drivers/edac/i82443bxgx_edac.c

   1 /*
   2  * Intel 82443BX/GX (440BX/GX chipset) Memory Controller EDAC kernel
   3  * module (C) 2006 Tim Small
   4  *
   5  * This file may be distributed under the terms of the GNU General
   6  * Public License.
   7  *
   8  * Written by Tim Small <tim@buttersideup.com>, based on work by Linux
   9  * Networx, Thayne Harbaugh, Dan Hollis <goemon at anime dot net> and
  10  * others.
  11  *
  12  * 440GX fix by Jason Uhlenkott <juhlenko@akamai.com>.
  13  *
  14  * Written with reference to 82443BX Host Bridge Datasheet:
  15  * http://download.intel.com/design/chipsets/datashts/29063301.pdf
  16  * references to this document given in [].
  17  *
  18  * This module doesn't support the 440LX, but it may be possible to
  19  * make it do so (the 440LX's register definitions are different, but
  20  * not completely so - I haven't studied them in enough detail to know
  21  * how easy this would be).
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/init.h>
  26
  27 #include <linux/pci.h>
  28 #include <linux/pci_ids.h>
  29
  30
  31 #include <linux/edac.h>
  32 #include "edac_core.h"
  33
  34 #define I82443_REVISION "0.1"
  35
  36 #define EDAC_MOD_STR    "i82443bxgx_edac"
  37
  38 /* The 82443BX supports SDRAM, or EDO (EDO for mobile only), "Memory
  39  * Size: 8 MB to 512 MB (1GB with Registered DIMMs) with eight memory
  40  * rows" "The 82443BX supports multiple-bit error detection and
  41  * single-bit error correction when ECC mode is enabled and
  42  * single/multi-bit error detection when correction is disabled.
  43  * During writes to the DRAM, the 82443BX generates ECC for the data
  44  * on a QWord basis. Partial QWord writes require a read-modify-write
  45  * cycle when ECC is enabled."
  46 */
  47
  48 /* "Additionally, the 82443BX ensures that the data is corrected in
  49  * main memory so that accumulation of errors is prevented. Another
  50  * error within the same QWord would result in a double-bit error
  51  * which is unrecoverable. This is known as hardware scrubbing since
  52  * it requires no software intervention to correct the data in memory."
  53  */
  54
  55 /* [Also see page 100 (section 4.3), "DRAM Interface"]
  56  * [Also see page 112 (section 4.6.1.4), ECC]
  57  */
  58
  59 #define I82443BXGX_NR_CSROWS 8
  60 #define I82443BXGX_NR_CHANS  1
  61 #define I82443BXGX_NR_DIMMS  4
  62
  63 /* 82443 PCI Device 0 */
  64 #define I82443BXGX_NBXCFG 0x50  /* 32bit register starting at this PCI
  65                                  * config space offset */
  66 #define I82443BXGX_NBXCFG_OFFSET_NON_ECCROW 24  /* Array of bits, zero if
  67                                                  * row is non-ECC */
  68 #define I82443BXGX_NBXCFG_OFFSET_DRAM_FREQ 12   /* 2 bits,00=100MHz,10=66 MHz */
  69
  70 #define I82443BXGX_NBXCFG_OFFSET_DRAM_INTEGRITY 7       /* 2 bits:       */
  71 #define I82443BXGX_NBXCFG_INTEGRITY_NONE   0x0  /* 00 = Non-ECC */
  72 #define I82443BXGX_NBXCFG_INTEGRITY_EC     0x1  /* 01 = EC (only) */
  73 #define I82443BXGX_NBXCFG_INTEGRITY_ECC    0x2  /* 10 = ECC */
  74 #define I82443BXGX_NBXCFG_INTEGRITY_SCRUB  0x3  /* 11 = ECC + HW Scrub */
  75
  76 #define I82443BXGX_NBXCFG_OFFSET_ECC_DIAG_ENABLE  6
  77
  78 /* 82443 PCI Device 0 */
  79 #define I82443BXGX_EAP   0x80   /* 32bit register starting at this PCI
  80                                  * config space offset, Error Address
  81                                  * Pointer Register */
  82 #define I82443BXGX_EAP_OFFSET_EAP  12   /* High 20 bits of error address */
  83 #define I82443BXGX_EAP_OFFSET_MBE  BIT(1)       /* Err at EAP was multi-bit (W1TC) */
  84 #define I82443BXGX_EAP_OFFSET_SBE  BIT(0)       /* Err at EAP was single-bit (W1TC) */
  85
  86 #define I82443BXGX_ERRCMD  0x90 /* 8bit register starting at this PCI
  87                                  * config space offset. */
  88 #define I82443BXGX_ERRCMD_OFFSET_SERR_ON_MBE BIT(1)     /* 1 = enable */
  89 #define I82443BXGX_ERRCMD_OFFSET_SERR_ON_SBE BIT(0)     /* 1 = enable */
  90
  91 #define I82443BXGX_ERRSTS  0x91 /* 16bit register starting at this PCI
  92                                  * config space offset. */
  93 #define I82443BXGX_ERRSTS_OFFSET_MBFRE 5        /* 3 bits - first err row multibit */
  94 #define I82443BXGX_ERRSTS_OFFSET_MEF   BIT(4)   /* 1 = MBE occurred */
  95 #define I82443BXGX_ERRSTS_OFFSET_SBFRE 1        /* 3 bits - first err row singlebit */
  96 #define I82443BXGX_ERRSTS_OFFSET_SEF   BIT(0)   /* 1 = SBE occurred */
  97
  98 #define I82443BXGX_DRAMC 0x57   /* 8bit register starting at this PCI
  99                                  * config space offset. */
 100 #define I82443BXGX_DRAMC_OFFSET_DT 3    /* 2 bits, DRAM Type */
 101 #define I82443BXGX_DRAMC_DRAM_IS_EDO 0  /* 00 = EDO */
 102 #define I82443BXGX_DRAMC_DRAM_IS_SDRAM 1        /* 01 = SDRAM */
 103 #define I82443BXGX_DRAMC_DRAM_IS_RSDRAM 2       /* 10 = Registered SDRAM */
 104
 105 #define I82443BXGX_DRB 0x60     /* 8x 8bit registers starting at this PCI
 106                                  * config space offset. */
 107
 108 /* FIXME - don't poll when ECC disabled? */
 109
 110 struct i82443bxgx_edacmc_error_info {
 111         u32 eap;
 112 };
 113
 114 static struct edac_pci_ctl_info *i82443bxgx_pci;
 115
 116 static struct pci_dev *mci_pdev;        /* init dev: in case that AGP code has
 117                                          * already registered driver
 118                                          */
 119
 120 static int i82443bxgx_registered = 1;
 121
 122 static void i82443bxgx_edacmc_get_error_info(struct mem_ctl_info *mci,
 123                                 struct i82443bxgx_edacmc_error_info
 124                                 *info)
 125 {
 126         struct pci_dev *pdev;
 127         pdev = to_pci_dev(mci->dev);
 128         pci_read_config_dword(pdev, I82443BXGX_EAP, &info->eap);
 129         if (info->eap & I82443BXGX_EAP_OFFSET_SBE)
 130                 /* Clear error to allow next error to be reported [p.61] */
 131                 pci_write_bits32(pdev, I82443BXGX_EAP,
 132                                  I82443BXGX_EAP_OFFSET_SBE,
 133                                  I82443BXGX_EAP_OFFSET_SBE);
 134
 135         if (info->eap & I82443BXGX_EAP_OFFSET_MBE)
 136                 /* Clear error to allow next error to be reported [p.61] */
 137                 pci_write_bits32(pdev, I82443BXGX_EAP,
 138                                  I82443BXGX_EAP_OFFSET_MBE,
 139                                  I82443BXGX_EAP_OFFSET_MBE);
 140 }
 141
 142 static int i82443bxgx_edacmc_process_error_info(struct mem_ctl_info *mci,
 143                                                 struct
 144                                                 i82443bxgx_edacmc_error_info
 145                                                 *info, int handle_errors)
 146 {
 147         int error_found = 0;
 148         u32 eapaddr, page, pageoffset;
 149
 150         /* bits 30:12 hold the 4kb block in which the error occurred
 151          * [p.61] */
 152         eapaddr = (info->eap & 0xfffff000);
 153         page = eapaddr >> PAGE_SHIFT;
 154         pageoffset = eapaddr - (page << PAGE_SHIFT);
 155
 156         if (info->eap & I82443BXGX_EAP_OFFSET_SBE) {
 157                 error_found = 1;
 158                 if (handle_errors)
 159                         edac_mc_handle_ce(mci, page, pageoffset,
 160                                 /* 440BX/GX don't make syndrome information
 161                                  * available */
 162                                 0, edac_mc_find_csrow_by_page(mci, page), 0,
 163                                 mci->ctl_name);
 164         }
 165
 166         if (info->eap & I82443BXGX_EAP_OFFSET_MBE) {
 167                 error_found = 1;
 168                 if (handle_errors)
 169                         edac_mc_handle_ue(mci, page, pageoffset,
 170                                         edac_mc_find_csrow_by_page(mci, page),
 171                                         mci->ctl_name);
 172         }
 173
 174         return error_found;
 175 }
 176
 177 static void i82443bxgx_edacmc_check(struct mem_ctl_info *mci)
 178 {
 179         struct i82443bxgx_edacmc_error_info info;
 180
 181         debugf1("MC%d: %s: %s()\n", mci->mc_idx, __FILE__, __func__);
 182         i82443bxgx_edacmc_get_error_info(mci, &info);
 183         i82443bxgx_edacmc_process_error_info(mci, &info, 1);
 184 }
 185
 186 static void i82443bxgx_init_csrows(struct mem_ctl_info *mci,
 187                                 struct pci_dev *pdev,
 188                                 enum edac_type edac_mode,
 189                                 enum mem_type mtype)
 190 {
 191         struct csrow_info *csrow;
 192         int index;
 193         u8 drbar, dramc;
 194         u32 row_base, row_high_limit, row_high_limit_last;
 195
 196         pci_read_config_byte(pdev, I82443BXGX_DRAMC, &dramc);
 197         row_high_limit_last = 0;
 198         for (index = 0; index < mci->nr_csrows; index++) {
 199                 csrow = &mci->csrows[index];
 200                 pci_read_config_byte(pdev, I82443BXGX_DRB + index, &drbar);
 201                 debugf1("MC%d: %s: %s() Row=%d DRB = %#0x\n",
 202                         mci->mc_idx, __FILE__, __func__, index, drbar);
 203                 row_high_limit = ((u32) drbar << 23);
 204                 /* find the DRAM Chip Select Base address and mask */
 205                 debugf1("MC%d: %s: %s() Row=%d, "
 206                         "Boundry Address=%#0x, Last = %#0x\n",
 207                         mci->mc_idx, __FILE__, __func__, index, row_high_limit,
 208                         row_high_limit_last);
 209
 210                 /* 440GX goes to 2GB, represented with a DRB of 0. */
 211                 if (row_high_limit_last && !row_high_limit)
 212                         row_high_limit = 1UL << 31;
 213
 214                 /* This row is empty [p.49] */
 215                 if (row_high_limit == row_high_limit_last)
 216                         continue;
 217                 row_base = row_high_limit_last;
 218                 csrow->first_page = row_base >> PAGE_SHIFT;
 219                 csrow->last_page = (row_high_limit >> PAGE_SHIFT) - 1;
 220                 csrow->nr_pages = csrow->last_page - csrow->first_page + 1;
 221                 /* EAP reports in 4kilobyte granularity [61] */
 222                 csrow->grain = 1 << 12;
 223                 csrow->mtype = mtype;
 224                 /* I don't think 440BX can tell you device type? FIXME? */
 225                 csrow->dtype = DEV_UNKNOWN;
 226                 /* Mode is global to all rows on 440BX */
 227                 csrow->edac_mode = edac_mode;
 228                 row_high_limit_last = row_high_limit;
 229         }
 230 }
 231
 232 static int i82443bxgx_edacmc_probe1(struct pci_dev *pdev, int dev_idx)
 233 {
 234         struct mem_ctl_info *mci;
 235         u8 dramc;
 236         u32 nbxcfg, ecc_mode;
 237         enum mem_type mtype;
 238         enum edac_type edac_mode;
 239
 240         debugf0("MC: %s: %s()\n", __FILE__, __func__);
 241
 242         /* Something is really hosed if PCI config space reads from
 243          * the MC aren't working.
 244          */
 245         if (pci_read_config_dword(pdev, I82443BXGX_NBXCFG, &nbxcfg))
 246                 return -EIO;
 247
 248         mci = edac_mc_alloc(0, I82443BXGX_NR_CSROWS, I82443BXGX_NR_CHANS, 0);
 249
 250         if (mci == NULL)
 251                 return -ENOMEM;
 252
 253         debugf0("MC: %s: %s(): mci = %p\n", __FILE__, __func__, mci);
 254         mci->dev = &pdev->dev;
 255         mci->mtype_cap = MEM_FLAG_EDO | MEM_FLAG_SDR | MEM_FLAG_RDR;
 256         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 257         pci_read_config_byte(pdev, I82443BXGX_DRAMC, &dramc);
 258         switch ((dramc >> I82443BXGX_DRAMC_OFFSET_DT) & (BIT(0) | BIT(1))) {
 259         case I82443BXGX_DRAMC_DRAM_IS_EDO:
 260                 mtype = MEM_EDO;
 261                 break;
 262         case I82443BXGX_DRAMC_DRAM_IS_SDRAM:
 263                 mtype = MEM_SDR;
 264                 break;
 265         case I82443BXGX_DRAMC_DRAM_IS_RSDRAM:
 266                 mtype = MEM_RDR;
 267                 break;
 268         default:
 269                 debugf0("Unknown/reserved DRAM type value "
 270                         "in DRAMC register!\n");
 271                 mtype = -MEM_UNKNOWN;
 272         }
 273
 274         if ((mtype == MEM_SDR) || (mtype == MEM_RDR))
 275                 mci->edac_cap = mci->edac_ctl_cap;
 276         else
 277                 mci->edac_cap = EDAC_FLAG_NONE;
 278
 279         mci->scrub_cap = SCRUB_FLAG_HW_SRC;
 280         pci_read_config_dword(pdev, I82443BXGX_NBXCFG, &nbxcfg);
 281         ecc_mode = ((nbxcfg >> I82443BXGX_NBXCFG_OFFSET_DRAM_INTEGRITY) &
 282                 (BIT(0) | BIT(1)));
 283
 284         mci->scrub_mode = (ecc_mode == I82443BXGX_NBXCFG_INTEGRITY_SCRUB)
 285                 ? SCRUB_HW_SRC : SCRUB_NONE;
 286
 287         switch (ecc_mode) {
 288         case I82443BXGX_NBXCFG_INTEGRITY_NONE:
 289                 edac_mode = EDAC_NONE;
 290                 break;
 291         case I82443BXGX_NBXCFG_INTEGRITY_EC:
 292                 edac_mode = EDAC_EC;
 293                 break;
 294         case I82443BXGX_NBXCFG_INTEGRITY_ECC:
 295         case I82443BXGX_NBXCFG_INTEGRITY_SCRUB:
 296                 edac_mode = EDAC_SECDED;
 297                 break;
 298         default:
 299                 debugf0("%s(): Unknown/reserved ECC state "
 300                         "in NBXCFG register!\n", __func__);
 301                 edac_mode = EDAC_UNKNOWN;
 302                 break;
 303         }
 304
 305         i82443bxgx_init_csrows(mci, pdev, edac_mode, mtype);
 306
 307         /* Many BIOSes don't clear error flags on boot, so do this
 308          * here, or we get "phantom" errors occuring at module-load
 309          * time. */
 310         pci_write_bits32(pdev, I82443BXGX_EAP,
 311                         (I82443BXGX_EAP_OFFSET_SBE |
 312                                 I82443BXGX_EAP_OFFSET_MBE),
 313                         (I82443BXGX_EAP_OFFSET_SBE |
 314                                 I82443BXGX_EAP_OFFSET_MBE));
 315
 316         mci->mod_name = EDAC_MOD_STR;
 317         mci->mod_ver = I82443_REVISION;
 318         mci->ctl_name = "I82443BXGX";
 319         mci->dev_name = pci_name(pdev);
 320         mci->edac_check = i82443bxgx_edacmc_check;
 321         mci->ctl_page_to_phys = NULL;
 322
 323         if (edac_mc_add_mc(mci)) {
 324                 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
 325                 goto fail;
 326         }
 327
 328         /* allocating generic PCI control info */
 329         i82443bxgx_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
 330         if (!i82443bxgx_pci) {
 331                 printk(KERN_WARNING
 332                         "%s(): Unable to create PCI control\n",
 333                         __func__);
 334                 printk(KERN_WARNING
 335                         "%s(): PCI error report via EDAC not setup\n",
 336                         __func__);
 337         }
 338
 339         debugf3("MC: %s: %s(): success\n", __FILE__, __func__);
 340         return 0;
 341
 342 fail:
 343         edac_mc_free(mci);
 344         return -ENODEV;
 345 }
 346
 347 EXPORT_SYMBOL_GPL(i82443bxgx_edacmc_probe1);
 348
 349 /* returns count (>= 0), or negative on error */
 350 static int __devinit i82443bxgx_edacmc_init_one(struct pci_dev *pdev,
 351                                                 const struct pci_device_id *ent)
 352 {
 353         int rc;
 354
 355         debugf0("MC: %s: %s()\n", __FILE__, __func__);
 356
 357         /* don't need to call pci_enable_device() */
 358         rc = i82443bxgx_edacmc_probe1(pdev, ent->driver_data);
 359
 360         if (mci_pdev == NULL)
 361                 mci_pdev = pci_dev_get(pdev);
 362
 363         return rc;
 364 }
 365
 366 static void __devexit i82443bxgx_edacmc_remove_one(struct pci_dev *pdev)
 367 {
 368         struct mem_ctl_info *mci;
 369
 370         debugf0("%s: %s()\n", __FILE__, __func__);
 371
 372         if (i82443bxgx_pci)
 373                 edac_pci_release_generic_ctl(i82443bxgx_pci);
 374
 375         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
 376                 return;
 377
 378         edac_mc_free(mci);
 379 }
 380
 381 EXPORT_SYMBOL_GPL(i82443bxgx_edacmc_remove_one);
 382
 383 static const struct pci_device_id i82443bxgx_pci_tbl[] __devinitdata = {
 384         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443BX_0)},
 385         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443BX_2)},
 386         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443GX_0)},
 387         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82443GX_2)},
 388         {0,}                    /* 0 terminated list. */
 389 };
 390
 391 MODULE_DEVICE_TABLE(pci, i82443bxgx_pci_tbl);
 392
 393 static struct pci_driver i82443bxgx_edacmc_driver = {
 394         .name = EDAC_MOD_STR,
 395         .probe = i82443bxgx_edacmc_init_one,
 396         .remove = __devexit_p(i82443bxgx_edacmc_remove_one),
 397         .id_table = i82443bxgx_pci_tbl,
 398 };
 399
 400 static int __init i82443bxgx_edacmc_init(void)
 401 {
 402         int pci_rc;
 403        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 404        opstate_init();
 405
 406         pci_rc = pci_register_driver(&i82443bxgx_edacmc_driver);
 407         if (pci_rc < 0)
 408                 goto fail0;
 409
 410         if (mci_pdev == NULL) {
 411                 const struct pci_device_id *id = &i82443bxgx_pci_tbl[0];
 412                 int i = 0;
 413                 i82443bxgx_registered = 0;
 414
 415                 while (mci_pdev == NULL && id->vendor != 0) {
 416                         mci_pdev = pci_get_device(id->vendor,
 417                                         id->device, NULL);
 418                         i++;
 419                         id = &i82443bxgx_pci_tbl[i];
 420                 }
 421                 if (!mci_pdev) {
 422                         debugf0("i82443bxgx pci_get_device fail\n");
 423                         pci_rc = -ENODEV;
 424                         goto fail1;
 425                 }
 426
 427                 pci_rc = i82443bxgx_edacmc_init_one(mci_pdev, i82443bxgx_pci_tbl);
 428
 429                 if (pci_rc < 0) {
 430                         debugf0("i82443bxgx init fail\n");
 431                         pci_rc = -ENODEV;
 432                         goto fail1;
 433                 }
 434         }
 435
 436         return 0;
 437
 438 fail1:
 439         pci_unregister_driver(&i82443bxgx_edacmc_driver);
 440
 441 fail0:
 442         if (mci_pdev != NULL)
 443                 pci_dev_put(mci_pdev);
 444
 445         return pci_rc;
 446 }
 447
 448 static void __exit i82443bxgx_edacmc_exit(void)
 449 {
 450         pci_unregister_driver(&i82443bxgx_edacmc_driver);
 451
 452         if (!i82443bxgx_registered)
 453                 i82443bxgx_edacmc_remove_one(mci_pdev);
 454
 455         if (mci_pdev)
 456                 pci_dev_put(mci_pdev);
 457 }
 458
 459 module_init(i82443bxgx_edacmc_init);
 460 module_exit(i82443bxgx_edacmc_exit);
 461
 462 MODULE_LICENSE("GPL");
 463 MODULE_AUTHOR("Tim Small <tim@buttersideup.com> - WPAD");
 464 MODULE_DESCRIPTION("EDAC MC support for Intel 82443BX/GX memory controllers");
 465
 466 module_param(edac_op_state, int, 0444);
 467 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");