drivers/edac/e752x_edac.c

   1 /*
   2  * Intel e752x Memory Controller kernel module
   3  * (C) 2004 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 e752x_chips" below for supported chipsets
   8  *
   9  * Written by Tom Zimmerman
  10  *
  11  * Contributors:
  12  *      Thayne Harbaugh at realmsys.com (?)
  13  *      Wang Zhenyu at intel.com
  14  *      Dave Jiang at mvista.com
  15  *
  16  * $Id: edac_e752x.c,v 1.5.2.11 2005/10/05 00:43:44 dsp_llnl Exp $
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/init.h>
  22 #include <linux/pci.h>
  23 #include <linux/pci_ids.h>
  24 #include <linux/slab.h>
  25 #include <linux/edac.h>
  26 #include "edac_core.h"
  27
  28 #define E752X_REVISION  " Ver: 2.0.2 " __DATE__
  29 #define EDAC_MOD_STR    "e752x_edac"
  30
  31 static int force_function_unhide;
  32
  33 static struct edac_pci_ctl_info *e752x_pci;
  34
  35 #define e752x_printk(level, fmt, arg...) \
  36         edac_printk(level, "e752x", fmt, ##arg)
  37
  38 #define e752x_mc_printk(mci, level, fmt, arg...) \
  39         edac_mc_chipset_printk(mci, level, "e752x", fmt, ##arg)
  40
  41 #ifndef PCI_DEVICE_ID_INTEL_7520_0
  42 #define PCI_DEVICE_ID_INTEL_7520_0      0x3590
  43 #endif                          /* PCI_DEVICE_ID_INTEL_7520_0      */
  44
  45 #ifndef PCI_DEVICE_ID_INTEL_7520_1_ERR
  46 #define PCI_DEVICE_ID_INTEL_7520_1_ERR  0x3591
  47 #endif                          /* PCI_DEVICE_ID_INTEL_7520_1_ERR  */
  48
  49 #ifndef PCI_DEVICE_ID_INTEL_7525_0
  50 #define PCI_DEVICE_ID_INTEL_7525_0      0x359E
  51 #endif                          /* PCI_DEVICE_ID_INTEL_7525_0      */
  52
  53 #ifndef PCI_DEVICE_ID_INTEL_7525_1_ERR
  54 #define PCI_DEVICE_ID_INTEL_7525_1_ERR  0x3593
  55 #endif                          /* PCI_DEVICE_ID_INTEL_7525_1_ERR  */
  56
  57 #ifndef PCI_DEVICE_ID_INTEL_7320_0
  58 #define PCI_DEVICE_ID_INTEL_7320_0      0x3592
  59 #endif                          /* PCI_DEVICE_ID_INTEL_7320_0 */
  60
  61 #ifndef PCI_DEVICE_ID_INTEL_7320_1_ERR
  62 #define PCI_DEVICE_ID_INTEL_7320_1_ERR  0x3593
  63 #endif                          /* PCI_DEVICE_ID_INTEL_7320_1_ERR */
  64
  65 #define E752X_NR_CSROWS         8       /* number of csrows */
  66
  67 /* E752X register addresses - device 0 function 0 */
  68 #define E752X_DRB               0x60    /* DRAM row boundary register (8b) */
  69 #define E752X_DRA               0x70    /* DRAM row attribute register (8b) */
  70                                         /*
  71                                          * 31:30   Device width row 7
  72                                          *      01=x8 10=x4 11=x8 DDR2
  73                                          * 27:26   Device width row 6
  74                                          * 23:22   Device width row 5
  75                                          * 19:20   Device width row 4
  76                                          * 15:14   Device width row 3
  77                                          * 11:10   Device width row 2
  78                                          *  7:6    Device width row 1
  79                                          *  3:2    Device width row 0
  80                                          */
  81 #define E752X_DRC               0x7C    /* DRAM controller mode reg (32b) */
  82                                         /* FIXME:IS THIS RIGHT? */
  83                                         /*
  84                                          * 22    Number channels 0=1,1=2
  85                                          * 19:18 DRB Granularity 32/64MB
  86                                          */
  87 #define E752X_DRM               0x80    /* Dimm mapping register */
  88 #define E752X_DDRCSR            0x9A    /* DDR control and status reg (16b) */
  89                                         /*
  90                                          * 14:12 1 single A, 2 single B, 3 dual
  91                                          */
  92 #define E752X_TOLM              0xC4    /* DRAM top of low memory reg (16b) */
  93 #define E752X_REMAPBASE         0xC6    /* DRAM remap base address reg (16b) */
  94 #define E752X_REMAPLIMIT        0xC8    /* DRAM remap limit address reg (16b) */
  95 #define E752X_REMAPOFFSET       0xCA    /* DRAM remap limit offset reg (16b) */
  96
  97 /* E752X register addresses - device 0 function 1 */
  98 #define E752X_FERR_GLOBAL       0x40    /* Global first error register (32b) */
  99 #define E752X_NERR_GLOBAL       0x44    /* Global next error register (32b) */
 100 #define E752X_HI_FERR           0x50    /* Hub interface first error reg (8b) */
 101 #define E752X_HI_NERR           0x52    /* Hub interface next error reg (8b) */
 102 #define E752X_HI_ERRMASK        0x54    /* Hub interface error mask reg (8b) */
 103 #define E752X_HI_SMICMD         0x5A    /* Hub interface SMI command reg (8b) */
 104 #define E752X_SYSBUS_FERR       0x60    /* System buss first error reg (16b) */
 105 #define E752X_SYSBUS_NERR       0x62    /* System buss next error reg (16b) */
 106 #define E752X_SYSBUS_ERRMASK    0x64    /* System buss error mask reg (16b) */
 107 #define E752X_SYSBUS_SMICMD     0x6A    /* System buss SMI command reg (16b) */
 108 #define E752X_BUF_FERR          0x70    /* Memory buffer first error reg (8b) */
 109 #define E752X_BUF_NERR          0x72    /* Memory buffer next error reg (8b) */
 110 #define E752X_BUF_ERRMASK       0x74    /* Memory buffer error mask reg (8b) */
 111 #define E752X_BUF_SMICMD        0x7A    /* Memory buffer SMI command reg (8b) */
 112 #define E752X_DRAM_FERR         0x80    /* DRAM first error register (16b) */
 113 #define E752X_DRAM_NERR         0x82    /* DRAM next error register (16b) */
 114 #define E752X_DRAM_ERRMASK      0x84    /* DRAM error mask register (8b) */
 115 #define E752X_DRAM_SMICMD       0x8A    /* DRAM SMI command register (8b) */
 116 #define E752X_DRAM_RETR_ADD     0xAC    /* DRAM Retry address register (32b) */
 117 #define E752X_DRAM_SEC1_ADD     0xA0    /* DRAM first correctable memory */
 118                                         /*     error address register (32b) */
 119                                         /*
 120                                          * 31    Reserved
 121                                          * 30:2  CE address (64 byte block 34:6)
 122                                          * 1     Reserved
 123                                          * 0     HiLoCS
 124                                          */
 125 #define E752X_DRAM_SEC2_ADD     0xC8    /* DRAM first correctable memory */
 126                                         /*     error address register (32b) */
 127                                         /*
 128                                          * 31    Reserved
 129                                          * 30:2  CE address (64 byte block 34:6)
 130                                          * 1     Reserved
 131                                          * 0     HiLoCS
 132                                          */
 133 #define E752X_DRAM_DED_ADD      0xA4    /* DRAM first uncorrectable memory */
 134                                         /*     error address register (32b) */
 135                                         /*
 136                                          * 31    Reserved
 137                                          * 30:2  CE address (64 byte block 34:6)
 138                                          * 1     Reserved
 139                                          * 0     HiLoCS
 140                                          */
 141 #define E752X_DRAM_SCRB_ADD     0xA8    /* DRAM first uncorrectable scrub memory */
 142                                         /*     error address register (32b) */
 143                                         /*
 144                                          * 31    Reserved
 145                                          * 30:2  CE address (64 byte block 34:6)
 146                                          * 1     Reserved
 147                                          * 0     HiLoCS
 148                                          */
 149 #define E752X_DRAM_SEC1_SYNDROME 0xC4   /* DRAM first correctable memory */
 150                                         /*     error syndrome register (16b) */
 151 #define E752X_DRAM_SEC2_SYNDROME 0xC6   /* DRAM second correctable memory */
 152                                         /*     error syndrome register (16b) */
 153 #define E752X_DEVPRES1          0xF4    /* Device Present 1 register (8b) */
 154
 155 /* ICH5R register addresses - device 30 function 0 */
 156 #define ICH5R_PCI_STAT          0x06    /* PCI status register (16b) */
 157 #define ICH5R_PCI_2ND_STAT      0x1E    /* PCI status secondary reg (16b) */
 158 #define ICH5R_PCI_BRIDGE_CTL    0x3E    /* PCI bridge control register (16b) */
 159
 160 enum e752x_chips {
 161         E7520 = 0,
 162         E7525 = 1,
 163         E7320 = 2
 164 };
 165
 166 struct e752x_pvt {
 167         struct pci_dev *bridge_ck;
 168         struct pci_dev *dev_d0f0;
 169         struct pci_dev *dev_d0f1;
 170         u32 tolm;
 171         u32 remapbase;
 172         u32 remaplimit;
 173         int mc_symmetric;
 174         u8 map[8];
 175         int map_type;
 176         const struct e752x_dev_info *dev_info;
 177 };
 178
 179 struct e752x_dev_info {
 180         u16 err_dev;
 181         u16 ctl_dev;
 182         const char *ctl_name;
 183 };
 184
 185 struct e752x_error_info {
 186         u32 ferr_global;
 187         u32 nerr_global;
 188         u8 hi_ferr;
 189         u8 hi_nerr;
 190         u16 sysbus_ferr;
 191         u16 sysbus_nerr;
 192         u8 buf_ferr;
 193         u8 buf_nerr;
 194         u16 dram_ferr;
 195         u16 dram_nerr;
 196         u32 dram_sec1_add;
 197         u32 dram_sec2_add;
 198         u16 dram_sec1_syndrome;
 199         u16 dram_sec2_syndrome;
 200         u32 dram_ded_add;
 201         u32 dram_scrb_add;
 202         u32 dram_retr_add;
 203 };
 204
 205 static const struct e752x_dev_info e752x_devs[] = {
 206         [E7520] = {
 207                 .err_dev = PCI_DEVICE_ID_INTEL_7520_1_ERR,
 208                 .ctl_dev = PCI_DEVICE_ID_INTEL_7520_0,
 209                 .ctl_name = "E7520"},
 210         [E7525] = {
 211                 .err_dev = PCI_DEVICE_ID_INTEL_7525_1_ERR,
 212                 .ctl_dev = PCI_DEVICE_ID_INTEL_7525_0,
 213                 .ctl_name = "E7525"},
 214         [E7320] = {
 215                 .err_dev = PCI_DEVICE_ID_INTEL_7320_1_ERR,
 216                 .ctl_dev = PCI_DEVICE_ID_INTEL_7320_0,
 217                 .ctl_name = "E7320"},
 218 };
 219
 220 static unsigned long ctl_page_to_phys(struct mem_ctl_info *mci,
 221                                 unsigned long page)
 222 {
 223         u32 remap;
 224         struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info;
 225
 226         debugf3("%s()\n", __func__);
 227
 228         if (page < pvt->tolm)
 229                 return page;
 230
 231         if ((page >= 0x100000) && (page < pvt->remapbase))
 232                 return page;
 233
 234         remap = (page - pvt->tolm) + pvt->remapbase;
 235
 236         if (remap < pvt->remaplimit)
 237                 return remap;
 238
 239         e752x_printk(KERN_ERR, "Invalid page %lx - out of range\n", page);
 240         return pvt->tolm - 1;
 241 }
 242
 243 static void do_process_ce(struct mem_ctl_info *mci, u16 error_one,
 244                         u32 sec1_add, u16 sec1_syndrome)
 245 {
 246         u32 page;
 247         int row;
 248         int channel;
 249         int i;
 250         struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info;
 251
 252         debugf3("%s()\n", __func__);
 253
 254         /* convert the addr to 4k page */
 255         page = sec1_add >> (PAGE_SHIFT - 4);
 256
 257         /* FIXME - check for -1 */
 258         if (pvt->mc_symmetric) {
 259                 /* chip select are bits 14 & 13 */
 260                 row = ((page >> 1) & 3);
 261                 e752x_printk(KERN_WARNING,
 262                         "Test row %d Table %d %d %d %d %d %d %d %d\n", row,
 263                         pvt->map[0], pvt->map[1], pvt->map[2], pvt->map[3],
 264                         pvt->map[4], pvt->map[5], pvt->map[6],
 265                         pvt->map[7]);
 266
 267                 /* test for channel remapping */
 268                 for (i = 0; i < 8; i++) {
 269                         if (pvt->map[i] == row)
 270                                 break;
 271                 }
 272
 273                 e752x_printk(KERN_WARNING, "Test computed row %d\n", i);
 274
 275                 if (i < 8)
 276                         row = i;
 277                 else
 278                         e752x_mc_printk(mci, KERN_WARNING,
 279                                         "row %d not found in remap table\n",
 280                                         row);
 281         } else
 282                 row = edac_mc_find_csrow_by_page(mci, page);
 283
 284         /* 0 = channel A, 1 = channel B */
 285         channel = !(error_one & 1);
 286
 287         /* e752x mc reads 34:6 of the DRAM linear address */
 288         edac_mc_handle_ce(mci, page, offset_in_page(sec1_add << 4),
 289                         sec1_syndrome, row, channel, "e752x CE");
 290 }
 291
 292 static inline void process_ce(struct mem_ctl_info *mci, u16 error_one,
 293                         u32 sec1_add, u16 sec1_syndrome, int *error_found,
 294                         int handle_error)
 295 {
 296         *error_found = 1;
 297
 298         if (handle_error)
 299                 do_process_ce(mci, error_one, sec1_add, sec1_syndrome);
 300 }
 301
 302 static void do_process_ue(struct mem_ctl_info *mci, u16 error_one,
 303                         u32 ded_add, u32 scrb_add)
 304 {
 305         u32 error_2b, block_page;
 306         int row;
 307         struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info;
 308
 309         debugf3("%s()\n", __func__);
 310
 311         if (error_one & 0x0202) {
 312                 error_2b = ded_add;
 313
 314                 /* convert to 4k address */
 315                 block_page = error_2b >> (PAGE_SHIFT - 4);
 316
 317                 row = pvt->mc_symmetric ?
 318                 /* chip select are bits 14 & 13 */
 319                         ((block_page >> 1) & 3) :
 320                         edac_mc_find_csrow_by_page(mci, block_page);
 321
 322                 /* e752x mc reads 34:6 of the DRAM linear address */
 323                 edac_mc_handle_ue(mci, block_page,
 324                                 offset_in_page(error_2b << 4),
 325                                 row, "e752x UE from Read");
 326         }
 327         if (error_one & 0x0404) {
 328                 error_2b = scrb_add;
 329
 330                 /* convert to 4k address */
 331                 block_page = error_2b >> (PAGE_SHIFT - 4);
 332
 333                 row = pvt->mc_symmetric ?
 334                 /* chip select are bits 14 & 13 */
 335                         ((block_page >> 1) & 3) :
 336                         edac_mc_find_csrow_by_page(mci, block_page);
 337
 338                 /* e752x mc reads 34:6 of the DRAM linear address */
 339                 edac_mc_handle_ue(mci, block_page,
 340                                 offset_in_page(error_2b << 4),
 341                                 row, "e752x UE from Scruber");
 342         }
 343 }
 344
 345 static inline void process_ue(struct mem_ctl_info *mci, u16 error_one,
 346                         u32 ded_add, u32 scrb_add, int *error_found,
 347                         int handle_error)
 348 {
 349         *error_found = 1;
 350
 351         if (handle_error)
 352                 do_process_ue(mci, error_one, ded_add, scrb_add);
 353 }
 354
 355 static inline void process_ue_no_info_wr(struct mem_ctl_info *mci,
 356                                          int *error_found, int handle_error)
 357 {
 358         *error_found = 1;
 359
 360         if (!handle_error)
 361                 return;
 362
 363         debugf3("%s()\n", __func__);
 364         edac_mc_handle_ue_no_info(mci, "e752x UE log memory write");
 365 }
 366
 367 static void do_process_ded_retry(struct mem_ctl_info *mci, u16 error,
 368                                  u32 retry_add)
 369 {
 370         u32 error_1b, page;
 371         int row;
 372         struct e752x_pvt *pvt = (struct e752x_pvt *)mci->pvt_info;
 373
 374         error_1b = retry_add;
 375         page = error_1b >> (PAGE_SHIFT - 4);    /* convert the addr to 4k page */
 376         row = pvt->mc_symmetric ? ((page >> 1) & 3) :   /* chip select are bits 14 & 13 */
 377                 edac_mc_find_csrow_by_page(mci, page);
 378         e752x_mc_printk(mci, KERN_WARNING,
 379                         "CE page 0x%lx, row %d : Memory read retry\n",
 380                         (long unsigned int)page, row);
 381 }
 382
 383 static inline void process_ded_retry(struct mem_ctl_info *mci, u16 error,
 384                                 u32 retry_add, int *error_found,
 385                                 int handle_error)
 386 {
 387         *error_found = 1;
 388
 389         if (handle_error)
 390                 do_process_ded_retry(mci, error, retry_add);
 391 }
 392
 393 static inline void process_threshold_ce(struct mem_ctl_info *mci, u16 error,
 394                                         int *error_found, int handle_error)
 395 {
 396         *error_found = 1;
 397
 398         if (handle_error)
 399                 e752x_mc_printk(mci, KERN_WARNING, "Memory threshold CE\n");
 400 }
 401
 402 static char *global_message[11] = {
 403         "PCI Express C1", "PCI Express C", "PCI Express B1",
 404         "PCI Express B", "PCI Express A1", "PCI Express A",
 405         "DMA Controler", "HUB Interface", "System Bus",
 406         "DRAM Controler", "Internal Buffer"
 407 };
 408
 409 static char *fatal_message[2] = { "Non-Fatal ", "Fatal " };
 410
 411 static void do_global_error(int fatal, u32 errors)
 412 {
 413         int i;
 414
 415         for (i = 0; i < 11; i++) {
 416                 if (errors & (1 << i))
 417                         e752x_printk(KERN_WARNING, "%sError %s\n",
 418                                 fatal_message[fatal], global_message[i]);
 419         }
 420 }
 421
 422 static inline void global_error(int fatal, u32 errors, int *error_found,
 423                                 int handle_error)
 424 {
 425         *error_found = 1;
 426
 427         if (handle_error)
 428                 do_global_error(fatal, errors);
 429 }
 430
 431 static char *hub_message[7] = {
 432         "HI Address or Command Parity", "HI Illegal Access",
 433         "HI Internal Parity", "Out of Range Access",
 434         "HI Data Parity", "Enhanced Config Access",
 435         "Hub Interface Target Abort"
 436 };
 437
 438 static void do_hub_error(int fatal, u8 errors)
 439 {
 440         int i;
 441
 442         for (i = 0; i < 7; i++) {
 443                 if (errors & (1 << i))
 444                         e752x_printk(KERN_WARNING, "%sError %s\n",
 445                                 fatal_message[fatal], hub_message[i]);
 446         }
 447 }
 448
 449 static inline void hub_error(int fatal, u8 errors, int *error_found,
 450                         int handle_error)
 451 {
 452         *error_found = 1;
 453
 454         if (handle_error)
 455                 do_hub_error(fatal, errors);
 456 }
 457
 458 static char *membuf_message[4] = {
 459         "Internal PMWB to DRAM parity",
 460         "Internal PMWB to System Bus Parity",
 461         "Internal System Bus or IO to PMWB Parity",
 462         "Internal DRAM to PMWB Parity"
 463 };
 464
 465 static void do_membuf_error(u8 errors)
 466 {
 467         int i;
 468
 469         for (i = 0; i < 4; i++) {
 470                 if (errors & (1 << i))
 471                         e752x_printk(KERN_WARNING, "Non-Fatal Error %s\n",
 472                                 membuf_message[i]);
 473         }
 474 }
 475
 476 static inline void membuf_error(u8 errors, int *error_found, int handle_error)
 477 {
 478         *error_found = 1;
 479
 480         if (handle_error)
 481                 do_membuf_error(errors);
 482 }
 483
 484 static char *sysbus_message[10] = {
 485         "Addr or Request Parity",
 486         "Data Strobe Glitch",
 487         "Addr Strobe Glitch",
 488         "Data Parity",
 489         "Addr Above TOM",
 490         "Non DRAM Lock Error",
 491         "MCERR", "BINIT",
 492         "Memory Parity",
 493         "IO Subsystem Parity"
 494 };
 495
 496 static void do_sysbus_error(int fatal, u32 errors)
 497 {
 498         int i;
 499
 500         for (i = 0; i < 10; i++) {
 501                 if (errors & (1 << i))
 502                         e752x_printk(KERN_WARNING, "%sError System Bus %s\n",
 503                                 fatal_message[fatal], sysbus_message[i]);
 504         }
 505 }
 506
 507 static inline void sysbus_error(int fatal, u32 errors, int *error_found,
 508                                 int handle_error)
 509 {
 510         *error_found = 1;
 511
 512         if (handle_error)
 513                 do_sysbus_error(fatal, errors);
 514 }
 515
 516 static void e752x_check_hub_interface(struct e752x_error_info *info,
 517                                 int *error_found, int handle_error)
 518 {
 519         u8 stat8;
 520
 521         //pci_read_config_byte(dev,E752X_HI_FERR,&stat8);
 522
 523         stat8 = info->hi_ferr;
 524
 525         if (stat8 & 0x7f) {     /* Error, so process */
 526                 stat8 &= 0x7f;
 527
 528                 if (stat8 & 0x2b)
 529                         hub_error(1, stat8 & 0x2b, error_found, handle_error);
 530
 531                 if (stat8 & 0x54)
 532                         hub_error(0, stat8 & 0x54, error_found, handle_error);
 533         }
 534         //pci_read_config_byte(dev,E752X_HI_NERR,&stat8);
 535
 536         stat8 = info->hi_nerr;
 537
 538         if (stat8 & 0x7f) {     /* Error, so process */
 539                 stat8 &= 0x7f;
 540
 541                 if (stat8 & 0x2b)
 542                         hub_error(1, stat8 & 0x2b, error_found, handle_error);
 543
 544                 if (stat8 & 0x54)
 545                         hub_error(0, stat8 & 0x54, error_found, handle_error);
 546         }
 547 }
 548
 549 static void e752x_check_sysbus(struct e752x_error_info *info,
 550                         int *error_found, int handle_error)
 551 {
 552         u32 stat32, error32;
 553
 554         //pci_read_config_dword(dev,E752X_SYSBUS_FERR,&stat32);
 555         stat32 = info->sysbus_ferr + (info->sysbus_nerr << 16);
 556
 557         if (stat32 == 0)
 558                 return;         /* no errors */
 559
 560         error32 = (stat32 >> 16) & 0x3ff;
 561         stat32 = stat32 & 0x3ff;
 562
 563         if (stat32 & 0x087)
 564                 sysbus_error(1, stat32 & 0x087, error_found, handle_error);
 565
 566         if (stat32 & 0x378)
 567                 sysbus_error(0, stat32 & 0x378, error_found, handle_error);
 568
 569         if (error32 & 0x087)
 570                 sysbus_error(1, error32 & 0x087, error_found, handle_error);
 571
 572         if (error32 & 0x378)
 573                 sysbus_error(0, error32 & 0x378, error_found, handle_error);
 574 }
 575
 576 static void e752x_check_membuf(struct e752x_error_info *info,
 577                         int *error_found, int handle_error)
 578 {
 579         u8 stat8;
 580
 581         stat8 = info->buf_ferr;
 582
 583         if (stat8 & 0x0f) {     /* Error, so process */
 584                 stat8 &= 0x0f;
 585                 membuf_error(stat8, error_found, handle_error);
 586         }
 587
 588         stat8 = info->buf_nerr;
 589
 590         if (stat8 & 0x0f) {     /* Error, so process */
 591                 stat8 &= 0x0f;
 592                 membuf_error(stat8, error_found, handle_error);
 593         }
 594 }
 595
 596 static void e752x_check_dram(struct mem_ctl_info *mci,
 597                         struct e752x_error_info *info, int *error_found,
 598                         int handle_error)
 599 {
 600         u16 error_one, error_next;
 601
 602         error_one = info->dram_ferr;
 603         error_next = info->dram_nerr;
 604
 605         /* decode and report errors */
 606         if (error_one & 0x0101) /* check first error correctable */
 607                 process_ce(mci, error_one, info->dram_sec1_add,
 608                         info->dram_sec1_syndrome, error_found, handle_error);
 609
 610         if (error_next & 0x0101)        /* check next error correctable */
 611                 process_ce(mci, error_next, info->dram_sec2_add,
 612                         info->dram_sec2_syndrome, error_found, handle_error);
 613
 614         if (error_one & 0x4040)
 615                 process_ue_no_info_wr(mci, error_found, handle_error);
 616
 617         if (error_next & 0x4040)
 618                 process_ue_no_info_wr(mci, error_found, handle_error);
 619
 620         if (error_one & 0x2020)
 621                 process_ded_retry(mci, error_one, info->dram_retr_add,
 622                                 error_found, handle_error);
 623
 624         if (error_next & 0x2020)
 625                 process_ded_retry(mci, error_next, info->dram_retr_add,
 626                                 error_found, handle_error);
 627
 628         if (error_one & 0x0808)
 629                 process_threshold_ce(mci, error_one, error_found, handle_error);
 630
 631         if (error_next & 0x0808)
 632                 process_threshold_ce(mci, error_next, error_found,
 633                                 handle_error);
 634
 635         if (error_one & 0x0606)
 636                 process_ue(mci, error_one, info->dram_ded_add,
 637                         info->dram_scrb_add, error_found, handle_error);
 638
 639         if (error_next & 0x0606)
 640                 process_ue(mci, error_next, info->dram_ded_add,
 641                         info->dram_scrb_add, error_found, handle_error);
 642 }
 643
 644 static void e752x_get_error_info(struct mem_ctl_info *mci,
 645                                  struct e752x_error_info *info)
 646 {
 647         struct pci_dev *dev;
 648         struct e752x_pvt *pvt;
 649
 650         memset(info, 0, sizeof(*info));
 651         pvt = (struct e752x_pvt *)mci->pvt_info;
 652         dev = pvt->dev_d0f1;
 653         pci_read_config_dword(dev, E752X_FERR_GLOBAL, &info->ferr_global);
 654
 655         if (info->ferr_global) {
 656                 pci_read_config_byte(dev, E752X_HI_FERR, &info->hi_ferr);
 657                 pci_read_config_word(dev, E752X_SYSBUS_FERR,
 658                                 &info->sysbus_ferr);
 659                 pci_read_config_byte(dev, E752X_BUF_FERR, &info->buf_ferr);
 660                 pci_read_config_word(dev, E752X_DRAM_FERR, &info->dram_ferr);
 661                 pci_read_config_dword(dev, E752X_DRAM_SEC1_ADD,
 662                                 &info->dram_sec1_add);
 663                 pci_read_config_word(dev, E752X_DRAM_SEC1_SYNDROME,
 664                                 &info->dram_sec1_syndrome);
 665                 pci_read_config_dword(dev, E752X_DRAM_DED_ADD,
 666                                 &info->dram_ded_add);
 667                 pci_read_config_dword(dev, E752X_DRAM_SCRB_ADD,
 668                                 &info->dram_scrb_add);
 669                 pci_read_config_dword(dev, E752X_DRAM_RETR_ADD,
 670                                 &info->dram_retr_add);
 671
 672                 if (info->hi_ferr & 0x7f)
 673                         pci_write_config_byte(dev, E752X_HI_FERR,
 674                                         info->hi_ferr);
 675
 676                 if (info->sysbus_ferr)
 677                         pci_write_config_word(dev, E752X_SYSBUS_FERR,
 678                                         info->sysbus_ferr);
 679
 680                 if (info->buf_ferr & 0x0f)
 681                         pci_write_config_byte(dev, E752X_BUF_FERR,
 682                                         info->buf_ferr);
 683
 684                 if (info->dram_ferr)
 685                         pci_write_bits16(pvt->bridge_ck, E752X_DRAM_FERR,
 686                                          info->dram_ferr, info->dram_ferr);
 687
 688                 pci_write_config_dword(dev, E752X_FERR_GLOBAL,
 689                                 info->ferr_global);
 690         }
 691
 692         pci_read_config_dword(dev, E752X_NERR_GLOBAL, &info->nerr_global);
 693
 694         if (info->nerr_global) {
 695                 pci_read_config_byte(dev, E752X_HI_NERR, &info->hi_nerr);
 696                 pci_read_config_word(dev, E752X_SYSBUS_NERR,
 697                                 &info->sysbus_nerr);
 698                 pci_read_config_byte(dev, E752X_BUF_NERR, &info->buf_nerr);
 699                 pci_read_config_word(dev, E752X_DRAM_NERR, &info->dram_nerr);
 700                 pci_read_config_dword(dev, E752X_DRAM_SEC2_ADD,
 701                                 &info->dram_sec2_add);
 702                 pci_read_config_word(dev, E752X_DRAM_SEC2_SYNDROME,
 703                                 &info->dram_sec2_syndrome);
 704
 705                 if (info->hi_nerr & 0x7f)
 706                         pci_write_config_byte(dev, E752X_HI_NERR,
 707                                         info->hi_nerr);
 708
 709                 if (info->sysbus_nerr)
 710                         pci_write_config_word(dev, E752X_SYSBUS_NERR,
 711                                         info->sysbus_nerr);
 712
 713                 if (info->buf_nerr & 0x0f)
 714                         pci_write_config_byte(dev, E752X_BUF_NERR,
 715                                         info->buf_nerr);
 716
 717                 if (info->dram_nerr)
 718                         pci_write_bits16(pvt->bridge_ck, E752X_DRAM_NERR,
 719                                          info->dram_nerr, info->dram_nerr);
 720
 721                 pci_write_config_dword(dev, E752X_NERR_GLOBAL,
 722                                 info->nerr_global);
 723         }
 724 }
 725
 726 static int e752x_process_error_info(struct mem_ctl_info *mci,
 727                                 struct e752x_error_info *info,
 728                                 int handle_errors)
 729 {
 730         u32 error32, stat32;
 731         int error_found;
 732
 733         error_found = 0;
 734         error32 = (info->ferr_global >> 18) & 0x3ff;
 735         stat32 = (info->ferr_global >> 4) & 0x7ff;
 736
 737         if (error32)
 738                 global_error(1, error32, &error_found, handle_errors);
 739
 740         if (stat32)
 741                 global_error(0, stat32, &error_found, handle_errors);
 742
 743         error32 = (info->nerr_global >> 18) & 0x3ff;
 744         stat32 = (info->nerr_global >> 4) & 0x7ff;
 745
 746         if (error32)
 747                 global_error(1, error32, &error_found, handle_errors);
 748
 749         if (stat32)
 750                 global_error(0, stat32, &error_found, handle_errors);
 751
 752         e752x_check_hub_interface(info, &error_found, handle_errors);
 753         e752x_check_sysbus(info, &error_found, handle_errors);
 754         e752x_check_membuf(info, &error_found, handle_errors);
 755         e752x_check_dram(mci, info, &error_found, handle_errors);
 756         return error_found;
 757 }
 758
 759 static void e752x_check(struct mem_ctl_info *mci)
 760 {
 761         struct e752x_error_info info;
 762
 763         debugf3("%s()\n", __func__);
 764         e752x_get_error_info(mci, &info);
 765         e752x_process_error_info(mci, &info, 1);
 766 }
 767
 768 /* Return 1 if dual channel mode is active.  Else return 0. */
 769 static inline int dual_channel_active(u16 ddrcsr)
 770 {
 771         return (((ddrcsr >> 12) & 3) == 3);
 772 }
 773
 774 /* Remap csrow index numbers if map_type is "reverse"
 775  */
 776 static inline int remap_csrow_index(struct mem_ctl_info *mci, int index)
 777 {
 778         struct e752x_pvt *pvt = mci->pvt_info;
 779
 780         if (!pvt->map_type)
 781                 return (7 - index);
 782
 783         return (index);
 784 }
 785
 786 static void e752x_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
 787                         u16 ddrcsr)
 788 {
 789         struct csrow_info *csrow;
 790         unsigned long last_cumul_size;
 791         int index, mem_dev, drc_chan;
 792         int drc_drbg;           /* DRB granularity 0=64mb, 1=128mb */
 793         int drc_ddim;           /* DRAM Data Integrity Mode 0=none, 2=edac */
 794         u8 value;
 795         u32 dra, drc, cumul_size;
 796
 797         dra = 0;
 798         for (index = 0; index < 4; index++) {
 799                 u8 dra_reg;
 800                 pci_read_config_byte(pdev, E752X_DRA + index, &dra_reg);
 801                 dra |= dra_reg << (index * 8);
 802         }
 803         pci_read_config_dword(pdev, E752X_DRC, &drc);
 804         drc_chan = dual_channel_active(ddrcsr);
 805         drc_drbg = drc_chan + 1;        /* 128 in dual mode, 64 in single */
 806         drc_ddim = (drc >> 20) & 0x3;
 807
 808         /* The dram row boundary (DRB) reg values are boundary address for
 809          * each DRAM row with a granularity of 64 or 128MB (single/dual
 810          * channel operation).  DRB regs are cumulative; therefore DRB7 will
 811          * contain the total memory contained in all eight rows.
 812          */
 813         for (last_cumul_size = index = 0; index < mci->nr_csrows; index++) {
 814                 /* mem_dev 0=x8, 1=x4 */
 815                 mem_dev = (dra >> (index * 4 + 2)) & 0x3;
 816                 csrow = &mci->csrows[remap_csrow_index(mci, index)];
 817
 818                 mem_dev = (mem_dev == 2);
 819                 pci_read_config_byte(pdev, E752X_DRB + index, &value);
 820                 /* convert a 128 or 64 MiB DRB to a page size. */
 821                 cumul_size = value << (25 + drc_drbg - PAGE_SHIFT);
 822                 debugf3("%s(): (%d) cumul_size 0x%x\n", __func__, index,
 823                         cumul_size);
 824                 if (cumul_size == last_cumul_size)
 825                         continue;       /* not populated */
 826
 827                 csrow->first_page = last_cumul_size;
 828                 csrow->last_page = cumul_size - 1;
 829                 csrow->nr_pages = cumul_size - last_cumul_size;
 830                 last_cumul_size = cumul_size;
 831                 csrow->grain = 1 << 12; /* 4KiB - resolution of CELOG */
 832                 csrow->mtype = MEM_RDDR;        /* only one type supported */
 833                 csrow->dtype = mem_dev ? DEV_X4 : DEV_X8;
 834
 835                 /*
 836                  * if single channel or x8 devices then SECDED
 837                  * if dual channel and x4 then S4ECD4ED
 838                  */
 839                 if (drc_ddim) {
 840                         if (drc_chan && mem_dev) {
 841                                 csrow->edac_mode = EDAC_S4ECD4ED;
 842                                 mci->edac_cap |= EDAC_FLAG_S4ECD4ED;
 843                         } else {
 844                                 csrow->edac_mode = EDAC_SECDED;
 845                                 mci->edac_cap |= EDAC_FLAG_SECDED;
 846                         }
 847                 } else
 848                         csrow->edac_mode = EDAC_NONE;
 849         }
 850 }
 851
 852 static void e752x_init_mem_map_table(struct pci_dev *pdev,
 853                                 struct e752x_pvt *pvt)
 854 {
 855         int index;
 856         u8 value, last, row;
 857
 858         last = 0;
 859         row = 0;
 860
 861         for (index = 0; index < 8; index += 2) {
 862                 pci_read_config_byte(pdev, E752X_DRB + index, &value);
 863                 /* test if there is a dimm in this slot */
 864                 if (value == last) {
 865                         /* no dimm in the slot, so flag it as empty */
 866                         pvt->map[index] = 0xff;
 867                         pvt->map[index + 1] = 0xff;
 868                 } else {        /* there is a dimm in the slot */
 869                         pvt->map[index] = row;
 870                         row++;
 871                         last = value;
 872                         /* test the next value to see if the dimm is double
 873                          * sided
 874                          */
 875                         pci_read_config_byte(pdev, E752X_DRB + index + 1,
 876                                         &value);
 877
 878                         /* the dimm is single sided, so flag as empty */
 879                         /* this is a double sided dimm to save the next row #*/
 880                         pvt->map[index + 1] = (value == last) ? 0xff :  row;
 881                         row++;
 882                         last = value;
 883                 }
 884         }
 885 }
 886
 887 /* Return 0 on success or 1 on failure. */
 888 static int e752x_get_devs(struct pci_dev *pdev, int dev_idx,
 889                         struct e752x_pvt *pvt)
 890 {
 891         struct pci_dev *dev;
 892
 893         pvt->bridge_ck = pci_get_device(PCI_VENDOR_ID_INTEL,
 894                                         pvt->dev_info->err_dev, pvt->bridge_ck);
 895
 896         if (pvt->bridge_ck == NULL)
 897                 pvt->bridge_ck = pci_scan_single_device(pdev->bus,
 898                                                         PCI_DEVFN(0, 1));
 899
 900         if (pvt->bridge_ck == NULL) {
 901                 e752x_printk(KERN_ERR, "error reporting device not found:"
 902                         "vendor %x device 0x%x (broken BIOS?)\n",
 903                         PCI_VENDOR_ID_INTEL, e752x_devs[dev_idx].err_dev);
 904                 return 1;
 905         }
 906
 907         dev = pci_get_device(PCI_VENDOR_ID_INTEL, e752x_devs[dev_idx].ctl_dev,
 908                         NULL);
 909
 910         if (dev == NULL)
 911                 goto fail;
 912
 913         pvt->dev_d0f0 = dev;
 914         pvt->dev_d0f1 = pci_dev_get(pvt->bridge_ck);
 915
 916         return 0;
 917
 918 fail:
 919         pci_dev_put(pvt->bridge_ck);
 920         return 1;
 921 }
 922
 923 static void e752x_init_error_reporting_regs(struct e752x_pvt *pvt)
 924 {
 925         struct pci_dev *dev;
 926
 927         dev = pvt->dev_d0f1;
 928         /* Turn off error disable & SMI in case the BIOS turned it on */
 929         pci_write_config_byte(dev, E752X_HI_ERRMASK, 0x00);
 930         pci_write_config_byte(dev, E752X_HI_SMICMD, 0x00);
 931         pci_write_config_word(dev, E752X_SYSBUS_ERRMASK, 0x00);
 932         pci_write_config_word(dev, E752X_SYSBUS_SMICMD, 0x00);
 933         pci_write_config_byte(dev, E752X_BUF_ERRMASK, 0x00);
 934         pci_write_config_byte(dev, E752X_BUF_SMICMD, 0x00);
 935         pci_write_config_byte(dev, E752X_DRAM_ERRMASK, 0x00);
 936         pci_write_config_byte(dev, E752X_DRAM_SMICMD, 0x00);
 937 }
 938
 939 static int e752x_probe1(struct pci_dev *pdev, int dev_idx)
 940 {
 941         u16 pci_data;
 942         u8 stat8;
 943         struct mem_ctl_info *mci;
 944         struct e752x_pvt *pvt;
 945         u16 ddrcsr;
 946         int drc_chan;           /* Number of channels 0=1chan,1=2chan */
 947         struct e752x_error_info discard;
 948
 949         debugf0("%s(): mci\n", __func__);
 950         debugf0("Starting Probe1\n");
 951
 952         /* make sure error reporting method is sane */
 953         switch (edac_op_state) {
 954         case EDAC_OPSTATE_POLL:
 955         case EDAC_OPSTATE_NMI:
 956                 break;
 957         default:
 958                 edac_op_state = EDAC_OPSTATE_POLL;
 959                 break;
 960         }
 961
 962         /* check to see if device 0 function 1 is enabled; if it isn't, we
 963          * assume the BIOS has reserved it for a reason and is expecting
 964          * exclusive access, we take care not to violate that assumption and
 965          * fail the probe. */
 966         pci_read_config_byte(pdev, E752X_DEVPRES1, &stat8);
 967         if (!force_function_unhide && !(stat8 & (1 << 5))) {
 968                 printk(KERN_INFO "Contact your BIOS vendor to see if the "
 969                         "E752x error registers can be safely un-hidden\n");
 970                 return -ENODEV;
 971         }
 972         stat8 |= (1 << 5);
 973         pci_write_config_byte(pdev, E752X_DEVPRES1, stat8);
 974
 975         pci_read_config_word(pdev, E752X_DDRCSR, &ddrcsr);
 976         /* FIXME: should check >>12 or 0xf, true for all? */
 977         /* Dual channel = 1, Single channel = 0 */
 978         drc_chan = dual_channel_active(ddrcsr);
 979
 980         mci = edac_mc_alloc(sizeof(*pvt), E752X_NR_CSROWS, drc_chan + 1, 0);
 981
 982         if (mci == NULL) {
 983                 return -ENOMEM;
 984         }
 985
 986         debugf3("%s(): init mci\n", __func__);
 987         mci->mtype_cap = MEM_FLAG_RDDR;
 988         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED |
 989                 EDAC_FLAG_S4ECD4ED;
 990         /* FIXME - what if different memory types are in different csrows? */
 991         mci->mod_name = EDAC_MOD_STR;
 992         mci->mod_ver = E752X_REVISION;
 993         mci->dev = &pdev->dev;
 994
 995         debugf3("%s(): init pvt\n", __func__);
 996         pvt = (struct e752x_pvt *)mci->pvt_info;
 997         pvt->dev_info = &e752x_devs[dev_idx];
 998         pvt->mc_symmetric = ((ddrcsr & 0x10) != 0);
 999
1000         if (e752x_get_devs(pdev, dev_idx, pvt)) {
1001                 edac_mc_free(mci);
1002                 return -ENODEV;
1003         }
1004
1005         debugf3("%s(): more mci init\n", __func__);
1006         mci->ctl_name = pvt->dev_info->ctl_name;
1007         mci->dev_name = pci_name(pdev);
1008         mci->edac_check = e752x_check;
1009         mci->ctl_page_to_phys = ctl_page_to_phys;
1010
1011         /* set the map type.  1 = normal, 0 = reversed
1012          * Must be set before e752x_init_csrows in case csrow mapping
1013          * is reversed.
1014          */
1015         pci_read_config_byte(pdev, E752X_DRM, &stat8);
1016         pvt->map_type = ((stat8 & 0x0f) > ((stat8 >> 4) & 0x0f));
1017
1018         e752x_init_csrows(mci, pdev, ddrcsr);
1019         e752x_init_mem_map_table(pdev, pvt);
1020
1021         mci->edac_cap |= EDAC_FLAG_NONE;
1022         debugf3("%s(): tolm, remapbase, remaplimit\n", __func__);
1023
1024         /* load the top of low memory, remap base, and remap limit vars */
1025         pci_read_config_word(pdev, E752X_TOLM, &pci_data);
1026         pvt->tolm = ((u32) pci_data) << 4;
1027         pci_read_config_word(pdev, E752X_REMAPBASE, &pci_data);
1028         pvt->remapbase = ((u32) pci_data) << 14;
1029         pci_read_config_word(pdev, E752X_REMAPLIMIT, &pci_data);
1030         pvt->remaplimit = ((u32) pci_data) << 14;
1031         e752x_printk(KERN_INFO,
1032                         "tolm = %x, remapbase = %x, remaplimit = %x\n",
1033                         pvt->tolm, pvt->remapbase, pvt->remaplimit);
1034
1035         /* Here we assume that we will never see multiple instances of this
1036          * type of memory controller.  The ID is therefore hardcoded to 0.
1037          */
1038         if (edac_mc_add_mc(mci)) {
1039                 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
1040                 goto fail;
1041         }
1042
1043         e752x_init_error_reporting_regs(pvt);
1044         e752x_get_error_info(mci, &discard);    /* clear other MCH errors */
1045
1046         /* allocating generic PCI control info */
1047         e752x_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
1048         if (!e752x_pci) {
1049                 printk(KERN_WARNING
1050                         "%s(): Unable to create PCI control\n", __func__);
1051                 printk(KERN_WARNING
1052                         "%s(): PCI error report via EDAC not setup\n",
1053                         __func__);
1054         }
1055
1056         /* get this far and it's successful */
1057         debugf3("%s(): success\n", __func__);
1058         return 0;
1059
1060 fail:
1061         pci_dev_put(pvt->dev_d0f0);
1062         pci_dev_put(pvt->dev_d0f1);
1063         pci_dev_put(pvt->bridge_ck);
1064         edac_mc_free(mci);
1065
1066         return -ENODEV;
1067 }
1068
1069 /* returns count (>= 0), or negative on error */
1070 static int __devinit e752x_init_one(struct pci_dev *pdev,
1071                                 const struct pci_device_id *ent)
1072 {
1073         debugf0("%s()\n", __func__);
1074
1075         /* wake up and enable device */
1076         if (pci_enable_device(pdev) < 0)
1077                 return -EIO;
1078
1079         return e752x_probe1(pdev, ent->driver_data);
1080 }
1081
1082 static void __devexit e752x_remove_one(struct pci_dev *pdev)
1083 {
1084         struct mem_ctl_info *mci;
1085         struct e752x_pvt *pvt;
1086
1087         debugf0("%s()\n", __func__);
1088
1089         if (e752x_pci)
1090                 edac_pci_release_generic_ctl(e752x_pci);
1091
1092         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
1093                 return;
1094
1095         pvt = (struct e752x_pvt *)mci->pvt_info;
1096         pci_dev_put(pvt->dev_d0f0);
1097         pci_dev_put(pvt->dev_d0f1);
1098         pci_dev_put(pvt->bridge_ck);
1099         edac_mc_free(mci);
1100 }
1101
1102 static const struct pci_device_id e752x_pci_tbl[] __devinitdata = {
1103         {
1104          PCI_VEND_DEV(INTEL, 7520_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
1105          E7520},
1106         {
1107          PCI_VEND_DEV(INTEL, 7525_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
1108          E7525},
1109         {
1110          PCI_VEND_DEV(INTEL, 7320_0), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
1111          E7320},
1112         {
1113          0,
1114          }                      /* 0 terminated list. */
1115 };
1116
1117 MODULE_DEVICE_TABLE(pci, e752x_pci_tbl);
1118
1119 static struct pci_driver e752x_driver = {
1120         .name = EDAC_MOD_STR,
1121         .probe = e752x_init_one,
1122         .remove = __devexit_p(e752x_remove_one),
1123         .id_table = e752x_pci_tbl,
1124 };
1125
1126 static int __init e752x_init(void)
1127 {
1128         int pci_rc;
1129
1130         debugf3("%s()\n", __func__);
1131         pci_rc = pci_register_driver(&e752x_driver);
1132         return (pci_rc < 0) ? pci_rc : 0;
1133 }
1134
1135 static void __exit e752x_exit(void)
1136 {
1137         debugf3("%s()\n", __func__);
1138         pci_unregister_driver(&e752x_driver);
1139 }
1140
1141 module_init(e752x_init);
1142 module_exit(e752x_exit);
1143
1144 MODULE_LICENSE("GPL");
1145 MODULE_AUTHOR("Linux Networx (http://lnxi.com) Tom Zimmerman\n");
1146 MODULE_DESCRIPTION("MC support for Intel e752x memory controllers");
1147
1148 module_param(force_function_unhide, int, 0444);
1149 MODULE_PARM_DESC(force_function_unhide, "if BIOS sets Dev0:Fun1 up as hidden:"
1150                  " 1=force unhide and hope BIOS doesn't fight driver for Dev0:Fun1 access");
1151 module_param(edac_op_state, int, 0444);
1152 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");