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