linux/drivers/edac/r82600_edac.c
<<
>>
Prefs
   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/edac.h>
  23#include "edac_module.h"
  24
  25#define EDAC_MOD_STR    "r82600_edac"
  26
  27#define r82600_printk(level, fmt, arg...) \
  28        edac_printk(level, "r82600", fmt, ##arg)
  29
  30#define r82600_mc_printk(mci, level, fmt, arg...) \
  31        edac_mc_chipset_printk(mci, level, "r82600", fmt, ##arg)
  32
  33/* Radisys say "The 82600 integrates a main memory SDRAM controller that
  34 * supports up to four banks of memory. The four banks can support a mix of
  35 * sizes of 64 bit wide (72 bits with ECC) Synchronous DRAM (SDRAM) DIMMs,
  36 * each of which can be any size from 16MB to 512MB. Both registered (control
  37 * signals buffered) and unbuffered DIMM types are supported. Mixing of
  38 * registered and unbuffered DIMMs as well as mixing of ECC and non-ECC DIMMs
  39 * is not allowed. The 82600 SDRAM interface operates at the same frequency as
  40 * the CPU bus, 66MHz, 100MHz or 133MHz."
  41 */
  42
  43#define R82600_NR_CSROWS 4
  44#define R82600_NR_CHANS  1
  45#define R82600_NR_DIMMS  4
  46
  47#define R82600_BRIDGE_ID  0x8200
  48
  49/* Radisys 82600 register addresses - device 0 function 0 - PCI bridge */
  50#define R82600_DRAMC    0x57    /* Various SDRAM related control bits
  51                                 * all bits are R/W
  52                                 *
  53                                 * 7    SDRAM ISA Hole Enable
  54                                 * 6    Flash Page Mode Enable
  55                                 * 5    ECC Enable: 1=ECC 0=noECC
  56                                 * 4    DRAM DIMM Type: 1=
  57                                 * 3    BIOS Alias Disable
  58                                 * 2    SDRAM BIOS Flash Write Enable
  59                                 * 1:0  SDRAM Refresh Rate: 00=Disabled
  60                                 *          01=7.8usec (256Mbit SDRAMs)
  61                                 *          10=15.6us 11=125usec
  62                                 */
  63
  64#define R82600_SDRAMC   0x76    /* "SDRAM Control Register"
  65                                 * More SDRAM related control bits
  66                                 * all bits are R/W
  67                                 *
  68                                 * 15:8 Reserved.
  69                                 *
  70                                 * 7:5  Special SDRAM Mode Select
  71                                 *
  72                                 * 4    Force ECC
  73                                 *
  74                                 *        1=Drive ECC bits to 0 during
  75                                 *          write cycles (i.e. ECC test mode)
  76                                 *
  77                                 *        0=Normal ECC functioning
  78                                 *
  79                                 * 3    Enhanced Paging Enable
  80                                 *
  81                                 * 2    CAS# Latency 0=3clks 1=2clks
  82                                 *
  83                                 * 1    RAS# to CAS# Delay 0=3 1=2
  84                                 *
  85                                 * 0    RAS# Precharge     0=3 1=2
  86                                 */
  87
  88#define R82600_EAP      0x80    /* ECC Error Address Pointer Register
  89                                 *
  90                                 * 31    Disable Hardware Scrubbing (RW)
  91                                 *        0=Scrub on corrected read
  92                                 *        1=Don't scrub on corrected read
  93                                 *
  94                                 * 30:12 Error Address Pointer (RO)
  95                                 *        Upper 19 bits of error address
  96                                 *
  97                                 * 11:4  Syndrome Bits (RO)
  98                                 *
  99                                 * 3     BSERR# on multibit error (RW)
 100                                 *        1=enable 0=disable
 101                                 *
 102                                 * 2     NMI on Single Bit Eror (RW)
 103                                 *        1=NMI triggered by SBE n.b. other
 104                                 *          prerequeists
 105                                 *        0=NMI not triggered
 106                                 *
 107                                 * 1     MBE (R/WC)
 108                                 *        read 1=MBE at EAP (see above)
 109                                 *        read 0=no MBE, or SBE occurred first
 110                                 *        write 1=Clear MBE status (must also
 111                                 *          clear SBE)
 112                                 *        write 0=NOP
 113                                 *
 114                                 * 1     SBE (R/WC)
 115                                 *        read 1=SBE at EAP (see above)
 116                                 *        read 0=no SBE, or MBE occurred first
 117                                 *        write 1=Clear SBE status (must also
 118                                 *          clear MBE)
 119                                 *        write 0=NOP
 120                                 */
 121
 122#define R82600_DRBA     0x60    /* + 0x60..0x63 SDRAM Row Boundary Address
 123                                 *  Registers
 124                                 *
 125                                 * 7:0  Address lines 30:24 - upper limit of
 126                                 * each row [p57]
 127                                 */
 128
 129struct r82600_error_info {
 130        u32 eapr;
 131};
 132
 133static bool disable_hardware_scrub;
 134
 135static struct edac_pci_ctl_info *r82600_pci;
 136
 137static void r82600_get_error_info(struct mem_ctl_info *mci,
 138                                struct r82600_error_info *info)
 139{
 140        struct pci_dev *pdev;
 141
 142        pdev = to_pci_dev(mci->pdev);
 143        pci_read_config_dword(pdev, R82600_EAP, &info->eapr);
 144
 145        if (info->eapr & BIT(0))
 146                /* Clear error to allow next error to be reported [p.62] */
 147                pci_write_bits32(pdev, R82600_EAP,
 148                                 ((u32) BIT(0) & (u32) BIT(1)),
 149                                 ((u32) BIT(0) & (u32) BIT(1)));
 150
 151        if (info->eapr & BIT(1))
 152                /* Clear error to allow next error to be reported [p.62] */
 153                pci_write_bits32(pdev, R82600_EAP,
 154                                 ((u32) BIT(0) & (u32) BIT(1)),
 155                                 ((u32) BIT(0) & (u32) BIT(1)));
 156}
 157
 158static int r82600_process_error_info(struct mem_ctl_info *mci,
 159                                struct r82600_error_info *info,
 160                                int handle_errors)
 161{
 162        int error_found;
 163        u32 eapaddr, page;
 164        u32 syndrome;
 165
 166        error_found = 0;
 167
 168        /* bits 30:12 store the upper 19 bits of the 32 bit error address */
 169        eapaddr = ((info->eapr >> 12) & 0x7FFF) << 13;
 170        /* Syndrome in bits 11:4 [p.62]       */
 171        syndrome = (info->eapr >> 4) & 0xFF;
 172
 173        /* the R82600 reports at less than page *
 174         * granularity (upper 19 bits only)     */
 175        page = eapaddr >> PAGE_SHIFT;
 176
 177        if (info->eapr & BIT(0)) {      /* CE? */
 178                error_found = 1;
 179
 180                if (handle_errors)
 181                        edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
 182                                             page, 0, syndrome,
 183                                             edac_mc_find_csrow_by_page(mci, page),
 184                                             0, -1,
 185                                             mci->ctl_name, "");
 186        }
 187
 188        if (info->eapr & BIT(1)) {      /* UE? */
 189                error_found = 1;
 190
 191                if (handle_errors)
 192                        /* 82600 doesn't give enough info */
 193                        edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 194                                             page, 0, 0,
 195                                             edac_mc_find_csrow_by_page(mci, page),
 196                                             0, -1,
 197                                             mci->ctl_name, "");
 198        }
 199
 200        return error_found;
 201}
 202
 203static void r82600_check(struct mem_ctl_info *mci)
 204{
 205        struct r82600_error_info info;
 206
 207        edac_dbg(1, "MC%d\n", mci->mc_idx);
 208        r82600_get_error_info(mci, &info);
 209        r82600_process_error_info(mci, &info, 1);
 210}
 211
 212static inline int ecc_enabled(u8 dramcr)
 213{
 214        return dramcr & BIT(5);
 215}
 216
 217static void r82600_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev,
 218                        u8 dramcr)
 219{
 220        struct csrow_info *csrow;
 221        struct dimm_info *dimm;
 222        int index;
 223        u8 drbar;               /* SDRAM Row Boundary Address Register */
 224        u32 row_high_limit, row_high_limit_last;
 225        u32 reg_sdram, ecc_on, row_base;
 226
 227        ecc_on = ecc_enabled(dramcr);
 228        reg_sdram = dramcr & BIT(4);
 229        row_high_limit_last = 0;
 230
 231        for (index = 0; index < mci->nr_csrows; index++) {
 232                csrow = mci->csrows[index];
 233                dimm = csrow->channels[0]->dimm;
 234
 235                /* find the DRAM Chip Select Base address and mask */
 236                pci_read_config_byte(pdev, R82600_DRBA + index, &drbar);
 237
 238                edac_dbg(1, "Row=%d DRBA = %#0x\n", index, drbar);
 239
 240                row_high_limit = ((u32) drbar << 24);
 241/*              row_high_limit = ((u32)drbar << 24) | 0xffffffUL; */
 242
 243                edac_dbg(1, "Row=%d, Boundary Address=%#0x, Last = %#0x\n",
 244                         index, row_high_limit, row_high_limit_last);
 245
 246                /* Empty row [p.57] */
 247                if (row_high_limit == row_high_limit_last)
 248                        continue;
 249
 250                row_base = row_high_limit_last;
 251
 252                csrow->first_page = row_base >> PAGE_SHIFT;
 253                csrow->last_page = (row_high_limit >> PAGE_SHIFT) - 1;
 254
 255                dimm->nr_pages = csrow->last_page - csrow->first_page + 1;
 256                /* Error address is top 19 bits - so granularity is      *
 257                 * 14 bits                                               */
 258                dimm->grain = 1 << 14;
 259                dimm->mtype = reg_sdram ? MEM_RDDR : MEM_DDR;
 260                /* FIXME - check that this is unknowable with this chipset */
 261                dimm->dtype = DEV_UNKNOWN;
 262
 263                /* Mode is global on 82600 */
 264                dimm->edac_mode = ecc_on ? EDAC_SECDED : EDAC_NONE;
 265                row_high_limit_last = row_high_limit;
 266        }
 267}
 268
 269static int r82600_probe1(struct pci_dev *pdev, int dev_idx)
 270{
 271        struct mem_ctl_info *mci;
 272        struct edac_mc_layer layers[2];
 273        u8 dramcr;
 274        u32 eapr;
 275        u32 scrub_disabled;
 276        u32 sdram_refresh_rate;
 277        struct r82600_error_info discard;
 278
 279        edac_dbg(0, "\n");
 280        pci_read_config_byte(pdev, R82600_DRAMC, &dramcr);
 281        pci_read_config_dword(pdev, R82600_EAP, &eapr);
 282        scrub_disabled = eapr & BIT(31);
 283        sdram_refresh_rate = dramcr & (BIT(0) | BIT(1));
 284        edac_dbg(2, "sdram refresh rate = %#0x\n", sdram_refresh_rate);
 285        edac_dbg(2, "DRAMC register = %#0x\n", dramcr);
 286        layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 287        layers[0].size = R82600_NR_CSROWS;
 288        layers[0].is_virt_csrow = true;
 289        layers[1].type = EDAC_MC_LAYER_CHANNEL;
 290        layers[1].size = R82600_NR_CHANS;
 291        layers[1].is_virt_csrow = false;
 292        mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0);
 293        if (mci == NULL)
 294                return -ENOMEM;
 295
 296        edac_dbg(0, "mci = %p\n", mci);
 297        mci->pdev = &pdev->dev;
 298        mci->mtype_cap = MEM_FLAG_RDDR | MEM_FLAG_DDR;
 299        mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 300        /* FIXME try to work out if the chip leads have been used for COM2
 301         * instead on this board? [MA6?] MAYBE:
 302         */
 303
 304        /* On the R82600, the pins for memory bits 72:65 - i.e. the   *
 305         * EC bits are shared with the pins for COM2 (!), so if COM2  *
 306         * is enabled, we assume COM2 is wired up, and thus no EDAC   *
 307         * is possible.                                               */
 308        mci->edac_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 309
 310        if (ecc_enabled(dramcr)) {
 311                if (scrub_disabled)
 312                        edac_dbg(3, "mci = %p - Scrubbing disabled! EAP: %#0x\n",
 313                                 mci, eapr);
 314        } else
 315                mci->edac_cap = EDAC_FLAG_NONE;
 316
 317        mci->mod_name = EDAC_MOD_STR;
 318        mci->ctl_name = "R82600";
 319        mci->dev_name = pci_name(pdev);
 320        mci->edac_check = r82600_check;
 321        mci->ctl_page_to_phys = NULL;
 322        r82600_init_csrows(mci, pdev, dramcr);
 323        r82600_get_error_info(mci, &discard);   /* clear counters */
 324
 325        /* Here we assume that we will never see multiple instances of this
 326         * type of memory controller.  The ID is therefore hardcoded to 0.
 327         */
 328        if (edac_mc_add_mc(mci)) {
 329                edac_dbg(3, "failed edac_mc_add_mc()\n");
 330                goto fail;
 331        }
 332
 333        /* get this far and it's successful */
 334
 335        if (disable_hardware_scrub) {
 336                edac_dbg(3, "Disabling Hardware Scrub (scrub on error)\n");
 337                pci_write_bits32(pdev, R82600_EAP, BIT(31), BIT(31));
 338        }
 339
 340        /* allocating generic PCI control info */
 341        r82600_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR);
 342        if (!r82600_pci) {
 343                printk(KERN_WARNING
 344                        "%s(): Unable to create PCI control\n",
 345                        __func__);
 346                printk(KERN_WARNING
 347                        "%s(): PCI error report via EDAC not setup\n",
 348                        __func__);
 349        }
 350
 351        edac_dbg(3, "success\n");
 352        return 0;
 353
 354fail:
 355        edac_mc_free(mci);
 356        return -ENODEV;
 357}
 358
 359/* returns count (>= 0), or negative on error */
 360static int r82600_init_one(struct pci_dev *pdev,
 361                           const struct pci_device_id *ent)
 362{
 363        edac_dbg(0, "\n");
 364
 365        /* don't need to call pci_enable_device() */
 366        return r82600_probe1(pdev, ent->driver_data);
 367}
 368
 369static void r82600_remove_one(struct pci_dev *pdev)
 370{
 371        struct mem_ctl_info *mci;
 372
 373        edac_dbg(0, "\n");
 374
 375        if (r82600_pci)
 376                edac_pci_release_generic_ctl(r82600_pci);
 377
 378        if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL)
 379                return;
 380
 381        edac_mc_free(mci);
 382}
 383
 384static const struct pci_device_id r82600_pci_tbl[] = {
 385        {
 386         PCI_DEVICE(PCI_VENDOR_ID_RADISYS, R82600_BRIDGE_ID)
 387         },
 388        {
 389         0,
 390         }                      /* 0 terminated list. */
 391};
 392
 393MODULE_DEVICE_TABLE(pci, r82600_pci_tbl);
 394
 395static struct pci_driver r82600_driver = {
 396        .name = EDAC_MOD_STR,
 397        .probe = r82600_init_one,
 398        .remove = r82600_remove_one,
 399        .id_table = r82600_pci_tbl,
 400};
 401
 402static int __init r82600_init(void)
 403{
 404       /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 405       opstate_init();
 406
 407        return pci_register_driver(&r82600_driver);
 408}
 409
 410static void __exit r82600_exit(void)
 411{
 412        pci_unregister_driver(&r82600_driver);
 413}
 414
 415module_init(r82600_init);
 416module_exit(r82600_exit);
 417
 418MODULE_LICENSE("GPL");
 419MODULE_AUTHOR("Tim Small <tim@buttersideup.com> - WPAD Ltd. "
 420                "on behalf of EADS Astrium");
 421MODULE_DESCRIPTION("MC support for Radisys 82600 memory controllers");
 422
 423module_param(disable_hardware_scrub, bool, 0644);
 424MODULE_PARM_DESC(disable_hardware_scrub,
 425                 "If set, disable the chipset's automatic scrub for CEs");
 426
 427module_param(edac_op_state, int, 0444);
 428MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
 429