drivers/edac/e7xxx_edac.c

   1 /*
   2  * Intel e7xxx 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  * See "enum e7xxx_chips" below for supported chipsets
   8  *
   9  * Written by Thayne Harbaugh
  10  * Based on work by Dan Hollis <goemon at anime dot net> and others.
  11  *      http://www.anime.net/~goemon/linux-ecc/
  12  *
  13  * Contributors:
  14  *      Eric Biederman (Linux Networx)
  15  *      Tom Zimmerman (Linux Networx)
  16  *      Jim Garlick (Lawrence Livermore National Labs)
  17  *      Dave Peterson (Lawrence Livermore National Labs)
  18  *      That One Guy (Some other place)
  19  *      Wang Zhenyu (intel.com)
  20  *
  21  * $Id: edac_e7xxx.c,v 1.5.2.9 2005/10/05 00:43:44 dsp_llnl Exp $
  22  *
  23  */
  24
  25
  26 #include <linux/config.h>
  27 #include <linux/module.h>
  28 #include <linux/init.h>
  29 #include <linux/pci.h>
  30 #include <linux/pci_ids.h>
  31 #include <linux/slab.h>
  32 #include "edac_mc.h"
  33
  34
  35 #ifndef PCI_DEVICE_ID_INTEL_7205_0
  36 #define PCI_DEVICE_ID_INTEL_7205_0      0x255d
  37 #endif                          /* PCI_DEVICE_ID_INTEL_7205_0 */
  38
  39 #ifndef PCI_DEVICE_ID_INTEL_7205_1_ERR
  40 #define PCI_DEVICE_ID_INTEL_7205_1_ERR  0x2551
  41 #endif                          /* PCI_DEVICE_ID_INTEL_7205_1_ERR */
  42
  43 #ifndef PCI_DEVICE_ID_INTEL_7500_0
  44 #define PCI_DEVICE_ID_INTEL_7500_0      0x2540
  45 #endif                          /* PCI_DEVICE_ID_INTEL_7500_0 */
  46
  47 #ifndef PCI_DEVICE_ID_INTEL_7500_1_ERR
  48 #define PCI_DEVICE_ID_INTEL_7500_1_ERR  0x2541
  49 #endif                          /* PCI_DEVICE_ID_INTEL_7500_1_ERR */
  50
  51 #ifndef PCI_DEVICE_ID_INTEL_7501_0
  52 #define PCI_DEVICE_ID_INTEL_7501_0      0x254c
  53 #endif                          /* PCI_DEVICE_ID_INTEL_7501_0 */
  54
  55 #ifndef PCI_DEVICE_ID_INTEL_7501_1_ERR
  56 #define PCI_DEVICE_ID_INTEL_7501_1_ERR  0x2541
  57 #endif                          /* PCI_DEVICE_ID_INTEL_7501_1_ERR */
  58
  59 #ifndef PCI_DEVICE_ID_INTEL_7505_0
  60 #define PCI_DEVICE_ID_INTEL_7505_0      0x2550
  61 #endif                          /* PCI_DEVICE_ID_INTEL_7505_0 */
  62
  63 #ifndef PCI_DEVICE_ID_INTEL_7505_1_ERR
  64 #define PCI_DEVICE_ID_INTEL_7505_1_ERR  0x2551
  65 #endif                          /* PCI_DEVICE_ID_INTEL_7505_1_ERR */
  66
  67
  68 #define E7XXX_NR_CSROWS         8       /* number of csrows */
  69 #define E7XXX_NR_DIMMS          8       /* FIXME - is this correct? */
  70
  71
  72 /* E7XXX register addresses - device 0 function 0 */
  73 #define E7XXX_DRB               0x60    /* DRAM row boundary register (8b) */
  74 #define E7XXX_DRA               0x70    /* DRAM row attribute register (8b) */
  75                                         /*
  76                                          * 31   Device width row 7 0=x8 1=x4
  77                                          * 27   Device width row 6
  78                                          * 23   Device width row 5
  79                                          * 19   Device width row 4
  80                                          * 15   Device width row 3
  81                                          * 11   Device width row 2
  82                                          *  7   Device width row 1
  83                                          *  3   Device width row 0
  84                                          */
  85 #define E7XXX_DRC               0x7C    /* DRAM controller mode reg (32b) */
  86                                         /*
  87                                          * 22    Number channels 0=1,1=2
  88                                          * 19:18 DRB Granularity 32/64MB
  89                                          */
  90 #define E7XXX_TOLM              0xC4    /* DRAM top of low memory reg (16b) */
  91 #define E7XXX_REMAPBASE         0xC6    /* DRAM remap base address reg (16b) */
  92 #define E7XXX_REMAPLIMIT        0xC8    /* DRAM remap limit address reg (16b) */
  93
  94 /* E7XXX register addresses - device 0 function 1 */
  95 #define E7XXX_DRAM_FERR         0x80    /* DRAM first error register (8b) */
  96 #define E7XXX_DRAM_NERR         0x82    /* DRAM next error register (8b) */
  97 #define E7XXX_DRAM_CELOG_ADD    0xA0    /* DRAM first correctable memory */
  98                                         /*     error address register (32b) */
  99                                         /*
 100                                          * 31:28 Reserved
 101                                          * 27:6  CE address (4k block 33:12)
 102                                          *  5:0  Reserved
 103                                          */
 104 #define E7XXX_DRAM_UELOG_ADD    0xB0    /* DRAM first uncorrectable memory */
 105                                         /*     error address register (32b) */
 106                                         /*
 107                                          * 31:28 Reserved
 108                                          * 27:6  CE address (4k block 33:12)
 109                                          *  5:0  Reserved
 110                                          */
 111 #define E7XXX_DRAM_CELOG_SYNDROME 0xD0  /* DRAM first correctable memory */
 112                                         /*     error syndrome register (16b) */
 113
 114 enum e7xxx_chips {
 115         E7500 = 0,
 116         E7501,
 117         E7505,
 118         E7205,
 119 };
 120
 121
 122 struct e7xxx_pvt {
 123         struct pci_dev *bridge_ck;
 124         u32 tolm;
 125         u32 remapbase;
 126         u32 remaplimit;
 127         const struct e7xxx_dev_info *dev_info;
 128 };
 129
 130
 131 struct e7xxx_dev_info {
 132         u16 err_dev;
 133         const char *ctl_name;
 134 };
 135
 136
 137 struct e7xxx_error_info {
 138         u8 dram_ferr;
 139         u8 dram_nerr;
 140         u32 dram_celog_add;
 141         u16 dram_celog_syndrome;
 142         u32 dram_uelog_add;
 143 };
 144
 145 static const struct e7xxx_dev_info e7xxx_devs[] = {
 146         [E7500] = {
 147                    .err_dev = PCI_DEVICE_ID_INTEL_7500_1_ERR,
 148                    .ctl_name = "E7500"},
 149         [E7501] = {
 150                    .err_dev = PCI_DEVICE_ID_INTEL_7501_1_ERR,
 151                    .ctl_name = "E7501"},
 152         [E7505] = {
 153                    .err_dev = PCI_DEVICE_ID_INTEL_7505_1_ERR,
 154                    .ctl_name = "E7505"},
 155         [E7205] = {
 156                    .err_dev = PCI_DEVICE_ID_INTEL_7205_1_ERR,
 157                    .ctl_name = "E7205"},
 158 };
 159
 160
 161 /* FIXME - is this valid for both SECDED and S4ECD4ED? */
 162 static inline int e7xxx_find_channel(u16 syndrome)
 163 {
 164         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 165
 166         if ((syndrome & 0xff00) == 0)
 167                 return 0;
 168         if ((syndrome & 0x00ff) == 0)
 169                 return 1;
 170         if ((syndrome & 0xf000) == 0 || (syndrome & 0x0f00) == 0)
 171                 return 0;
 172         return 1;
 173 }
 174
 175
 176 static unsigned long
 177 ctl_page_to_phys(struct mem_ctl_info *mci, unsigned long page)
 178 {
 179         u32 remap;
 180         struct e7xxx_pvt *pvt = (struct e7xxx_pvt *) mci->pvt_info;
 181
 182         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 183
 184         if ((page < pvt->tolm) ||
 185             ((page >= 0x100000) && (page < pvt->remapbase)))
 186                 return page;
 187         remap = (page - pvt->tolm) + pvt->remapbase;
 188         if (remap < pvt->remaplimit)
 189                 return remap;
 190         printk(KERN_ERR "Invalid page %lx - out of range\n", page);
 191         return pvt->tolm - 1;
 192 }
 193
 194
 195 static void process_ce(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
 196 {
 197         u32 error_1b, page;
 198         u16 syndrome;
 199         int row;
 200         int channel;
 201
 202         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 203
 204         /* read the error address */
 205         error_1b = info->dram_celog_add;
 206         /* FIXME - should use PAGE_SHIFT */
 207         page = error_1b >> 6;   /* convert the address to 4k page */
 208         /* read the syndrome */
 209         syndrome = info->dram_celog_syndrome;
 210         /* FIXME - check for -1 */
 211         row = edac_mc_find_csrow_by_page(mci, page);
 212         /* convert syndrome to channel */
 213         channel = e7xxx_find_channel(syndrome);
 214         edac_mc_handle_ce(mci, page, 0, syndrome, row, channel,
 215                                "e7xxx CE");
 216 }
 217
 218
 219 static void process_ce_no_info(struct mem_ctl_info *mci)
 220 {
 221         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 222         edac_mc_handle_ce_no_info(mci, "e7xxx CE log register overflow");
 223 }
 224
 225
 226 static void process_ue(struct mem_ctl_info *mci, struct e7xxx_error_info *info)
 227 {
 228         u32 error_2b, block_page;
 229         int row;
 230
 231         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 232
 233         /* read the error address */
 234         error_2b = info->dram_uelog_add;
 235         /* FIXME - should use PAGE_SHIFT */
 236         block_page = error_2b >> 6;     /* convert to 4k address */
 237         row = edac_mc_find_csrow_by_page(mci, block_page);
 238         edac_mc_handle_ue(mci, block_page, 0, row, "e7xxx UE");
 239 }
 240
 241
 242 static void process_ue_no_info(struct mem_ctl_info *mci)
 243 {
 244         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 245         edac_mc_handle_ue_no_info(mci, "e7xxx UE log register overflow");
 246 }
 247
 248
 249 static void e7xxx_get_error_info (struct mem_ctl_info *mci,
 250                 struct e7xxx_error_info *info)
 251 {
 252         struct e7xxx_pvt *pvt;
 253
 254         pvt = (struct e7xxx_pvt *) mci->pvt_info;
 255         pci_read_config_byte(pvt->bridge_ck, E7XXX_DRAM_FERR,
 256             &info->dram_ferr);
 257         pci_read_config_byte(pvt->bridge_ck, E7XXX_DRAM_NERR,
 258             &info->dram_nerr);
 259
 260         if ((info->dram_ferr & 1) || (info->dram_nerr & 1)) {
 261                 pci_read_config_dword(pvt->bridge_ck, E7XXX_DRAM_CELOG_ADD,
 262                     &info->dram_celog_add);
 263                 pci_read_config_word(pvt->bridge_ck,
 264                     E7XXX_DRAM_CELOG_SYNDROME, &info->dram_celog_syndrome);
 265         }
 266
 267         if ((info->dram_ferr & 2) || (info->dram_nerr & 2))
 268                 pci_read_config_dword(pvt->bridge_ck, E7XXX_DRAM_UELOG_ADD,
 269                     &info->dram_uelog_add);
 270
 271         if (info->dram_ferr & 3)
 272                 pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_FERR, 0x03,
 273                     0x03);
 274
 275         if (info->dram_nerr & 3)
 276                 pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_NERR, 0x03,
 277                     0x03);
 278 }
 279
 280
 281 static int e7xxx_process_error_info (struct mem_ctl_info *mci,
 282                 struct e7xxx_error_info *info, int handle_errors)
 283 {
 284         int error_found;
 285
 286         error_found = 0;
 287
 288         /* decode and report errors */
 289         if (info->dram_ferr & 1) {      /* check first error correctable */
 290                 error_found = 1;
 291
 292                 if (handle_errors)
 293                         process_ce(mci, info);
 294         }
 295
 296         if (info->dram_ferr & 2) {      /* check first error uncorrectable */
 297                 error_found = 1;
 298
 299                 if (handle_errors)
 300                         process_ue(mci, info);
 301         }
 302
 303         if (info->dram_nerr & 1) {      /* check next error correctable */
 304                 error_found = 1;
 305
 306                 if (handle_errors) {
 307                         if (info->dram_ferr & 1)
 308                                 process_ce_no_info(mci);
 309                         else
 310                                 process_ce(mci, info);
 311                 }
 312         }
 313
 314         if (info->dram_nerr & 2) {      /* check next error uncorrectable */
 315                 error_found = 1;
 316
 317                 if (handle_errors) {
 318                         if (info->dram_ferr & 2)
 319                                 process_ue_no_info(mci);
 320                         else
 321                                 process_ue(mci, info);
 322                 }
 323         }
 324
 325         return error_found;
 326 }
 327
 328
 329 static void e7xxx_check(struct mem_ctl_info *mci)
 330 {
 331         struct e7xxx_error_info info;
 332
 333         debugf3("MC: " __FILE__ ": %s()\n", __func__);
 334         e7xxx_get_error_info(mci, &info);
 335         e7xxx_process_error_info(mci, &info, 1);
 336 }
 337
 338
 339 static int e7xxx_probe1(struct pci_dev *pdev, int dev_idx)
 340 {
 341         int rc = -ENODEV;
 342         int index;
 343         u16 pci_data;
 344         struct mem_ctl_info *mci = NULL;
 345         struct e7xxx_pvt *pvt = NULL;
 346         u32 drc;
 347         int drc_chan = 1;       /* Number of channels 0=1chan,1=2chan */
 348         int drc_drbg = 1;       /* DRB granularity 0=32mb,1=64mb */
 349         int drc_ddim;           /* DRAM Data Integrity Mode 0=none,2=edac */
 350         u32 dra;
 351         unsigned long last_cumul_size;
 352
 353
 354         debugf0("MC: " __FILE__ ": %s(): mci\n", __func__);
 355
 356         /* need to find out the number of channels */
 357         pci_read_config_dword(pdev, E7XXX_DRC, &drc);
 358         /* only e7501 can be single channel */
 359         if (dev_idx == E7501) {
 360                 drc_chan = ((drc >> 22) & 0x1);
 361                 drc_drbg = (drc >> 18) & 0x3;
 362         }
 363         drc_ddim = (drc >> 20) & 0x3;
 364
 365         mci = edac_mc_alloc(sizeof(*pvt), E7XXX_NR_CSROWS, drc_chan + 1);
 366
 367         if (mci == NULL) {
 368                 rc = -ENOMEM;
 369                 goto fail;
 370         }
 371
 372         debugf3("MC: " __FILE__ ": %s(): init mci\n", __func__);
 373
 374         mci->mtype_cap = MEM_FLAG_RDDR;
 375         mci->edac_ctl_cap =
 376             EDAC_FLAG_NONE | EDAC_FLAG_SECDED | EDAC_FLAG_S4ECD4ED;
 377         /* FIXME - what if different memory types are in different csrows? */
 378         mci->mod_name = BS_MOD_STR;
 379         mci->mod_ver = "$Revision: 1.5.2.9 $";
 380         mci->pdev = pdev;
 381
 382         debugf3("MC: " __FILE__ ": %s(): init pvt\n", __func__);
 383         pvt = (struct e7xxx_pvt *) mci->pvt_info;
 384         pvt->dev_info = &e7xxx_devs[dev_idx];
 385         pvt->bridge_ck = pci_get_device(PCI_VENDOR_ID_INTEL,
 386                                          pvt->dev_info->err_dev,
 387                                          pvt->bridge_ck);
 388         if (!pvt->bridge_ck) {
 389                 printk(KERN_ERR
 390                        "MC: error reporting device not found:"
 391                        "vendor %x device 0x%x (broken BIOS?)\n",
 392                        PCI_VENDOR_ID_INTEL, e7xxx_devs[dev_idx].err_dev);
 393                 goto fail;
 394         }
 395
 396         debugf3("MC: " __FILE__ ": %s(): more mci init\n", __func__);
 397         mci->ctl_name = pvt->dev_info->ctl_name;
 398
 399         mci->edac_check = e7xxx_check;
 400         mci->ctl_page_to_phys = ctl_page_to_phys;
 401
 402         /* find out the device types */
 403         pci_read_config_dword(pdev, E7XXX_DRA, &dra);
 404
 405         /*
 406          * The dram row boundary (DRB) reg values are boundary address
 407          * for each DRAM row with a granularity of 32 or 64MB (single/dual
 408          * channel operation).  DRB regs are cumulative; therefore DRB7 will
 409          * contain the total memory contained in all eight rows.
 410          */
 411         for (last_cumul_size = index = 0; index < mci->nr_csrows; index++) {
 412                 u8 value;
 413                 u32 cumul_size;
 414                 /* mem_dev 0=x8, 1=x4 */
 415                 int mem_dev = (dra >> (index * 4 + 3)) & 0x1;
 416                 struct csrow_info *csrow = &mci->csrows[index];
 417
 418                 pci_read_config_byte(mci->pdev, E7XXX_DRB + index, &value);
 419                 /* convert a 64 or 32 MiB DRB to a page size. */
 420                 cumul_size = value << (25 + drc_drbg - PAGE_SHIFT);
 421                 debugf3("MC: " __FILE__ ": %s(): (%d) cumul_size 0x%x\n",
 422                         __func__, index, cumul_size);
 423                 if (cumul_size == last_cumul_size)
 424                         continue;       /* not populated */
 425
 426                 csrow->first_page = last_cumul_size;
 427                 csrow->last_page = cumul_size - 1;
 428                 csrow->nr_pages = cumul_size - last_cumul_size;
 429                 last_cumul_size = cumul_size;
 430                 csrow->grain = 1 << 12; /* 4KiB - resolution of CELOG */
 431                 csrow->mtype = MEM_RDDR;        /* only one type supported */
 432                 csrow->dtype = mem_dev ? DEV_X4 : DEV_X8;
 433
 434                 /*
 435                  * if single channel or x8 devices then SECDED
 436                  * if dual channel and x4 then S4ECD4ED
 437                  */
 438                 if (drc_ddim) {
 439                         if (drc_chan && mem_dev) {
 440                                 csrow->edac_mode = EDAC_S4ECD4ED;
 441                                 mci->edac_cap |= EDAC_FLAG_S4ECD4ED;
 442                         } else {
 443                                 csrow->edac_mode = EDAC_SECDED;
 444                                 mci->edac_cap |= EDAC_FLAG_SECDED;
 445                         }
 446                 } else
 447                         csrow->edac_mode = EDAC_NONE;
 448         }
 449
 450         mci->edac_cap |= EDAC_FLAG_NONE;
 451
 452         debugf3("MC: " __FILE__ ": %s(): tolm, remapbase, remaplimit\n",
 453                 __func__);
 454         /* load the top of low memory, remap base, and remap limit vars */
 455         pci_read_config_word(mci->pdev, E7XXX_TOLM, &pci_data);
 456         pvt->tolm = ((u32) pci_data) << 4;
 457         pci_read_config_word(mci->pdev, E7XXX_REMAPBASE, &pci_data);
 458         pvt->remapbase = ((u32) pci_data) << 14;
 459         pci_read_config_word(mci->pdev, E7XXX_REMAPLIMIT, &pci_data);
 460         pvt->remaplimit = ((u32) pci_data) << 14;
 461         printk("tolm = %x, remapbase = %x, remaplimit = %x\n", pvt->tolm,
 462                pvt->remapbase, pvt->remaplimit);
 463
 464         /* clear any pending errors, or initial state bits */
 465         pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_FERR, 0x03, 0x03);
 466         pci_write_bits8(pvt->bridge_ck, E7XXX_DRAM_NERR, 0x03, 0x03);
 467
 468         if (edac_mc_add_mc(mci) != 0) {
 469                 debugf3("MC: " __FILE__
 470                         ": %s(): failed edac_mc_add_mc()\n",
 471                         __func__);
 472                 goto fail;
 473         }
 474
 475         /* get this far and it's successful */
 476         debugf3("MC: " __FILE__ ": %s(): success\n", __func__);
 477         return 0;
 478
 479 fail:
 480         if (mci != NULL) {
 481                 if(pvt != NULL && pvt->bridge_ck)
 482                         pci_dev_put(pvt->bridge_ck);
 483                 edac_mc_free(mci);
 484         }
 485
 486         return rc;
 487 }
 488
 489 /* returns count (>= 0), or negative on error */
 490 static int __devinit
 491 e7xxx_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 492 {
 493         debugf0("MC: " __FILE__ ": %s()\n", __func__);
 494
 495         /* wake up and enable device */
 496         return pci_enable_device(pdev) ?
 497             -EIO : e7xxx_probe1(pdev, ent->driver_data);
 498 }
 499
 500
 501 static void __devexit e7xxx_remove_one(struct pci_dev *pdev)
 502 {
 503         struct mem_ctl_info *mci;
 504         struct e7xxx_pvt *pvt;
 505
 506         debugf0(__FILE__ ": %s()\n", __func__);
 507
 508         if (((mci = edac_mc_find_mci_by_pdev(pdev)) != 0) &&
 509             edac_mc_del_mc(mci)) {
 510                 pvt = (struct e7xxx_pvt *) mci->pvt_info;
 511                 pci_dev_put(pvt->bridge_ck);
 512                 edac_mc_free(mci);
 513         }
 514 }
 515
 516
 517 static const struct pci_device_id e7xxx_pci_tbl[] __devinitdata = {
 518         {PCI_VEND_DEV(INTEL, 7205_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 519          E7205},
 520         {PCI_VEND_DEV(INTEL, 7500_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 521          E7500},
 522         {PCI_VEND_DEV(INTEL, 7501_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 523          E7501},
 524         {PCI_VEND_DEV(INTEL, 7505_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 525          E7505},
 526         {0,}                    /* 0 terminated list. */
 527 };
 528
 529 MODULE_DEVICE_TABLE(pci, e7xxx_pci_tbl);
 530
 531
 532 static struct pci_driver e7xxx_driver = {
 533         .name = BS_MOD_STR,
 534         .probe = e7xxx_init_one,
 535         .remove = __devexit_p(e7xxx_remove_one),
 536         .id_table = e7xxx_pci_tbl,
 537 };
 538
 539
 540 static int __init e7xxx_init(void)
 541 {
 542         return pci_register_driver(&e7xxx_driver);
 543 }
 544
 545
 546 static void __exit e7xxx_exit(void)
 547 {
 548         pci_unregister_driver(&e7xxx_driver);
 549 }
 550
 551 module_init(e7xxx_init);
 552 module_exit(e7xxx_exit);
 553
 554
 555 MODULE_LICENSE("GPL");
 556 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n"
 557               "Based on.work by Dan Hollis et al");
 558 MODULE_DESCRIPTION("MC support for Intel e7xxx memory controllers");