drivers/edac/r82600_edac.c

   1 /*
   2  * Radisys 82600 Embedded chipset Memory Controller kernel module
   3  * (C) 2005 EADS Astrium
   4  * This file may be distributed under the terms of the
   5  * GNU General Public License.
   6  *
   7  * Written by Tim Small <tim@buttersideup.com>, based on work by Thayne
   8  * Harbaugh, Dan Hollis <goemon at anime dot net> and others.
   9  *
  10  * $Id: edac_r82600.c,v 1.1.2.6 2005/10/05 00:43:44 dsp_llnl Exp $
  11  *
  12  * Written with reference to 82600 High Integration Dual PCI System
  13  * Controller Data Book:
  14  * www.radisys.com/files/support_downloads/007-01277-0002.82600DataBook.pdf
  15  * references to this document given in []
  16  */
  17
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/pci.h>
  21 #include <linux/pci_ids.h>
  22 #include <linux/slab.h>
  23 #include <linux/edac.h>
  24 #include "edac_core.h"
  25
  26 #define R82600_REVISION " Ver: 2.0.2 " __DATE__
  27 #define EDAC_MOD_STR    "r82600_edac"
  28
  29 #define r82600_printk(level, fmt, arg...) \
  30         edac_printk(level, "r82600", fmt, ##arg)
  31
  32 #define r82600_mc_printk(mci, level, fmt, arg...) \
  33         edac_mc_chipset_printk(mci, level, "r82600", fmt, ##arg)
  34
  35 /* Radisys say "The 82600 integrates a main memory SDRAM controller that
  36  * supports up to four banks of memory. The four banks can support a mix of
  37  * sizes of 64 bit wide (72 bits with ECC) Synchronous DRAM (SDRAM) DIMMs,
  38  * each of which can be any size from 16MB to 512MB. Both registered (control
  39  * signals buffered) and unbuffered DIMM types are supported. Mixing of
  40  * registered and unbuffered DIMMs as well as mixing of ECC and non-ECC DIMMs
  41  * is not allowed. The 82600 SDRAM interface operates at the same frequency as
  42  * the CPU bus, 66MHz, 100MHz or 133MHz."
  43  */
  44
  45 #define R82600_NR_CSROWS 4
  46 #define R82600_NR_CHANS  1
  47 #define R82600_NR_DIMMS  4
  48
  49 #define R82600_BRIDGE_ID  0x8200
  50
  51 /* Radisys 82600 register addresses - device 0 function 0 - PCI bridge */
  52 #define R82600_DRAMC    0x57    /* Various SDRAM related control bits
  53                                  * all bits are R/W
  54                                  *
  55                                  * 7    SDRAM ISA Hole Enable
  56                                  * 6    Flash Page Mode Enable
  57                                  * 5    ECC Enable: 1=ECC 0=noECC
  58                                  * 4    DRAM DIMM Type: 1=
  59                                  * 3    BIOS Alias Disable
  60                                  * 2    SDRAM BIOS Flash Write Enable
  61                                  * 1:0  SDRAM Refresh Rate: 00=Disabled
  62                                  *          01=7.8usec (256Mbit SDRAMs)
  63                                  *          10=15.6us 11=125usec
  64                                  */
  65
  66 #define R82600_SDRAMC   0x76    /* "SDRAM Control Register"
  67                                  * More SDRAM related control bits
  68                                  * all bits are R/W
  69                                  *
  70                                  * 15:8 Reserved.
  71                                  *
  72                                  * 7:5  Special SDRAM Mode Select
  73                                  *
  74                                  * 4    Force ECC
  75                                  *
  76                                  *        1=Drive ECC bits to 0 during
  77                                  *          write cycles (i.e. ECC test mode)
  78                                  *
  79                                  *        0=Normal ECC functioning
  80                                  *
  81                                  * 3    Enhanced Paging Enable
  82                                  *
  83                                  * 2    CAS# Latency 0=3clks 1=2clks
  84                                  *
  85                                  * 1    RAS# to CAS# Delay 0=3 1=2
  86                                  *
  87                                  * 0    RAS# Precharge     0=3 1=2
  88                                  */
  89
  90 #define R82600_EAP      0x80    /* ECC Error Address Pointer Register
  91                                  *
  92                                  * 31    Disable Hardware Scrubbing (RW)
  93                                  *        0=Scrub on corrected read
  94                                  *        1=Don't scrub on corrected read
  95                                  *
  96                                  * 30:12 Error Address Pointer (RO)
  97                                  *        Upper 19 bits of error address
  98                                  *
  99                                  * 11:4  Syndrome Bits (RO)
 100                                  *
 101                                  * 3     BSERR# on multibit error (RW)
 102                                  *        1=enable 0=disable
 103                                  *
 104                                  * 2     NMI on Single Bit Eror (RW)
 105                                  *        1=NMI triggered by SBE n.b. other
 106                                  *          prerequeists
 107                                  *        0=NMI not triggered
 108                                  *
 109                                  * 1     MBE (R/WC)
 110                                  *        read 1=MBE at EAP (see above)
 111                                  *        read 0=no MBE, or SBE occurred first
 112                                  *        write 1=Clear MBE status (must also
 113                                  *          clear SBE)
 114                                  *        write 0=NOP
 115                                  *
 116                                  * 1     SBE (R/WC)
 117                                  *        read 1=SBE at EAP (see above)
 118                                  *        read 0=no SBE, or MBE occurred first
 119                                  *        write 1=Clear SBE status (must also
 120                                  *          clear MBE)
 121                                  *        write 0=NOP
 122                                  */
 123
 124 #define R82600_DRBA     0x60    /* + 0x60..0x63 SDRAM Row Boundry Address
 125                                  *  Registers
 126                                  *
 127                                  * 7:0  Address lines 30:24 - upper limit of
 128                                  * each row [p57]
 129                                  */
 130
 131 struct r82600_error_info {
 132         u32 eapr;
 133 };
 134
 135 static unsigned int disable_hardware_scrub;
 136
 137 static struct edac_pci_ctl_info *r82600_pci;
 138
 139 static void r82600_get_error_info(struct mem_ctl_info *mci,
 140                                 struct r82600_error_info *info)
 141 {
 142         struct pci_dev *pdev;
 143
 144         pdev = to_pci_dev(mci->dev);
 145         pci_read_config_dword(pdev, R82600_EAP, &info->eapr);
 146
 147         if (info->eapr & BIT(0))
 148                 /* Clear error to allow next error to be reported [p.62] */
 149                 pci_write_bits32(pdev, R82600_EAP,
 150                                  ((u32) BIT(0) & (u32) BIT(1)),
 151                                  ((u32) BIT(0) & (u32) BIT(1)));
 152
 153         if (info->eapr & BIT(1))
 154                 /* Clear error to allow next error to be reported [p.62] */
 155                 pci_write_bits32(pdev, R82600_EAP,
 156                                  ((u32) BIT(0) & (u32) BIT(1)),
 157                                  ((u32) BIT(0) & (u32) BIT(1)));
 158 }
 159
 160 static int r82600_process_error_info(struct mem_ctl_info *mci,
 161                                 struct r82600_error_info *info,
 162                                 int handle_errors)
 163 {
 164         int error_found;
 165         u32 eapaddr, page;
 166         u32 syndrome;
 167
 168         error_found = 0;
 169
 170         /* bits 30:12 store the upper 19 bits of the 32 bit error address */
 171         eapaddr = ((info->eapr >> 12) & 0x7FFF) << 13;
 172         /* Syndrome in bits 11:4 [p.62]       */
 173         syndrome = (info->eapr >> 4) & 0xFF;
 174
 175         /* the R82600 reports at less than page *
 176          * granularity (upper 19 bits only)     */
 177         page = eapaddr >> PAGE_SHIFT;
 178
 179         if (info->eapr & BIT(0)) {      /* CE? */
 180                 error_found = 1;
 181
 182                 if (handle_errors)
 183                         edac_mc_handle_ce(mci, page, 0, /* not avail */
 184                                         syndrome,
 185                                         edac_mc_find_csrow_by_page(mci, page),
 186                                         0, mci->ctl_name);
 187         }
 188
 189         if (info->eapr & BIT(1)) {      /* UE? */
 190                 error_found = 1;
 191
 192                 if (handle_errors)
 193                         /* 82600 doesn't give enough info */
 194                         edac_mc_handle_ue(mci, page, 0,
 195                                         edac_mc_find_csrow_by_page(mci, page),
 196                                         mci->ctl_name);
 197         }
 198
 199         return error_found;
 200 }
 201
 202 static void r82600_check(struct mem_ctl_info *mci)
 203 {
 204         struct r82600_error_info info;
 205
 206         debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 207         r82600_get_error_info(mci, &info);
 208         r82600_process_error_info(mci, &info, 1);
 209 }
 210
 211 static inline int ecc_enabled(u8 dramcr)
 212 {
 213         return dramcr & BIT(5);
 214 }
 215
 216 static void r82600_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
 217                         u8 dramcr)
 218 {
 219         struct csrow_info *csrow;
 220         int index;
 221         u8 drbar;               /* SDRAM Row Boundry Address Register */
 222         u32 row_high_limit, row_high_limit_last;
 223         u32 reg_sdram, ecc_on, row_base;
 224
 225         ecc_on = ecc_enabled(dramcr);
 226         reg_sdram = dramcr & BIT(4);
 227         row_high_limit_last = 0;
 228
 229         for (index = 0; index < mci->nr_csrows; index++) {
 230                 csrow = &mci->csrows[index];
 231
 232                 /* find the DRAM Chip Select Base address and mask */
 233                 pci_read_config_byte(pdev, R82600_DRBA + index, &drbar);
 234
 235                 debugf1("%s() Row=%d DRBA = %#0x\n", __func__, index, drbar);
 236
 237                 row_high_limit = ((u32) drbar << 24);
 238 /*              row_high_limit = ((u32)drbar << 24) | 0xffffffUL; */
 239
 240                 debugf1("%s() Row=%d, Boundry Address=%#0x, Last = %#0x\n",
 241                         __func__, index, row_high_limit, row_high_limit_last);
 242
 243                 /* Empty row [p.57] */
 244                 if (row_high_limit == row_high_limit_last)
 245                         continue;
 246
 247                 row_base = row_high_limit_last;
 248
 249                 csrow->first_page = row_base >> PAGE_SHIFT;
 250                 csrow->last_page = (row_high_limit >> PAGE_SHIFT) - 1;
 251                 csrow->nr_pages = csrow->last_page - csrow->first_page + 1;
 252                 /* Error address is top 19 bits - so granularity is      *
 253                  * 14 bits                                               */
 254                 csrow->grain = 1 << 14;
 255                 csrow->mtype = reg_sdram ? MEM_RDDR : MEM_DDR;
 256                 /* FIXME - check that this is unknowable with this chipset */
 257                 csrow->dtype = DEV_UNKNOWN;
 258
 259                 /* Mode is global on 82600 */
 260                 csrow->edac_mode = ecc_on ? EDAC_SECDED : EDAC_NONE;
 261                 row_high_limit_last = row_high_limit;
 262         }
 263 }
 264
 265 static int r82600_probe1(struct pci_dev *pdev, int dev_idx)
 266 {
 267         struct mem_ctl_info *mci;
 268         u8 dramcr;
 269         u32 eapr;
 270         u32 scrub_disabled;
 271         u32 sdram_refresh_rate;
 272         struct r82600_error_info discard;
 273
 274         debugf0("%s()\n", __func__);
 275         pci_read_config_byte(pdev, R82600_DRAMC, &dramcr);
 276         pci_read_config_dword(pdev, R82600_EAP, &eapr);
 277         scrub_disabled = eapr & BIT(31);
 278         sdram_refresh_rate = dramcr & (BIT(0) | BIT(1));
 279         debugf2("%s(): sdram refresh rate = %#0x\n", __func__,
 280                 sdram_refresh_rate);
 281         debugf2("%s(): DRAMC register = %#0x\n", __func__, dramcr);
 282         mci = edac_mc_alloc(0, R82600_NR_CSROWS, R82600_NR_CHANS, 0);
 283
 284         if (mci == NULL)
 285                 return -ENOMEM;
 286
 287         debugf0("%s(): mci = %p\n", __func__, mci);
 288         mci->dev = &pdev->dev;
 289         mci->mtype_cap = MEM_FLAG_RDDR | MEM_FLAG_DDR;
 290         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 291         /* FIXME try to work out if the chip leads have been used for COM2
 292          * instead on this board? [MA6?] MAYBE:
 293          */
 294
 295         /* On the R82600, the pins for memory bits 72:65 - i.e. the   *
 296          * EC bits are shared with the pins for COM2 (!), so if COM2  *
 297          * is enabled, we assume COM2 is wired up, and thus no EDAC   *
 298          * is possible.                                               */
 299         mci->edac_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 300
 301         if (ecc_enabled(dramcr)) {
 302                 if (scrub_disabled)
 303                         debugf3("%s(): mci = %p - Scrubbing disabled! EAP: "
 304                                 "%#0x\n", __func__, mci, eapr);
 305         } else
 306                 mci->edac_cap = EDAC_FLAG_NONE;
 307
 308         mci->mod_name = EDAC_MOD_STR;
 309         mci->mod_ver = R82600_REVISION;
 310         mci->ctl_name = "R82600";
 311         mci->dev_name = pci_name(pdev);
 312         mci->edac_check = r82600_check;
 313         mci->ctl_page_to_phys = NULL;
 314         r82600_init_csrows(mci, pdev, dramcr);
 315         r82600_get_error_info(mci, &discard);   /* clear counters */
 316
 317         /* Here we assume that we will never see multiple instances of this
 318          * type of memory controller.  The ID is therefore hardcoded to 0.
 319          */
 320         if (edac_mc_add_mc(mci)) {
 321                 debugf3("%s(): failed edac_mc_add_mc()\n", __func__);
 322                 goto fail;
 323         }
 324
 325         /* get this far and it's successful */
 326
 327         if (disable_hardware_scrub) {
 328                 debugf3("%s(): Disabling Hardware Scrub (scrub on error)\n",
 329                         __func__);
 330                 pci_write_bits32(pdev, R82600_EAP, BIT(31), BIT(31));
 331         }
 332
 333         /* allocating generic PCI control info */
 334         r82600_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
 335         if (!r82600_pci) {
 336                 printk(KERN_WARNING
 337                         "%s(): Unable to create PCI control\n",
 338                         __func__);
 339                 printk(KERN_WARNING
 340                         "%s(): PCI error report via EDAC not setup\n",
 341                         __func__);
 342         }
 343
 344         debugf3("%s(): success\n", __func__);
 345         return 0;
 346
 347 fail:
 348         edac_mc_free(mci);
 349         return -ENODEV;
 350 }
 351
 352 /* returns count (>= 0), or negative on error */
 353 static int __devinit r82600_init_one(struct pci_dev *pdev,
 354                                 const struct pci_device_id *ent)
 355 {
 356         debugf0("%s()\n", __func__);
 357
 358         /* don't need to call pci_device_enable() */
 359         return r82600_probe1(pdev, ent->driver_data);
 360 }
 361
 362 static void __devexit r82600_remove_one(struct pci_dev *pdev)
 363 {
 364         struct mem_ctl_info *mci;
 365
 366         debugf0("%s()\n", __func__);
 367
 368         if (r82600_pci)
 369                 edac_pci_release_generic_ctl(r82600_pci);
 370
 371         if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
 372                 return;
 373
 374         edac_mc_free(mci);
 375 }
 376
 377 static const struct pci_device_id r82600_pci_tbl[] __devinitdata = {
 378         {
 379          PCI_DEVICE(PCI_VENDOR_ID_RADISYS, R82600_BRIDGE_ID)
 380          },
 381         {
 382          0,
 383          }                      /* 0 terminated list. */
 384 };
 385
 386 MODULE_DEVICE_TABLE(pci, r82600_pci_tbl);
 387
 388 static struct pci_driver r82600_driver = {
 389         .name = EDAC_MOD_STR,
 390         .probe = r82600_init_one,
 391         .remove = __devexit_p(r82600_remove_one),
 392         .id_table = r82600_pci_tbl,
 393 };
 394
 395 static int __init r82600_init(void)
 396 {
 397        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 398        opstate_init();
 399
 400         return pci_register_driver(&r82600_driver);
 401 }
 402
 403 static void __exit r82600_exit(void)
 404 {
 405         pci_unregister_driver(&r82600_driver);
 406 }
 407
 408 module_init(r82600_init);
 409 module_exit(r82600_exit);
 410
 411 MODULE_LICENSE("GPL");
 412 MODULE_AUTHOR("Tim Small <tim@buttersideup.com> - WPAD Ltd. "
 413                 "on behalf of EADS Astrium");
 414 MODULE_DESCRIPTION("MC support for Radisys 82600 memory controllers");
 415
 416 module_param(disable_hardware_scrub, bool, 0644);
 417 MODULE_PARM_DESC(disable_hardware_scrub,
 418                  "If set, disable the chipset's automatic scrub for CEs");
 419
 420 module_param(edac_op_state, int, 0444);
 421 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");