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/config.h>
  17 #include <linux/module.h>
  18 #include <linux/init.h>
  19 #include <linux/pci.h>
  20 #include <linux/pci_ids.h>
  21 #include <linux/slab.h>
  22 #include "edac_mc.h"
  23
  24 #define I82875P_REVISION        " Ver: 2.0.0 " __DATE__
  25
  26 #define i82875p_printk(level, fmt, arg...) \
  27         edac_printk(level, "i82875p", fmt, ##arg)
  28
  29 #define i82875p_mc_printk(mci, level, fmt, arg...) \
  30         edac_mc_chipset_printk(mci, level, "i82875p", fmt, ##arg)
  31
  32 #ifndef PCI_DEVICE_ID_INTEL_82875_0
  33 #define PCI_DEVICE_ID_INTEL_82875_0     0x2578
  34 #endif                          /* PCI_DEVICE_ID_INTEL_82875_0 */
  35
  36 #ifndef PCI_DEVICE_ID_INTEL_82875_6
  37 #define PCI_DEVICE_ID_INTEL_82875_6     0x257e
  38 #endif                          /* PCI_DEVICE_ID_INTEL_82875_6 */
  39
  40 /* four csrows in dual channel, eight in single channel */
  41 #define I82875P_NR_CSROWS(nr_chans) (8/(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
 181 static struct pci_dev *mci_pdev = NULL; /* init dev: in case that AGP code has
 182                                          * already registered driver
 183                                          */
 184
 185 static int i82875p_registered = 1;
 186
 187 static void i82875p_get_error_info(struct mem_ctl_info *mci,
 188                 struct i82875p_error_info *info)
 189 {
 190         struct pci_dev *pdev;
 191
 192         pdev = to_pci_dev(mci->dev);
 193
 194         /*
 195          * This is a mess because there is no atomic way to read all the
 196          * registers at once and the registers can transition from CE being
 197          * overwritten by UE.
 198          */
 199         pci_read_config_word(pdev, I82875P_ERRSTS, &info->errsts);
 200         pci_read_config_dword(pdev, I82875P_EAP, &info->eap);
 201         pci_read_config_byte(pdev, I82875P_DES, &info->des);
 202         pci_read_config_byte(pdev, I82875P_DERRSYN, &info->derrsyn);
 203         pci_read_config_word(pdev, I82875P_ERRSTS, &info->errsts2);
 204
 205         pci_write_bits16(pdev, I82875P_ERRSTS, 0x0081, 0x0081);
 206
 207         /*
 208          * If the error is the same then we can for both reads then
 209          * the first set of reads is valid.  If there is a change then
 210          * there is a CE no info and the second set of reads is valid
 211          * and should be UE info.
 212          */
 213         if (!(info->errsts2 & 0x0081))
 214                 return;
 215
 216         if ((info->errsts ^ info->errsts2) & 0x0081) {
 217                 pci_read_config_dword(pdev, I82875P_EAP, &info->eap);
 218                 pci_read_config_byte(pdev, I82875P_DES, &info->des);
 219                 pci_read_config_byte(pdev, I82875P_DERRSYN,
 220                                 &info->derrsyn);
 221         }
 222 }
 223
 224 static int i82875p_process_error_info(struct mem_ctl_info *mci,
 225                 struct i82875p_error_info *info, int handle_errors)
 226 {
 227         int row, multi_chan;
 228
 229         multi_chan = mci->csrows[0].nr_channels - 1;
 230
 231         if (!(info->errsts2 & 0x0081))
 232                 return 0;
 233
 234         if (!handle_errors)
 235                 return 1;
 236
 237         if ((info->errsts ^ info->errsts2) & 0x0081) {
 238                 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 239                 info->errsts = info->errsts2;
 240         }
 241
 242         info->eap >>= PAGE_SHIFT;
 243         row = edac_mc_find_csrow_by_page(mci, info->eap);
 244
 245         if (info->errsts & 0x0080)
 246                 edac_mc_handle_ue(mci, info->eap, 0, row, "i82875p UE");
 247         else
 248                 edac_mc_handle_ce(mci, info->eap, 0, info->derrsyn, row,
 249                                 multi_chan ? (info->des & 0x1) : 0,
 250                                 "i82875p CE");
 251
 252         return 1;
 253 }
 254
 255 static void i82875p_check(struct mem_ctl_info *mci)
 256 {
 257         struct i82875p_error_info info;
 258
 259         debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 260         i82875p_get_error_info(mci, &info);
 261         i82875p_process_error_info(mci, &info, 1);
 262 }
 263
 264 #ifdef CONFIG_PROC_FS
 265 extern int pci_proc_attach_device(struct pci_dev *);
 266 #endif
 267
 268 static int i82875p_probe1(struct pci_dev *pdev, int dev_idx)
 269 {
 270         int rc = -ENODEV;
 271         int index;
 272         struct mem_ctl_info *mci = NULL;
 273         struct i82875p_pvt *pvt = NULL;
 274         unsigned long last_cumul_size;
 275         struct pci_dev *ovrfl_pdev;
 276         void __iomem *ovrfl_window = NULL;
 277         u32 drc;
 278         u32 drc_chan;           /* Number of channels 0=1chan,1=2chan */
 279         u32 nr_chans;
 280         u32 drc_ddim;           /* DRAM Data Integrity Mode 0=none,2=edac */
 281         struct i82875p_error_info discard;
 282
 283         debugf0("%s()\n", __func__);
 284         ovrfl_pdev = pci_get_device(PCI_VEND_DEV(INTEL, 82875_6), NULL);
 285
 286         if (!ovrfl_pdev) {
 287                 /*
 288                  * Intel tells BIOS developers to hide device 6 which
 289                  * configures the overflow device access containing
 290                  * the DRBs - this is where we expose device 6.
 291                  * http://www.x86-secret.com/articles/tweak/pat/patsecrets-2.htm
 292                  */
 293                 pci_write_bits8(pdev, 0xf4, 0x2, 0x2);
 294                 ovrfl_pdev =
 295                         pci_scan_single_device(pdev->bus, PCI_DEVFN(6, 0));
 296
 297                 if (!ovrfl_pdev)
 298                         return -ENODEV;
 299         }
 300
 301 #ifdef CONFIG_PROC_FS
 302         if (!ovrfl_pdev->procent && pci_proc_attach_device(ovrfl_pdev)) {
 303                 i82875p_printk(KERN_ERR,
 304                         "%s(): Failed to attach overflow device\n", __func__);
 305                 return -ENODEV;
 306         }
 307 #endif
 308                                 /* CONFIG_PROC_FS */
 309         if (pci_enable_device(ovrfl_pdev)) {
 310                 i82875p_printk(KERN_ERR,
 311                         "%s(): Failed to enable overflow device\n", __func__);
 312                 return -ENODEV;
 313         }
 314
 315         if (pci_request_regions(ovrfl_pdev, pci_name(ovrfl_pdev))) {
 316 #ifdef CORRECT_BIOS
 317                 goto fail0;
 318 #endif
 319         }
 320
 321         /* cache is irrelevant for PCI bus reads/writes */
 322         ovrfl_window = ioremap_nocache(pci_resource_start(ovrfl_pdev, 0),
 323                                 pci_resource_len(ovrfl_pdev, 0));
 324
 325         if (!ovrfl_window) {
 326                 i82875p_printk(KERN_ERR, "%s(): Failed to ioremap bar6\n",
 327                         __func__);
 328                 goto fail1;
 329         }
 330
 331         /* need to find out the number of channels */
 332         drc = readl(ovrfl_window + I82875P_DRC);
 333         drc_chan = ((drc >> 21) & 0x1);
 334         nr_chans = drc_chan + 1;
 335
 336         drc_ddim = (drc >> 18) & 0x1;
 337         mci = edac_mc_alloc(sizeof(*pvt), I82875P_NR_CSROWS(nr_chans),
 338                                 nr_chans);
 339
 340         if (!mci) {
 341                 rc = -ENOMEM;
 342                 goto fail2;
 343         }
 344
 345         debugf3("%s(): init mci\n", __func__);
 346         mci->dev = &pdev->dev;
 347         mci->mtype_cap = MEM_FLAG_DDR;
 348         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
 349         mci->edac_cap = EDAC_FLAG_UNKNOWN;
 350         /* adjust FLAGS */
 351
 352         mci->mod_name = EDAC_MOD_STR;
 353         mci->mod_ver = I82875P_REVISION;
 354         mci->ctl_name = i82875p_devs[dev_idx].ctl_name;
 355         mci->edac_check = i82875p_check;
 356         mci->ctl_page_to_phys = NULL;
 357         debugf3("%s(): init pvt\n", __func__);
 358         pvt = (struct i82875p_pvt *) mci->pvt_info;
 359         pvt->ovrfl_pdev = ovrfl_pdev;
 360         pvt->ovrfl_window = ovrfl_window;
 361
 362         /*
 363          * The dram row boundary (DRB) reg values are boundary address
 364          * for each DRAM row with a granularity of 32 or 64MB (single/dual
 365          * channel operation).  DRB regs are cumulative; therefore DRB7 will
 366          * contain the total memory contained in all eight rows.
 367          */
 368         for (last_cumul_size = index = 0; index < mci->nr_csrows; index++) {
 369                 u8 value;
 370                 u32 cumul_size;
 371                 struct csrow_info *csrow = &mci->csrows[index];
 372
 373                 value = readb(ovrfl_window + I82875P_DRB + index);
 374                 cumul_size = value << (I82875P_DRB_SHIFT - PAGE_SHIFT);
 375                 debugf3("%s(): (%d) cumul_size 0x%x\n", __func__, index,
 376                         cumul_size);
 377
 378                 if (cumul_size == last_cumul_size)
 379                         continue;       /* not populated */
 380
 381                 csrow->first_page = last_cumul_size;
 382                 csrow->last_page = cumul_size - 1;
 383                 csrow->nr_pages = cumul_size - last_cumul_size;
 384                 last_cumul_size = cumul_size;
 385                 csrow->grain = 1 << 12;  /* I82875P_EAP has 4KiB reolution */
 386                 csrow->mtype = MEM_DDR;
 387                 csrow->dtype = DEV_UNKNOWN;
 388                 csrow->edac_mode = drc_ddim ? EDAC_SECDED : EDAC_NONE;
 389         }
 390
 391         i82875p_get_error_info(mci, &discard);  /* clear counters */
 392
 393         /* Here we assume that we will never see multiple instances of this
 394          * type of memory controller.  The ID is therefore hardcoded to 0.
 395          */
 396         if (edac_mc_add_mc(mci,0)) {
 397                 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
 398                 goto fail3;
 399         }
 400
 401         /* get this far and it's successful */
 402         debugf3("%s(): success\n", __func__);
 403         return 0;
 404
 405 fail3:
 406         edac_mc_free(mci);
 407
 408 fail2:
 409         iounmap(ovrfl_window);
 410
 411 fail1:
 412         pci_release_regions(ovrfl_pdev);
 413
 414 #ifdef CORRECT_BIOS
 415 fail0:
 416 #endif
 417         pci_disable_device(ovrfl_pdev);
 418         /* NOTE: the ovrfl proc entry and pci_dev are intentionally left */
 419         return rc;
 420 }
 421
 422 /* returns count (>= 0), or negative on error */
 423 static int __devinit i82875p_init_one(struct pci_dev *pdev,
 424                 const struct pci_device_id *ent)
 425 {
 426         int rc;
 427
 428         debugf0("%s()\n", __func__);
 429         i82875p_printk(KERN_INFO, "i82875p init one\n");
 430
 431         if (pci_enable_device(pdev) < 0)
 432                 return -EIO;
 433
 434         rc = i82875p_probe1(pdev, ent->driver_data);
 435
 436         if (mci_pdev == NULL)
 437                 mci_pdev = pci_dev_get(pdev);
 438
 439         return rc;
 440 }
 441
 442 static void __devexit i82875p_remove_one(struct pci_dev *pdev)
 443 {
 444         struct mem_ctl_info *mci;
 445         struct i82875p_pvt *pvt = NULL;
 446
 447         debugf0("%s()\n", __func__);
 448
 449         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
 450                 return;
 451
 452         pvt = (struct i82875p_pvt *) mci->pvt_info;
 453
 454         if (pvt->ovrfl_window)
 455                 iounmap(pvt->ovrfl_window);
 456
 457         if (pvt->ovrfl_pdev) {
 458 #ifdef CORRECT_BIOS
 459                 pci_release_regions(pvt->ovrfl_pdev);
 460 #endif                          /*CORRECT_BIOS */
 461                 pci_disable_device(pvt->ovrfl_pdev);
 462                 pci_dev_put(pvt->ovrfl_pdev);
 463         }
 464
 465         edac_mc_free(mci);
 466 }
 467
 468 static const struct pci_device_id i82875p_pci_tbl[] __devinitdata = {
 469         {
 470                 PCI_VEND_DEV(INTEL, 82875_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 471                 I82875P
 472         },
 473         {
 474                 0,
 475         }       /* 0 terminated list. */
 476 };
 477
 478 MODULE_DEVICE_TABLE(pci, i82875p_pci_tbl);
 479
 480 static struct pci_driver i82875p_driver = {
 481         .name = EDAC_MOD_STR,
 482         .probe = i82875p_init_one,
 483         .remove = __devexit_p(i82875p_remove_one),
 484         .id_table = i82875p_pci_tbl,
 485 };
 486
 487 static int __init i82875p_init(void)
 488 {
 489         int pci_rc;
 490
 491         debugf3("%s()\n", __func__);
 492         pci_rc = pci_register_driver(&i82875p_driver);
 493
 494         if (pci_rc < 0)
 495                 goto fail0;
 496
 497         if (mci_pdev == NULL) {
 498                 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 499                                 PCI_DEVICE_ID_INTEL_82875_0, NULL);
 500
 501                 if (!mci_pdev) {
 502                         debugf0("875p pci_get_device fail\n");
 503                         pci_rc = -ENODEV;
 504                         goto fail1;
 505                 }
 506
 507                 pci_rc = i82875p_init_one(mci_pdev, i82875p_pci_tbl);
 508
 509                 if (pci_rc < 0) {
 510                         debugf0("875p init fail\n");
 511                         pci_rc = -ENODEV;
 512                         goto fail1;
 513                 }
 514         }
 515
 516         return 0;
 517
 518 fail1:
 519         pci_unregister_driver(&i82875p_driver);
 520
 521 fail0:
 522         if (mci_pdev != NULL)
 523                 pci_dev_put(mci_pdev);
 524
 525         return pci_rc;
 526 }
 527
 528 static void __exit i82875p_exit(void)
 529 {
 530         debugf3("%s()\n", __func__);
 531
 532         pci_unregister_driver(&i82875p_driver);
 533
 534         if (!i82875p_registered) {
 535                 i82875p_remove_one(mci_pdev);
 536                 pci_dev_put(mci_pdev);
 537         }
 538 }
 539
 540 module_init(i82875p_init);
 541 module_exit(i82875p_exit);
 542
 543 MODULE_LICENSE("GPL");
 544 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh");
 545 MODULE_DESCRIPTION("MC support for Intel 82875 memory hub controllers");