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/slab.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 #define I3000_DEAP_PFN(edeap, deap)     ((((edeap) & 1) << (32 - PAGE_SHIFT)) | \
  46                                         ((deap) >> PAGE_SHIFT))
  47 #define I3000_DEAP_OFFSET(deap)         ((deap) & ~(I3000_DEAP_GRAIN-1) & ~PAGE_MASK)
  48 #define I3000_DEAP_CHANNEL(deap)        ((deap) & 1)
  49
  50 #define I3000_DERRSYN           0x5c    /* DRAM Error Syndrome (8b)
  51                                          *
  52                                          *  7:0  DRAM ECC Syndrome
  53                                          */
  54
  55 #define I3000_ERRSTS            0xc8    /* Error Status Register (16b)
  56                                          *
  57                                          * 15:12 reserved
  58                                          * 11    MCH Thermal Sensor Event for SMI/SCI/SERR
  59                                          * 10    reserved
  60                                          *  9    LOCK to non-DRAM Memory Flag (LCKF)
  61                                          *  8    Received Refresh Timeout Flag (RRTOF)
  62                                          *  7:2  reserved
  63                                          *  1    Multiple-bit DRAM ECC Error Flag (DMERR)
  64                                          *  0    Single-bit DRAM ECC Error Flag (DSERR)
  65                                          */
  66 #define I3000_ERRSTS_BITS       0x0b03  /* bits which indicate errors */
  67 #define I3000_ERRSTS_UE         0x0002
  68 #define I3000_ERRSTS_CE         0x0001
  69
  70 #define I3000_ERRCMD            0xca    /* Error Command (16b)
  71                                          *
  72                                          * 15:12 reserved
  73                                          * 11    SERR on MCH Thermal Sensor Event (TSESERR)
  74                                          * 10    reserved
  75                                          *  9    SERR on LOCK to non-DRAM Memory (LCKERR)
  76                                          *  8    SERR on DRAM Refresh Timeout (DRTOERR)
  77                                          *  7:2  reserved
  78                                          *  1    SERR Multiple-Bit DRAM ECC Error (DMERR)
  79                                          *  0    SERR on Single-Bit ECC Error (DSERR)
  80                                          */
  81
  82 /* Intel  MMIO register space - device 0 function 0 - MMR space */
  83
  84 #define I3000_DRB_SHIFT 25      /* 32MiB grain */
  85
  86 #define I3000_C0DRB             0x100   /* Channel 0 DRAM Rank Boundary (8b x 4)
  87                                          *
  88                                          * 7:0   Channel 0 DRAM Rank Boundary Address
  89                                          */
  90 #define I3000_C1DRB             0x180   /* Channel 1 DRAM Rank Boundary (8b x 4)
  91                                          *
  92                                          * 7:0   Channel 1 DRAM Rank Boundary Address
  93                                          */
  94
  95 #define I3000_C0DRA             0x108   /* Channel 0 DRAM Rank Attribute (8b x 2)
  96                                          *
  97                                          * 7     reserved
  98                                          * 6:4   DRAM odd Rank Attribute
  99                                          * 3     reserved
 100                                          * 2:0   DRAM even Rank Attribute
 101                                          *
 102                                          * Each attribute defines the page
 103                                          * size of the corresponding rank:
 104                                          *     000: unpopulated
 105                                          *     001: reserved
 106                                          *     010: 4 KB
 107                                          *     011: 8 KB
 108                                          *     100: 16 KB
 109                                          *     Others: reserved
 110                                          */
 111 #define I3000_C1DRA             0x188   /* Channel 1 DRAM Rank Attribute (8b x 2) */
 112 #define ODD_RANK_ATTRIB(dra) (((dra) & 0x70) >> 4)
 113 #define EVEN_RANK_ATTRIB(dra) ((dra) & 0x07)
 114
 115 #define I3000_C0DRC0            0x120   /* DRAM Controller Mode 0 (32b)
 116                                          *
 117                                          * 31:30 reserved
 118                                          * 29    Initialization Complete (IC)
 119                                          * 28:11 reserved
 120                                          * 10:8  Refresh Mode Select (RMS)
 121                                          * 7     reserved
 122                                          * 6:4   Mode Select (SMS)
 123                                          * 3:2   reserved
 124                                          * 1:0   DRAM Type (DT)
 125                                          */
 126
 127 #define I3000_C0DRC1            0x124   /* DRAM Controller Mode 1 (32b)
 128                                          *
 129                                          * 31    Enhanced Addressing Enable (ENHADE)
 130                                          * 30:0  reserved
 131                                          */
 132
 133 enum i3000p_chips {
 134         I3000 = 0,
 135 };
 136
 137 struct i3000_dev_info {
 138         const char *ctl_name;
 139 };
 140
 141 struct i3000_error_info {
 142         u16 errsts;
 143         u8 derrsyn;
 144         u8 edeap;
 145         u32 deap;
 146         u16 errsts2;
 147 };
 148
 149 static const struct i3000_dev_info i3000_devs[] = {
 150         [I3000] = {
 151                    .ctl_name = "i3000"},
 152 };
 153
 154 static struct pci_dev *mci_pdev = NULL;
 155 static int i3000_registered = 1;
 156
 157 static void i3000_get_error_info(struct mem_ctl_info *mci,
 158                                  struct i3000_error_info *info)
 159 {
 160         struct pci_dev *pdev;
 161
 162         pdev = to_pci_dev(mci->dev);
 163
 164         /*
 165          * This is a mess because there is no atomic way to read all the
 166          * registers at once and the registers can transition from CE being
 167          * overwritten by UE.
 168          */
 169         pci_read_config_word(pdev, I3000_ERRSTS, &info->errsts);
 170         if (!(info->errsts & I3000_ERRSTS_BITS))
 171                 return;
 172         pci_read_config_byte(pdev, I3000_EDEAP, &info->edeap);
 173         pci_read_config_dword(pdev, I3000_DEAP, &info->deap);
 174         pci_read_config_byte(pdev, I3000_DERRSYN, &info->derrsyn);
 175         pci_read_config_word(pdev, I3000_ERRSTS, &info->errsts2);
 176
 177         /*
 178          * If the error is the same for both reads then the first set
 179          * of reads is valid.  If there is a change then there is a CE
 180          * with no info and the second set of reads is valid and
 181          * should be UE info.
 182          */
 183         if ((info->errsts ^ info->errsts2) & I3000_ERRSTS_BITS) {
 184                 pci_read_config_byte(pdev, I3000_EDEAP, &info->edeap);
 185                 pci_read_config_dword(pdev, I3000_DEAP, &info->deap);
 186                 pci_read_config_byte(pdev, I3000_DERRSYN, &info->derrsyn);
 187         }
 188
 189         /* Clear any error bits.
 190          * (Yes, we really clear bits by writing 1 to them.)
 191          */
 192         pci_write_bits16(pdev, I3000_ERRSTS, I3000_ERRSTS_BITS,
 193                          I3000_ERRSTS_BITS);
 194 }
 195
 196 static int i3000_process_error_info(struct mem_ctl_info *mci,
 197                                     struct i3000_error_info *info,
 198                                     int handle_errors)
 199 {
 200         int row, multi_chan;
 201         int pfn, offset, channel;
 202
 203         multi_chan = mci->csrows[0].nr_channels - 1;
 204
 205         if (!(info->errsts & I3000_ERRSTS_BITS))
 206                 return 0;
 207
 208         if (!handle_errors)
 209                 return 1;
 210
 211         if ((info->errsts ^ info->errsts2) & I3000_ERRSTS_BITS) {
 212                 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 213                 info->errsts = info->errsts2;
 214         }
 215
 216         pfn = I3000_DEAP_PFN(info->edeap, info->deap);
 217         offset = I3000_DEAP_OFFSET(info->deap);
 218         channel = I3000_DEAP_CHANNEL(info->deap);
 219
 220         row = edac_mc_find_csrow_by_page(mci, pfn);
 221
 222         if (info->errsts & I3000_ERRSTS_UE)
 223                 edac_mc_handle_ue(mci, pfn, offset, row, "i3000 UE");
 224         else
 225                 edac_mc_handle_ce(mci, pfn, offset, info->derrsyn, row,
 226                                   multi_chan ? channel : 0, "i3000 CE");
 227
 228         return 1;
 229 }
 230
 231 static void i3000_check(struct mem_ctl_info *mci)
 232 {
 233         struct i3000_error_info info;
 234
 235         debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 236         i3000_get_error_info(mci, &info);
 237         i3000_process_error_info(mci, &info, 1);
 238 }
 239
 240 static int i3000_is_interleaved(const unsigned char *c0dra,
 241                                 const unsigned char *c1dra,
 242                                 const unsigned char *c0drb,
 243                                 const unsigned char *c1drb)
 244 {
 245         int i;
 246
 247         /* If the channels aren't populated identically then
 248          * we're not interleaved.
 249          */
 250         for (i = 0; i < I3000_RANKS_PER_CHANNEL / 2; i++)
 251                 if (ODD_RANK_ATTRIB(c0dra[i]) != ODD_RANK_ATTRIB(c1dra[i]) ||
 252                     EVEN_RANK_ATTRIB(c0dra[i]) != EVEN_RANK_ATTRIB(c1dra[i]))
 253                         return 0;
 254
 255         /* If the rank boundaries for the two channels are different
 256          * then we're not interleaved.
 257          */
 258         for (i = 0; i < I3000_RANKS_PER_CHANNEL; i++)
 259                 if (c0drb[i] != c1drb[i])
 260                         return 0;
 261
 262         return 1;
 263 }
 264
 265 static int i3000_probe1(struct pci_dev *pdev, int dev_idx)
 266 {
 267         int rc;
 268         int i;
 269         struct mem_ctl_info *mci = NULL;
 270         unsigned long last_cumul_size;
 271         int interleaved, nr_channels;
 272         unsigned char dra[I3000_RANKS / 2], drb[I3000_RANKS];
 273         unsigned char *c0dra = dra, *c1dra = &dra[I3000_RANKS_PER_CHANNEL / 2];
 274         unsigned char *c0drb = drb, *c1drb = &drb[I3000_RANKS_PER_CHANNEL];
 275         unsigned long mchbar;
 276         void *window;
 277
 278         debugf0("MC: %s()\n", __func__);
 279
 280         pci_read_config_dword(pdev, I3000_MCHBAR, (u32 *) & mchbar);
 281         mchbar &= I3000_MCHBAR_MASK;
 282         window = ioremap_nocache(mchbar, I3000_MMR_WINDOW_SIZE);
 283         if (!window) {
 284                 printk(KERN_ERR "i3000: cannot map mmio space at 0x%lx\n",
 285                        mchbar);
 286                 return -ENODEV;
 287         }
 288
 289         c0dra[0] = readb(window + I3000_C0DRA + 0);     /* ranks 0,1 */
 290         c0dra[1] = readb(window + I3000_C0DRA + 1);     /* ranks 2,3 */
 291         c1dra[0] = readb(window + I3000_C1DRA + 0);     /* ranks 0,1 */
 292         c1dra[1] = readb(window + I3000_C1DRA + 1);     /* ranks 2,3 */
 293
 294         for (i = 0; i < I3000_RANKS_PER_CHANNEL; i++) {
 295                 c0drb[i] = readb(window + I3000_C0DRB + i);
 296                 c1drb[i] = readb(window + I3000_C1DRB + i);
 297         }
 298
 299         iounmap(window);
 300
 301         /* Figure out how many channels we have.
 302          *
 303          * If we have what the datasheet calls "asymmetric channels"
 304          * (essentially the same as what was called "virtual single
 305          * channel mode" in the i82875) then it's a single channel as
 306          * far as EDAC is concerned.
 307          */
 308         interleaved = i3000_is_interleaved(c0dra, c1dra, c0drb, c1drb);
 309         nr_channels = interleaved ? 2 : 1;
 310         mci = edac_mc_alloc(0, I3000_RANKS / nr_channels, nr_channels);
 311         if (!mci)
 312                 return -ENOMEM;
 313
 314         debugf3("MC: %s(): init mci\n", __func__);
 315
 316         mci->dev = &pdev->dev;
 317         mci->mtype_cap = MEM_FLAG_DDR2;
 318
 319         mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 320         mci->edac_cap = EDAC_FLAG_SECDED;
 321
 322         mci->mod_name = EDAC_MOD_STR;
 323         mci->mod_ver = I3000_REVISION;
 324         mci->ctl_name = i3000_devs[dev_idx].ctl_name;
 325         mci->dev_name = pci_name(pdev);
 326         mci->edac_check = i3000_check;
 327         mci->ctl_page_to_phys = NULL;
 328
 329         /*
 330          * The dram rank boundary (DRB) reg values are boundary addresses
 331          * for each DRAM rank with a granularity of 32MB.  DRB regs are
 332          * cumulative; the last one will contain the total memory
 333          * contained in all ranks.
 334          *
 335          * If we're in interleaved mode then we're only walking through
 336          * the ranks of controller 0, so we double all the values we see.
 337          */
 338         for (last_cumul_size = i = 0; i < mci->nr_csrows; i++) {
 339                 u8 value;
 340                 u32 cumul_size;
 341                 struct csrow_info *csrow = &mci->csrows[i];
 342
 343                 value = drb[i];
 344                 cumul_size = value << (I3000_DRB_SHIFT - PAGE_SHIFT);
 345                 if (interleaved)
 346                         cumul_size <<= 1;
 347                 debugf3("MC: %s(): (%d) cumul_size 0x%x\n",
 348                         __func__, i, cumul_size);
 349                 if (cumul_size == last_cumul_size) {
 350                         csrow->mtype = MEM_EMPTY;
 351                         continue;
 352                 }
 353
 354                 csrow->first_page = last_cumul_size;
 355                 csrow->last_page = cumul_size - 1;
 356                 csrow->nr_pages = cumul_size - last_cumul_size;
 357                 last_cumul_size = cumul_size;
 358                 csrow->grain = I3000_DEAP_GRAIN;
 359                 csrow->mtype = MEM_DDR2;
 360                 csrow->dtype = DEV_UNKNOWN;
 361                 csrow->edac_mode = EDAC_UNKNOWN;
 362         }
 363
 364         /* Clear any error bits.
 365          * (Yes, we really clear bits by writing 1 to them.)
 366          */
 367         pci_write_bits16(pdev, I3000_ERRSTS, I3000_ERRSTS_BITS,
 368                          I3000_ERRSTS_BITS);
 369
 370         rc = -ENODEV;
 371         if (edac_mc_add_mc(mci, 0)) {
 372                 debugf3("MC: %s(): failed edac_mc_add_mc()\n", __func__);
 373                 goto fail;
 374         }
 375
 376         /* get this far and it's successful */
 377         debugf3("MC: %s(): success\n", __func__);
 378         return 0;
 379
 380       fail:
 381         if (mci)
 382                 edac_mc_free(mci);
 383
 384         return rc;
 385 }
 386
 387 /* returns count (>= 0), or negative on error */
 388 static int __devinit i3000_init_one(struct pci_dev *pdev,
 389                                     const struct pci_device_id *ent)
 390 {
 391         int rc;
 392
 393         debugf0("MC: %s()\n", __func__);
 394
 395         if (pci_enable_device(pdev) < 0)
 396                 return -EIO;
 397
 398         rc = i3000_probe1(pdev, ent->driver_data);
 399         if (mci_pdev == NULL)
 400                 mci_pdev = pci_dev_get(pdev);
 401
 402         return rc;
 403 }
 404
 405 static void __devexit i3000_remove_one(struct pci_dev *pdev)
 406 {
 407         struct mem_ctl_info *mci;
 408
 409         debugf0("%s()\n", __func__);
 410
 411         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
 412                 return;
 413
 414         edac_mc_free(mci);
 415 }
 416
 417 static const struct pci_device_id i3000_pci_tbl[] __devinitdata = {
 418         {
 419          PCI_VEND_DEV(INTEL, 3000_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 420          I3000},
 421         {
 422          0,
 423          }                      /* 0 terminated list. */
 424 };
 425
 426 MODULE_DEVICE_TABLE(pci, i3000_pci_tbl);
 427
 428 static struct pci_driver i3000_driver = {
 429         .name = EDAC_MOD_STR,
 430         .probe = i3000_init_one,
 431         .remove = __devexit_p(i3000_remove_one),
 432         .id_table = i3000_pci_tbl,
 433 };
 434
 435 static int __init i3000_init(void)
 436 {
 437         int pci_rc;
 438
 439         debugf3("MC: %s()\n", __func__);
 440         pci_rc = pci_register_driver(&i3000_driver);
 441         if (pci_rc < 0)
 442                 goto fail0;
 443
 444         if (mci_pdev == NULL) {
 445                 i3000_registered = 0;
 446                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 447                                           PCI_DEVICE_ID_INTEL_3000_HB, NULL);
 448                 if (!mci_pdev) {
 449                         debugf0("i3000 pci_get_device fail\n");
 450                         pci_rc = -ENODEV;
 451                         goto fail1;
 452                 }
 453
 454                 pci_rc = i3000_init_one(mci_pdev, i3000_pci_tbl);
 455                 if (pci_rc < 0) {
 456                         debugf0("i3000 init fail\n");
 457                         pci_rc = -ENODEV;
 458                         goto fail1;
 459                 }
 460         }
 461
 462         return 0;
 463
 464       fail1:
 465         pci_unregister_driver(&i3000_driver);
 466
 467       fail0:
 468         if (mci_pdev)
 469                 pci_dev_put(mci_pdev);
 470
 471         return pci_rc;
 472 }
 473
 474 static void __exit i3000_exit(void)
 475 {
 476         debugf3("MC: %s()\n", __func__);
 477
 478         pci_unregister_driver(&i3000_driver);
 479         if (!i3000_registered) {
 480                 i3000_remove_one(mci_pdev);
 481                 pci_dev_put(mci_pdev);
 482         }
 483 }
 484
 485 module_init(i3000_init);
 486 module_exit(i3000_exit);
 487
 488 MODULE_LICENSE("GPL");
 489 MODULE_AUTHOR("Akamai Technologies Arthur Ulfeldt/Jason Uhlenkott");
 490 MODULE_DESCRIPTION("MC support for Intel 3000 memory hub controllers");