linux/drivers/edac/i3200_edac.c
<<
>>
Prefs
   1/*
   2 * Intel 3200/3210 Memory Controller kernel module
   3 * Copyright (C) 2008-2009 Akamai Technologies, Inc.
   4 * Portions by Hitoshi Mitake <h.mitake@gmail.com>.
   5 *
   6 * This file may be distributed under the terms of the
   7 * GNU General Public License.
   8 */
   9
  10#include <linux/module.h>
  11#include <linux/init.h>
  12#include <linux/pci.h>
  13#include <linux/pci_ids.h>
  14#include <linux/slab.h>
  15#include <linux/edac.h>
  16#include <linux/io.h>
  17#include "edac_core.h"
  18
  19#define I3200_REVISION        "1.1"
  20
  21#define EDAC_MOD_STR        "i3200_edac"
  22
  23#define PCI_DEVICE_ID_INTEL_3200_HB    0x29f0
  24
  25#define I3200_RANKS             8
  26#define I3200_RANKS_PER_CHANNEL 4
  27#define I3200_CHANNELS          2
  28
  29/* Intel 3200 register addresses - device 0 function 0 - DRAM Controller */
  30
  31#define I3200_MCHBAR_LOW        0x48    /* MCH Memory Mapped Register BAR */
  32#define I3200_MCHBAR_HIGH       0x4c
  33#define I3200_MCHBAR_MASK       0xfffffc000ULL  /* bits 35:14 */
  34#define I3200_MMR_WINDOW_SIZE   16384
  35
  36#define I3200_TOM               0xa0    /* Top of Memory (16b)
  37                 *
  38                 * 15:10 reserved
  39                 *  9:0  total populated physical memory
  40                 */
  41#define I3200_TOM_MASK          0x3ff   /* bits 9:0 */
  42#define I3200_TOM_SHIFT         26      /* 64MiB grain */
  43
  44#define I3200_ERRSTS            0xc8    /* Error Status Register (16b)
  45                 *
  46                 * 15    reserved
  47                 * 14    Isochronous TBWRR Run Behind FIFO Full
  48                 *       (ITCV)
  49                 * 13    Isochronous TBWRR Run Behind FIFO Put
  50                 *       (ITSTV)
  51                 * 12    reserved
  52                 * 11    MCH Thermal Sensor Event
  53                 *       for SMI/SCI/SERR (GTSE)
  54                 * 10    reserved
  55                 *  9    LOCK to non-DRAM Memory Flag (LCKF)
  56                 *  8    reserved
  57                 *  7    DRAM Throttle Flag (DTF)
  58                 *  6:2  reserved
  59                 *  1    Multi-bit DRAM ECC Error Flag (DMERR)
  60                 *  0    Single-bit DRAM ECC Error Flag (DSERR)
  61                 */
  62#define I3200_ERRSTS_UE         0x0002
  63#define I3200_ERRSTS_CE         0x0001
  64#define I3200_ERRSTS_BITS       (I3200_ERRSTS_UE | I3200_ERRSTS_CE)
  65
  66
  67/* Intel  MMIO register space - device 0 function 0 - MMR space */
  68
  69#define I3200_C0DRB     0x200   /* Channel 0 DRAM Rank Boundary (16b x 4)
  70                 *
  71                 * 15:10 reserved
  72                 *  9:0  Channel 0 DRAM Rank Boundary Address
  73                 */
  74#define I3200_C1DRB     0x600   /* Channel 1 DRAM Rank Boundary (16b x 4) */
  75#define I3200_DRB_MASK  0x3ff   /* bits 9:0 */
  76#define I3200_DRB_SHIFT 26      /* 64MiB grain */
  77
  78#define I3200_C0ECCERRLOG       0x280   /* Channel 0 ECC Error Log (64b)
  79                 *
  80                 * 63:48 Error Column Address (ERRCOL)
  81                 * 47:32 Error Row Address (ERRROW)
  82                 * 31:29 Error Bank Address (ERRBANK)
  83                 * 28:27 Error Rank Address (ERRRANK)
  84                 * 26:24 reserved
  85                 * 23:16 Error Syndrome (ERRSYND)
  86                 * 15: 2 reserved
  87                 *    1  Multiple Bit Error Status (MERRSTS)
  88                 *    0  Correctable Error Status (CERRSTS)
  89                 */
  90#define I3200_C1ECCERRLOG               0x680   /* Chan 1 ECC Error Log (64b) */
  91#define I3200_ECCERRLOG_CE              0x1
  92#define I3200_ECCERRLOG_UE              0x2
  93#define I3200_ECCERRLOG_RANK_BITS       0x18000000
  94#define I3200_ECCERRLOG_RANK_SHIFT      27
  95#define I3200_ECCERRLOG_SYNDROME_BITS   0xff0000
  96#define I3200_ECCERRLOG_SYNDROME_SHIFT  16
  97#define I3200_CAPID0                    0xe0    /* P.95 of spec for details */
  98
  99struct i3200_priv {
 100        void __iomem *window;
 101};
 102
 103static int nr_channels;
 104
 105static int how_many_channels(struct pci_dev *pdev)
 106{
 107        unsigned char capid0_8b; /* 8th byte of CAPID0 */
 108
 109        pci_read_config_byte(pdev, I3200_CAPID0 + 8, &capid0_8b);
 110        if (capid0_8b & 0x20) { /* check DCD: Dual Channel Disable */
 111                debugf0("In single channel mode.\n");
 112                return 1;
 113        } else {
 114                debugf0("In dual channel mode.\n");
 115                return 2;
 116        }
 117}
 118
 119static unsigned long eccerrlog_syndrome(u64 log)
 120{
 121        return (log & I3200_ECCERRLOG_SYNDROME_BITS) >>
 122                I3200_ECCERRLOG_SYNDROME_SHIFT;
 123}
 124
 125static int eccerrlog_row(int channel, u64 log)
 126{
 127        u64 rank = ((log & I3200_ECCERRLOG_RANK_BITS) >>
 128                I3200_ECCERRLOG_RANK_SHIFT);
 129        return rank | (channel * I3200_RANKS_PER_CHANNEL);
 130}
 131
 132enum i3200_chips {
 133        I3200 = 0,
 134};
 135
 136struct i3200_dev_info {
 137        const char *ctl_name;
 138};
 139
 140struct i3200_error_info {
 141        u16 errsts;
 142        u16 errsts2;
 143        u64 eccerrlog[I3200_CHANNELS];
 144};
 145
 146static const struct i3200_dev_info i3200_devs[] = {
 147        [I3200] = {
 148                .ctl_name = "i3200"
 149        },
 150};
 151
 152static struct pci_dev *mci_pdev;
 153static int i3200_registered = 1;
 154
 155
 156static void i3200_clear_error_info(struct mem_ctl_info *mci)
 157{
 158        struct pci_dev *pdev;
 159
 160        pdev = to_pci_dev(mci->dev);
 161
 162        /*
 163         * Clear any error bits.
 164         * (Yes, we really clear bits by writing 1 to them.)
 165         */
 166        pci_write_bits16(pdev, I3200_ERRSTS, I3200_ERRSTS_BITS,
 167                I3200_ERRSTS_BITS);
 168}
 169
 170static void i3200_get_and_clear_error_info(struct mem_ctl_info *mci,
 171                struct i3200_error_info *info)
 172{
 173        struct pci_dev *pdev;
 174        struct i3200_priv *priv = mci->pvt_info;
 175        void __iomem *window = priv->window;
 176
 177        pdev = to_pci_dev(mci->dev);
 178
 179        /*
 180         * This is a mess because there is no atomic way to read all the
 181         * registers at once and the registers can transition from CE being
 182         * overwritten by UE.
 183         */
 184        pci_read_config_word(pdev, I3200_ERRSTS, &info->errsts);
 185        if (!(info->errsts & I3200_ERRSTS_BITS))
 186                return;
 187
 188        info->eccerrlog[0] = readq(window + I3200_C0ECCERRLOG);
 189        if (nr_channels == 2)
 190                info->eccerrlog[1] = readq(window + I3200_C1ECCERRLOG);
 191
 192        pci_read_config_word(pdev, I3200_ERRSTS, &info->errsts2);
 193
 194        /*
 195         * If the error is the same for both reads then the first set
 196         * of reads is valid.  If there is a change then there is a CE
 197         * with no info and the second set of reads is valid and
 198         * should be UE info.
 199         */
 200        if ((info->errsts ^ info->errsts2) & I3200_ERRSTS_BITS) {
 201                info->eccerrlog[0] = readq(window + I3200_C0ECCERRLOG);
 202                if (nr_channels == 2)
 203                        info->eccerrlog[1] = readq(window + I3200_C1ECCERRLOG);
 204        }
 205
 206        i3200_clear_error_info(mci);
 207}
 208
 209static void i3200_process_error_info(struct mem_ctl_info *mci,
 210                struct i3200_error_info *info)
 211{
 212        int channel;
 213        u64 log;
 214
 215        if (!(info->errsts & I3200_ERRSTS_BITS))
 216                return;
 217
 218        if ((info->errsts ^ info->errsts2) & I3200_ERRSTS_BITS) {
 219                edac_mc_handle_ce_no_info(mci, "UE overwrote CE");
 220                info->errsts = info->errsts2;
 221        }
 222
 223        for (channel = 0; channel < nr_channels; channel++) {
 224                log = info->eccerrlog[channel];
 225                if (log & I3200_ECCERRLOG_UE) {
 226                        edac_mc_handle_ue(mci, 0, 0,
 227                                eccerrlog_row(channel, log),
 228                                "i3200 UE");
 229                } else if (log & I3200_ECCERRLOG_CE) {
 230                        edac_mc_handle_ce(mci, 0, 0,
 231                                eccerrlog_syndrome(log),
 232                                eccerrlog_row(channel, log), 0,
 233                                "i3200 CE");
 234                }
 235        }
 236}
 237
 238static void i3200_check(struct mem_ctl_info *mci)
 239{
 240        struct i3200_error_info info;
 241
 242        debugf1("MC%d: %s()\n", mci->mc_idx, __func__);
 243        i3200_get_and_clear_error_info(mci, &info);
 244        i3200_process_error_info(mci, &info);
 245}
 246
 247
 248void __iomem *i3200_map_mchbar(struct pci_dev *pdev)
 249{
 250        union {
 251                u64 mchbar;
 252                struct {
 253                        u32 mchbar_low;
 254                        u32 mchbar_high;
 255                };
 256        } u;
 257        void __iomem *window;
 258
 259        pci_read_config_dword(pdev, I3200_MCHBAR_LOW, &u.mchbar_low);
 260        pci_read_config_dword(pdev, I3200_MCHBAR_HIGH, &u.mchbar_high);
 261        u.mchbar &= I3200_MCHBAR_MASK;
 262
 263        if (u.mchbar != (resource_size_t)u.mchbar) {
 264                printk(KERN_ERR
 265                        "i3200: mmio space beyond accessible range (0x%llx)\n",
 266                        (unsigned long long)u.mchbar);
 267                return NULL;
 268        }
 269
 270        window = ioremap_nocache(u.mchbar, I3200_MMR_WINDOW_SIZE);
 271        if (!window)
 272                printk(KERN_ERR "i3200: cannot map mmio space at 0x%llx\n",
 273                        (unsigned long long)u.mchbar);
 274
 275        return window;
 276}
 277
 278
 279static void i3200_get_drbs(void __iomem *window,
 280        u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL])
 281{
 282        int i;
 283
 284        for (i = 0; i < I3200_RANKS_PER_CHANNEL; i++) {
 285                drbs[0][i] = readw(window + I3200_C0DRB + 2*i) & I3200_DRB_MASK;
 286                drbs[1][i] = readw(window + I3200_C1DRB + 2*i) & I3200_DRB_MASK;
 287        }
 288}
 289
 290static bool i3200_is_stacked(struct pci_dev *pdev,
 291        u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL])
 292{
 293        u16 tom;
 294
 295        pci_read_config_word(pdev, I3200_TOM, &tom);
 296        tom &= I3200_TOM_MASK;
 297
 298        return drbs[I3200_CHANNELS - 1][I3200_RANKS_PER_CHANNEL - 1] == tom;
 299}
 300
 301static unsigned long drb_to_nr_pages(
 302        u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL], bool stacked,
 303        int channel, int rank)
 304{
 305        int n;
 306
 307        n = drbs[channel][rank];
 308        if (rank > 0)
 309                n -= drbs[channel][rank - 1];
 310        if (stacked && (channel == 1) &&
 311        drbs[channel][rank] == drbs[channel][I3200_RANKS_PER_CHANNEL - 1])
 312                n -= drbs[0][I3200_RANKS_PER_CHANNEL - 1];
 313
 314        n <<= (I3200_DRB_SHIFT - PAGE_SHIFT);
 315        return n;
 316}
 317
 318static int i3200_probe1(struct pci_dev *pdev, int dev_idx)
 319{
 320        int rc;
 321        int i;
 322        struct mem_ctl_info *mci = NULL;
 323        unsigned long last_page;
 324        u16 drbs[I3200_CHANNELS][I3200_RANKS_PER_CHANNEL];
 325        bool stacked;
 326        void __iomem *window;
 327        struct i3200_priv *priv;
 328
 329        debugf0("MC: %s()\n", __func__);
 330
 331        window = i3200_map_mchbar(pdev);
 332        if (!window)
 333                return -ENODEV;
 334
 335        i3200_get_drbs(window, drbs);
 336        nr_channels = how_many_channels(pdev);
 337
 338        mci = edac_mc_alloc(sizeof(struct i3200_priv), I3200_RANKS,
 339                nr_channels, 0);
 340        if (!mci)
 341                return -ENOMEM;
 342
 343        debugf3("MC: %s(): init mci\n", __func__);
 344
 345        mci->dev = &pdev->dev;
 346        mci->mtype_cap = MEM_FLAG_DDR2;
 347
 348        mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 349        mci->edac_cap = EDAC_FLAG_SECDED;
 350
 351        mci->mod_name = EDAC_MOD_STR;
 352        mci->mod_ver = I3200_REVISION;
 353        mci->ctl_name = i3200_devs[dev_idx].ctl_name;
 354        mci->dev_name = pci_name(pdev);
 355        mci->edac_check = i3200_check;
 356        mci->ctl_page_to_phys = NULL;
 357        priv = mci->pvt_info;
 358        priv->window = window;
 359
 360        stacked = i3200_is_stacked(pdev, drbs);
 361
 362        /*
 363         * The dram rank boundary (DRB) reg values are boundary addresses
 364         * for each DRAM rank with a granularity of 64MB.  DRB regs are
 365         * cumulative; the last one will contain the total memory
 366         * contained in all ranks.
 367         */
 368        last_page = -1UL;
 369        for (i = 0; i < mci->nr_csrows; i++) {
 370                unsigned long nr_pages;
 371                struct csrow_info *csrow = &mci->csrows[i];
 372
 373                nr_pages = drb_to_nr_pages(drbs, stacked,
 374                        i / I3200_RANKS_PER_CHANNEL,
 375                        i % I3200_RANKS_PER_CHANNEL);
 376
 377                if (nr_pages == 0) {
 378                        csrow->mtype = MEM_EMPTY;
 379                        continue;
 380                }
 381
 382                csrow->first_page = last_page + 1;
 383                last_page += nr_pages;
 384                csrow->last_page = last_page;
 385                csrow->nr_pages = nr_pages;
 386
 387                csrow->grain = nr_pages << PAGE_SHIFT;
 388                csrow->mtype = MEM_DDR2;
 389                csrow->dtype = DEV_UNKNOWN;
 390                csrow->edac_mode = EDAC_UNKNOWN;
 391        }
 392
 393        i3200_clear_error_info(mci);
 394
 395        rc = -ENODEV;
 396        if (edac_mc_add_mc(mci)) {
 397                debugf3("MC: %s(): failed edac_mc_add_mc()\n", __func__);
 398                goto fail;
 399        }
 400
 401        /* get this far and it's successful */
 402        debugf3("MC: %s(): success\n", __func__);
 403        return 0;
 404
 405fail:
 406        iounmap(window);
 407        if (mci)
 408                edac_mc_free(mci);
 409
 410        return rc;
 411}
 412
 413static int __devinit i3200_init_one(struct pci_dev *pdev,
 414                const struct pci_device_id *ent)
 415{
 416        int rc;
 417
 418        debugf0("MC: %s()\n", __func__);
 419
 420        if (pci_enable_device(pdev) < 0)
 421                return -EIO;
 422
 423        rc = i3200_probe1(pdev, ent->driver_data);
 424        if (!mci_pdev)
 425                mci_pdev = pci_dev_get(pdev);
 426
 427        return rc;
 428}
 429
 430static void __devexit i3200_remove_one(struct pci_dev *pdev)
 431{
 432        struct mem_ctl_info *mci;
 433        struct i3200_priv *priv;
 434
 435        debugf0("%s()\n", __func__);
 436
 437        mci = edac_mc_del_mc(&pdev->dev);
 438        if (!mci)
 439                return;
 440
 441        priv = mci->pvt_info;
 442        iounmap(priv->window);
 443
 444        edac_mc_free(mci);
 445}
 446
 447static const struct pci_device_id i3200_pci_tbl[] __devinitdata = {
 448        {
 449                PCI_VEND_DEV(INTEL, 3200_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 450                I3200},
 451        {
 452                0,
 453        }            /* 0 terminated list. */
 454};
 455
 456MODULE_DEVICE_TABLE(pci, i3200_pci_tbl);
 457
 458static struct pci_driver i3200_driver = {
 459        .name = EDAC_MOD_STR,
 460        .probe = i3200_init_one,
 461        .remove = __devexit_p(i3200_remove_one),
 462        .id_table = i3200_pci_tbl,
 463};
 464
 465static int __init i3200_init(void)
 466{
 467        int pci_rc;
 468
 469        debugf3("MC: %s()\n", __func__);
 470
 471        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 472        opstate_init();
 473
 474        pci_rc = pci_register_driver(&i3200_driver);
 475        if (pci_rc < 0)
 476                goto fail0;
 477
 478        if (!mci_pdev) {
 479                i3200_registered = 0;
 480                mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
 481                                PCI_DEVICE_ID_INTEL_3200_HB, NULL);
 482                if (!mci_pdev) {
 483                        debugf0("i3200 pci_get_device fail\n");
 484                        pci_rc = -ENODEV;
 485                        goto fail1;
 486                }
 487
 488                pci_rc = i3200_init_one(mci_pdev, i3200_pci_tbl);
 489                if (pci_rc < 0) {
 490                        debugf0("i3200 init fail\n");
 491                        pci_rc = -ENODEV;
 492                        goto fail1;
 493                }
 494        }
 495
 496        return 0;
 497
 498fail1:
 499        pci_unregister_driver(&i3200_driver);
 500
 501fail0:
 502        if (mci_pdev)
 503                pci_dev_put(mci_pdev);
 504
 505        return pci_rc;
 506}
 507
 508static void __exit i3200_exit(void)
 509{
 510        debugf3("MC: %s()\n", __func__);
 511
 512        pci_unregister_driver(&i3200_driver);
 513        if (!i3200_registered) {
 514                i3200_remove_one(mci_pdev);
 515                pci_dev_put(mci_pdev);
 516        }
 517}
 518
 519module_init(i3200_init);
 520module_exit(i3200_exit);
 521
 522MODULE_LICENSE("GPL");
 523MODULE_AUTHOR("Akamai Technologies, Inc.");
 524MODULE_DESCRIPTION("MC support for Intel 3200 memory hub controllers");
 525
 526module_param(edac_op_state, int, 0444);
 527MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
 528