drivers/edac/i3000_edac.c

   1 /*
   2  * Intel 3000/3010 Memory Controller kernel module
   3  * Copyright (C) 2007 Akamai Technologies, Inc.
   4  * Shamelessly copied from:
   5  *      Intel D82875P Memory Controller kernel module
   6  *      (C) 2003 Linux Networx (http://lnxi.com)
   7  *
   8  * This file may be distributed under the terms of the
   9  * GNU General Public License.
  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 #include "edac_core.h"
  18
  19 #define I3000_REVISION          "1.1"
  20
  21 #define EDAC_MOD_STR            "i3000_edac"
  22
  23 #define I3000_RANKS             8
  24 #define I3000_RANKS_PER_CHANNEL 4
  25 #define I3000_CHANNELS          2
  26
  27 /* Intel 3000 register addresses - device 0 function 0 - DRAM Controller */
  28
  29 #define I3000_MCHBAR            0x44    /* MCH Memory Mapped Register BAR */
  30 #define I3000_MCHBAR_MASK       0xffffc000
  31 #define I3000_MMR_WINDOW_SIZE   16384
  32
  33 #define I3000_EDEAP     0x70    /* Extended DRAM Error Address Pointer (8b)
  34                                  *
  35                                  * 7:1   reserved
  36                                  * 0     bit 32 of address
  37                                  */
  38 #define I3000_DEAP      0x58    /* DRAM Error Address Pointer (32b)
  39                                  *
  40                                  * 31:7  address
  41                                  * 6:1   reserved
  42                                  * 0     Error channel 0/1
  43                                  */
  44 #define I3000_DEAP_GRAIN                (1 << 7)
  45
  46 /*
  47  * Helper functions to decode the DEAP/EDEAP hardware registers.
  48  *
  49  * The type promotion here is deliberate; we're deriving an
  50  * unsigned long pfn and offset from hardware regs which are u8/u32.
  51  */
  52
  53 static inline unsigned long deap_pfn(u8 edeap, u32 deap)
  54 {
  55         deap >>= PAGE_SHIFT;
  56         deap |= (edeap & 1) << (32 - PAGE_SHIFT);
  57         return deap;
  58 }
  59
  60 static inline unsigned long deap_offset(u32 deap)
  61 {
  62         return deap & ~(I3000_DEAP_GRAIN - 1) & ~PAGE_MASK;
  63 }
  64
  65 static inline int deap_channel(u32 deap)
  66 {
  67         return deap & 1;
  68 }
  69
  70 #define I3000_DERRSYN   0x5c    /* DRAM Error Syndrome (8b)
  71                                  *
  72                                  *  7:0  DRAM ECC Syndrome
  73                                  */
  74
  75 #define I3000_ERRSTS    0xc8    /* Error Status Register (16b)
  76                                  *
  77                                  * 15:12 reserved
  78                                  * 11    MCH Thermal Sensor Event
  79                                  *         for SMI/SCI/SERR
  80                                  * 10    reserved
  81                                  *  9    LOCK to non-DRAM Memory Flag (LCKF)
  82                                  *  8    Received Refresh Timeout Flag (RRTOF)
  83                                  *  7:2  reserved
  84                                  *  1    Multi-bit DRAM ECC Error Flag (DMERR)
  85                                  *  0    Single-bit DRAM ECC Error Flag (DSERR)
  86                                  */
  87 #define I3000_ERRSTS_BITS       0x0b03  /* bits which indicate errors */
  88 #define I3000_ERRSTS_UE         0x0002
  89 #define I3000_ERRSTS_CE         0x0001
  90
  91 #define I3000_ERRCMD    0xca    /* Error Command (16b)
  92                                  *
  93                                  * 15:12 reserved
  94                                  * 11    SERR on MCH Thermal Sensor Event
  95                                  *         (TSESERR)
  96                                  * 10    reserved
  97                                  *  9    SERR on LOCK to non-DRAM Memory
  98                                  *         (LCKERR)
  99                                  *  8    SERR on DRAM Refresh Timeout
 100                                  *         (DRTOERR)
 101                                  *  7:2  reserved
 102                                  *  1    SERR Multi-Bit DRAM ECC Error
 103                                  *         (DMERR)
 104                                  *  0    SERR on Single-Bit ECC Error
 105                                  *         (DSERR)
 106                                  */
 107
 108 /* Intel  MMIO register space - device 0 function 0 - MMR space */
 109
 110 #define I3000_DRB_SHIFT 25      /* 32MiB grain */
 111
 112 #define I3000_C0DRB     0x100   /* Channel 0 DRAM Rank Boundary (8b x 4)
 113                                  *
 114                                  * 7:0   Channel 0 DRAM Rank Boundary Address
 115                                  */
 116 #define I3000_C1DRB     0x180   /* Channel 1 DRAM Rank Boundary (8b x 4)
 117                                  *
 118                                  * 7:0   Channel 1 DRAM Rank Boundary Address
 119                                  */
 120
 121 #define I3000_C0DRA     0x108   /* Channel 0 DRAM Rank Attribute (8b x 2)
 122                                  *
 123                                  * 7     reserved
 124                                  * 6:4   DRAM odd Rank Attribute
 125                                  * 3     reserved
 126                                  * 2:0   DRAM even Rank Attribute
 127                                  *
 128                                  * Each attribute defines the page
 129                                  * size of the corresponding rank:
 130                                  *     000: unpopulated
 131                                  *     001: reserved
 132                                  *     010: 4 KB
 133                                  *     011: 8 KB
 134                                  *     100: 16 KB
 135                                  *     Others: reserved
 136                                  */
 137 #define I3000_C1DRA     0x188   /* Channel 1 DRAM Rank Attribute (8b x 2) */
 138
 139 static inline unsigned char odd_rank_attrib(unsigned char dra)
 140 {
 141         return (dra & 0x70) >> 4;
 142 }
 143
 144 static inline unsigned char even_rank_attrib(unsigned char dra)
 145 {
 146         return dra & 0x07;
 147 }
 148
 149 #define I3000_C0DRC0    0x120   /* DRAM Controller Mode 0 (32b)
 150                                  *
 151                                  * 31:30 reserved
 152                                  * 29    Initialization Complete (IC)
 153                                  * 28:11 reserved
 154                                  * 10:8  Refresh Mode Select (RMS)
 155                                  * 7     reserved
 156                                  * 6:4   Mode Select (SMS)
 157                                  * 3:2   reserved
 158                                  * 1:0   DRAM Type (DT)
 159                                  */
 160
 161 #define I3000_C0DRC1    0x124   /* DRAM Controller Mode 1 (32b)
 162                                  *
 163                                  * 31    Enhanced Addressing Enable (ENHADE)
 164                                  * 30:0  reserved
 165                                  */
 166
 167 enum i3000p_chips {
 168         I3000 = 0,
 169 };
 170
 171 struct i3000_dev_info {
 172         const char *ctl_name;
 173 };
 174
 175 struct i3000_error_info {
 176         u16 errsts;
 177         u8 derrsyn;
 178         u8 edeap;
 179         u32 deap;
 180         u16 errsts2;
 181 };
 182
 183 static const struct i3000_dev_info i3000_devs[] = {
 184         [I3000] = {
 185                 .ctl_name = "i3000"},
 186 };
 187
 188 static struct pci_dev *mci_pdev;
 189 static int i3000_registered = 1;
 190 static struct edac_pci_ctl_info *i3000_pci;
 191
 192 static void i3000_get_error_info(struct mem_ctl_info *mci,
 193                                  struct i3000_error_info *info)
 194 {
 195         struct pci_dev *pdev;
 196
 197         pdev = to_pci_dev(mci->dev);
 198
 199         /*
 200          * This is a mess because there is no atomic way to read all the
 201          * registers at once and the registers can transition from CE being
 202          * overwritten by UE.
 203          */
 204         pci_read_config_word(pdev, I3000_ERRSTS, &info->errsts);
 205         if (!(info->errsts & I3000_ERRSTS_BITS))
 206                 return;
 207         pci_read_config_byte(pdev, I3000_EDEAP, &info->edeap);
 208         pci_read_config_dword(pdev, I3000_DEAP, &info->deap);
 209         pci_read_config_byte(pdev, I3000_DERRSYN, &info->derrsyn);
 210         pci_read_config_word(pdev, I3000_ERRSTS, &info->errsts2);
 211
 212         /*
 213          * If the error is the same for both reads then the first set
 214          * of reads is valid.  If there is a change then there is a CE
 215          * with no info and the second set of reads is valid and
 216          * should be UE info.
 217          */
 218         if ((info->errsts ^ info->errsts2) & I3000_ERRSTS_BITS) {
 219                 pci_read_config_byte(pdev, I3000_EDEAP, &info->edeap);
 220                 pci_read_config_dword(pdev, I3000_DEAP, &info->deap);
 221                 pci_read_config_byte(pdev, I3000_DERRSYN, &info->derrsyn);
 222         }
 223
 224         /*
 225          * Clear any error bits.
 226          * (Yes, we really clear bits by writing 1 to them.)
 227          */
 228         pci_write_bits16(pdev, I3000_ERRSTS, I3000_ERRSTS_BITS,
 229                          I3000_ERRSTS_BITS);
 230 }
 231
 232 static int i3000_process_error_info(struct mem_ctl_info *mci,
 233                                 struct i3000_error_info *info,
 234                                 int handle_errors)
 235 {
 236         int row, multi_chan, channel;
 237         unsigned long pfn, offset;
 238
 239         multi_chan = mci->csrows[0].nr_channels - 1;
 240
 241         if (!(info->errsts & I3000_ERRSTS_BITS))
 242                 return 0;
 243
 244         if (!handle_errors)
 245                 return 1;
 246
 247         if ((info->errsts ^ info->errsts2) & I3000_ERRSTS_BITS) {
 248                 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 249                 info->errsts = info->errsts2;
 250         }
 251
 252         pfn = deap_pfn(info->edeap, info->deap);
 253         offset = deap_offset(info->deap);
 254         channel = deap_channel(info->deap);
 255
 256         row = edac_mc_find_csrow_by_page(mci, pfn);
 257
 258         if (info->errsts & I3000_ERRSTS_UE)
 259                 edac_mc_handle_ue(mci, pfn, offset, row, "i3000 UE");
 260         else
 261                 edac_mc_handle_ce(mci, pfn, offset, info->derrsyn, row,
 262                                 multi_chan ? channel : 0, "i3000 CE");
 263
 264         return 1;
 265 }
 266
 267 static void i3000_check(struct mem_ctl_info *mci)
 268 {
 269         struct i3000_error_info info;
 270
 271         debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 272         i3000_get_error_info(mci, &info);
 273         i3000_process_error_info(mci, &info, 1);
 274 }
 275
 276 static int i3000_is_interleaved(const unsigned char *c0dra,
 277                                 const unsigned char *c1dra,
 278                                 const unsigned char *c0drb,
 279                                 const unsigned char *c1drb)
 280 {
 281         int i;
 282
 283         /*
 284          * If the channels aren't populated identically then
 285          * we're not interleaved.
 286          */
 287         for (i = 0; i < I3000_RANKS_PER_CHANNEL / 2; i++)
 288                 if (odd_rank_attrib(c0dra[i]) != odd_rank_attrib(c1dra[i]) ||
 289                         even_rank_attrib(c0dra[i]) !=
 290                                                 even_rank_attrib(c1dra[i]))
 291                         return 0;
 292
 293         /*
 294          * If the rank boundaries for the two channels are different
 295          * then we're not interleaved.
 296          */
 297         for (i = 0; i < I3000_RANKS_PER_CHANNEL; i++)
 298                 if (c0drb[i] != c1drb[i])
 299                         return 0;
 300
 301         return 1;
 302 }
 303
 304 static int i3000_probe1(struct pci_dev *pdev, int dev_idx)
 305 {
 306         int rc;
 307         int i;
 308         struct mem_ctl_info *mci = NULL;
 309         unsigned long last_cumul_size;
 310         int interleaved, nr_channels;
 311         unsigned char dra[I3000_RANKS / 2], drb[I3000_RANKS];
 312         unsigned char *c0dra = dra, *c1dra = &dra[I3000_RANKS_PER_CHANNEL / 2];
 313         unsigned char *c0drb = drb, *c1drb = &drb[I3000_RANKS_PER_CHANNEL];
 314         unsigned long mchbar;
 315         void __iomem *window;
 316
 317         debugf0("MC: %s()\n", __func__);
 318
 319         pci_read_config_dword(pdev, I3000_MCHBAR, (u32 *) & mchbar);
 320         mchbar &= I3000_MCHBAR_MASK;
 321         window = ioremap_nocache(mchbar, I3000_MMR_WINDOW_SIZE);
 322         if (!window) {
 323                 printk(KERN_ERR "i3000: cannot map mmio space at 0x%lx\n",
 324                         mchbar);
 325                 return -ENODEV;
 326         }
 327
 328         c0dra[0] = readb(window + I3000_C0DRA + 0);     /* ranks 0,1 */
 329         c0dra[1] = readb(window + I3000_C0DRA + 1);     /* ranks 2,3 */
 330         c1dra[0] = readb(window + I3000_C1DRA + 0);     /* ranks 0,1 */
 331         c1dra[1] = readb(window + I3000_C1DRA + 1);     /* ranks 2,3 */
 332
 333         for (i = 0; i < I3000_RANKS_PER_CHANNEL; i++) {
 334                 c0drb[i] = readb(window + I3000_C0DRB + i);
 335                 c1drb[i] = readb(window + I3000_C1DRB + i);
 336         }
 337
 338         iounmap(window);
 339
 340         /*
 341          * Figure out how many channels we have.
 342          *
 343          * If we have what the datasheet calls "asymmetric channels"
 344          * (essentially the same as what was called "virtual single
 345          * channel mode" in the i82875) then it's a single channel as
 346          * far as EDAC is concerned.
 347          */
 348         interleaved = i3000_is_interleaved(c0dra, c1dra, c0drb, c1drb);
 349         nr_channels = interleaved ? 2 : 1;
 350         mci = edac_mc_alloc(0, I3000_RANKS / nr_channels, nr_channels, 0);
 351         if (!mci)
 352                 return -ENOMEM;
 353
 354         debugf3("MC: %s(): init mci\n", __func__);
 355
 356         mci->dev = &pdev->dev;
 357         mci->mtype_cap = MEM_FLAG_DDR2;
 358
 359         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 360         mci->edac_cap = EDAC_FLAG_SECDED;
 361
 362         mci->mod_name = EDAC_MOD_STR;
 363         mci->mod_ver = I3000_REVISION;
 364         mci->ctl_name = i3000_devs[dev_idx].ctl_name;
 365         mci->dev_name = pci_name(pdev);
 366         mci->edac_check = i3000_check;
 367         mci->ctl_page_to_phys = NULL;
 368
 369         /*
 370          * The dram rank boundary (DRB) reg values are boundary addresses
 371          * for each DRAM rank with a granularity of 32MB.  DRB regs are
 372          * cumulative; the last one will contain the total memory
 373          * contained in all ranks.
 374          *
 375          * If we're in interleaved mode then we're only walking through
 376          * the ranks of controller 0, so we double all the values we see.
 377          */
 378         for (last_cumul_size = i = 0; i < mci->nr_csrows; i++) {
 379                 u8 value;
 380                 u32 cumul_size;
 381                 struct csrow_info *csrow = &mci->csrows[i];
 382
 383                 value = drb[i];
 384                 cumul_size = value << (I3000_DRB_SHIFT - PAGE_SHIFT);
 385                 if (interleaved)
 386                         cumul_size <<= 1;
 387                 debugf3("MC: %s(): (%d) cumul_size 0x%x\n",
 388                         __func__, i, cumul_size);
 389                 if (cumul_size == last_cumul_size) {
 390                         csrow->mtype = MEM_EMPTY;
 391                         continue;
 392                 }
 393
 394                 csrow->first_page = last_cumul_size;
 395                 csrow->last_page = cumul_size - 1;
 396                 csrow->nr_pages = cumul_size - last_cumul_size;
 397                 last_cumul_size = cumul_size;
 398                 csrow->grain = I3000_DEAP_GRAIN;
 399                 csrow->mtype = MEM_DDR2;
 400                 csrow->dtype = DEV_UNKNOWN;
 401                 csrow->edac_mode = EDAC_UNKNOWN;
 402         }
 403
 404         /*
 405          * Clear any error bits.
 406          * (Yes, we really clear bits by writing 1 to them.)
 407          */
 408         pci_write_bits16(pdev, I3000_ERRSTS, I3000_ERRSTS_BITS,
 409                          I3000_ERRSTS_BITS);
 410
 411         rc = -ENODEV;
 412         if (edac_mc_add_mc(mci)) {
 413                 debugf3("MC: %s(): failed edac_mc_add_mc()\n", __func__);
 414                 goto fail;
 415         }
 416
 417         /* allocating generic PCI control info */
 418         i3000_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
 419         if (!i3000_pci) {
 420                 printk(KERN_WARNING
 421                         "%s(): Unable to create PCI control\n",
 422                         __func__);
 423                 printk(KERN_WARNING
 424                         "%s(): PCI error report via EDAC not setup\n",
 425                         __func__);
 426         }
 427
 428         /* get this far and it's successful */
 429         debugf3("MC: %s(): success\n", __func__);
 430         return 0;
 431
 432 fail:
 433         if (mci)
 434                 edac_mc_free(mci);
 435
 436         return rc;
 437 }
 438
 439 /* returns count (>= 0), or negative on error */
 440 static int __devinit i3000_init_one(struct pci_dev *pdev,
 441                                 const struct pci_device_id *ent)
 442 {
 443         int rc;
 444
 445         debugf0("MC: %s()\n", __func__);
 446
 447         if (pci_enable_device(pdev) < 0)
 448                 return -EIO;
 449
 450         rc = i3000_probe1(pdev, ent->driver_data);
 451         if (!mci_pdev)
 452                 mci_pdev = pci_dev_get(pdev);
 453
 454         return rc;
 455 }
 456
 457 static void __devexit i3000_remove_one(struct pci_dev *pdev)
 458 {
 459         struct mem_ctl_info *mci;
 460
 461         debugf0("%s()\n", __func__);
 462
 463         if (i3000_pci)
 464                 edac_pci_release_generic_ctl(i3000_pci);
 465
 466         mci = edac_mc_del_mc(&pdev->dev);
 467         if (!mci)
 468                 return;
 469
 470         edac_mc_free(mci);
 471 }
 472
 473 static DEFINE_PCI_DEVICE_TABLE(i3000_pci_tbl) = {
 474         {
 475          PCI_VEND_DEV(INTEL, 3000_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 476          I3000},
 477         {
 478          0,
 479          }                      /* 0 terminated list. */
 480 };
 481
 482 MODULE_DEVICE_TABLE(pci, i3000_pci_tbl);
 483
 484 static struct pci_driver i3000_driver = {
 485         .name = EDAC_MOD_STR,
 486         .probe = i3000_init_one,
 487         .remove = __devexit_p(i3000_remove_one),
 488         .id_table = i3000_pci_tbl,
 489 };
 490
 491 static int __init i3000_init(void)
 492 {
 493         int pci_rc;
 494
 495         debugf3("MC: %s()\n", __func__);
 496
 497        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 498        opstate_init();
 499
 500         pci_rc = pci_register_driver(&i3000_driver);
 501         if (pci_rc < 0)
 502                 goto fail0;
 503
 504         if (!mci_pdev) {
 505                 i3000_registered = 0;
 506                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 507                                         PCI_DEVICE_ID_INTEL_3000_HB, NULL);
 508                 if (!mci_pdev) {
 509                         debugf0("i3000 pci_get_device fail\n");
 510                         pci_rc = -ENODEV;
 511                         goto fail1;
 512                 }
 513
 514                 pci_rc = i3000_init_one(mci_pdev, i3000_pci_tbl);
 515                 if (pci_rc < 0) {
 516                         debugf0("i3000 init fail\n");
 517                         pci_rc = -ENODEV;
 518                         goto fail1;
 519                 }
 520         }
 521
 522         return 0;
 523
 524 fail1:
 525         pci_unregister_driver(&i3000_driver);
 526
 527 fail0:
 528         if (mci_pdev)
 529                 pci_dev_put(mci_pdev);
 530
 531         return pci_rc;
 532 }
 533
 534 static void __exit i3000_exit(void)
 535 {
 536         debugf3("MC: %s()\n", __func__);
 537
 538         pci_unregister_driver(&i3000_driver);
 539         if (!i3000_registered) {
 540                 i3000_remove_one(mci_pdev);
 541                 pci_dev_put(mci_pdev);
 542         }
 543 }
 544
 545 module_init(i3000_init);
 546 module_exit(i3000_exit);
 547
 548 MODULE_LICENSE("GPL");
 549 MODULE_AUTHOR("Akamai Technologies Arthur Ulfeldt/Jason Uhlenkott");
 550 MODULE_DESCRIPTION("MC support for Intel 3000 memory hub controllers");
 551
 552 module_param(edac_op_state, int, 0444);
 553 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");