linux/drivers/edac/ie31200_edac.c
<<
>>
Prefs
   1/*
   2 * Intel E3-1200
   3 * Copyright (C) 2014 Jason Baron <jbaron@akamai.com>
   4 *
   5 * Support for the E3-1200 processor family. Heavily based on previous
   6 * Intel EDAC drivers.
   7 *
   8 * Since the DRAM controller is on the cpu chip, we can use its PCI device
   9 * id to identify these processors.
  10 *
  11 * PCI DRAM controller device ids (Taken from The PCI ID Repository - http://pci-ids.ucw.cz/)
  12 *
  13 * 0108: Xeon E3-1200 Processor Family DRAM Controller
  14 * 010c: Xeon E3-1200/2nd Generation Core Processor Family DRAM Controller
  15 * 0150: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
  16 * 0158: Xeon E3-1200 v2/Ivy Bridge DRAM Controller
  17 * 015c: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
  18 * 0c04: Xeon E3-1200 v3/4th Gen Core Processor DRAM Controller
  19 * 0c08: Xeon E3-1200 v3 Processor DRAM Controller
  20 *
  21 * Based on Intel specification:
  22 * http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e3-1200v3-vol-2-datasheet.pdf
  23 * http://www.intel.com/content/www/us/en/processors/xeon/xeon-e3-1200-family-vol-2-datasheet.html
  24 *
  25 * According to the above datasheet (p.16):
  26 * "
  27 * 6. Software must not access B0/D0/F0 32-bit memory-mapped registers with
  28 * requests that cross a DW boundary.
  29 * "
  30 *
  31 * Thus, we make use of the explicit: lo_hi_readq(), which breaks the readq into
  32 * 2 readl() calls. This restriction may be lifted in subsequent chip releases,
  33 * but lo_hi_readq() ensures that we are safe across all e3-1200 processors.
  34 */
  35
  36#include <linux/module.h>
  37#include <linux/init.h>
  38#include <linux/pci.h>
  39#include <linux/pci_ids.h>
  40#include <linux/edac.h>
  41
  42#include <asm-generic/io-64-nonatomic-lo-hi.h>
  43#include "edac_core.h"
  44
  45#define IE31200_REVISION "1.0"
  46#define EDAC_MOD_STR "ie31200_edac"
  47
  48#define ie31200_printk(level, fmt, arg...) \
  49        edac_printk(level, "ie31200", fmt, ##arg)
  50
  51#define PCI_DEVICE_ID_INTEL_IE31200_HB_1 0x0108
  52#define PCI_DEVICE_ID_INTEL_IE31200_HB_2 0x010c
  53#define PCI_DEVICE_ID_INTEL_IE31200_HB_3 0x0150
  54#define PCI_DEVICE_ID_INTEL_IE31200_HB_4 0x0158
  55#define PCI_DEVICE_ID_INTEL_IE31200_HB_5 0x015c
  56#define PCI_DEVICE_ID_INTEL_IE31200_HB_6 0x0c04
  57#define PCI_DEVICE_ID_INTEL_IE31200_HB_7 0x0c08
  58
  59#define IE31200_DIMMS                   4
  60#define IE31200_RANKS                   8
  61#define IE31200_RANKS_PER_CHANNEL       4
  62#define IE31200_DIMMS_PER_CHANNEL       2
  63#define IE31200_CHANNELS                2
  64
  65/* Intel IE31200 register addresses - device 0 function 0 - DRAM Controller */
  66#define IE31200_MCHBAR_LOW              0x48
  67#define IE31200_MCHBAR_HIGH             0x4c
  68#define IE31200_MCHBAR_MASK             GENMASK_ULL(38, 15)
  69#define IE31200_MMR_WINDOW_SIZE         BIT(15)
  70
  71/*
  72 * Error Status Register (16b)
  73 *
  74 * 15    reserved
  75 * 14    Isochronous TBWRR Run Behind FIFO Full
  76 *       (ITCV)
  77 * 13    Isochronous TBWRR Run Behind FIFO Put
  78 *       (ITSTV)
  79 * 12    reserved
  80 * 11    MCH Thermal Sensor Event
  81 *       for SMI/SCI/SERR (GTSE)
  82 * 10    reserved
  83 *  9    LOCK to non-DRAM Memory Flag (LCKF)
  84 *  8    reserved
  85 *  7    DRAM Throttle Flag (DTF)
  86 *  6:2  reserved
  87 *  1    Multi-bit DRAM ECC Error Flag (DMERR)
  88 *  0    Single-bit DRAM ECC Error Flag (DSERR)
  89 */
  90#define IE31200_ERRSTS                  0xc8
  91#define IE31200_ERRSTS_UE               BIT(1)
  92#define IE31200_ERRSTS_CE               BIT(0)
  93#define IE31200_ERRSTS_BITS             (IE31200_ERRSTS_UE | IE31200_ERRSTS_CE)
  94
  95/*
  96 * Channel 0 ECC Error Log (64b)
  97 *
  98 * 63:48 Error Column Address (ERRCOL)
  99 * 47:32 Error Row Address (ERRROW)
 100 * 31:29 Error Bank Address (ERRBANK)
 101 * 28:27 Error Rank Address (ERRRANK)
 102 * 26:24 reserved
 103 * 23:16 Error Syndrome (ERRSYND)
 104 * 15: 2 reserved
 105 *    1  Multiple Bit Error Status (MERRSTS)
 106 *    0  Correctable Error Status (CERRSTS)
 107 */
 108#define IE31200_C0ECCERRLOG                     0x40c8
 109#define IE31200_C1ECCERRLOG                     0x44c8
 110#define IE31200_ECCERRLOG_CE                    BIT(0)
 111#define IE31200_ECCERRLOG_UE                    BIT(1)
 112#define IE31200_ECCERRLOG_RANK_BITS             GENMASK_ULL(28, 27)
 113#define IE31200_ECCERRLOG_RANK_SHIFT            27
 114#define IE31200_ECCERRLOG_SYNDROME_BITS         GENMASK_ULL(23, 16)
 115#define IE31200_ECCERRLOG_SYNDROME_SHIFT        16
 116
 117#define IE31200_ECCERRLOG_SYNDROME(log)            \
 118        ((log & IE31200_ECCERRLOG_SYNDROME_BITS) >> \
 119         IE31200_ECCERRLOG_SYNDROME_SHIFT)
 120
 121#define IE31200_CAPID0                  0xe4
 122#define IE31200_CAPID0_PDCD             BIT(4)
 123#define IE31200_CAPID0_DDPCD            BIT(6)
 124#define IE31200_CAPID0_ECC              BIT(1)
 125
 126#define IE31200_MAD_DIMM_0_OFFSET       0x5004
 127#define IE31200_MAD_DIMM_SIZE           GENMASK_ULL(7, 0)
 128#define IE31200_MAD_DIMM_A_RANK         BIT(17)
 129#define IE31200_MAD_DIMM_A_WIDTH        BIT(19)
 130
 131#define IE31200_PAGES(n)                (n << (28 - PAGE_SHIFT))
 132
 133static int nr_channels;
 134
 135struct ie31200_priv {
 136        void __iomem *window;
 137};
 138
 139enum ie31200_chips {
 140        IE31200 = 0,
 141};
 142
 143struct ie31200_dev_info {
 144        const char *ctl_name;
 145};
 146
 147struct ie31200_error_info {
 148        u16 errsts;
 149        u16 errsts2;
 150        u64 eccerrlog[IE31200_CHANNELS];
 151};
 152
 153static const struct ie31200_dev_info ie31200_devs[] = {
 154        [IE31200] = {
 155                .ctl_name = "IE31200"
 156        },
 157};
 158
 159struct dimm_data {
 160        u8 size; /* in 256MB multiples */
 161        u8 dual_rank : 1,
 162           x16_width : 1; /* 0 means x8 width */
 163};
 164
 165static int how_many_channels(struct pci_dev *pdev)
 166{
 167        int n_channels;
 168        unsigned char capid0_2b; /* 2nd byte of CAPID0 */
 169
 170        pci_read_config_byte(pdev, IE31200_CAPID0 + 1, &capid0_2b);
 171
 172        /* check PDCD: Dual Channel Disable */
 173        if (capid0_2b & IE31200_CAPID0_PDCD) {
 174                edac_dbg(0, "In single channel mode\n");
 175                n_channels = 1;
 176        } else {
 177                edac_dbg(0, "In dual channel mode\n");
 178                n_channels = 2;
 179        }
 180
 181        /* check DDPCD - check if both channels are filled */
 182        if (capid0_2b & IE31200_CAPID0_DDPCD)
 183                edac_dbg(0, "2 DIMMS per channel disabled\n");
 184        else
 185                edac_dbg(0, "2 DIMMS per channel enabled\n");
 186
 187        return n_channels;
 188}
 189
 190static bool ecc_capable(struct pci_dev *pdev)
 191{
 192        unsigned char capid0_4b; /* 4th byte of CAPID0 */
 193
 194        pci_read_config_byte(pdev, IE31200_CAPID0 + 3, &capid0_4b);
 195        if (capid0_4b & IE31200_CAPID0_ECC)
 196                return false;
 197        return true;
 198}
 199
 200static int eccerrlog_row(int channel, u64 log)
 201{
 202        int rank = ((log & IE31200_ECCERRLOG_RANK_BITS) >>
 203                IE31200_ECCERRLOG_RANK_SHIFT);
 204        return rank | (channel * IE31200_RANKS_PER_CHANNEL);
 205}
 206
 207static void ie31200_clear_error_info(struct mem_ctl_info *mci)
 208{
 209        /*
 210         * Clear any error bits.
 211         * (Yes, we really clear bits by writing 1 to them.)
 212         */
 213        pci_write_bits16(to_pci_dev(mci->pdev), IE31200_ERRSTS,
 214                         IE31200_ERRSTS_BITS, IE31200_ERRSTS_BITS);
 215}
 216
 217static void ie31200_get_and_clear_error_info(struct mem_ctl_info *mci,
 218                                             struct ie31200_error_info *info)
 219{
 220        struct pci_dev *pdev;
 221        struct ie31200_priv *priv = mci->pvt_info;
 222        void __iomem *window = priv->window;
 223
 224        pdev = to_pci_dev(mci->pdev);
 225
 226        /*
 227         * This is a mess because there is no atomic way to read all the
 228         * registers at once and the registers can transition from CE being
 229         * overwritten by UE.
 230         */
 231        pci_read_config_word(pdev, IE31200_ERRSTS, &info->errsts);
 232        if (!(info->errsts & IE31200_ERRSTS_BITS))
 233                return;
 234
 235        info->eccerrlog[0] = lo_hi_readq(window + IE31200_C0ECCERRLOG);
 236        if (nr_channels == 2)
 237                info->eccerrlog[1] = lo_hi_readq(window + IE31200_C1ECCERRLOG);
 238
 239        pci_read_config_word(pdev, IE31200_ERRSTS, &info->errsts2);
 240
 241        /*
 242         * If the error is the same for both reads then the first set
 243         * of reads is valid.  If there is a change then there is a CE
 244         * with no info and the second set of reads is valid and
 245         * should be UE info.
 246         */
 247        if ((info->errsts ^ info->errsts2) & IE31200_ERRSTS_BITS) {
 248                info->eccerrlog[0] = lo_hi_readq(window + IE31200_C0ECCERRLOG);
 249                if (nr_channels == 2)
 250                        info->eccerrlog[1] =
 251                                lo_hi_readq(window + IE31200_C1ECCERRLOG);
 252        }
 253
 254        ie31200_clear_error_info(mci);
 255}
 256
 257static void ie31200_process_error_info(struct mem_ctl_info *mci,
 258                                       struct ie31200_error_info *info)
 259{
 260        int channel;
 261        u64 log;
 262
 263        if (!(info->errsts & IE31200_ERRSTS_BITS))
 264                return;
 265
 266        if ((info->errsts ^ info->errsts2) & IE31200_ERRSTS_BITS) {
 267                edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
 268                                     -1, -1, -1, "UE overwrote CE", "");
 269                info->errsts = info->errsts2;
 270        }
 271
 272        for (channel = 0; channel < nr_channels; channel++) {
 273                log = info->eccerrlog[channel];
 274                if (log & IE31200_ECCERRLOG_UE) {
 275                        edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
 276                                             0, 0, 0,
 277                                             eccerrlog_row(channel, log),
 278                                             channel, -1,
 279                                             "ie31200 UE", "");
 280                } else if (log & IE31200_ECCERRLOG_CE) {
 281                        edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
 282                                             0, 0,
 283                                             IE31200_ECCERRLOG_SYNDROME(log),
 284                                             eccerrlog_row(channel, log),
 285                                             channel, -1,
 286                                             "ie31200 CE", "");
 287                }
 288        }
 289}
 290
 291static void ie31200_check(struct mem_ctl_info *mci)
 292{
 293        struct ie31200_error_info info;
 294
 295        edac_dbg(1, "MC%d\n", mci->mc_idx);
 296        ie31200_get_and_clear_error_info(mci, &info);
 297        ie31200_process_error_info(mci, &info);
 298}
 299
 300static void __iomem *ie31200_map_mchbar(struct pci_dev *pdev)
 301{
 302        union {
 303                u64 mchbar;
 304                struct {
 305                        u32 mchbar_low;
 306                        u32 mchbar_high;
 307                };
 308        } u;
 309        void __iomem *window;
 310
 311        pci_read_config_dword(pdev, IE31200_MCHBAR_LOW, &u.mchbar_low);
 312        pci_read_config_dword(pdev, IE31200_MCHBAR_HIGH, &u.mchbar_high);
 313        u.mchbar &= IE31200_MCHBAR_MASK;
 314
 315        if (u.mchbar != (resource_size_t)u.mchbar) {
 316                ie31200_printk(KERN_ERR, "mmio space beyond accessible range (0x%llx)\n",
 317                               (unsigned long long)u.mchbar);
 318                return NULL;
 319        }
 320
 321        window = ioremap_nocache(u.mchbar, IE31200_MMR_WINDOW_SIZE);
 322        if (!window)
 323                ie31200_printk(KERN_ERR, "Cannot map mmio space at 0x%llx\n",
 324                               (unsigned long long)u.mchbar);
 325
 326        return window;
 327}
 328
 329static int ie31200_probe1(struct pci_dev *pdev, int dev_idx)
 330{
 331        int i, j, ret;
 332        struct mem_ctl_info *mci = NULL;
 333        struct edac_mc_layer layers[2];
 334        struct dimm_data dimm_info[IE31200_CHANNELS][IE31200_DIMMS_PER_CHANNEL];
 335        void __iomem *window;
 336        struct ie31200_priv *priv;
 337        u32 addr_decode;
 338
 339        edac_dbg(0, "MC:\n");
 340
 341        if (!ecc_capable(pdev)) {
 342                ie31200_printk(KERN_INFO, "No ECC support\n");
 343                return -ENODEV;
 344        }
 345
 346        nr_channels = how_many_channels(pdev);
 347        layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 348        layers[0].size = IE31200_DIMMS;
 349        layers[0].is_virt_csrow = true;
 350        layers[1].type = EDAC_MC_LAYER_CHANNEL;
 351        layers[1].size = nr_channels;
 352        layers[1].is_virt_csrow = false;
 353        mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
 354                            sizeof(struct ie31200_priv));
 355        if (!mci)
 356                return -ENOMEM;
 357
 358        window = ie31200_map_mchbar(pdev);
 359        if (!window) {
 360                ret = -ENODEV;
 361                goto fail_free;
 362        }
 363
 364        edac_dbg(3, "MC: init mci\n");
 365        mci->pdev = &pdev->dev;
 366        mci->mtype_cap = MEM_FLAG_DDR3;
 367        mci->edac_ctl_cap = EDAC_FLAG_SECDED;
 368        mci->edac_cap = EDAC_FLAG_SECDED;
 369        mci->mod_name = EDAC_MOD_STR;
 370        mci->mod_ver = IE31200_REVISION;
 371        mci->ctl_name = ie31200_devs[dev_idx].ctl_name;
 372        mci->dev_name = pci_name(pdev);
 373        mci->edac_check = ie31200_check;
 374        mci->ctl_page_to_phys = NULL;
 375        priv = mci->pvt_info;
 376        priv->window = window;
 377
 378        /* populate DIMM info */
 379        for (i = 0; i < IE31200_CHANNELS; i++) {
 380                addr_decode = readl(window + IE31200_MAD_DIMM_0_OFFSET +
 381                                        (i * 4));
 382                edac_dbg(0, "addr_decode: 0x%x\n", addr_decode);
 383                for (j = 0; j < IE31200_DIMMS_PER_CHANNEL; j++) {
 384                        dimm_info[i][j].size = (addr_decode >> (j * 8)) &
 385                                                IE31200_MAD_DIMM_SIZE;
 386                        dimm_info[i][j].dual_rank = (addr_decode &
 387                                (IE31200_MAD_DIMM_A_RANK << j)) ? 1 : 0;
 388                        dimm_info[i][j].x16_width = (addr_decode &
 389                                (IE31200_MAD_DIMM_A_WIDTH << j)) ? 1 : 0;
 390                        edac_dbg(0, "size: 0x%x, rank: %d, width: %d\n",
 391                                 dimm_info[i][j].size,
 392                                 dimm_info[i][j].dual_rank,
 393                                 dimm_info[i][j].x16_width);
 394                }
 395        }
 396
 397        /*
 398         * The dram rank boundary (DRB) reg values are boundary addresses
 399         * for each DRAM rank with a granularity of 64MB.  DRB regs are
 400         * cumulative; the last one will contain the total memory
 401         * contained in all ranks.
 402         */
 403        for (i = 0; i < IE31200_DIMMS_PER_CHANNEL; i++) {
 404                for (j = 0; j < IE31200_CHANNELS; j++) {
 405                        struct dimm_info *dimm;
 406                        unsigned long nr_pages;
 407
 408                        nr_pages = IE31200_PAGES(dimm_info[j][i].size);
 409                        if (nr_pages == 0)
 410                                continue;
 411
 412                        if (dimm_info[j][i].dual_rank) {
 413                                nr_pages = nr_pages / 2;
 414                                dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
 415                                                     mci->n_layers, (i * 2) + 1,
 416                                                     j, 0);
 417                                dimm->nr_pages = nr_pages;
 418                                edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages);
 419                                dimm->grain = 8; /* just a guess */
 420                                dimm->mtype = MEM_DDR3;
 421                                dimm->dtype = DEV_UNKNOWN;
 422                                dimm->edac_mode = EDAC_UNKNOWN;
 423                        }
 424                        dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
 425                                             mci->n_layers, i * 2, j, 0);
 426                        dimm->nr_pages = nr_pages;
 427                        edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages);
 428                        dimm->grain = 8; /* same guess */
 429                        dimm->mtype = MEM_DDR3;
 430                        dimm->dtype = DEV_UNKNOWN;
 431                        dimm->edac_mode = EDAC_UNKNOWN;
 432                }
 433        }
 434
 435        ie31200_clear_error_info(mci);
 436
 437        if (edac_mc_add_mc(mci)) {
 438                edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
 439                ret = -ENODEV;
 440                goto fail_unmap;
 441        }
 442
 443        /* get this far and it's successful */
 444        edac_dbg(3, "MC: success\n");
 445        return 0;
 446
 447fail_unmap:
 448        iounmap(window);
 449
 450fail_free:
 451        edac_mc_free(mci);
 452
 453        return ret;
 454}
 455
 456static int ie31200_init_one(struct pci_dev *pdev,
 457                            const struct pci_device_id *ent)
 458{
 459        edac_dbg(0, "MC:\n");
 460
 461        if (pci_enable_device(pdev) < 0)
 462                return -EIO;
 463
 464        return ie31200_probe1(pdev, ent->driver_data);
 465}
 466
 467static void ie31200_remove_one(struct pci_dev *pdev)
 468{
 469        struct mem_ctl_info *mci;
 470        struct ie31200_priv *priv;
 471
 472        edac_dbg(0, "\n");
 473        mci = edac_mc_del_mc(&pdev->dev);
 474        if (!mci)
 475                return;
 476        priv = mci->pvt_info;
 477        iounmap(priv->window);
 478        edac_mc_free(mci);
 479}
 480
 481static const struct pci_device_id ie31200_pci_tbl[] = {
 482        {
 483                PCI_VEND_DEV(INTEL, IE31200_HB_1), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 484                IE31200},
 485        {
 486                PCI_VEND_DEV(INTEL, IE31200_HB_2), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 487                IE31200},
 488        {
 489                PCI_VEND_DEV(INTEL, IE31200_HB_3), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 490                IE31200},
 491        {
 492                PCI_VEND_DEV(INTEL, IE31200_HB_4), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 493                IE31200},
 494        {
 495                PCI_VEND_DEV(INTEL, IE31200_HB_5), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 496                IE31200},
 497        {
 498                PCI_VEND_DEV(INTEL, IE31200_HB_6), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 499                IE31200},
 500        {
 501                PCI_VEND_DEV(INTEL, IE31200_HB_7), PCI_ANY_ID, PCI_ANY_ID, 0, 0,
 502                IE31200},
 503        {
 504                0,
 505        }            /* 0 terminated list. */
 506};
 507MODULE_DEVICE_TABLE(pci, ie31200_pci_tbl);
 508
 509static struct pci_driver ie31200_driver = {
 510        .name = EDAC_MOD_STR,
 511        .probe = ie31200_init_one,
 512        .remove = ie31200_remove_one,
 513        .id_table = ie31200_pci_tbl,
 514};
 515
 516static int __init ie31200_init(void)
 517{
 518        edac_dbg(3, "MC:\n");
 519        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
 520        opstate_init();
 521
 522        return pci_register_driver(&ie31200_driver);
 523}
 524
 525static void __exit ie31200_exit(void)
 526{
 527        edac_dbg(3, "MC:\n");
 528        pci_unregister_driver(&ie31200_driver);
 529}
 530
 531module_init(ie31200_init);
 532module_exit(ie31200_exit);
 533
 534MODULE_LICENSE("GPL");
 535MODULE_AUTHOR("Jason Baron <jbaron@akamai.com>");
 536MODULE_DESCRIPTION("MC support for Intel Processor E31200 memory hub controllers");
 537