drivers/edac/i82875p_edac.c

   1 /*
   2  * Intel D82875P Memory Controller kernel module
   3  * (C) 2003 Linux Networx (http://lnxi.com)
   4  * This file may be distributed under the terms of the
   5  * GNU General Public License.
   6  *
   7  * Written by Thayne Harbaugh
   8  * Contributors:
   9  *      Wang Zhenyu at intel.com
  10  *
  11  * $Id: edac_i82875p.c,v 1.5.2.11 2005/10/05 00:43:44 dsp_llnl Exp $
  12  *
  13  * Note: E7210 appears same as D82875P - zhenyu.z.wang at intel.com
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/init.h>
  18 #include <linux/pci.h>
  19 #include <linux/pci_ids.h>
  20 #include <linux/edac.h>
  21 #include "edac_module.h"
  22
  23 #define EDAC_MOD_STR            "i82875p_edac"
  24
  25 #define i82875p_printk(level, fmt, arg...) \
  26         edac_printk(level, "i82875p", fmt, ##arg)
  27
  28 #define i82875p_mc_printk(mci, level, fmt, arg...) \
  29         edac_mc_chipset_printk(mci, level, "i82875p", fmt, ##arg)
  30
  31 #ifndef PCI_DEVICE_ID_INTEL_82875_0
  32 #define PCI_DEVICE_ID_INTEL_82875_0     0x2578
  33 #endif                          /* PCI_DEVICE_ID_INTEL_82875_0 */
  34
  35 #ifndef PCI_DEVICE_ID_INTEL_82875_6
  36 #define PCI_DEVICE_ID_INTEL_82875_6     0x257e
  37 #endif                          /* PCI_DEVICE_ID_INTEL_82875_6 */
  38
  39 /* four csrows in dual channel, eight in single channel */
  40 #define I82875P_NR_DIMMS                8
  41 #define I82875P_NR_CSROWS(nr_chans)     (I82875P_NR_DIMMS / (nr_chans))
  42
  43 /* Intel 82875p register addresses - device 0 function 0 - DRAM Controller */
  44 #define I82875P_EAP             0x58    /* Error Address Pointer (32b)
  45                                          *
  46                                          * 31:12 block address
  47                                          * 11:0  reserved
  48                                          */
  49
  50 #define I82875P_DERRSYN         0x5c    /* DRAM Error Syndrome (8b)
  51                                          *
  52                                          *  7:0  DRAM ECC Syndrome
  53                                          */
  54
  55 #define I82875P_DES             0x5d    /* DRAM Error Status (8b)
  56                                          *
  57                                          *  7:1  reserved
  58                                          *  0    Error channel 0/1
  59                                          */
  60
  61 #define I82875P_ERRSTS          0xc8    /* Error Status Register (16b)
  62                                          *
  63                                          * 15:10 reserved
  64                                          *  9    non-DRAM lock error (ndlock)
  65                                          *  8    Sftwr Generated SMI
  66                                          *  7    ECC UE
  67                                          *  6    reserved
  68                                          *  5    MCH detects unimplemented cycle
  69                                          *  4    AGP access outside GA
  70                                          *  3    Invalid AGP access
  71                                          *  2    Invalid GA translation table
  72                                          *  1    Unsupported AGP command
  73                                          *  0    ECC CE
  74                                          */
  75
  76 #define I82875P_ERRCMD          0xca    /* Error Command (16b)
  77                                          *
  78                                          * 15:10 reserved
  79                                          *  9    SERR on non-DRAM lock
  80                                          *  8    SERR on ECC UE
  81                                          *  7    SERR on ECC CE
  82                                          *  6    target abort on high exception
  83                                          *  5    detect unimplemented cyc
  84                                          *  4    AGP access outside of GA
  85                                          *  3    SERR on invalid AGP access
  86                                          *  2    invalid translation table
  87                                          *  1    SERR on unsupported AGP command
  88                                          *  0    reserved
  89                                          */
  90
  91 /* Intel 82875p register addresses - device 6 function 0 - DRAM Controller */
  92 #define I82875P_PCICMD6         0x04    /* PCI Command Register (16b)
  93                                          *
  94                                          * 15:10 reserved
  95                                          *  9    fast back-to-back - ro 0
  96                                          *  8    SERR enable - ro 0
  97                                          *  7    addr/data stepping - ro 0
  98                                          *  6    parity err enable - ro 0
  99                                          *  5    VGA palette snoop - ro 0
 100                                          *  4    mem wr & invalidate - ro 0
 101                                          *  3    special cycle - ro 0
 102                                          *  2    bus master - ro 0
 103                                          *  1    mem access dev6 - 0(dis),1(en)
 104                                          *  0    IO access dev3 - 0(dis),1(en)
 105                                          */
 106
 107 #define I82875P_BAR6            0x10    /* Mem Delays Base ADDR Reg (32b)
 108                                          *
 109                                          * 31:12 mem base addr [31:12]
 110                                          * 11:4  address mask - ro 0
 111                                          *  3    prefetchable - ro 0(non),1(pre)
 112                                          *  2:1  mem type - ro 0
 113                                          *  0    mem space - ro 0
 114                                          */
 115
 116 /* Intel 82875p MMIO register space - device 0 function 0 - MMR space */
 117
 118 #define I82875P_DRB_SHIFT 26    /* 64MiB grain */
 119 #define I82875P_DRB             0x00    /* DRAM Row Boundary (8b x 8)
 120                                          *
 121                                          *  7    reserved
 122                                          *  6:0  64MiB row boundary addr
 123                                          */
 124
 125 #define I82875P_DRA             0x10    /* DRAM Row Attribute (4b x 8)
 126                                          *
 127                                          *  7    reserved
 128                                          *  6:4  row attr row 1
 129                                          *  3    reserved
 130                                          *  2:0  row attr row 0
 131                                          *
 132                                          * 000 =  4KiB
 133                                          * 001 =  8KiB
 134                                          * 010 = 16KiB
 135                                          * 011 = 32KiB
 136                                          */
 137
 138 #define I82875P_DRC             0x68    /* DRAM Controller Mode (32b)
 139                                          *
 140                                          * 31:30 reserved
 141                                          * 29    init complete
 142                                          * 28:23 reserved
 143                                          * 22:21 nr chan 00=1,01=2
 144                                          * 20    reserved
 145                                          * 19:18 Data Integ Mode 00=none,01=ecc
 146                                          * 17:11 reserved
 147                                          * 10:8  refresh mode
 148                                          *  7    reserved
 149                                          *  6:4  mode select
 150                                          *  3:2  reserved
 151                                          *  1:0  DRAM type 01=DDR
 152                                          */
 153
 154 enum i82875p_chips {
 155         I82875P = 0,
 156 };
 157
 158 struct i82875p_pvt {
 159         struct pci_dev *ovrfl_pdev;
 160         void __iomem *ovrfl_window;
 161 };
 162
 163 struct i82875p_dev_info {
 164         const char *ctl_name;
 165 };
 166
 167 struct i82875p_error_info {
 168         u16 errsts;
 169         u32 eap;
 170         u8 des;
 171         u8 derrsyn;
 172         u16 errsts2;
 173 };
 174
 175 static const struct i82875p_dev_info i82875p_devs[] = {
 176         [I82875P] = {
 177                 .ctl_name = "i82875p"},
 178 };
 179
 180 static struct pci_dev *mci_pdev;        /* init dev: in case that AGP code has
 181                                          * already registered driver
 182                                          */
 183
 184 static struct edac_pci_ctl_info *i82875p_pci;
 185
 186 static void i82875p_get_error_info(struct mem_ctl_info *mci,
 187                                 struct i82875p_error_info *info)
 188 {
 189         struct pci_dev *pdev;
 190
 191         pdev = to_pci_dev(mci->pdev);
 192
 193         /*
 194          * This is a mess because there is no atomic way to read all the
 195          * registers at once and the registers can transition from CE being
 196          * overwritten by UE.
 197          */
 198         pci_read_config_word(pdev, I82875P_ERRSTS, &info->errsts);
 199
 200         if (!(info->errsts & 0x0081))
 201                 return;
 202
 203         pci_read_config_dword(pdev, I82875P_EAP, &info->eap);
 204         pci_read_config_byte(pdev, I82875P_DES, &info->des);
 205         pci_read_config_byte(pdev, I82875P_DERRSYN, &info->derrsyn);
 206         pci_read_config_word(pdev, I82875P_ERRSTS, &info->errsts2);
 207
 208         /*
 209          * If the error is the same then we can for both reads then
 210          * the first set of reads is valid.  If there is a change then
 211          * there is a CE no info and the second set of reads is valid
 212          * and should be UE info.
 213          */
 214         if ((info->errsts ^ info->errsts2) & 0x0081) {
 215                 pci_read_config_dword(pdev, I82875P_EAP, &info->eap);
 216                 pci_read_config_byte(pdev, I82875P_DES, &info->des);
 217                 pci_read_config_byte(pdev, I82875P_DERRSYN, &info->derrsyn);
 218         }
 219
 220         pci_write_bits16(pdev, I82875P_ERRSTS, 0x0081, 0x0081);
 221 }
 222
 223 static int i82875p_process_error_info(struct mem_ctl_info *mci,
 224                                 struct i82875p_error_info *info,
 225                                 int handle_errors)
 226 {
 227         int row, multi_chan;
 228
 229         multi_chan = mci->csrows[0]->nr_channels - 1;
 230
 231         if (!(info->errsts & 0x0081))
 232                 return 0;
 233
 234         if (!handle_errors)
 235                 return 1;
 236
 237         if ((info->errsts ^ info->errsts2) & 0x0081) {
 238                 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
 239                                      -1, -1, -1,
 240                                      "UE overwrote CE", "");
 241                 info->errsts = info->errsts2;
 242         }
 243
 244         info->eap >>= PAGE_SHIFT;
 245         row = edac_mc_find_csrow_by_page(mci, info->eap);
 246
 247         if (info->errsts & 0x0080)
 248                 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 249                                      info->eap, 0, 0,
 250                                      row, -1, -1,
 251                                      "i82875p UE", "");
 252         else
 253                 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
 254                                      info->eap, 0, info->derrsyn,
 255                                      row, multi_chan ? (info->des & 0x1) : 0,
 256                                      -1, "i82875p CE", "");
 257
 258         return 1;
 259 }
 260
 261 static void i82875p_check(struct mem_ctl_info *mci)
 262 {
 263         struct i82875p_error_info info;
 264
 265         edac_dbg(1, "MC%d\n", mci->mc_idx);
 266         i82875p_get_error_info(mci, &info);
 267         i82875p_process_error_info(mci, &info, 1);
 268 }
 269
 270 /* Return 0 on success or 1 on failure. */
 271 static int i82875p_setup_overfl_dev(struct pci_dev *pdev,
 272                                 struct pci_dev **ovrfl_pdev,
 273                                 void __iomem **ovrfl_window)
 274 {
 275         struct pci_dev *dev;
 276         void __iomem *window;
 277
 278         *ovrfl_pdev = NULL;
 279         *ovrfl_window = NULL;
 280         dev = pci_get_device(PCI_VEND_DEV(INTEL, 82875_6), NULL);
 281
 282         if (dev == NULL) {
 283                 /* Intel tells BIOS developers to hide device 6 which
 284                  * configures the overflow device access containing
 285                  * the DRBs - this is where we expose device 6.
 286                  * http://www.x86-secret.com/articles/tweak/pat/patsecrets-2.htm
 287                  */
 288                 pci_write_bits8(pdev, 0xf4, 0x2, 0x2);
 289                 dev = pci_scan_single_device(pdev->bus, PCI_DEVFN(6, 0));
 290
 291                 if (dev == NULL)
 292                         return 1;
 293
 294                 pci_bus_assign_resources(dev->bus);
 295                 pci_bus_add_device(dev);
 296         }
 297
 298         *ovrfl_pdev = dev;
 299
 300         if (pci_enable_device(dev)) {
 301                 i82875p_printk(KERN_ERR, "%s(): Failed to enable overflow "
 302                         "device\n", __func__);
 303                 return 1;
 304         }
 305
 306         if (pci_request_regions(dev, pci_name(dev))) {
 307 #ifdef CORRECT_BIOS
 308                 goto fail0;
 309 #endif
 310         }
 311
 312         /* cache is irrelevant for PCI bus reads/writes */
 313         window = pci_ioremap_bar(dev, 0);
 314         if (window == NULL) {
 315                 i82875p_printk(KERN_ERR, "%s(): Failed to ioremap bar6\n",
 316                         __func__);
 317                 goto fail1;
 318         }
 319
 320         *ovrfl_window = window;
 321         return 0;
 322
 323 fail1:
 324         pci_release_regions(dev);
 325
 326 #ifdef CORRECT_BIOS
 327 fail0:
 328         pci_disable_device(dev);
 329 #endif
 330         /* NOTE: the ovrfl proc entry and pci_dev are intentionally left */
 331         return 1;
 332 }
 333
 334 /* Return 1 if dual channel mode is active.  Else return 0. */
 335 static inline int dual_channel_active(u32 drc)
 336 {
 337         return (drc >> 21) & 0x1;
 338 }
 339
 340 static void i82875p_init_csrows(struct mem_ctl_info *mci,
 341                                 struct pci_dev *pdev,
 342                                 void __iomem * ovrfl_window, u32 drc)
 343 {
 344         struct csrow_info *csrow;
 345         struct dimm_info *dimm;
 346         unsigned nr_chans = dual_channel_active(drc) + 1;
 347         unsigned long last_cumul_size;
 348         u8 value;
 349         u32 drc_ddim;           /* DRAM Data Integrity Mode 0=none,2=edac */
 350         u32 cumul_size, nr_pages;
 351         int index, j;
 352
 353         drc_ddim = (drc >> 18) & 0x1;
 354         last_cumul_size = 0;
 355
 356         /* The dram row boundary (DRB) reg values are boundary address
 357          * for each DRAM row with a granularity of 32 or 64MB (single/dual
 358          * channel operation).  DRB regs are cumulative; therefore DRB7 will
 359          * contain the total memory contained in all eight rows.
 360          */
 361
 362         for (index = 0; index < mci->nr_csrows; index++) {
 363                 csrow = mci->csrows[index];
 364
 365                 value = readb(ovrfl_window + I82875P_DRB + index);
 366                 cumul_size = value << (I82875P_DRB_SHIFT - PAGE_SHIFT);
 367                 edac_dbg(3, "(%d) cumul_size 0x%x\n", index, cumul_size);
 368                 if (cumul_size == last_cumul_size)
 369                         continue;       /* not populated */
 370
 371                 csrow->first_page = last_cumul_size;
 372                 csrow->last_page = cumul_size - 1;
 373                 nr_pages = cumul_size - last_cumul_size;
 374                 last_cumul_size = cumul_size;
 375
 376                 for (j = 0; j < nr_chans; j++) {
 377                         dimm = csrow->channels[j]->dimm;
 378
 379                         dimm->nr_pages = nr_pages / nr_chans;
 380                         dimm->grain = 1 << 12;  /* I82875P_EAP has 4KiB reolution */
 381                         dimm->mtype = MEM_DDR;
 382                         dimm->dtype = DEV_UNKNOWN;
 383                         dimm->edac_mode = drc_ddim ? EDAC_SECDED : EDAC_NONE;
 384                 }
 385         }
 386 }
 387
 388 static int i82875p_probe1(struct pci_dev *pdev, int dev_idx)
 389 {
 390         int rc = -ENODEV;
 391         struct mem_ctl_info *mci;
 392         struct edac_mc_layer layers[2];
 393         struct i82875p_pvt *pvt;
 394         struct pci_dev *ovrfl_pdev;
 395         void __iomem *ovrfl_window;
 396         u32 drc;
 397         u32 nr_chans;
 398         struct i82875p_error_info discard;
 399
 400         edac_dbg(0, "\n");
 401
 402         if (i82875p_setup_overfl_dev(pdev, &ovrfl_pdev, &ovrfl_window))
 403                 return -ENODEV;
 404         drc = readl(ovrfl_window + I82875P_DRC);
 405         nr_chans = dual_channel_active(drc) + 1;
 406
 407         layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 408         layers[0].size = I82875P_NR_CSROWS(nr_chans);
 409         layers[0].is_virt_csrow = true;
 410         layers[1].type = EDAC_MC_LAYER_CHANNEL;
 411         layers[1].size = nr_chans;
 412         layers[1].is_virt_csrow = false;
 413         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(*pvt));
 414         if (!mci) {
 415                 rc = -ENOMEM;
 416                 goto fail0;
 417         }
 418
 419         edac_dbg(3, "init mci\n");
 420         mci->pdev = &pdev->dev;
 421         mci->mtype_cap = MEM_FLAG_DDR;
 422         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
 423         mci->edac_cap = EDAC_FLAG_UNKNOWN;
 424         mci->mod_name = EDAC_MOD_STR;
 425         mci->ctl_name = i82875p_devs[dev_idx].ctl_name;
 426         mci->dev_name = pci_name(pdev);
 427         mci->edac_check = i82875p_check;
 428         mci->ctl_page_to_phys = NULL;
 429         edac_dbg(3, "init pvt\n");
 430         pvt = (struct i82875p_pvt *)mci->pvt_info;
 431         pvt->ovrfl_pdev = ovrfl_pdev;
 432         pvt->ovrfl_window = ovrfl_window;
 433         i82875p_init_csrows(mci, pdev, ovrfl_window, drc);
 434         i82875p_get_error_info(mci, &discard);  /* clear counters */
 435
 436         /* Here we assume that we will never see multiple instances of this
 437          * type of memory controller.  The ID is therefore hardcoded to 0.
 438          */
 439         if (edac_mc_add_mc(mci)) {
 440                 edac_dbg(3, "failed edac_mc_add_mc()\n");
 441                 goto fail1;
 442         }
 443
 444         /* allocating generic PCI control info */
 445         i82875p_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
 446         if (!i82875p_pci) {
 447                 printk(KERN_WARNING
 448                         "%s(): Unable to create PCI control\n",
 449                         __func__);
 450                 printk(KERN_WARNING
 451                         "%s(): PCI error report via EDAC not setup\n",
 452                         __func__);
 453         }
 454
 455         /* get this far and it's successful */
 456         edac_dbg(3, "success\n");
 457         return 0;
 458
 459 fail1:
 460         edac_mc_free(mci);
 461
 462 fail0:
 463         iounmap(ovrfl_window);
 464         pci_release_regions(ovrfl_pdev);
 465
 466         pci_disable_device(ovrfl_pdev);
 467         /* NOTE: the ovrfl proc entry and pci_dev are intentionally left */
 468         return rc;
 469 }
 470
 471 /* returns count (>= 0), or negative on error */
 472 static int i82875p_init_one(struct pci_dev *pdev,
 473                             const struct pci_device_id *ent)
 474 {
 475         int rc;
 476
 477         edac_dbg(0, "\n");
 478         i82875p_printk(KERN_INFO, "i82875p init one\n");
 479
 480         if (pci_enable_device(pdev) < 0)
 481                 return -EIO;
 482
 483         rc = i82875p_probe1(pdev, ent->driver_data);
 484
 485         if (mci_pdev == NULL)
 486                 mci_pdev = pci_dev_get(pdev);
 487
 488         return rc;
 489 }
 490
 491 static void i82875p_remove_one(struct pci_dev *pdev)
 492 {
 493         struct mem_ctl_info *mci;
 494         struct i82875p_pvt *pvt = NULL;
 495
 496         edac_dbg(0, "\n");
 497
 498         if (i82875p_pci)
 499                 edac_pci_release_generic_ctl(i82875p_pci);
 500
 501         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
 502                 return;
 503
 504         pvt = (struct i82875p_pvt *)mci->pvt_info;
 505
 506         if (pvt->ovrfl_window)
 507                 iounmap(pvt->ovrfl_window);
 508
 509         if (pvt->ovrfl_pdev) {
 510 #ifdef CORRECT_BIOS
 511                 pci_release_regions(pvt->ovrfl_pdev);
 512 #endif                          /*CORRECT_BIOS */
 513                 pci_disable_device(pvt->ovrfl_pdev);
 514                 pci_dev_put(pvt->ovrfl_pdev);
 515         }
 516
 517         edac_mc_free(mci);
 518 }
 519
 520 static const struct pci_device_id i82875p_pci_tbl[] = {
 521         {
 522          PCI_VEND_DEV(INTEL, 82875_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 523          I82875P},
 524         {
 525          0,
 526          }                      /* 0 terminated list. */
 527 };
 528
 529 MODULE_DEVICE_TABLE(pci, i82875p_pci_tbl);
 530
 531 static struct pci_driver i82875p_driver = {
 532         .name = EDAC_MOD_STR,
 533         .probe = i82875p_init_one,
 534         .remove = i82875p_remove_one,
 535         .id_table = i82875p_pci_tbl,
 536 };
 537
 538 static int __init i82875p_init(void)
 539 {
 540         int pci_rc;
 541
 542         edac_dbg(3, "\n");
 543
 544        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 545        opstate_init();
 546
 547         pci_rc = pci_register_driver(&i82875p_driver);
 548
 549         if (pci_rc < 0)
 550                 goto fail0;
 551
 552         if (mci_pdev == NULL) {
 553                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 554                                         PCI_DEVICE_ID_INTEL_82875_0, NULL);
 555
 556                 if (!mci_pdev) {
 557                         edac_dbg(0, "875p pci_get_device fail\n");
 558                         pci_rc = -ENODEV;
 559                         goto fail1;
 560                 }
 561
 562                 pci_rc = i82875p_init_one(mci_pdev, i82875p_pci_tbl);
 563
 564                 if (pci_rc < 0) {
 565                         edac_dbg(0, "875p init fail\n");
 566                         pci_rc = -ENODEV;
 567                         goto fail1;
 568                 }
 569         }
 570
 571         return 0;
 572
 573 fail1:
 574         pci_unregister_driver(&i82875p_driver);
 575
 576 fail0:
 577         pci_dev_put(mci_pdev);
 578         return pci_rc;
 579 }
 580
 581 static void __exit i82875p_exit(void)
 582 {
 583         edac_dbg(3, "\n");
 584
 585         i82875p_remove_one(mci_pdev);
 586         pci_dev_put(mci_pdev);
 587
 588         pci_unregister_driver(&i82875p_driver);
 589
 590 }
 591
 592 module_init(i82875p_init);
 593 module_exit(i82875p_exit);
 594
 595 MODULE_LICENSE("GPL");
 596 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh");
 597 MODULE_DESCRIPTION("MC support for Intel 82875 memory hub controllers");
 598
 599 module_param(edac_op_state, int, 0444);
 600 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");