SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  17 #include <linux/config.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20
  21 #include <linux/pci.h>
  22 #include <linux/pci_ids.h>
  23
  24 #include <linux/slab.h>
  25
  26 #include "edac_mc.h"
  27
  28
  29 #define i82875p_printk(level, fmt, arg...) \
  30     edac_printk(level, "i82875p", fmt, ##arg)
  31
  32
  33 #define i82875p_mc_printk(mci, level, fmt, arg...) \
  34     edac_mc_chipset_printk(mci, level, "i82875p", fmt, ##arg)
  35
  36
  37 #ifndef PCI_DEVICE_ID_INTEL_82875_0
  38 #define PCI_DEVICE_ID_INTEL_82875_0     0x2578
  39 #endif                          /* PCI_DEVICE_ID_INTEL_82875_0 */
  40
  41 #ifndef PCI_DEVICE_ID_INTEL_82875_6
  42 #define PCI_DEVICE_ID_INTEL_82875_6     0x257e
  43 #endif                          /* PCI_DEVICE_ID_INTEL_82875_6 */
  44
  45
  46 /* four csrows in dual channel, eight in single channel */
  47 #define I82875P_NR_CSROWS(nr_chans) (8/(nr_chans))
  48
  49
  50 /* Intel 82875p register addresses - device 0 function 0 - DRAM Controller */
  51 #define I82875P_EAP             0x58    /* Error Address Pointer (32b)
  52                                          *
  53                                          * 31:12 block address
  54                                          * 11:0  reserved
  55                                          */
  56
  57 #define I82875P_DERRSYN         0x5c    /* DRAM Error Syndrome (8b)
  58                                          *
  59                                          *  7:0  DRAM ECC Syndrome
  60                                          */
  61
  62 #define I82875P_DES             0x5d    /* DRAM Error Status (8b)
  63                                          *
  64                                          *  7:1  reserved
  65                                          *  0    Error channel 0/1
  66                                          */
  67
  68 #define I82875P_ERRSTS          0xc8    /* Error Status Register (16b)
  69                                          *
  70                                          * 15:10 reserved
  71                                          *  9    non-DRAM lock error (ndlock)
  72                                          *  8    Sftwr Generated SMI
  73                                          *  7    ECC UE
  74                                          *  6    reserved
  75                                          *  5    MCH detects unimplemented cycle
  76                                          *  4    AGP access outside GA
  77                                          *  3    Invalid AGP access
  78                                          *  2    Invalid GA translation table
  79                                          *  1    Unsupported AGP command
  80                                          *  0    ECC CE
  81                                          */
  82
  83 #define I82875P_ERRCMD          0xca    /* Error Command (16b)
  84                                          *
  85                                          * 15:10 reserved
  86                                          *  9    SERR on non-DRAM lock
  87                                          *  8    SERR on ECC UE
  88                                          *  7    SERR on ECC CE
  89                                          *  6    target abort on high exception
  90                                          *  5    detect unimplemented cyc
  91                                          *  4    AGP access outside of GA
  92                                          *  3    SERR on invalid AGP access
  93                                          *  2    invalid translation table
  94                                          *  1    SERR on unsupported AGP command
  95                                          *  0    reserved
  96                                          */
  97
  98
  99 /* Intel 82875p register addresses - device 6 function 0 - DRAM Controller */
 100 #define I82875P_PCICMD6         0x04    /* PCI Command Register (16b)
 101                                          *
 102                                          * 15:10 reserved
 103                                          *  9    fast back-to-back - ro 0
 104                                          *  8    SERR enable - ro 0
 105                                          *  7    addr/data stepping - ro 0
 106                                          *  6    parity err enable - ro 0
 107                                          *  5    VGA palette snoop - ro 0
 108                                          *  4    mem wr & invalidate - ro 0
 109                                          *  3    special cycle - ro 0
 110                                          *  2    bus master - ro 0
 111                                          *  1    mem access dev6 - 0(dis),1(en)
 112                                          *  0    IO access dev3 - 0(dis),1(en)
 113                                          */
 114
 115 #define I82875P_BAR6            0x10    /* Mem Delays Base ADDR Reg (32b)
 116                                          *
 117                                          * 31:12 mem base addr [31:12]
 118                                          * 11:4  address mask - ro 0
 119                                          *  3    prefetchable - ro 0(non),1(pre)
 120                                          *  2:1  mem type - ro 0
 121                                          *  0    mem space - ro 0
 122                                          */
 123
 124 /* Intel 82875p MMIO register space - device 0 function 0 - MMR space */
 125
 126 #define I82875P_DRB_SHIFT 26    /* 64MiB grain */
 127 #define I82875P_DRB             0x00    /* DRAM Row Boundary (8b x 8)
 128                                          *
 129                                          *  7    reserved
 130                                          *  6:0  64MiB row boundary addr
 131                                          */
 132
 133 #define I82875P_DRA             0x10    /* DRAM Row Attribute (4b x 8)
 134                                          *
 135                                          *  7    reserved
 136                                          *  6:4  row attr row 1
 137                                          *  3    reserved
 138                                          *  2:0  row attr row 0
 139                                          *
 140                                          * 000 =  4KiB
 141                                          * 001 =  8KiB
 142                                          * 010 = 16KiB
 143                                          * 011 = 32KiB
 144                                          */
 145
 146 #define I82875P_DRC             0x68    /* DRAM Controller Mode (32b)
 147                                          *
 148                                          * 31:30 reserved
 149                                          * 29    init complete
 150                                          * 28:23 reserved
 151                                          * 22:21 nr chan 00=1,01=2
 152                                          * 20    reserved
 153                                          * 19:18 Data Integ Mode 00=none,01=ecc
 154                                          * 17:11 reserved
 155                                          * 10:8  refresh mode
 156                                          *  7    reserved
 157                                          *  6:4  mode select
 158                                          *  3:2  reserved
 159                                          *  1:0  DRAM type 01=DDR
 160                                          */
 161
 162
 163 enum i82875p_chips {
 164         I82875P = 0,
 165 };
 166
 167
 168 struct i82875p_pvt {
 169         struct pci_dev *ovrfl_pdev;
 170         void __iomem *ovrfl_window;
 171 };
 172
 173
 174 struct i82875p_dev_info {
 175         const char *ctl_name;
 176 };
 177
 178
 179 struct i82875p_error_info {
 180         u16 errsts;
 181         u32 eap;
 182         u8 des;
 183         u8 derrsyn;
 184         u16 errsts2;
 185 };
 186
 187
 188 static const struct i82875p_dev_info i82875p_devs[] = {
 189         [I82875P] = {
 190                      .ctl_name = "i82875p"},
 191 };
 192
 193 static struct pci_dev *mci_pdev = NULL; /* init dev: in case that AGP code
 194                                            has already registered driver */
 195 static int i82875p_registered = 1;
 196
 197 static void i82875p_get_error_info (struct mem_ctl_info *mci,
 198                 struct i82875p_error_info *info)
 199 {
 200         /*
 201          * This is a mess because there is no atomic way to read all the
 202          * registers at once and the registers can transition from CE being
 203          * overwritten by UE.
 204          */
 205         pci_read_config_word(mci->pdev, I82875P_ERRSTS, &info->errsts);
 206         pci_read_config_dword(mci->pdev, I82875P_EAP, &info->eap);
 207         pci_read_config_byte(mci->pdev, I82875P_DES, &info->des);
 208         pci_read_config_byte(mci->pdev, I82875P_DERRSYN, &info->derrsyn);
 209         pci_read_config_word(mci->pdev, I82875P_ERRSTS, &info->errsts2);
 210
 211         pci_write_bits16(mci->pdev, I82875P_ERRSTS, 0x0081, 0x0081);
 212
 213         /*
 214          * If the error is the same then we can for both reads then
 215          * the first set of reads is valid.  If there is a change then
 216          * there is a CE no info and the second set of reads is valid
 217          * and should be UE info.
 218          */
 219         if (!(info->errsts2 & 0x0081))
 220                 return;
 221         if ((info->errsts ^ info->errsts2) & 0x0081) {
 222                 pci_read_config_dword(mci->pdev, I82875P_EAP, &info->eap);
 223                 pci_read_config_byte(mci->pdev, I82875P_DES, &info->des);
 224                 pci_read_config_byte(mci->pdev, I82875P_DERRSYN,
 225                     &info->derrsyn);
 226         }
 227 }
 228
 229 static int i82875p_process_error_info (struct mem_ctl_info *mci,
 230                 struct i82875p_error_info *info, int handle_errors)
 231 {
 232         int row, multi_chan;
 233
 234         multi_chan = mci->csrows[0].nr_channels - 1;
 235
 236         if (!(info->errsts2 & 0x0081))
 237                 return 0;
 238
 239         if (!handle_errors)
 240                 return 1;
 241
 242         if ((info->errsts ^ info->errsts2) & 0x0081) {
 243                 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 244                 info->errsts = info->errsts2;
 245         }
 246
 247         info->eap >>= PAGE_SHIFT;
 248         row = edac_mc_find_csrow_by_page(mci, info->eap);
 249
 250         if (info->errsts & 0x0080)
 251                 edac_mc_handle_ue(mci, info->eap, 0, row, "i82875p UE");
 252         else
 253                 edac_mc_handle_ce(mci, info->eap, 0, info->derrsyn, row,
 254                                        multi_chan ? (info->des & 0x1) : 0,
 255                                        "i82875p CE");
 256
 257         return 1;
 258 }
 259
 260
 261 static void i82875p_check(struct mem_ctl_info *mci)
 262 {
 263         struct i82875p_error_info info;
 264
 265         debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 266         i82875p_get_error_info(mci, &info);
 267         i82875p_process_error_info(mci, &info, 1);
 268 }
 269
 270
 271 #ifdef CONFIG_PROC_FS
 272 extern int pci_proc_attach_device(struct pci_dev *);
 273 #endif
 274
 275 static int i82875p_probe1(struct pci_dev *pdev, int dev_idx)
 276 {
 277         int rc = -ENODEV;
 278         int index;
 279         struct mem_ctl_info *mci = NULL;
 280         struct i82875p_pvt *pvt = NULL;
 281         unsigned long last_cumul_size;
 282         struct pci_dev *ovrfl_pdev;
 283         void __iomem *ovrfl_window = NULL;
 284
 285         u32 drc;
 286         u32 drc_chan;           /* Number of channels 0=1chan,1=2chan */
 287         u32 nr_chans;
 288         u32 drc_ddim;           /* DRAM Data Integrity Mode 0=none,2=edac */
 289         struct i82875p_error_info discard;
 290
 291         debugf0("%s()\n", __func__);
 292
 293         ovrfl_pdev = pci_get_device(PCI_VEND_DEV(INTEL, 82875_6), NULL);
 294
 295         if (!ovrfl_pdev) {
 296                 /*
 297                  * Intel tells BIOS developers to hide device 6 which
 298                  * configures the overflow device access containing
 299                  * the DRBs - this is where we expose device 6.
 300                  * http://www.x86-secret.com/articles/tweak/pat/patsecrets-2.htm
 301                  */
 302                 pci_write_bits8(pdev, 0xf4, 0x2, 0x2);
 303                 ovrfl_pdev =
 304                     pci_scan_single_device(pdev->bus, PCI_DEVFN(6, 0));
 305                 if (!ovrfl_pdev)
 306                         return -ENODEV;
 307         }
 308 #ifdef CONFIG_PROC_FS
 309         if (!ovrfl_pdev->procent && pci_proc_attach_device(ovrfl_pdev)) {
 310                 i82875p_printk(KERN_ERR,
 311                                "%s(): Failed to attach overflow device\n",
 312                                __func__);
 313                 return -ENODEV;
 314         }
 315 #endif                          /* CONFIG_PROC_FS */
 316         if (pci_enable_device(ovrfl_pdev)) {
 317                 i82875p_printk(KERN_ERR,
 318                                "%s(): Failed to enable overflow device\n",
 319                                __func__);
 320                 return -ENODEV;
 321         }
 322
 323         if (pci_request_regions(ovrfl_pdev, pci_name(ovrfl_pdev))) {
 324 #ifdef CORRECT_BIOS
 325                 goto fail0;
 326 #endif
 327         }
 328         /* cache is irrelevant for PCI bus reads/writes */
 329         ovrfl_window = ioremap_nocache(pci_resource_start(ovrfl_pdev, 0),
 330                                        pci_resource_len(ovrfl_pdev, 0));
 331
 332         if (!ovrfl_window) {
 333                 i82875p_printk(KERN_ERR, "%s(): Failed to ioremap bar6\n",
 334                                __func__);
 335                 goto fail1;
 336         }
 337
 338         /* need to find out the number of channels */
 339         drc = readl(ovrfl_window + I82875P_DRC);
 340         drc_chan = ((drc >> 21) & 0x1);
 341         nr_chans = drc_chan + 1;
 342         drc_ddim = (drc >> 18) & 0x1;
 343
 344         mci = edac_mc_alloc(sizeof(*pvt), I82875P_NR_CSROWS(nr_chans),
 345                                  nr_chans);
 346
 347         if (!mci) {
 348                 rc = -ENOMEM;
 349                 goto fail2;
 350         }
 351
 352         debugf3("%s(): init mci\n", __func__);
 353
 354         mci->pdev = pdev;
 355         mci->mtype_cap = MEM_FLAG_DDR;
 356
 357         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
 358         mci->edac_cap = EDAC_FLAG_UNKNOWN;
 359         /* adjust FLAGS */
 360
 361         mci->mod_name = EDAC_MOD_STR;
 362         mci->mod_ver = "$Revision: 1.5.2.11 $";
 363         mci->ctl_name = i82875p_devs[dev_idx].ctl_name;
 364         mci->edac_check = i82875p_check;
 365         mci->ctl_page_to_phys = NULL;
 366
 367         debugf3("%s(): init pvt\n", __func__);
 368
 369         pvt = (struct i82875p_pvt *) mci->pvt_info;
 370         pvt->ovrfl_pdev = ovrfl_pdev;
 371         pvt->ovrfl_window = ovrfl_window;
 372
 373         /*
 374          * The dram row boundary (DRB) reg values are boundary address
 375          * for each DRAM row with a granularity of 32 or 64MB (single/dual
 376          * channel operation).  DRB regs are cumulative; therefore DRB7 will
 377          * contain the total memory contained in all eight rows.
 378          */
 379         for (last_cumul_size = index = 0; index < mci->nr_csrows; index++) {
 380                 u8 value;
 381                 u32 cumul_size;
 382                 struct csrow_info *csrow = &mci->csrows[index];
 383
 384                 value = readb(ovrfl_window + I82875P_DRB + index);
 385                 cumul_size = value << (I82875P_DRB_SHIFT - PAGE_SHIFT);
 386                 debugf3("%s(): (%d) cumul_size 0x%x\n", __func__, index,
 387                         cumul_size);
 388                 if (cumul_size == last_cumul_size)
 389                         continue;       /* not populated */
 390
 391                 csrow->first_page = last_cumul_size;
 392                 csrow->last_page = cumul_size - 1;
 393                 csrow->nr_pages = cumul_size - last_cumul_size;
 394                 last_cumul_size = cumul_size;
 395                 csrow->grain = 1 << 12; /* I82875P_EAP has 4KiB reolution */
 396                 csrow->mtype = MEM_DDR;
 397                 csrow->dtype = DEV_UNKNOWN;
 398                 csrow->edac_mode = drc_ddim ? EDAC_SECDED : EDAC_NONE;
 399         }
 400
 401         i82875p_get_error_info(mci, &discard);  /* clear counters */
 402
 403         if (edac_mc_add_mc(mci)) {
 404                 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
 405                 goto fail3;
 406         }
 407
 408         /* get this far and it's successful */
 409         debugf3("%s(): success\n", __func__);
 410         return 0;
 411
 412 fail3:
 413         edac_mc_free(mci);
 414
 415 fail2:
 416         iounmap(ovrfl_window);
 417
 418 fail1:
 419         pci_release_regions(ovrfl_pdev);
 420
 421 #ifdef CORRECT_BIOS
 422 fail0:
 423 #endif
 424         pci_disable_device(ovrfl_pdev);
 425         /* NOTE: the ovrfl proc entry and pci_dev are intentionally left */
 426         return rc;
 427 }
 428
 429
 430 /* returns count (>= 0), or negative on error */
 431 static int __devinit i82875p_init_one(struct pci_dev *pdev,
 432                                       const struct pci_device_id *ent)
 433 {
 434         int rc;
 435
 436         debugf0("%s()\n", __func__);
 437
 438         i82875p_printk(KERN_INFO, "i82875p init one\n");
 439         if(pci_enable_device(pdev) < 0)
 440                 return -EIO;
 441         rc = i82875p_probe1(pdev, ent->driver_data);
 442         if (mci_pdev == NULL)
 443                 mci_pdev = pci_dev_get(pdev);
 444         return rc;
 445 }
 446
 447
 448 static void __devexit i82875p_remove_one(struct pci_dev *pdev)
 449 {
 450         struct mem_ctl_info *mci;
 451         struct i82875p_pvt *pvt = NULL;
 452
 453         debugf0("%s()\n", __func__);
 454
 455         if ((mci = edac_mc_del_mc(pdev)) == NULL)
 456                 return;
 457
 458         pvt = (struct i82875p_pvt *) mci->pvt_info;
 459         if (pvt->ovrfl_window)
 460                 iounmap(pvt->ovrfl_window);
 461
 462         if (pvt->ovrfl_pdev) {
 463 #ifdef CORRECT_BIOS
 464                 pci_release_regions(pvt->ovrfl_pdev);
 465 #endif                          /*CORRECT_BIOS */
 466                 pci_disable_device(pvt->ovrfl_pdev);
 467                 pci_dev_put(pvt->ovrfl_pdev);
 468         }
 469
 470         edac_mc_free(mci);
 471 }
 472
 473
 474 static const struct pci_device_id i82875p_pci_tbl[] __devinitdata = {
 475         {PCI_VEND_DEV(INTEL, 82875_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 476          I82875P},
 477         {0,}                    /* 0 terminated list. */
 478 };
 479
 480 MODULE_DEVICE_TABLE(pci, i82875p_pci_tbl);
 481
 482
 483 static struct pci_driver i82875p_driver = {
 484         .name = EDAC_MOD_STR,
 485         .probe = i82875p_init_one,
 486         .remove = __devexit_p(i82875p_remove_one),
 487         .id_table = i82875p_pci_tbl,
 488 };
 489
 490
 491 static int __init i82875p_init(void)
 492 {
 493         int pci_rc;
 494
 495         debugf3("%s()\n", __func__);
 496         pci_rc = pci_register_driver(&i82875p_driver);
 497         if (pci_rc < 0)
 498                 goto fail0;
 499         if (mci_pdev == NULL) {
 500                 mci_pdev =
 501                     pci_get_device(PCI_VENDOR_ID_INTEL,
 502                                    PCI_DEVICE_ID_INTEL_82875_0, NULL);
 503                 if (!mci_pdev) {
 504                         debugf0("875p pci_get_device fail\n");
 505                         pci_rc = -ENODEV;
 506                         goto fail1;
 507                 }
 508                 pci_rc = i82875p_init_one(mci_pdev, i82875p_pci_tbl);
 509                 if (pci_rc < 0) {
 510                         debugf0("875p init fail\n");
 511                         pci_rc = -ENODEV;
 512                         goto fail1;
 513                 }
 514         }
 515         return 0;
 516
 517 fail1:
 518         pci_unregister_driver(&i82875p_driver);
 519
 520 fail0:
 521         if (mci_pdev != NULL)
 522                 pci_dev_put(mci_pdev);
 523
 524         return pci_rc;
 525 }
 526
 527
 528 static void __exit i82875p_exit(void)
 529 {
 530         debugf3("%s()\n", __func__);
 531
 532         pci_unregister_driver(&i82875p_driver);
 533         if (!i82875p_registered) {
 534                 i82875p_remove_one(mci_pdev);
 535                 pci_dev_put(mci_pdev);
 536         }
 537 }
 538
 539
 540 module_init(i82875p_init);
 541 module_exit(i82875p_exit);
 542
 543
 544 MODULE_LICENSE("GPL");
 545 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh");
 546 MODULE_DESCRIPTION("MC support for Intel 82875 memory hub controllers");