + 8, &capid0_8b);
 106         if (capid0_8b & 0x20) { /* check DCD: Dual Channel Disable */
 107                 debugf0("In single channel mode.\n");
 108                 x38_channel_num = 1;
 109         } else {
 110                 debugf0("In dual channel mode.\n");
 111                 x38_channel_num = 2;
 112         }
 113
 114         return x38_channel_num;
 115 }
 116
 117 static unsigned long eccerrlog_syndrome(u64 log)
 118 {
 119         return (log & X38_ECCERRLOG_SYNDROME_BITS) >> 16;
 120 }
 121
 122 static int eccerrlog_row(int channel, u64 log)
 123 {
 124         return ((log & X38_ECCERRLOG_RANK_BITS) >> 27) |
 125                 (channel * X38_RANKS_PER_CHANNEL);
 126 }
 127
 128 enum x38_chips {
 129         X38 = 0,
 130 };
 131
 132 struct x38_dev_info {
 133         const char *ctl_name;
 134 };
 135
 136 struct x38_error_info {
 137         u16 errsts;
 138         u16 errsts2;
 139         u64 eccerrlog[X38_CHANNELS];
 140 };
 141
 142 static const struct x38_dev_info x38_devs[] = {
 143         [X38] = {
 144                 .ctl_name = "x38"},
 145 };
 146
 147 static struct pci_dev *mci_pdev;
 148 static int x38_registered = 1;
 149
 150
 151 static void x38_clear_error_info(struct mem_ctl_info *mci)
 152 {
 153         struct pci_dev *pdev;
 154
 155         pdev = to_pci_dev(mci->dev);
 156
 157         /*
 158          * Clear any error bits.
 159          * (Yes, we really clear bits by writing 1 to them.)
 160          */
 161         pci_write_bits16(pdev, X38_ERRSTS, X38_ERRSTS_BITS,
 162                          X38_ERRSTS_BITS);
 163 }
 164
 165 static u64 x38_readq(const void __iomem *addr)
 166 {
 167         return readl(addr) | (((u64)readl(addr + 4)) << 32);
 168 }
 169
 170 static void x38_get_and_clear_error_info(struct mem_ctl_info *mci,
 171                                  struct x38_error_info *info)
 172 {
 173         struct pci_dev *pdev;
 174         void __iomem *window = mci->pvt_info;
 175
 176         pdev = to_pci_dev(mci->dev);
 177
 178         /*
 179          * This is a mess because there is no atomic way to read all the
 180          * registers at once and the registers can transition from CE being
 181          * overwritten by UE.
 182          */
 183         pci_read_config_word(pdev, X38_ERRSTS, &info->errsts);
 184         if (!(info->errsts & X38_ERRSTS_BITS))
 185                 return;
 186
 187         info->eccerrlog[0] = x38_readq(window + X38_C0ECCERRLOG);
 188         if (x38_channel_num == 2)
 189                 info->eccerrlog[1] = x38_readq(window + X38_C1ECCERRLOG);
 190
 191         pci_read_config_word(pdev, X38_ERRSTS, &info->errsts2);
 192
 193         /*
 194          * If the error is the same for both reads then the first set
 195          * of reads is valid.  If there is a change then there is a CE
 196          * with no info and the second set of reads is valid and
 197          * should be UE info.
 198          */
 199         if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) {
 200                 info->eccerrlog[0] = x38_readq(window + X38_C0ECCERRLOG);
 201                 if (x38_channel_num == 2)
 202                         info->eccerrlog[1] =
 203                                 x38_readq(window + X38_C1ECCERRLOG);
 204         }
 205
 206         x38_clear_error_info(mci);
 207 }
 208
 209 static void x38_process_error_info(struct mem_ctl_info *mci,
 210                                 struct x38_error_info *info)
 211 {
 212         int channel;
 213         u64 log;
 214
 215         if (!(info->errsts & X38_ERRSTS_BITS))
 216                 return;
 217
 218         if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) {
 219                 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 220                 info->errsts = info->errsts2;
 221         }
 222
 223         for (channel = 0; channel < x38_channel_num; channel++) {
 224                 log = info->eccerrlog[channel];
 225                 if (log & X38_ECCERRLOG_UE) {
 226                         edac_mc_handle_ue(mci, 0, 0,
 227                                 eccerrlog_row(channel, log), "x38 UE");
 228                 } else if (log & X38_ECCERRLOG_CE) {
 229                         edac_mc_handle_ce(mci, 0, 0,
 230                                 eccerrlog_syndrome(log),
 231                                 eccerrlog_row(channel, log), 0, "x38 CE");
 232                 }
 233         }
 234 }
 235
 236 static void x38_check(struct mem_ctl_info *mci)
 237 {
 238         struct x38_error_info info;
 239
 240         debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 241         x38_get_and_clear_error_info(mci, &info);
 242         x38_process_error_info(mci, &info);
 243 }
 244
 245
 246 void __iomem *x38_map_mchbar(struct pci_dev *pdev)
 247 {
 248         union {
 249                 u64 mchbar;
 250                 struct {
 251                         u32 mchbar_low;
 252                         u32 mchbar_high;
 253                 };
 254         } u;
 255         void __iomem *window;
 256
 257         pci_read_config_dword(pdev, X38_MCHBAR_LOW, &u.mchbar_low);
 258         pci_write_config_dword(pdev, X38_MCHBAR_LOW, u.mchbar_low | 0x1);
 259         pci_read_config_dword(pdev, X38_MCHBAR_HIGH, &u.mchbar_high);
 260         u.mchbar &= X38_MCHBAR_MASK;
 261
 262         if (u.mchbar != (resource_size_t)u.mchbar) {
 263                 printk(KERN_ERR
 264                         "x38: mmio space beyond accessible range (0x%llx)\n",
 265                         (unsigned long long)u.mchbar);
 266                 return NULL;
 267         }
 268
 269         window = ioremap_nocache(u.mchbar, X38_MMR_WINDOW_SIZE);
 270         if (!window)
 271                 printk(KERN_ERR "x38: cannot map mmio space at 0x%llx\n",
 272                         (unsigned long long)u.mchbar);
 273
 274         return window;
 275 }
 276
 277
 278 static void x38_get_drbs(void __iomem *window,
 279                         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])
 280 {
 281         int i;
 282
 283         for (i = 0; i < X38_RANKS_PER_CHANNEL; i++) {
 284                 drbs[0][i] = readw(window + X38_C0DRB + 2*i) & X38_DRB_MASK;
 285                 drbs[1][i] = readw(window + X38_C1DRB + 2*i) & X38_DRB_MASK;
 286         }
 287 }
 288
 289 static bool x38_is_stacked(struct pci_dev *pdev,
 290                         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL])
 291 {
 292         u16 tom;
 293
 294         pci_read_config_word(pdev, X38_TOM, &tom);
 295         tom &= X38_TOM_MASK;
 296
 297         return drbs[X38_CHANNELS - 1][X38_RANKS_PER_CHANNEL - 1] == tom;
 298 }
 299
 300 static unsigned long drb_to_nr_pages(
 301                         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL],
 302                         bool stacked, int channel, int rank)
 303 {
 304         int n;
 305
 306         n = drbs[channel][rank];
 307         if (rank > 0)
 308                 n -= drbs[channel][rank - 1];
 309         if (stacked && (channel == 1) && drbs[channel][rank] ==
 310                                 drbs[channel][X38_RANKS_PER_CHANNEL - 1]) {
 311                 n -= drbs[0][X38_RANKS_PER_CHANNEL - 1];
 312         }
 313
 314         n <<= (X38_DRB_SHIFT - PAGE_SHIFT);
 315         return n;
 316 }
 317
 318 static int x38_probe1(struct pci_dev *pdev, int dev_idx)
 319 {
 320         int rc;
 321         int i;
 322         struct mem_ctl_info *mci = NULL;
 323         unsigned long last_page;
 324         u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL];
 325         bool stacked;
 326         void __iomem *window;
 327
 328         debugf0("MC: %s()\n", __func__);
 329
 330         window = x38_map_mchbar(pdev);
 331         if (!window)
 332                 return -ENODEV;
 333
 334         x38_get_drbs(window, drbs);
 335
 336         how_many_channel(pdev);
 337
 338         /* FIXME: unconventional pvt_info usage */
 339         mci = edac_mc_alloc(0, X38_RANKS, x38_channel_num, 0);
 340         if (!mci)
 341                 return -ENOMEM;
 342
 343         debugf3("MC: %s(): init mci\n", __func__);
 344
 345         mci->dev = &pdev->dev;
 346         mci->mtype_cap = MEM_FLAG_DDR2;
 347
 348         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 349         mci->edac_cap = EDAC_FLAG_SECDED;
 350
 351         mci->mod_name = EDAC_MOD_STR;
 352         mci->mod_ver = X38_REVISION;
 353         mci->ctl_name = x38_devs[dev_idx].ctl_name;
 354         mci->dev_name = pci_name(pdev);
 355         mci->edac_check = x38_check;
 356         mci->ctl_page_to_phys = NULL;
 357         mci->pvt_info = window;
 358
 359         stacked = x38_is_stacked(pdev, drbs);
 360
 361         /*
 362          * The dram rank boundary (DRB) reg values are boundary addresses
 363          * for each DRAM rank with a granularity of 64MB.  DRB regs are
 364          * cumulative; the last one will contain the total memory
 365          * contained in all ranks.
 366          */
 367         last_page = -1UL;
 368         for (i = 0; i < mci->nr_csrows; i++) {
 369                 unsigned long nr_pages;
 370                 struct csrow_info *csrow = &mci->csrows[i];
 371
 372                 nr_pages = drb_to_nr_pages(drbs, stacked,
 373                         i / X38_RANKS_PER_CHANNEL,
 374                         i % X38_RANKS_PER_CHANNEL);
 375
 376                 if (nr_pages == 0) {
 377                         csrow->mtype = MEM_EMPTY;
 378                         continue;
 379                 }
 380
 381                 csrow->first_page = last_page + 1;
 382                 last_page += nr_pages;
 383                 csrow->last_page = last_page;
 384                 csrow->nr_pages = nr_pages;
 385
 386                 csrow->grain = nr_pages << PAGE_SHIFT;
 387                 csrow->mtype = MEM_DDR2;
 388                 csrow->dtype = DEV_UNKNOWN;
 389                 csrow->edac_mode = EDAC_UNKNOWN;
 390         }
 391
 392         x38_clear_error_info(mci);
 393
 394         rc = -ENODEV;
 395         if (edac_mc_add_mc(mci)) {
 396                 debugf3("MC: %s(): failed edac_mc_add_mc()\n", __func__);
 397                 goto fail;
 398         }
 399
 400         /* get this far and it's successful */
 401         debugf3("MC: %s(): success\n", __func__);
 402         return 0;
 403
 404 fail:
 405         iounmap(window);
 406         if (mci)
 407                 edac_mc_free(mci);
 408
 409         return rc;
 410 }
 411
 412 static int __devinit x38_init_one(struct pci_dev *pdev,
 413                                 const struct pci_device_id *ent)
 414 {
 415         int rc;
 416
 417         debugf0("MC: %s()\n", __func__);
 418
 419         if (pci_enable_device(pdev) < 0)
 420                 return -EIO;
 421
 422         rc = x38_probe1(pdev, ent->driver_data);
 423         if (!mci_pdev)
 424                 mci_pdev = pci_dev_get(pdev);
 425
 426         return rc;
 427 }
 428
 429 static void __devexit x38_remove_one(struct pci_dev *pdev)
 430 {
 431         struct mem_ctl_info *mci;
 432
 433         debugf0("%s()\n", __func__);
 434
 435         mci = edac_mc_del_mc(&pdev->dev);
 436         if (!mci)
 437                 return;
 438
 439         iounmap(mci->pvt_info);
 440
 441         edac_mc_free(mci);
 442 }
 443
 444 static const struct pci_device_id x38_pci_tbl[] __devinitdata = {
 445         {
 446          PCI_VEND_DEV(INTEL, X38_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 447          X38},
 448         {
 449          0,
 450          }                      /* 0 terminated list. */
 451 };
 452
 453 MODULE_DEVICE_TABLE(pci, x38_pci_tbl);
 454
 455 static struct pci_driver x38_driver = {
 456         .name = EDAC_MOD_STR,
 457         .probe = x38_init_one,
 458         .remove = __devexit_p(x38_remove_one),
 459         .id_table = x38_pci_tbl,
 460 };
 461
 462 static int __init x38_init(void)
 463 {
 464         int pci_rc;
 465
 466         debugf3("MC: %s()\n", __func__);
 467
 468         /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 469         opstate_init();
 470
 471         pci_rc = pci_register_driver(&x38_driver);
 472         if (pci_rc < 0)
 473                 goto fail0;
 474
 475         if (!mci_pdev) {
 476                 x38_registered = 0;
 477                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 478                                         PCI_DEVICE_ID_INTEL_X38_HB, NULL);
 479                 if (!mci_pdev) {
 480                         debugf0("x38 pci_get_device fail\n");
 481                         pci_rc = -ENODEV;
 482                         goto fail1;
 483                 }
 484
 485                 pci_rc = x38_init_one(mci_pdev, x38_pci_tbl);
 486                 if (pci_rc < 0) {
 487                         debugf0("x38 init fail\n");
 488                         pci_rc = -ENODEV;
 489                         goto fail1;
 490                 }
 491         }
 492
 493         return 0;
 494
 495 fail1:
 496         pci_unregister_driver(&x38_driver);
 497
 498 fail0:
 499         if (mci_pdev)
 500                 pci_dev_put(mci_pdev);
 501
 502         return pci_rc;
 503 }
 504
 505 static void __exit x38_exit(void)
 506 {
 507         debugf3("MC: %s()\n", __func__);
 508
 509         pci_unregister_driver(&x38_driver);
 510         if (!x38_registered) {
 511                 x38_remove_one(mci_pdev);
 512                 pci_dev_put(mci_pdev);
 513         }
 514 }
 515
 516 module_init(x38_init);
 517 module_exit(x38_exit);
 518
 519 MODULE_LICENSE("GPL");
 520 MODULE_AUTHOR("Cluster Computing, Inc. Hitoshi Mitake");
 521 MODULE_DESCRIPTION("MC support for Intel X38 memory hub controllers");
 522
 523 module_param(edac_op_state, int, 0444);
 524 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");