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 #ifndef PCI_DEVICE_ID_INTEL_82875_0
  30 #define PCI_DEVICE_ID_INTEL_82875_0     0x2578
  31 #endif                          /* PCI_DEVICE_ID_INTEL_82875_0 */
  32
  33 #ifndef PCI_DEVICE_ID_INTEL_82875_6
  34 #define PCI_DEVICE_ID_INTEL_82875_6     0x257e
  35 #endif                          /* PCI_DEVICE_ID_INTEL_82875_6 */
  36
  37
  38 /* four csrows in dual channel, eight in single channel */
  39 #define I82875P_NR_CSROWS(nr_chans) (8/(nr_chans))
  40
  41
  42 /* Intel 82875p register addresses - device 0 function 0 - DRAM Controller */
  43 #define I82875P_EAP             0x58    /* Error Address Pointer (32b)
  44                                          *
  45                                          * 31:12 block address
  46                                          * 11:0  reserved
  47                                          */
  48
  49 #define I82875P_DERRSYN         0x5c    /* DRAM Error Syndrome (8b)
  50                                          *
  51                                          *  7:0  DRAM ECC Syndrome
  52                                          */
  53
  54 #define I82875P_DES             0x5d    /* DRAM Error Status (8b)
  55                                          *
  56                                          *  7:1  reserved
  57                                          *  0    Error channel 0/1
  58                                          */
  59
  60 #define I82875P_ERRSTS          0xc8    /* Error Status Register (16b)
  61                                          *
  62                                          * 15:10 reserved
  63                                          *  9    non-DRAM lock error (ndlock)
  64                                          *  8    Sftwr Generated SMI
  65                                          *  7    ECC UE
  66                                          *  6    reserved
  67                                          *  5    MCH detects unimplemented cycle
  68                                          *  4    AGP access outside GA
  69                                          *  3    Invalid AGP access
  70                                          *  2    Invalid GA translation table
  71                                          *  1    Unsupported AGP command
  72                                          *  0    ECC CE
  73                                          */
  74
  75 #define I82875P_ERRCMD          0xca    /* Error Command (16b)
  76                                          *
  77                                          * 15:10 reserved
  78                                          *  9    SERR on non-DRAM lock
  79                                          *  8    SERR on ECC UE
  80                                          *  7    SERR on ECC CE
  81                                          *  6    target abort on high exception
  82                                          *  5    detect unimplemented cyc
  83                                          *  4    AGP access outside of GA
  84                                          *  3    SERR on invalid AGP access
  85                                          *  2    invalid translation table
  86                                          *  1    SERR on unsupported AGP command
  87                                          *  0    reserved
  88                                          */
  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
 155 enum i82875p_chips {
 156         I82875P = 0,
 157 };
 158
 159
 160 struct i82875p_pvt {
 161         struct pci_dev *ovrfl_pdev;
 162         void __iomem *ovrfl_window;
 163 };
 164
 165
 166 struct i82875p_dev_info {
 167         const char *ctl_name;
 168 };
 169
 170
 171 struct i82875p_error_info {
 172         u16 errsts;
 173         u32 eap;
 174         u8 des;
 175         u8 derrsyn;
 176         u16 errsts2;
 177 };
 178
 179
 180 static const struct i82875p_dev_info i82875p_devs[] = {
 181         [I82875P] = {
 182                      .ctl_name = "i82875p"},
 183 };
 184
 185 static struct pci_dev *mci_pdev = NULL; /* init dev: in case that AGP code
 186                                            has already registered driver */
 187 static int i82875p_registered = 1;
 188
 189 static void i82875p_get_error_info (struct mem_ctl_info *mci,
 190                 struct i82875p_error_info *info)
 191 {
 192         /*
 193          * This is a mess because there is no atomic way to read all the
 194          * registers at once and the registers can transition from CE being
 195          * overwritten by UE.
 196          */
 197         pci_read_config_word(mci->pdev, I82875P_ERRSTS, &info->errsts);
 198         pci_read_config_dword(mci->pdev, I82875P_EAP, &info->eap);
 199         pci_read_config_byte(mci->pdev, I82875P_DES, &info->des);
 200         pci_read_config_byte(mci->pdev, I82875P_DERRSYN, &info->derrsyn);
 201         pci_read_config_word(mci->pdev, I82875P_ERRSTS, &info->errsts2);
 202
 203         pci_write_bits16(mci->pdev, I82875P_ERRSTS, 0x0081, 0x0081);
 204
 205         /*
 206          * If the error is the same then we can for both reads then
 207          * the first set of reads is valid.  If there is a change then
 208          * there is a CE no info and the second set of reads is valid
 209          * and should be UE info.
 210          */
 211         if (!(info->errsts2 & 0x0081))
 212                 return;
 213         if ((info->errsts ^ info->errsts2) & 0x0081) {
 214                 pci_read_config_dword(mci->pdev, I82875P_EAP, &info->eap);
 215                 pci_read_config_byte(mci->pdev, I82875P_DES, &info->des);
 216                 pci_read_config_byte(mci->pdev, I82875P_DERRSYN,
 217                     &info->derrsyn);
 218         }
 219 }
 220
 221 static int i82875p_process_error_info (struct mem_ctl_info *mci,
 222                 struct i82875p_error_info *info, int handle_errors)
 223 {
 224         int row, multi_chan;
 225
 226         multi_chan = mci->csrows[0].nr_channels - 1;
 227
 228         if (!(info->errsts2 & 0x0081))
 229                 return 0;
 230
 231         if (!handle_errors)
 232                 return 1;
 233
 234         if ((info->errsts ^ info->errsts2) & 0x0081) {
 235                 edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 236                 info->errsts = info->errsts2;
 237         }
 238
 239         info->eap >>= PAGE_SHIFT;
 240         row = edac_mc_find_csrow_by_page(mci, info->eap);
 241
 242         if (info->errsts & 0x0080)
 243                 edac_mc_handle_ue(mci, info->eap, 0, row, "i82875p UE");
 244         else
 245                 edac_mc_handle_ce(mci, info->eap, 0, info->derrsyn, row,
 246                                        multi_chan ? (info->des & 0x1) : 0,
 247                                        "i82875p CE");
 248
 249         return 1;
 250 }
 251
 252
 253 static void i82875p_check(struct mem_ctl_info *mci)
 254 {
 255         struct i82875p_error_info info;
 256
 257         debugf1("MC%d: " __FILE__ ": %s()\n", mci->mc_idx, __func__);
 258         i82875p_get_error_info(mci, &info);
 259         i82875p_process_error_info(mci, &info, 1);
 260 }
 261
 262
 263 #ifdef CONFIG_PROC_FS
 264 extern int pci_proc_attach_device(struct pci_dev *);
 265 #endif
 266
 267 static int i82875p_probe1(struct pci_dev *pdev, int dev_idx)
 268 {
 269         int rc = -ENODEV;
 270         int index;
 271         struct mem_ctl_info *mci = NULL;
 272         struct i82875p_pvt *pvt = NULL;
 273         unsigned long last_cumul_size;
 274         struct pci_dev *ovrfl_pdev;
 275         void __iomem *ovrfl_window = NULL;
 276
 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
 282         debugf0("MC: " __FILE__ ": %s()\n", __func__);
 283
 284         ovrfl_pdev = pci_find_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                 if (!ovrfl_pdev)
 297                         goto fail;
 298         }
 299 #ifdef CONFIG_PROC_FS
 300         if (!ovrfl_pdev->procent && pci_proc_attach_device(ovrfl_pdev)) {
 301                 printk(KERN_ERR "MC: " __FILE__
 302                        ": %s(): Failed to attach overflow device\n",
 303                        __func__);
 304                 goto fail;
 305         }
 306 #endif                          /* CONFIG_PROC_FS */
 307         if (pci_enable_device(ovrfl_pdev)) {
 308                 printk(KERN_ERR "MC: " __FILE__
 309                        ": %s(): Failed to enable overflow device\n",
 310                        __func__);
 311                 goto fail;
 312         }
 313
 314         if (pci_request_regions(ovrfl_pdev, pci_name(ovrfl_pdev))) {
 315 #ifdef CORRECT_BIOS
 316                 goto fail;
 317 #endif
 318         }
 319         /* cache is irrelevant for PCI bus reads/writes */
 320         ovrfl_window = ioremap_nocache(pci_resource_start(ovrfl_pdev, 0),
 321                                        pci_resource_len(ovrfl_pdev, 0));
 322
 323         if (!ovrfl_window) {
 324                 printk(KERN_ERR "MC: " __FILE__
 325                        ": %s(): Failed to ioremap bar6\n", __func__);
 326                 goto fail;
 327         }
 328
 329         /* need to find out the number of channels */
 330         drc = readl(ovrfl_window + I82875P_DRC);
 331         drc_chan = ((drc >> 21) & 0x1);
 332         nr_chans = drc_chan + 1;
 333         drc_ddim = (drc >> 18) & 0x1;
 334
 335         mci = edac_mc_alloc(sizeof(*pvt), I82875P_NR_CSROWS(nr_chans),
 336                                  nr_chans);
 337
 338         if (!mci) {
 339                 rc = -ENOMEM;
 340                 goto fail;
 341         }
 342
 343         debugf3("MC: " __FILE__ ": %s(): init mci\n", __func__);
 344
 345         mci->pdev = pdev;
 346         mci->mtype_cap = MEM_FLAG_DDR;
 347
 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 = BS_MOD_STR;
 353         mci->mod_ver = "$Revision: 1.5.2.11 $";
 354         mci->ctl_name = i82875p_devs[dev_idx].ctl_name;
 355         mci->edac_check = i82875p_check;
 356         mci->ctl_page_to_phys = NULL;
 357
 358         debugf3("MC: " __FILE__ ": %s(): init pvt\n", __func__);
 359
 360         pvt = (struct i82875p_pvt *) mci->pvt_info;
 361         pvt->ovrfl_pdev = ovrfl_pdev;
 362         pvt->ovrfl_window = ovrfl_window;
 363
 364         /*
 365          * The dram row boundary (DRB) reg values are boundary address
 366          * for each DRAM row with a granularity of 32 or 64MB (single/dual
 367          * channel operation).  DRB regs are cumulative; therefore DRB7 will
 368          * contain the total memory contained in all eight rows.
 369          */
 370         for (last_cumul_size = index = 0; index < mci->nr_csrows; index++) {
 371                 u8 value;
 372                 u32 cumul_size;
 373                 struct csrow_info *csrow = &mci->csrows[index];
 374
 375                 value = readb(ovrfl_window + I82875P_DRB + index);
 376                 cumul_size = value << (I82875P_DRB_SHIFT - PAGE_SHIFT);
 377                 debugf3("MC: " __FILE__ ": %s(): (%d) cumul_size 0x%x\n",
 378                         __func__, index, cumul_size);
 379                 if (cumul_size == last_cumul_size)
 380                         continue;       /* not populated */
 381
 382                 csrow->first_page = last_cumul_size;
 383                 csrow->last_page = cumul_size - 1;
 384                 csrow->nr_pages = cumul_size - last_cumul_size;
 385                 last_cumul_size = cumul_size;
 386                 csrow->grain = 1 << 12; /* I82875P_EAP has 4KiB reolution */
 387                 csrow->mtype = MEM_DDR;
 388                 csrow->dtype = DEV_UNKNOWN;
 389                 csrow->edac_mode = drc_ddim ? EDAC_SECDED : EDAC_NONE;
 390         }
 391
 392         /* clear counters */
 393         pci_write_bits16(mci->pdev, I82875P_ERRSTS, 0x0081, 0x0081);
 394
 395         if (edac_mc_add_mc(mci)) {
 396                 debugf3("MC: " __FILE__
 397                         ": %s(): failed edac_mc_add_mc()\n", __func__);
 398                 goto fail;
 399         }
 400
 401         /* get this far and it's successful */
 402         debugf3("MC: " __FILE__ ": %s(): success\n", __func__);
 403         return 0;
 404
 405       fail:
 406         if (mci)
 407                 edac_mc_free(mci);
 408
 409         if (ovrfl_window)
 410                 iounmap(ovrfl_window);
 411
 412         if (ovrfl_pdev) {
 413                 pci_release_regions(ovrfl_pdev);
 414                 pci_disable_device(ovrfl_pdev);
 415         }
 416
 417         /* NOTE: the ovrfl proc entry and pci_dev are intentionally left */
 418         return rc;
 419 }
 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("MC: " __FILE__ ": %s()\n", __func__);
 429
 430         printk(KERN_INFO "i82875p init one\n");
 431         if(pci_enable_device(pdev) < 0)
 432                 return -EIO;
 433         rc = i82875p_probe1(pdev, ent->driver_data);
 434         if (mci_pdev == NULL)
 435                 mci_pdev = pci_dev_get(pdev);
 436         return rc;
 437 }
 438
 439
 440 static void __devexit i82875p_remove_one(struct pci_dev *pdev)
 441 {
 442         struct mem_ctl_info *mci;
 443         struct i82875p_pvt *pvt = NULL;
 444
 445         debugf0(__FILE__ ": %s()\n", __func__);
 446
 447         if ((mci = edac_mc_find_mci_by_pdev(pdev)) == NULL)
 448                 return;
 449
 450         pvt = (struct i82875p_pvt *) mci->pvt_info;
 451         if (pvt->ovrfl_window)
 452                 iounmap(pvt->ovrfl_window);
 453
 454         if (pvt->ovrfl_pdev) {
 455 #ifdef CORRECT_BIOS
 456                 pci_release_regions(pvt->ovrfl_pdev);
 457 #endif                          /*CORRECT_BIOS */
 458                 pci_disable_device(pvt->ovrfl_pdev);
 459                 pci_dev_put(pvt->ovrfl_pdev);
 460         }
 461
 462         if (edac_mc_del_mc(mci))
 463                 return;
 464
 465         edac_mc_free(mci);
 466 }
 467
 468
 469 static const struct pci_device_id i82875p_pci_tbl[] __devinitdata = {
 470         {PCI_VEND_DEV(INTEL, 82875_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 471          I82875P},
 472         {0,}                    /* 0 terminated list. */
 473 };
 474
 475 MODULE_DEVICE_TABLE(pci, i82875p_pci_tbl);
 476
 477
 478 static struct pci_driver i82875p_driver = {
 479         .name = BS_MOD_STR,
 480         .probe = i82875p_init_one,
 481         .remove = __devexit_p(i82875p_remove_one),
 482         .id_table = i82875p_pci_tbl,
 483 };
 484
 485
 486 static int __init i82875p_init(void)
 487 {
 488         int pci_rc;
 489
 490         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 491         pci_rc = pci_register_driver(&i82875p_driver);
 492         if (pci_rc < 0)
 493                 return pci_rc;
 494         if (mci_pdev == NULL) {
 495                 i82875p_registered = 0;
 496                 mci_pdev =
 497                     pci_get_device(PCI_VENDOR_ID_INTEL,
 498                                    PCI_DEVICE_ID_INTEL_82875_0, NULL);
 499                 if (!mci_pdev) {
 500                         debugf0("875p pci_get_device fail\n");
 501                         return -ENODEV;
 502                 }
 503                 pci_rc = i82875p_init_one(mci_pdev, i82875p_pci_tbl);
 504                 if (pci_rc < 0) {
 505                         debugf0("875p init fail\n");
 506                         pci_dev_put(mci_pdev);
 507                         return -ENODEV;
 508                 }
 509         }
 510         return 0;
 511 }
 512
 513
 514 static void __exit i82875p_exit(void)
 515 {
 516         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 517
 518         pci_unregister_driver(&i82875p_driver);
 519         if (!i82875p_registered) {
 520                 i82875p_remove_one(mci_pdev);
 521                 pci_dev_put(mci_pdev);
 522         }
 523 }
 524
 525
 526 module_init(i82875p_init);
 527 module_exit(i82875p_exit);
 528
 529
 530 MODULE_LICENSE("GPL");
 531 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh");
 532 MODULE_DESCRIPTION("MC support for Intel 82875 memory hub controllers");