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