linux/drivers/edac/i7core_edac.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/* Intel i7 core/Nehalem Memory Controller kernel module
   3 *
   4 * This driver supports the memory controllers found on the Intel
   5 * processor families i7core, i7core 7xx/8xx, i5core, Xeon 35xx,
   6 * Xeon 55xx and Xeon 56xx also known as Nehalem, Nehalem-EP, Lynnfield
   7 * and Westmere-EP.
   8 *
   9 * Copyright (c) 2009-2010 by:
  10 *       Mauro Carvalho Chehab
  11 *
  12 * Red Hat Inc. http://www.redhat.com
  13 *
  14 * Forked and adapted from the i5400_edac driver
  15 *
  16 * Based on the following public Intel datasheets:
  17 * Intel Core i7 Processor Extreme Edition and Intel Core i7 Processor
  18 * Datasheet, Volume 2:
  19 *      http://download.intel.com/design/processor/datashts/320835.pdf
  20 * Intel Xeon Processor 5500 Series Datasheet Volume 2
  21 *      http://www.intel.com/Assets/PDF/datasheet/321322.pdf
  22 * also available at:
  23 *      http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
  24 */
  25
  26#include <linux/module.h>
  27#include <linux/init.h>
  28#include <linux/pci.h>
  29#include <linux/pci_ids.h>
  30#include <linux/slab.h>
  31#include <linux/delay.h>
  32#include <linux/dmi.h>
  33#include <linux/edac.h>
  34#include <linux/mmzone.h>
  35#include <linux/smp.h>
  36#include <asm/mce.h>
  37#include <asm/processor.h>
  38#include <asm/div64.h>
  39
  40#include "edac_module.h"
  41
  42/* Static vars */
  43static LIST_HEAD(i7core_edac_list);
  44static DEFINE_MUTEX(i7core_edac_lock);
  45static int probed;
  46
  47static int use_pci_fixup;
  48module_param(use_pci_fixup, int, 0444);
  49MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices");
  50/*
  51 * This is used for Nehalem-EP and Nehalem-EX devices, where the non-core
  52 * registers start at bus 255, and are not reported by BIOS.
  53 * We currently find devices with only 2 sockets. In order to support more QPI
  54 * Quick Path Interconnect, just increment this number.
  55 */
  56#define MAX_SOCKET_BUSES        2
  57
  58
  59/*
  60 * Alter this version for the module when modifications are made
  61 */
  62#define I7CORE_REVISION    " Ver: 1.0.0"
  63#define EDAC_MOD_STR      "i7core_edac"
  64
  65/*
  66 * Debug macros
  67 */
  68#define i7core_printk(level, fmt, arg...)                       \
  69        edac_printk(level, "i7core", fmt, ##arg)
  70
  71#define i7core_mc_printk(mci, level, fmt, arg...)               \
  72        edac_mc_chipset_printk(mci, level, "i7core", fmt, ##arg)
  73
  74/*
  75 * i7core Memory Controller Registers
  76 */
  77
  78        /* OFFSETS for Device 0 Function 0 */
  79
  80#define MC_CFG_CONTROL  0x90
  81  #define MC_CFG_UNLOCK         0x02
  82  #define MC_CFG_LOCK           0x00
  83
  84        /* OFFSETS for Device 3 Function 0 */
  85
  86#define MC_CONTROL      0x48
  87#define MC_STATUS       0x4c
  88#define MC_MAX_DOD      0x64
  89
  90/*
  91 * OFFSETS for Device 3 Function 4, as indicated on Xeon 5500 datasheet:
  92 * http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
  93 */
  94
  95#define MC_TEST_ERR_RCV1        0x60
  96  #define DIMM2_COR_ERR(r)                      ((r) & 0x7fff)
  97
  98#define MC_TEST_ERR_RCV0        0x64
  99  #define DIMM1_COR_ERR(r)                      (((r) >> 16) & 0x7fff)
 100  #define DIMM0_COR_ERR(r)                      ((r) & 0x7fff)
 101
 102/* OFFSETS for Device 3 Function 2, as indicated on Xeon 5500 datasheet */
 103#define MC_SSRCONTROL           0x48
 104  #define SSR_MODE_DISABLE      0x00
 105  #define SSR_MODE_ENABLE       0x01
 106  #define SSR_MODE_MASK         0x03
 107
 108#define MC_SCRUB_CONTROL        0x4c
 109  #define STARTSCRUB            (1 << 24)
 110  #define SCRUBINTERVAL_MASK    0xffffff
 111
 112#define MC_COR_ECC_CNT_0        0x80
 113#define MC_COR_ECC_CNT_1        0x84
 114#define MC_COR_ECC_CNT_2        0x88
 115#define MC_COR_ECC_CNT_3        0x8c
 116#define MC_COR_ECC_CNT_4        0x90
 117#define MC_COR_ECC_CNT_5        0x94
 118
 119#define DIMM_TOP_COR_ERR(r)                     (((r) >> 16) & 0x7fff)
 120#define DIMM_BOT_COR_ERR(r)                     ((r) & 0x7fff)
 121
 122
 123        /* OFFSETS for Devices 4,5 and 6 Function 0 */
 124
 125#define MC_CHANNEL_DIMM_INIT_PARAMS 0x58
 126  #define THREE_DIMMS_PRESENT           (1 << 24)
 127  #define SINGLE_QUAD_RANK_PRESENT      (1 << 23)
 128  #define QUAD_RANK_PRESENT             (1 << 22)
 129  #define REGISTERED_DIMM               (1 << 15)
 130
 131#define MC_CHANNEL_MAPPER       0x60
 132  #define RDLCH(r, ch)          ((((r) >> (3 + (ch * 6))) & 0x07) - 1)
 133  #define WRLCH(r, ch)          ((((r) >> (ch * 6)) & 0x07) - 1)
 134
 135#define MC_CHANNEL_RANK_PRESENT 0x7c
 136  #define RANK_PRESENT_MASK             0xffff
 137
 138#define MC_CHANNEL_ADDR_MATCH   0xf0
 139#define MC_CHANNEL_ERROR_MASK   0xf8
 140#define MC_CHANNEL_ERROR_INJECT 0xfc
 141  #define INJECT_ADDR_PARITY    0x10
 142  #define INJECT_ECC            0x08
 143  #define MASK_CACHELINE        0x06
 144  #define MASK_FULL_CACHELINE   0x06
 145  #define MASK_MSB32_CACHELINE  0x04
 146  #define MASK_LSB32_CACHELINE  0x02
 147  #define NO_MASK_CACHELINE     0x00
 148  #define REPEAT_EN             0x01
 149
 150        /* OFFSETS for Devices 4,5 and 6 Function 1 */
 151
 152#define MC_DOD_CH_DIMM0         0x48
 153#define MC_DOD_CH_DIMM1         0x4c
 154#define MC_DOD_CH_DIMM2         0x50
 155  #define RANKOFFSET_MASK       ((1 << 12) | (1 << 11) | (1 << 10))
 156  #define RANKOFFSET(x)         ((x & RANKOFFSET_MASK) >> 10)
 157  #define DIMM_PRESENT_MASK     (1 << 9)
 158  #define DIMM_PRESENT(x)       (((x) & DIMM_PRESENT_MASK) >> 9)
 159  #define MC_DOD_NUMBANK_MASK           ((1 << 8) | (1 << 7))
 160  #define MC_DOD_NUMBANK(x)             (((x) & MC_DOD_NUMBANK_MASK) >> 7)
 161  #define MC_DOD_NUMRANK_MASK           ((1 << 6) | (1 << 5))
 162  #define MC_DOD_NUMRANK(x)             (((x) & MC_DOD_NUMRANK_MASK) >> 5)
 163  #define MC_DOD_NUMROW_MASK            ((1 << 4) | (1 << 3) | (1 << 2))
 164  #define MC_DOD_NUMROW(x)              (((x) & MC_DOD_NUMROW_MASK) >> 2)
 165  #define MC_DOD_NUMCOL_MASK            3
 166  #define MC_DOD_NUMCOL(x)              ((x) & MC_DOD_NUMCOL_MASK)
 167
 168#define MC_RANK_PRESENT         0x7c
 169
 170#define MC_SAG_CH_0     0x80
 171#define MC_SAG_CH_1     0x84
 172#define MC_SAG_CH_2     0x88
 173#define MC_SAG_CH_3     0x8c
 174#define MC_SAG_CH_4     0x90
 175#define MC_SAG_CH_5     0x94
 176#define MC_SAG_CH_6     0x98
 177#define MC_SAG_CH_7     0x9c
 178
 179#define MC_RIR_LIMIT_CH_0       0x40
 180#define MC_RIR_LIMIT_CH_1       0x44
 181#define MC_RIR_LIMIT_CH_2       0x48
 182#define MC_RIR_LIMIT_CH_3       0x4C
 183#define MC_RIR_LIMIT_CH_4       0x50
 184#define MC_RIR_LIMIT_CH_5       0x54
 185#define MC_RIR_LIMIT_CH_6       0x58
 186#define MC_RIR_LIMIT_CH_7       0x5C
 187#define MC_RIR_LIMIT_MASK       ((1 << 10) - 1)
 188
 189#define MC_RIR_WAY_CH           0x80
 190  #define MC_RIR_WAY_OFFSET_MASK        (((1 << 14) - 1) & ~0x7)
 191  #define MC_RIR_WAY_RANK_MASK          0x7
 192
 193/*
 194 * i7core structs
 195 */
 196
 197#define NUM_CHANS 3
 198#define MAX_DIMMS 3             /* Max DIMMS per channel */
 199#define MAX_MCR_FUNC  4
 200#define MAX_CHAN_FUNC 3
 201
 202struct i7core_info {
 203        u32     mc_control;
 204        u32     mc_status;
 205        u32     max_dod;
 206        u32     ch_map;
 207};
 208
 209
 210struct i7core_inject {
 211        int     enable;
 212
 213        u32     section;
 214        u32     type;
 215        u32     eccmask;
 216
 217        /* Error address mask */
 218        int channel, dimm, rank, bank, page, col;
 219};
 220
 221struct i7core_channel {
 222        bool            is_3dimms_present;
 223        bool            is_single_4rank;
 224        bool            has_4rank;
 225        u32             dimms;
 226};
 227
 228struct pci_id_descr {
 229        int                     dev;
 230        int                     func;
 231        int                     dev_id;
 232        int                     optional;
 233};
 234
 235struct pci_id_table {
 236        const struct pci_id_descr       *descr;
 237        int                             n_devs;
 238};
 239
 240struct i7core_dev {
 241        struct list_head        list;
 242        u8                      socket;
 243        struct pci_dev          **pdev;
 244        int                     n_devs;
 245        struct mem_ctl_info     *mci;
 246};
 247
 248struct i7core_pvt {
 249        struct device *addrmatch_dev, *chancounts_dev;
 250
 251        struct pci_dev  *pci_noncore;
 252        struct pci_dev  *pci_mcr[MAX_MCR_FUNC + 1];
 253        struct pci_dev  *pci_ch[NUM_CHANS][MAX_CHAN_FUNC + 1];
 254
 255        struct i7core_dev *i7core_dev;
 256
 257        struct i7core_info      info;
 258        struct i7core_inject    inject;
 259        struct i7core_channel   channel[NUM_CHANS];
 260
 261        int             ce_count_available;
 262
 263                        /* ECC corrected errors counts per udimm */
 264        unsigned long   udimm_ce_count[MAX_DIMMS];
 265        int             udimm_last_ce_count[MAX_DIMMS];
 266                        /* ECC corrected errors counts per rdimm */
 267        unsigned long   rdimm_ce_count[NUM_CHANS][MAX_DIMMS];
 268        int             rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS];
 269
 270        bool            is_registered, enable_scrub;
 271
 272        /* DCLK Frequency used for computing scrub rate */
 273        int                     dclk_freq;
 274
 275        /* Struct to control EDAC polling */
 276        struct edac_pci_ctl_info *i7core_pci;
 277};
 278
 279#define PCI_DESCR(device, function, device_id)  \
 280        .dev = (device),                        \
 281        .func = (function),                     \
 282        .dev_id = (device_id)
 283
 284static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = {
 285                /* Memory controller */
 286        { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR)     },
 287        { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD)  },
 288                        /* Exists only for RDIMM */
 289        { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1  },
 290        { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) },
 291
 292                /* Channel 0 */
 293        { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH0_CTRL) },
 294        { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH0_ADDR) },
 295        { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH0_RANK) },
 296        { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH0_TC)   },
 297
 298                /* Channel 1 */
 299        { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH1_CTRL) },
 300        { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH1_ADDR) },
 301        { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH1_RANK) },
 302        { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH1_TC)   },
 303
 304                /* Channel 2 */
 305        { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_I7_MC_CH2_CTRL) },
 306        { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) },
 307        { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) },
 308        { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC)   },
 309
 310                /* Generic Non-core registers */
 311        /*
 312         * This is the PCI device on i7core and on Xeon 35xx (8086:2c41)
 313         * On Xeon 55xx, however, it has a different id (8086:2c40). So,
 314         * the probing code needs to test for the other address in case of
 315         * failure of this one
 316         */
 317        { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE)  },
 318
 319};
 320
 321static const struct pci_id_descr pci_dev_descr_lynnfield[] = {
 322        { PCI_DESCR( 3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR)         },
 323        { PCI_DESCR( 3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD)      },
 324        { PCI_DESCR( 3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST)     },
 325
 326        { PCI_DESCR( 4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL) },
 327        { PCI_DESCR( 4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR) },
 328        { PCI_DESCR( 4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK) },
 329        { PCI_DESCR( 4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC)   },
 330
 331        { PCI_DESCR( 5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL) },
 332        { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) },
 333        { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) },
 334        { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC)   },
 335
 336        /*
 337         * This is the PCI device has an alternate address on some
 338         * processors like Core i7 860
 339         */
 340        { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE)     },
 341};
 342
 343static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = {
 344                /* Memory controller */
 345        { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MCR_REV2)     },
 346        { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TAD_REV2)  },
 347                        /* Exists only for RDIMM */
 348        { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_RAS_REV2), .optional = 1  },
 349        { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_TEST_REV2) },
 350
 351                /* Channel 0 */
 352        { PCI_DESCR(4, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_CTRL_REV2) },
 353        { PCI_DESCR(4, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_ADDR_REV2) },
 354        { PCI_DESCR(4, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_RANK_REV2) },
 355        { PCI_DESCR(4, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH0_TC_REV2)   },
 356
 357                /* Channel 1 */
 358        { PCI_DESCR(5, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_CTRL_REV2) },
 359        { PCI_DESCR(5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR_REV2) },
 360        { PCI_DESCR(5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK_REV2) },
 361        { PCI_DESCR(5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC_REV2)   },
 362
 363                /* Channel 2 */
 364        { PCI_DESCR(6, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_CTRL_REV2) },
 365        { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) },
 366        { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) },
 367        { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2)   },
 368
 369                /* Generic Non-core registers */
 370        { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2)  },
 371
 372};
 373
 374#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) }
 375static const struct pci_id_table pci_dev_table[] = {
 376        PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_nehalem),
 377        PCI_ID_TABLE_ENTRY(pci_dev_descr_lynnfield),
 378        PCI_ID_TABLE_ENTRY(pci_dev_descr_i7core_westmere),
 379        {0,}                    /* 0 terminated list. */
 380};
 381
 382/*
 383 *      pci_device_id   table for which devices we are looking for
 384 */
 385static const struct pci_device_id i7core_pci_tbl[] = {
 386        {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_X58_HUB_MGMT)},
 387        {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LYNNFIELD_QPI_LINK0)},
 388        {0,}                    /* 0 terminated list. */
 389};
 390
 391/****************************************************************************
 392                        Ancillary status routines
 393 ****************************************************************************/
 394
 395        /* MC_CONTROL bits */
 396#define CH_ACTIVE(pvt, ch)      ((pvt)->info.mc_control & (1 << (8 + ch)))
 397#define ECCx8(pvt)              ((pvt)->info.mc_control & (1 << 1))
 398
 399        /* MC_STATUS bits */
 400#define ECC_ENABLED(pvt)        ((pvt)->info.mc_status & (1 << 4))
 401#define CH_DISABLED(pvt, ch)    ((pvt)->info.mc_status & (1 << ch))
 402
 403        /* MC_MAX_DOD read functions */
 404static inline int numdimms(u32 dimms)
 405{
 406        return (dimms & 0x3) + 1;
 407}
 408
 409static inline int numrank(u32 rank)
 410{
 411        static const int ranks[] = { 1, 2, 4, -EINVAL };
 412
 413        return ranks[rank & 0x3];
 414}
 415
 416static inline int numbank(u32 bank)
 417{
 418        static const int banks[] = { 4, 8, 16, -EINVAL };
 419
 420        return banks[bank & 0x3];
 421}
 422
 423static inline int numrow(u32 row)
 424{
 425        static const int rows[] = {
 426                1 << 12, 1 << 13, 1 << 14, 1 << 15,
 427                1 << 16, -EINVAL, -EINVAL, -EINVAL,
 428        };
 429
 430        return rows[row & 0x7];
 431}
 432
 433static inline int numcol(u32 col)
 434{
 435        static const int cols[] = {
 436                1 << 10, 1 << 11, 1 << 12, -EINVAL,
 437        };
 438        return cols[col & 0x3];
 439}
 440
 441static struct i7core_dev *get_i7core_dev(u8 socket)
 442{
 443        struct i7core_dev *i7core_dev;
 444
 445        list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
 446                if (i7core_dev->socket == socket)
 447                        return i7core_dev;
 448        }
 449
 450        return NULL;
 451}
 452
 453static struct i7core_dev *alloc_i7core_dev(u8 socket,
 454                                           const struct pci_id_table *table)
 455{
 456        struct i7core_dev *i7core_dev;
 457
 458        i7core_dev = kzalloc(sizeof(*i7core_dev), GFP_KERNEL);
 459        if (!i7core_dev)
 460                return NULL;
 461
 462        i7core_dev->pdev = kcalloc(table->n_devs, sizeof(*i7core_dev->pdev),
 463                                   GFP_KERNEL);
 464        if (!i7core_dev->pdev) {
 465                kfree(i7core_dev);
 466                return NULL;
 467        }
 468
 469        i7core_dev->socket = socket;
 470        i7core_dev->n_devs = table->n_devs;
 471        list_add_tail(&i7core_dev->list, &i7core_edac_list);
 472
 473        return i7core_dev;
 474}
 475
 476static void free_i7core_dev(struct i7core_dev *i7core_dev)
 477{
 478        list_del(&i7core_dev->list);
 479        kfree(i7core_dev->pdev);
 480        kfree(i7core_dev);
 481}
 482
 483/****************************************************************************
 484                        Memory check routines
 485 ****************************************************************************/
 486
 487static int get_dimm_config(struct mem_ctl_info *mci)
 488{
 489        struct i7core_pvt *pvt = mci->pvt_info;
 490        struct pci_dev *pdev;
 491        int i, j;
 492        enum edac_type mode;
 493        enum mem_type mtype;
 494        struct dimm_info *dimm;
 495
 496        /* Get data from the MC register, function 0 */
 497        pdev = pvt->pci_mcr[0];
 498        if (!pdev)
 499                return -ENODEV;
 500
 501        /* Device 3 function 0 reads */
 502        pci_read_config_dword(pdev, MC_CONTROL, &pvt->info.mc_control);
 503        pci_read_config_dword(pdev, MC_STATUS, &pvt->info.mc_status);
 504        pci_read_config_dword(pdev, MC_MAX_DOD, &pvt->info.max_dod);
 505        pci_read_config_dword(pdev, MC_CHANNEL_MAPPER, &pvt->info.ch_map);
 506
 507        edac_dbg(0, "QPI %d control=0x%08x status=0x%08x dod=0x%08x map=0x%08x\n",
 508                 pvt->i7core_dev->socket, pvt->info.mc_control,
 509                 pvt->info.mc_status, pvt->info.max_dod, pvt->info.ch_map);
 510
 511        if (ECC_ENABLED(pvt)) {
 512                edac_dbg(0, "ECC enabled with x%d SDCC\n", ECCx8(pvt) ? 8 : 4);
 513                if (ECCx8(pvt))
 514                        mode = EDAC_S8ECD8ED;
 515                else
 516                        mode = EDAC_S4ECD4ED;
 517        } else {
 518                edac_dbg(0, "ECC disabled\n");
 519                mode = EDAC_NONE;
 520        }
 521
 522        /* FIXME: need to handle the error codes */
 523        edac_dbg(0, "DOD Max limits: DIMMS: %d, %d-ranked, %d-banked x%x x 0x%x\n",
 524                 numdimms(pvt->info.max_dod),
 525                 numrank(pvt->info.max_dod >> 2),
 526                 numbank(pvt->info.max_dod >> 4),
 527                 numrow(pvt->info.max_dod >> 6),
 528                 numcol(pvt->info.max_dod >> 9));
 529
 530        for (i = 0; i < NUM_CHANS; i++) {
 531                u32 data, dimm_dod[3], value[8];
 532
 533                if (!pvt->pci_ch[i][0])
 534                        continue;
 535
 536                if (!CH_ACTIVE(pvt, i)) {
 537                        edac_dbg(0, "Channel %i is not active\n", i);
 538                        continue;
 539                }
 540                if (CH_DISABLED(pvt, i)) {
 541                        edac_dbg(0, "Channel %i is disabled\n", i);
 542                        continue;
 543                }
 544
 545                /* Devices 4-6 function 0 */
 546                pci_read_config_dword(pvt->pci_ch[i][0],
 547                                MC_CHANNEL_DIMM_INIT_PARAMS, &data);
 548
 549
 550                if (data & THREE_DIMMS_PRESENT)
 551                        pvt->channel[i].is_3dimms_present = true;
 552
 553                if (data & SINGLE_QUAD_RANK_PRESENT)
 554                        pvt->channel[i].is_single_4rank = true;
 555
 556                if (data & QUAD_RANK_PRESENT)
 557                        pvt->channel[i].has_4rank = true;
 558
 559                if (data & REGISTERED_DIMM)
 560                        mtype = MEM_RDDR3;
 561                else
 562                        mtype = MEM_DDR3;
 563
 564                /* Devices 4-6 function 1 */
 565                pci_read_config_dword(pvt->pci_ch[i][1],
 566                                MC_DOD_CH_DIMM0, &dimm_dod[0]);
 567                pci_read_config_dword(pvt->pci_ch[i][1],
 568                                MC_DOD_CH_DIMM1, &dimm_dod[1]);
 569                pci_read_config_dword(pvt->pci_ch[i][1],
 570                                MC_DOD_CH_DIMM2, &dimm_dod[2]);
 571
 572                edac_dbg(0, "Ch%d phy rd%d, wr%d (0x%08x): %s%s%s%cDIMMs\n",
 573                         i,
 574                         RDLCH(pvt->info.ch_map, i), WRLCH(pvt->info.ch_map, i),
 575                         data,
 576                         pvt->channel[i].is_3dimms_present ? "3DIMMS " : "",
 577                         pvt->channel[i].is_3dimms_present ? "SINGLE_4R " : "",
 578                         pvt->channel[i].has_4rank ? "HAS_4R " : "",
 579                         (data & REGISTERED_DIMM) ? 'R' : 'U');
 580
 581                for (j = 0; j < 3; j++) {
 582                        u32 banks, ranks, rows, cols;
 583                        u32 size, npages;
 584
 585                        if (!DIMM_PRESENT(dimm_dod[j]))
 586                                continue;
 587
 588                        dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers,
 589                                       i, j, 0);
 590                        banks = numbank(MC_DOD_NUMBANK(dimm_dod[j]));
 591                        ranks = numrank(MC_DOD_NUMRANK(dimm_dod[j]));
 592                        rows = numrow(MC_DOD_NUMROW(dimm_dod[j]));
 593                        cols = numcol(MC_DOD_NUMCOL(dimm_dod[j]));
 594
 595                        /* DDR3 has 8 I/O banks */
 596                        size = (rows * cols * banks * ranks) >> (20 - 3);
 597
 598                        edac_dbg(0, "\tdimm %d %d MiB offset: %x, bank: %d, rank: %d, row: %#x, col: %#x\n",
 599                                 j, size,
 600                                 RANKOFFSET(dimm_dod[j]),
 601                                 banks, ranks, rows, cols);
 602
 603                        npages = MiB_TO_PAGES(size);
 604
 605                        dimm->nr_pages = npages;
 606
 607                        switch (banks) {
 608                        case 4:
 609                                dimm->dtype = DEV_X4;
 610                                break;
 611                        case 8:
 612                                dimm->dtype = DEV_X8;
 613                                break;
 614                        case 16:
 615                                dimm->dtype = DEV_X16;
 616                                break;
 617                        default:
 618                                dimm->dtype = DEV_UNKNOWN;
 619                        }
 620
 621                        snprintf(dimm->label, sizeof(dimm->label),
 622                                 "CPU#%uChannel#%u_DIMM#%u",
 623                                 pvt->i7core_dev->socket, i, j);
 624                        dimm->grain = 8;
 625                        dimm->edac_mode = mode;
 626                        dimm->mtype = mtype;
 627                }
 628
 629                pci_read_config_dword(pdev, MC_SAG_CH_0, &value[0]);
 630                pci_read_config_dword(pdev, MC_SAG_CH_1, &value[1]);
 631                pci_read_config_dword(pdev, MC_SAG_CH_2, &value[2]);
 632                pci_read_config_dword(pdev, MC_SAG_CH_3, &value[3]);
 633                pci_read_config_dword(pdev, MC_SAG_CH_4, &value[4]);
 634                pci_read_config_dword(pdev, MC_SAG_CH_5, &value[5]);
 635                pci_read_config_dword(pdev, MC_SAG_CH_6, &value[6]);
 636                pci_read_config_dword(pdev, MC_SAG_CH_7, &value[7]);
 637                edac_dbg(1, "\t[%i] DIVBY3\tREMOVED\tOFFSET\n", i);
 638                for (j = 0; j < 8; j++)
 639                        edac_dbg(1, "\t\t%#x\t%#x\t%#x\n",
 640                                 (value[j] >> 27) & 0x1,
 641                                 (value[j] >> 24) & 0x7,
 642                                 (value[j] & ((1 << 24) - 1)));
 643        }
 644
 645        return 0;
 646}
 647
 648/****************************************************************************
 649                        Error insertion routines
 650 ****************************************************************************/
 651
 652#define to_mci(k) container_of(k, struct mem_ctl_info, dev)
 653
 654/* The i7core has independent error injection features per channel.
 655   However, to have a simpler code, we don't allow enabling error injection
 656   on more than one channel.
 657   Also, since a change at an inject parameter will be applied only at enable,
 658   we're disabling error injection on all write calls to the sysfs nodes that
 659   controls the error code injection.
 660 */
 661static int disable_inject(const struct mem_ctl_info *mci)
 662{
 663        struct i7core_pvt *pvt = mci->pvt_info;
 664
 665        pvt->inject.enable = 0;
 666
 667        if (!pvt->pci_ch[pvt->inject.channel][0])
 668                return -ENODEV;
 669
 670        pci_write_config_dword(pvt->pci_ch[pvt->inject.channel][0],
 671                                MC_CHANNEL_ERROR_INJECT, 0);
 672
 673        return 0;
 674}
 675
 676/*
 677 * i7core inject inject.section
 678 *
 679 *      accept and store error injection inject.section value
 680 *      bit 0 - refers to the lower 32-byte half cacheline
 681 *      bit 1 - refers to the upper 32-byte half cacheline
 682 */
 683static ssize_t i7core_inject_section_store(struct device *dev,
 684                                           struct device_attribute *mattr,
 685                                           const char *data, size_t count)
 686{
 687        struct mem_ctl_info *mci = to_mci(dev);
 688        struct i7core_pvt *pvt = mci->pvt_info;
 689        unsigned long value;
 690        int rc;
 691
 692        if (pvt->inject.enable)
 693                disable_inject(mci);
 694
 695        rc = kstrtoul(data, 10, &value);
 696        if ((rc < 0) || (value > 3))
 697                return -EIO;
 698
 699        pvt->inject.section = (u32) value;
 700        return count;
 701}
 702
 703static ssize_t i7core_inject_section_show(struct device *dev,
 704                                          struct device_attribute *mattr,
 705                                          char *data)
 706{
 707        struct mem_ctl_info *mci = to_mci(dev);
 708        struct i7core_pvt *pvt = mci->pvt_info;
 709        return sprintf(data, "0x%08x\n", pvt->inject.section);
 710}
 711
 712/*
 713 * i7core inject.type
 714 *
 715 *      accept and store error injection inject.section value
 716 *      bit 0 - repeat enable - Enable error repetition
 717 *      bit 1 - inject ECC error
 718 *      bit 2 - inject parity error
 719 */
 720static ssize_t i7core_inject_type_store(struct device *dev,
 721                                        struct device_attribute *mattr,
 722                                        const char *data, size_t count)
 723{
 724        struct mem_ctl_info *mci = to_mci(dev);
 725        struct i7core_pvt *pvt = mci->pvt_info;
 726        unsigned long value;
 727        int rc;
 728
 729        if (pvt->inject.enable)
 730                disable_inject(mci);
 731
 732        rc = kstrtoul(data, 10, &value);
 733        if ((rc < 0) || (value > 7))
 734                return -EIO;
 735
 736        pvt->inject.type = (u32) value;
 737        return count;
 738}
 739
 740static ssize_t i7core_inject_type_show(struct device *dev,
 741                                       struct device_attribute *mattr,
 742                                       char *data)
 743{
 744        struct mem_ctl_info *mci = to_mci(dev);
 745        struct i7core_pvt *pvt = mci->pvt_info;
 746
 747        return sprintf(data, "0x%08x\n", pvt->inject.type);
 748}
 749
 750/*
 751 * i7core_inject_inject.eccmask_store
 752 *
 753 * The type of error (UE/CE) will depend on the inject.eccmask value:
 754 *   Any bits set to a 1 will flip the corresponding ECC bit
 755 *   Correctable errors can be injected by flipping 1 bit or the bits within
 756 *   a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
 757 *   23:16 and 31:24). Flipping bits in two symbol pairs will cause an
 758 *   uncorrectable error to be injected.
 759 */
 760static ssize_t i7core_inject_eccmask_store(struct device *dev,
 761                                           struct device_attribute *mattr,
 762                                           const char *data, size_t count)
 763{
 764        struct mem_ctl_info *mci = to_mci(dev);
 765        struct i7core_pvt *pvt = mci->pvt_info;
 766        unsigned long value;
 767        int rc;
 768
 769        if (pvt->inject.enable)
 770                disable_inject(mci);
 771
 772        rc = kstrtoul(data, 10, &value);
 773        if (rc < 0)
 774                return -EIO;
 775
 776        pvt->inject.eccmask = (u32) value;
 777        return count;
 778}
 779
 780static ssize_t i7core_inject_eccmask_show(struct device *dev,
 781                                          struct device_attribute *mattr,
 782                                          char *data)
 783{
 784        struct mem_ctl_info *mci = to_mci(dev);
 785        struct i7core_pvt *pvt = mci->pvt_info;
 786
 787        return sprintf(data, "0x%08x\n", pvt->inject.eccmask);
 788}
 789
 790/*
 791 * i7core_addrmatch
 792 *
 793 * The type of error (UE/CE) will depend on the inject.eccmask value:
 794 *   Any bits set to a 1 will flip the corresponding ECC bit
 795 *   Correctable errors can be injected by flipping 1 bit or the bits within
 796 *   a symbol pair (2 consecutive aligned 8-bit pairs - i.e. 7:0 and 15:8 or
 797 *   23:16 and 31:24). Flipping bits in two symbol pairs will cause an
 798 *   uncorrectable error to be injected.
 799 */
 800
 801#define DECLARE_ADDR_MATCH(param, limit)                        \
 802static ssize_t i7core_inject_store_##param(                     \
 803        struct device *dev,                                     \
 804        struct device_attribute *mattr,                         \
 805        const char *data, size_t count)                         \
 806{                                                               \
 807        struct mem_ctl_info *mci = dev_get_drvdata(dev);        \
 808        struct i7core_pvt *pvt;                                 \
 809        long value;                                             \
 810        int rc;                                                 \
 811                                                                \
 812        edac_dbg(1, "\n");                                      \
 813        pvt = mci->pvt_info;                                    \
 814                                                                \
 815        if (pvt->inject.enable)                                 \
 816                disable_inject(mci);                            \
 817                                                                \
 818        if (!strcasecmp(data, "any") || !strcasecmp(data, "any\n"))\
 819                value = -1;                                     \
 820        else {                                                  \
 821                rc = kstrtoul(data, 10, &value);                \
 822                if ((rc < 0) || (value >= limit))               \
 823                        return -EIO;                            \
 824        }                                                       \
 825                                                                \
 826        pvt->inject.param = value;                              \
 827                                                                \
 828        return count;                                           \
 829}                                                               \
 830                                                                \
 831static ssize_t i7core_inject_show_##param(                      \
 832        struct device *dev,                                     \
 833        struct device_attribute *mattr,                         \
 834        char *data)                                             \
 835{                                                               \
 836        struct mem_ctl_info *mci = dev_get_drvdata(dev);        \
 837        struct i7core_pvt *pvt;                                 \
 838                                                                \
 839        pvt = mci->pvt_info;                                    \
 840        edac_dbg(1, "pvt=%p\n", pvt);                           \
 841        if (pvt->inject.param < 0)                              \
 842                return sprintf(data, "any\n");                  \
 843        else                                                    \
 844                return sprintf(data, "%d\n", pvt->inject.param);\
 845}
 846
 847#define ATTR_ADDR_MATCH(param)                                  \
 848        static DEVICE_ATTR(param, S_IRUGO | S_IWUSR,            \
 849                    i7core_inject_show_##param,                 \
 850                    i7core_inject_store_##param)
 851
 852DECLARE_ADDR_MATCH(channel, 3);
 853DECLARE_ADDR_MATCH(dimm, 3);
 854DECLARE_ADDR_MATCH(rank, 4);
 855DECLARE_ADDR_MATCH(bank, 32);
 856DECLARE_ADDR_MATCH(page, 0x10000);
 857DECLARE_ADDR_MATCH(col, 0x4000);
 858
 859ATTR_ADDR_MATCH(channel);
 860ATTR_ADDR_MATCH(dimm);
 861ATTR_ADDR_MATCH(rank);
 862ATTR_ADDR_MATCH(bank);
 863ATTR_ADDR_MATCH(page);
 864ATTR_ADDR_MATCH(col);
 865
 866static int write_and_test(struct pci_dev *dev, const int where, const u32 val)
 867{
 868        u32 read;
 869        int count;
 870
 871        edac_dbg(0, "setting pci %02x:%02x.%x reg=%02x value=%08x\n",
 872                 dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
 873                 where, val);
 874
 875        for (count = 0; count < 10; count++) {
 876                if (count)
 877                        msleep(100);
 878                pci_write_config_dword(dev, where, val);
 879                pci_read_config_dword(dev, where, &read);
 880
 881                if (read == val)
 882                        return 0;
 883        }
 884
 885        i7core_printk(KERN_ERR, "Error during set pci %02x:%02x.%x reg=%02x "
 886                "write=%08x. Read=%08x\n",
 887                dev->bus->number, PCI_SLOT(dev->devfn), PCI_FUNC(dev->devfn),
 888                where, val, read);
 889
 890        return -EINVAL;
 891}
 892
 893/*
 894 * This routine prepares the Memory Controller for error injection.
 895 * The error will be injected when some process tries to write to the
 896 * memory that matches the given criteria.
 897 * The criteria can be set in terms of a mask where dimm, rank, bank, page
 898 * and col can be specified.
 899 * A -1 value for any of the mask items will make the MCU to ignore
 900 * that matching criteria for error injection.
 901 *
 902 * It should be noticed that the error will only happen after a write operation
 903 * on a memory that matches the condition. if REPEAT_EN is not enabled at
 904 * inject mask, then it will produce just one error. Otherwise, it will repeat
 905 * until the injectmask would be cleaned.
 906 *
 907 * FIXME: This routine assumes that MAXNUMDIMMS value of MC_MAX_DOD
 908 *    is reliable enough to check if the MC is using the
 909 *    three channels. However, this is not clear at the datasheet.
 910 */
 911static ssize_t i7core_inject_enable_store(struct device *dev,
 912                                          struct device_attribute *mattr,
 913                                          const char *data, size_t count)
 914{
 915        struct mem_ctl_info *mci = to_mci(dev);
 916        struct i7core_pvt *pvt = mci->pvt_info;
 917        u32 injectmask;
 918        u64 mask = 0;
 919        int  rc;
 920        long enable;
 921
 922        if (!pvt->pci_ch[pvt->inject.channel][0])
 923                return 0;
 924
 925        rc = kstrtoul(data, 10, &enable);
 926        if ((rc < 0))
 927                return 0;
 928
 929        if (enable) {
 930                pvt->inject.enable = 1;
 931        } else {
 932                disable_inject(mci);
 933                return count;
 934        }
 935
 936        /* Sets pvt->inject.dimm mask */
 937        if (pvt->inject.dimm < 0)
 938                mask |= 1LL << 41;
 939        else {
 940                if (pvt->channel[pvt->inject.channel].dimms > 2)
 941                        mask |= (pvt->inject.dimm & 0x3LL) << 35;
 942                else
 943                        mask |= (pvt->inject.dimm & 0x1LL) << 36;
 944        }
 945
 946        /* Sets pvt->inject.rank mask */
 947        if (pvt->inject.rank < 0)
 948                mask |= 1LL << 40;
 949        else {
 950                if (pvt->channel[pvt->inject.channel].dimms > 2)
 951                        mask |= (pvt->inject.rank & 0x1LL) << 34;
 952                else
 953                        mask |= (pvt->inject.rank & 0x3LL) << 34;
 954        }
 955
 956        /* Sets pvt->inject.bank mask */
 957        if (pvt->inject.bank < 0)
 958                mask |= 1LL << 39;
 959        else
 960                mask |= (pvt->inject.bank & 0x15LL) << 30;
 961
 962        /* Sets pvt->inject.page mask */
 963        if (pvt->inject.page < 0)
 964                mask |= 1LL << 38;
 965        else
 966                mask |= (pvt->inject.page & 0xffff) << 14;
 967
 968        /* Sets pvt->inject.column mask */
 969        if (pvt->inject.col < 0)
 970                mask |= 1LL << 37;
 971        else
 972                mask |= (pvt->inject.col & 0x3fff);
 973
 974        /*
 975         * bit    0: REPEAT_EN
 976         * bits 1-2: MASK_HALF_CACHELINE
 977         * bit    3: INJECT_ECC
 978         * bit    4: INJECT_ADDR_PARITY
 979         */
 980
 981        injectmask = (pvt->inject.type & 1) |
 982                     (pvt->inject.section & 0x3) << 1 |
 983                     (pvt->inject.type & 0x6) << (3 - 1);
 984
 985        /* Unlock writes to registers - this register is write only */
 986        pci_write_config_dword(pvt->pci_noncore,
 987                               MC_CFG_CONTROL, 0x2);
 988
 989        write_and_test(pvt->pci_ch[pvt->inject.channel][0],
 990                               MC_CHANNEL_ADDR_MATCH, mask);
 991        write_and_test(pvt->pci_ch[pvt->inject.channel][0],
 992                               MC_CHANNEL_ADDR_MATCH + 4, mask >> 32L);
 993
 994        write_and_test(pvt->pci_ch[pvt->inject.channel][0],
 995                               MC_CHANNEL_ERROR_MASK, pvt->inject.eccmask);
 996
 997        write_and_test(pvt->pci_ch[pvt->inject.channel][0],
 998                               MC_CHANNEL_ERROR_INJECT, injectmask);
 999
1000        /*
1001         * This is something undocumented, based on my tests
1002         * Without writing 8 to this register, errors aren't injected. Not sure
1003         * why.
1004         */
1005        pci_write_config_dword(pvt->pci_noncore,
1006                               MC_CFG_CONTROL, 8);
1007
1008        edac_dbg(0, "Error inject addr match 0x%016llx, ecc 0x%08x, inject 0x%08x\n",
1009                 mask, pvt->inject.eccmask, injectmask);
1010
1011
1012        return count;
1013}
1014
1015static ssize_t i7core_inject_enable_show(struct device *dev,
1016                                         struct device_attribute *mattr,
1017                                         char *data)
1018{
1019        struct mem_ctl_info *mci = to_mci(dev);
1020        struct i7core_pvt *pvt = mci->pvt_info;
1021        u32 injectmask;
1022
1023        if (!pvt->pci_ch[pvt->inject.channel][0])
1024                return 0;
1025
1026        pci_read_config_dword(pvt->pci_ch[pvt->inject.channel][0],
1027                               MC_CHANNEL_ERROR_INJECT, &injectmask);
1028
1029        edac_dbg(0, "Inject error read: 0x%018x\n", injectmask);
1030
1031        if (injectmask & 0x0c)
1032                pvt->inject.enable = 1;
1033
1034        return sprintf(data, "%d\n", pvt->inject.enable);
1035}
1036
1037#define DECLARE_COUNTER(param)                                  \
1038static ssize_t i7core_show_counter_##param(                     \
1039        struct device *dev,                                     \
1040        struct device_attribute *mattr,                         \
1041        char *data)                                             \
1042{                                                               \
1043        struct mem_ctl_info *mci = dev_get_drvdata(dev);        \
1044        struct i7core_pvt *pvt = mci->pvt_info;                 \
1045                                                                \
1046        edac_dbg(1, "\n");                                      \
1047        if (!pvt->ce_count_available || (pvt->is_registered))   \
1048                return sprintf(data, "data unavailable\n");     \
1049        return sprintf(data, "%lu\n",                           \
1050                        pvt->udimm_ce_count[param]);            \
1051}
1052
1053#define ATTR_COUNTER(param)                                     \
1054        static DEVICE_ATTR(udimm##param, S_IRUGO | S_IWUSR,     \
1055                    i7core_show_counter_##param,                \
1056                    NULL)
1057
1058DECLARE_COUNTER(0);
1059DECLARE_COUNTER(1);
1060DECLARE_COUNTER(2);
1061
1062ATTR_COUNTER(0);
1063ATTR_COUNTER(1);
1064ATTR_COUNTER(2);
1065
1066/*
1067 * inject_addrmatch device sysfs struct
1068 */
1069
1070static struct attribute *i7core_addrmatch_attrs[] = {
1071        &dev_attr_channel.attr,
1072        &dev_attr_dimm.attr,
1073        &dev_attr_rank.attr,
1074        &dev_attr_bank.attr,
1075        &dev_attr_page.attr,
1076        &dev_attr_col.attr,
1077        NULL
1078};
1079
1080static const struct attribute_group addrmatch_grp = {
1081        .attrs  = i7core_addrmatch_attrs,
1082};
1083
1084static const struct attribute_group *addrmatch_groups[] = {
1085        &addrmatch_grp,
1086        NULL
1087};
1088
1089static void addrmatch_release(struct device *device)
1090{
1091        edac_dbg(1, "Releasing device %s\n", dev_name(device));
1092        kfree(device);
1093}
1094
1095static const struct device_type addrmatch_type = {
1096        .groups         = addrmatch_groups,
1097        .release        = addrmatch_release,
1098};
1099
1100/*
1101 * all_channel_counts sysfs struct
1102 */
1103
1104static struct attribute *i7core_udimm_counters_attrs[] = {
1105        &dev_attr_udimm0.attr,
1106        &dev_attr_udimm1.attr,
1107        &dev_attr_udimm2.attr,
1108        NULL
1109};
1110
1111static const struct attribute_group all_channel_counts_grp = {
1112        .attrs  = i7core_udimm_counters_attrs,
1113};
1114
1115static const struct attribute_group *all_channel_counts_groups[] = {
1116        &all_channel_counts_grp,
1117        NULL
1118};
1119
1120static void all_channel_counts_release(struct device *device)
1121{
1122        edac_dbg(1, "Releasing device %s\n", dev_name(device));
1123        kfree(device);
1124}
1125
1126static const struct device_type all_channel_counts_type = {
1127        .groups         = all_channel_counts_groups,
1128        .release        = all_channel_counts_release,
1129};
1130
1131/*
1132 * inject sysfs attributes
1133 */
1134
1135static DEVICE_ATTR(inject_section, S_IRUGO | S_IWUSR,
1136                   i7core_inject_section_show, i7core_inject_section_store);
1137
1138static DEVICE_ATTR(inject_type, S_IRUGO | S_IWUSR,
1139                   i7core_inject_type_show, i7core_inject_type_store);
1140
1141
1142static DEVICE_ATTR(inject_eccmask, S_IRUGO | S_IWUSR,
1143                   i7core_inject_eccmask_show, i7core_inject_eccmask_store);
1144
1145static DEVICE_ATTR(inject_enable, S_IRUGO | S_IWUSR,
1146                   i7core_inject_enable_show, i7core_inject_enable_store);
1147
1148static struct attribute *i7core_dev_attrs[] = {
1149        &dev_attr_inject_section.attr,
1150        &dev_attr_inject_type.attr,
1151        &dev_attr_inject_eccmask.attr,
1152        &dev_attr_inject_enable.attr,
1153        NULL
1154};
1155
1156ATTRIBUTE_GROUPS(i7core_dev);
1157
1158static int i7core_create_sysfs_devices(struct mem_ctl_info *mci)
1159{
1160        struct i7core_pvt *pvt = mci->pvt_info;
1161        int rc;
1162
1163        pvt->addrmatch_dev = kzalloc(sizeof(*pvt->addrmatch_dev), GFP_KERNEL);
1164        if (!pvt->addrmatch_dev)
1165                return -ENOMEM;
1166
1167        pvt->addrmatch_dev->type = &addrmatch_type;
1168        pvt->addrmatch_dev->bus = mci->dev.bus;
1169        device_initialize(pvt->addrmatch_dev);
1170        pvt->addrmatch_dev->parent = &mci->dev;
1171        dev_set_name(pvt->addrmatch_dev, "inject_addrmatch");
1172        dev_set_drvdata(pvt->addrmatch_dev, mci);
1173
1174        edac_dbg(1, "creating %s\n", dev_name(pvt->addrmatch_dev));
1175
1176        rc = device_add(pvt->addrmatch_dev);
1177        if (rc < 0)
1178                goto err_put_addrmatch;
1179
1180        if (!pvt->is_registered) {
1181                pvt->chancounts_dev = kzalloc(sizeof(*pvt->chancounts_dev),
1182                                              GFP_KERNEL);
1183                if (!pvt->chancounts_dev) {
1184                        rc = -ENOMEM;
1185                        goto err_del_addrmatch;
1186                }
1187
1188                pvt->chancounts_dev->type = &all_channel_counts_type;
1189                pvt->chancounts_dev->bus = mci->dev.bus;
1190                device_initialize(pvt->chancounts_dev);
1191                pvt->chancounts_dev->parent = &mci->dev;
1192                dev_set_name(pvt->chancounts_dev, "all_channel_counts");
1193                dev_set_drvdata(pvt->chancounts_dev, mci);
1194
1195                edac_dbg(1, "creating %s\n", dev_name(pvt->chancounts_dev));
1196
1197                rc = device_add(pvt->chancounts_dev);
1198                if (rc < 0)
1199                        goto err_put_chancounts;
1200        }
1201        return 0;
1202
1203err_put_chancounts:
1204        put_device(pvt->chancounts_dev);
1205err_del_addrmatch:
1206        device_del(pvt->addrmatch_dev);
1207err_put_addrmatch:
1208        put_device(pvt->addrmatch_dev);
1209
1210        return rc;
1211}
1212
1213static void i7core_delete_sysfs_devices(struct mem_ctl_info *mci)
1214{
1215        struct i7core_pvt *pvt = mci->pvt_info;
1216
1217        edac_dbg(1, "\n");
1218
1219        if (!pvt->is_registered) {
1220                device_del(pvt->chancounts_dev);
1221                put_device(pvt->chancounts_dev);
1222        }
1223        device_del(pvt->addrmatch_dev);
1224        put_device(pvt->addrmatch_dev);
1225}
1226
1227/****************************************************************************
1228        Device initialization routines: put/get, init/exit
1229 ****************************************************************************/
1230
1231/*
1232 *      i7core_put_all_devices  'put' all the devices that we have
1233 *                              reserved via 'get'
1234 */
1235static void i7core_put_devices(struct i7core_dev *i7core_dev)
1236{
1237        int i;
1238
1239        edac_dbg(0, "\n");
1240        for (i = 0; i < i7core_dev->n_devs; i++) {
1241                struct pci_dev *pdev = i7core_dev->pdev[i];
1242                if (!pdev)
1243                        continue;
1244                edac_dbg(0, "Removing dev %02x:%02x.%d\n",
1245                         pdev->bus->number,
1246                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
1247                pci_dev_put(pdev);
1248        }
1249}
1250
1251static void i7core_put_all_devices(void)
1252{
1253        struct i7core_dev *i7core_dev, *tmp;
1254
1255        list_for_each_entry_safe(i7core_dev, tmp, &i7core_edac_list, list) {
1256                i7core_put_devices(i7core_dev);
1257                free_i7core_dev(i7core_dev);
1258        }
1259}
1260
1261static void __init i7core_xeon_pci_fixup(const struct pci_id_table *table)
1262{
1263        struct pci_dev *pdev = NULL;
1264        int i;
1265
1266        /*
1267         * On Xeon 55xx, the Intel Quick Path Arch Generic Non-core pci buses
1268         * aren't announced by acpi. So, we need to use a legacy scan probing
1269         * to detect them
1270         */
1271        while (table && table->descr) {
1272                pdev = pci_get_device(PCI_VENDOR_ID_INTEL, table->descr[0].dev_id, NULL);
1273                if (unlikely(!pdev)) {
1274                        for (i = 0; i < MAX_SOCKET_BUSES; i++)
1275                                pcibios_scan_specific_bus(255-i);
1276                }
1277                pci_dev_put(pdev);
1278                table++;
1279        }
1280}
1281
1282static unsigned i7core_pci_lastbus(void)
1283{
1284        int last_bus = 0, bus;
1285        struct pci_bus *b = NULL;
1286
1287        while ((b = pci_find_next_bus(b)) != NULL) {
1288                bus = b->number;
1289                edac_dbg(0, "Found bus %d\n", bus);
1290                if (bus > last_bus)
1291                        last_bus = bus;
1292        }
1293
1294        edac_dbg(0, "Last bus %d\n", last_bus);
1295
1296        return last_bus;
1297}
1298
1299/*
1300 *      i7core_get_all_devices  Find and perform 'get' operation on the MCH's
1301 *                      device/functions we want to reference for this driver
1302 *
1303 *                      Need to 'get' device 16 func 1 and func 2
1304 */
1305static int i7core_get_onedevice(struct pci_dev **prev,
1306                                const struct pci_id_table *table,
1307                                const unsigned devno,
1308                                const unsigned last_bus)
1309{
1310        struct i7core_dev *i7core_dev;
1311        const struct pci_id_descr *dev_descr = &table->descr[devno];
1312
1313        struct pci_dev *pdev = NULL;
1314        u8 bus = 0;
1315        u8 socket = 0;
1316
1317        pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1318                              dev_descr->dev_id, *prev);
1319
1320        /*
1321         * On Xeon 55xx, the Intel QuickPath Arch Generic Non-core regs
1322         * is at addr 8086:2c40, instead of 8086:2c41. So, we need
1323         * to probe for the alternate address in case of failure
1324         */
1325        if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev) {
1326                pci_dev_get(*prev);     /* pci_get_device will put it */
1327                pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1328                                      PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev);
1329        }
1330
1331        if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE &&
1332            !pdev) {
1333                pci_dev_get(*prev);     /* pci_get_device will put it */
1334                pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
1335                                      PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT,
1336                                      *prev);
1337        }
1338
1339        if (!pdev) {
1340                if (*prev) {
1341                        *prev = pdev;
1342                        return 0;
1343                }
1344
1345                if (dev_descr->optional)
1346                        return 0;
1347
1348                if (devno == 0)
1349                        return -ENODEV;
1350
1351                i7core_printk(KERN_INFO,
1352                        "Device not found: dev %02x.%d PCI ID %04x:%04x\n",
1353                        dev_descr->dev, dev_descr->func,
1354                        PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1355
1356                /* End of list, leave */
1357                return -ENODEV;
1358        }
1359        bus = pdev->bus->number;
1360
1361        socket = last_bus - bus;
1362
1363        i7core_dev = get_i7core_dev(socket);
1364        if (!i7core_dev) {
1365                i7core_dev = alloc_i7core_dev(socket, table);
1366                if (!i7core_dev) {
1367                        pci_dev_put(pdev);
1368                        return -ENOMEM;
1369                }
1370        }
1371
1372        if (i7core_dev->pdev[devno]) {
1373                i7core_printk(KERN_ERR,
1374                        "Duplicated device for "
1375                        "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1376                        bus, dev_descr->dev, dev_descr->func,
1377                        PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1378                pci_dev_put(pdev);
1379                return -ENODEV;
1380        }
1381
1382        i7core_dev->pdev[devno] = pdev;
1383
1384        /* Sanity check */
1385        if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev ||
1386                        PCI_FUNC(pdev->devfn) != dev_descr->func)) {
1387                i7core_printk(KERN_ERR,
1388                        "Device PCI ID %04x:%04x "
1389                        "has dev %02x:%02x.%d instead of dev %02x:%02x.%d\n",
1390                        PCI_VENDOR_ID_INTEL, dev_descr->dev_id,
1391                        bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1392                        bus, dev_descr->dev, dev_descr->func);
1393                return -ENODEV;
1394        }
1395
1396        /* Be sure that the device is enabled */
1397        if (unlikely(pci_enable_device(pdev) < 0)) {
1398                i7core_printk(KERN_ERR,
1399                        "Couldn't enable "
1400                        "dev %02x:%02x.%d PCI ID %04x:%04x\n",
1401                        bus, dev_descr->dev, dev_descr->func,
1402                        PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1403                return -ENODEV;
1404        }
1405
1406        edac_dbg(0, "Detected socket %d dev %02x:%02x.%d PCI ID %04x:%04x\n",
1407                 socket, bus, dev_descr->dev,
1408                 dev_descr->func,
1409                 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
1410
1411        /*
1412         * As stated on drivers/pci/search.c, the reference count for
1413         * @from is always decremented if it is not %NULL. So, as we need
1414         * to get all devices up to null, we need to do a get for the device
1415         */
1416        pci_dev_get(pdev);
1417
1418        *prev = pdev;
1419
1420        return 0;
1421}
1422
1423static int i7core_get_all_devices(void)
1424{
1425        int i, rc, last_bus;
1426        struct pci_dev *pdev = NULL;
1427        const struct pci_id_table *table = pci_dev_table;
1428
1429        last_bus = i7core_pci_lastbus();
1430
1431        while (table && table->descr) {
1432                for (i = 0; i < table->n_devs; i++) {
1433                        pdev = NULL;
1434                        do {
1435                                rc = i7core_get_onedevice(&pdev, table, i,
1436                                                          last_bus);
1437                                if (rc < 0) {
1438                                        if (i == 0) {
1439                                                i = table->n_devs;
1440                                                break;
1441                                        }
1442                                        i7core_put_all_devices();
1443                                        return -ENODEV;
1444                                }
1445                        } while (pdev);
1446                }
1447                table++;
1448        }
1449
1450        return 0;
1451}
1452
1453static int mci_bind_devs(struct mem_ctl_info *mci,
1454                         struct i7core_dev *i7core_dev)
1455{
1456        struct i7core_pvt *pvt = mci->pvt_info;
1457        struct pci_dev *pdev;
1458        int i, func, slot;
1459        char *family;
1460
1461        pvt->is_registered = false;
1462        pvt->enable_scrub  = false;
1463        for (i = 0; i < i7core_dev->n_devs; i++) {
1464                pdev = i7core_dev->pdev[i];
1465                if (!pdev)
1466                        continue;
1467
1468                func = PCI_FUNC(pdev->devfn);
1469                slot = PCI_SLOT(pdev->devfn);
1470                if (slot == 3) {
1471                        if (unlikely(func > MAX_MCR_FUNC))
1472                                goto error;
1473                        pvt->pci_mcr[func] = pdev;
1474                } else if (likely(slot >= 4 && slot < 4 + NUM_CHANS)) {
1475                        if (unlikely(func > MAX_CHAN_FUNC))
1476                                goto error;
1477                        pvt->pci_ch[slot - 4][func] = pdev;
1478                } else if (!slot && !func) {
1479                        pvt->pci_noncore = pdev;
1480
1481                        /* Detect the processor family */
1482                        switch (pdev->device) {
1483                        case PCI_DEVICE_ID_INTEL_I7_NONCORE:
1484                                family = "Xeon 35xx/ i7core";
1485                                pvt->enable_scrub = false;
1486                                break;
1487                        case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT:
1488                                family = "i7-800/i5-700";
1489                                pvt->enable_scrub = false;
1490                                break;
1491                        case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE:
1492                                family = "Xeon 34xx";
1493                                pvt->enable_scrub = false;
1494                                break;
1495                        case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT:
1496                                family = "Xeon 55xx";
1497                                pvt->enable_scrub = true;
1498                                break;
1499                        case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2:
1500                                family = "Xeon 56xx / i7-900";
1501                                pvt->enable_scrub = true;
1502                                break;
1503                        default:
1504                                family = "unknown";
1505                                pvt->enable_scrub = false;
1506                        }
1507                        edac_dbg(0, "Detected a processor type %s\n", family);
1508                } else
1509                        goto error;
1510
1511                edac_dbg(0, "Associated fn %d.%d, dev = %p, socket %d\n",
1512                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
1513                         pdev, i7core_dev->socket);
1514
1515                if (PCI_SLOT(pdev->devfn) == 3 &&
1516                        PCI_FUNC(pdev->devfn) == 2)
1517                        pvt->is_registered = true;
1518        }
1519
1520        return 0;
1521
1522error:
1523        i7core_printk(KERN_ERR, "Device %d, function %d "
1524                      "is out of the expected range\n",
1525                      slot, func);
1526        return -EINVAL;
1527}
1528
1529/****************************************************************************
1530                        Error check routines
1531 ****************************************************************************/
1532
1533static void i7core_rdimm_update_ce_count(struct mem_ctl_info *mci,
1534                                         const int chan,
1535                                         const int new0,
1536                                         const int new1,
1537                                         const int new2)
1538{
1539        struct i7core_pvt *pvt = mci->pvt_info;
1540        int add0 = 0, add1 = 0, add2 = 0;
1541        /* Updates CE counters if it is not the first time here */
1542        if (pvt->ce_count_available) {
1543                /* Updates CE counters */
1544
1545                add2 = new2 - pvt->rdimm_last_ce_count[chan][2];
1546                add1 = new1 - pvt->rdimm_last_ce_count[chan][1];
1547                add0 = new0 - pvt->rdimm_last_ce_count[chan][0];
1548
1549                if (add2 < 0)
1550                        add2 += 0x7fff;
1551                pvt->rdimm_ce_count[chan][2] += add2;
1552
1553                if (add1 < 0)
1554                        add1 += 0x7fff;
1555                pvt->rdimm_ce_count[chan][1] += add1;
1556
1557                if (add0 < 0)
1558                        add0 += 0x7fff;
1559                pvt->rdimm_ce_count[chan][0] += add0;
1560        } else
1561                pvt->ce_count_available = 1;
1562
1563        /* Store the new values */
1564        pvt->rdimm_last_ce_count[chan][2] = new2;
1565        pvt->rdimm_last_ce_count[chan][1] = new1;
1566        pvt->rdimm_last_ce_count[chan][0] = new0;
1567
1568        /*updated the edac core */
1569        if (add0 != 0)
1570                edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add0,
1571                                     0, 0, 0,
1572                                     chan, 0, -1, "error", "");
1573        if (add1 != 0)
1574                edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add1,
1575                                     0, 0, 0,
1576                                     chan, 1, -1, "error", "");
1577        if (add2 != 0)
1578                edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, add2,
1579                                     0, 0, 0,
1580                                     chan, 2, -1, "error", "");
1581}
1582
1583static void i7core_rdimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1584{
1585        struct i7core_pvt *pvt = mci->pvt_info;
1586        u32 rcv[3][2];
1587        int i, new0, new1, new2;
1588
1589        /*Read DEV 3: FUN 2:  MC_COR_ECC_CNT regs directly*/
1590        pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_0,
1591                                                                &rcv[0][0]);
1592        pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_1,
1593                                                                &rcv[0][1]);
1594        pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_2,
1595                                                                &rcv[1][0]);
1596        pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_3,
1597                                                                &rcv[1][1]);
1598        pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_4,
1599                                                                &rcv[2][0]);
1600        pci_read_config_dword(pvt->pci_mcr[2], MC_COR_ECC_CNT_5,
1601                                                                &rcv[2][1]);
1602        for (i = 0 ; i < 3; i++) {
1603                edac_dbg(3, "MC_COR_ECC_CNT%d = 0x%x; MC_COR_ECC_CNT%d = 0x%x\n",
1604                         (i * 2), rcv[i][0], (i * 2) + 1, rcv[i][1]);
1605                /*if the channel has 3 dimms*/
1606                if (pvt->channel[i].dimms > 2) {
1607                        new0 = DIMM_BOT_COR_ERR(rcv[i][0]);
1608                        new1 = DIMM_TOP_COR_ERR(rcv[i][0]);
1609                        new2 = DIMM_BOT_COR_ERR(rcv[i][1]);
1610                } else {
1611                        new0 = DIMM_TOP_COR_ERR(rcv[i][0]) +
1612                                        DIMM_BOT_COR_ERR(rcv[i][0]);
1613                        new1 = DIMM_TOP_COR_ERR(rcv[i][1]) +
1614                                        DIMM_BOT_COR_ERR(rcv[i][1]);
1615                        new2 = 0;
1616                }
1617
1618                i7core_rdimm_update_ce_count(mci, i, new0, new1, new2);
1619        }
1620}
1621
1622/* This function is based on the device 3 function 4 registers as described on:
1623 * Intel Xeon Processor 5500 Series Datasheet Volume 2
1624 *      http://www.intel.com/Assets/PDF/datasheet/321322.pdf
1625 * also available at:
1626 *      http://www.arrownac.com/manufacturers/intel/s/nehalem/5500-datasheet-v2.pdf
1627 */
1628static void i7core_udimm_check_mc_ecc_err(struct mem_ctl_info *mci)
1629{
1630        struct i7core_pvt *pvt = mci->pvt_info;
1631        u32 rcv1, rcv0;
1632        int new0, new1, new2;
1633
1634        if (!pvt->pci_mcr[4]) {
1635                edac_dbg(0, "MCR registers not found\n");
1636                return;
1637        }
1638
1639        /* Corrected test errors */
1640        pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV1, &rcv1);
1641        pci_read_config_dword(pvt->pci_mcr[4], MC_TEST_ERR_RCV0, &rcv0);
1642
1643        /* Store the new values */
1644        new2 = DIMM2_COR_ERR(rcv1);
1645        new1 = DIMM1_COR_ERR(rcv0);
1646        new0 = DIMM0_COR_ERR(rcv0);
1647
1648        /* Updates CE counters if it is not the first time here */
1649        if (pvt->ce_count_available) {
1650                /* Updates CE counters */
1651                int add0, add1, add2;
1652
1653                add2 = new2 - pvt->udimm_last_ce_count[2];
1654                add1 = new1 - pvt->udimm_last_ce_count[1];
1655                add0 = new0 - pvt->udimm_last_ce_count[0];
1656
1657                if (add2 < 0)
1658                        add2 += 0x7fff;
1659                pvt->udimm_ce_count[2] += add2;
1660
1661                if (add1 < 0)
1662                        add1 += 0x7fff;
1663                pvt->udimm_ce_count[1] += add1;
1664
1665                if (add0 < 0)
1666                        add0 += 0x7fff;
1667                pvt->udimm_ce_count[0] += add0;
1668
1669                if (add0 | add1 | add2)
1670                        i7core_printk(KERN_ERR, "New Corrected error(s): "
1671                                      "dimm0: +%d, dimm1: +%d, dimm2 +%d\n",
1672                                      add0, add1, add2);
1673        } else
1674                pvt->ce_count_available = 1;
1675
1676        /* Store the new values */
1677        pvt->udimm_last_ce_count[2] = new2;
1678        pvt->udimm_last_ce_count[1] = new1;
1679        pvt->udimm_last_ce_count[0] = new0;
1680}
1681
1682/*
1683 * According with tables E-11 and E-12 of chapter E.3.3 of Intel 64 and IA-32
1684 * Architectures Software Developer’s Manual Volume 3B.
1685 * Nehalem are defined as family 0x06, model 0x1a
1686 *
1687 * The MCA registers used here are the following ones:
1688 *     struct mce field MCA Register
1689 *     m->status        MSR_IA32_MC8_STATUS
1690 *     m->addr          MSR_IA32_MC8_ADDR
1691 *     m->misc          MSR_IA32_MC8_MISC
1692 * In the case of Nehalem, the error information is masked at .status and .misc
1693 * fields
1694 */
1695static void i7core_mce_output_error(struct mem_ctl_info *mci,
1696                                    const struct mce *m)
1697{
1698        struct i7core_pvt *pvt = mci->pvt_info;
1699        char *optype, *err;
1700        enum hw_event_mc_err_type tp_event;
1701        unsigned long error = m->status & 0x1ff0000l;
1702        bool uncorrected_error = m->mcgstatus & 1ll << 61;
1703        bool ripv = m->mcgstatus & 1;
1704        u32 optypenum = (m->status >> 4) & 0x07;
1705        u32 core_err_cnt = (m->status >> 38) & 0x7fff;
1706        u32 dimm = (m->misc >> 16) & 0x3;
1707        u32 channel = (m->misc >> 18) & 0x3;
1708        u32 syndrome = m->misc >> 32;
1709        u32 errnum = find_first_bit(&error, 32);
1710
1711        if (uncorrected_error) {
1712                core_err_cnt = 1;
1713                if (ripv)
1714                        tp_event = HW_EVENT_ERR_FATAL;
1715                else
1716                        tp_event = HW_EVENT_ERR_UNCORRECTED;
1717        } else {
1718                tp_event = HW_EVENT_ERR_CORRECTED;
1719        }
1720
1721        switch (optypenum) {
1722        case 0:
1723                optype = "generic undef request";
1724                break;
1725        case 1:
1726                optype = "read error";
1727                break;
1728        case 2:
1729                optype = "write error";
1730                break;
1731        case 3:
1732                optype = "addr/cmd error";
1733                break;
1734        case 4:
1735                optype = "scrubbing error";
1736                break;
1737        default:
1738                optype = "reserved";
1739                break;
1740        }
1741
1742        switch (errnum) {
1743        case 16:
1744                err = "read ECC error";
1745                break;
1746        case 17:
1747                err = "RAS ECC error";
1748                break;
1749        case 18:
1750                err = "write parity error";
1751                break;
1752        case 19:
1753                err = "redundancy loss";
1754                break;
1755        case 20:
1756                err = "reserved";
1757                break;
1758        case 21:
1759                err = "memory range error";
1760                break;
1761        case 22:
1762                err = "RTID out of range";
1763                break;
1764        case 23:
1765                err = "address parity error";
1766                break;
1767        case 24:
1768                err = "byte enable parity error";
1769                break;
1770        default:
1771                err = "unknown";
1772        }
1773
1774        /*
1775         * Call the helper to output message
1776         * FIXME: what to do if core_err_cnt > 1? Currently, it generates
1777         * only one event
1778         */
1779        if (uncorrected_error || !pvt->is_registered)
1780                edac_mc_handle_error(tp_event, mci, core_err_cnt,
1781                                     m->addr >> PAGE_SHIFT,
1782                                     m->addr & ~PAGE_MASK,
1783                                     syndrome,
1784                                     channel, dimm, -1,
1785                                     err, optype);
1786}
1787
1788/*
1789 *      i7core_check_error      Retrieve and process errors reported by the
1790 *                              hardware. Called by the Core module.
1791 */
1792static void i7core_check_error(struct mem_ctl_info *mci, struct mce *m)
1793{
1794        struct i7core_pvt *pvt = mci->pvt_info;
1795
1796        i7core_mce_output_error(mci, m);
1797
1798        /*
1799         * Now, let's increment CE error counts
1800         */
1801        if (!pvt->is_registered)
1802                i7core_udimm_check_mc_ecc_err(mci);
1803        else
1804                i7core_rdimm_check_mc_ecc_err(mci);
1805}
1806
1807/*
1808 * Check that logging is enabled and that this is the right type
1809 * of error for us to handle.
1810 */
1811static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val,
1812                                  void *data)
1813{
1814        struct mce *mce = (struct mce *)data;
1815        struct i7core_dev *i7_dev;
1816        struct mem_ctl_info *mci;
1817
1818        i7_dev = get_i7core_dev(mce->socketid);
1819        if (!i7_dev)
1820                return NOTIFY_DONE;
1821
1822        mci = i7_dev->mci;
1823
1824        /*
1825         * Just let mcelog handle it if the error is
1826         * outside the memory controller
1827         */
1828        if (((mce->status & 0xffff) >> 7) != 1)
1829                return NOTIFY_DONE;
1830
1831        /* Bank 8 registers are the only ones that we know how to handle */
1832        if (mce->bank != 8)
1833                return NOTIFY_DONE;
1834
1835        i7core_check_error(mci, mce);
1836
1837        /* Advise mcelog that the errors were handled */
1838        return NOTIFY_STOP;
1839}
1840
1841static struct notifier_block i7_mce_dec = {
1842        .notifier_call  = i7core_mce_check_error,
1843        .priority       = MCE_PRIO_EDAC,
1844};
1845
1846struct memdev_dmi_entry {
1847        u8 type;
1848        u8 length;
1849        u16 handle;
1850        u16 phys_mem_array_handle;
1851        u16 mem_err_info_handle;
1852        u16 total_width;
1853        u16 data_width;
1854        u16 size;
1855        u8 form;
1856        u8 device_set;
1857        u8 device_locator;
1858        u8 bank_locator;
1859        u8 memory_type;
1860        u16 type_detail;
1861        u16 speed;
1862        u8 manufacturer;
1863        u8 serial_number;
1864        u8 asset_tag;
1865        u8 part_number;
1866        u8 attributes;
1867        u32 extended_size;
1868        u16 conf_mem_clk_speed;
1869} __attribute__((__packed__));
1870
1871
1872/*
1873 * Decode the DRAM Clock Frequency, be paranoid, make sure that all
1874 * memory devices show the same speed, and if they don't then consider
1875 * all speeds to be invalid.
1876 */
1877static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq)
1878{
1879        int *dclk_freq = _dclk_freq;
1880        u16 dmi_mem_clk_speed;
1881
1882        if (*dclk_freq == -1)
1883                return;
1884
1885        if (dh->type == DMI_ENTRY_MEM_DEVICE) {
1886                struct memdev_dmi_entry *memdev_dmi_entry =
1887                        (struct memdev_dmi_entry *)dh;
1888                unsigned long conf_mem_clk_speed_offset =
1889                        (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed -
1890                        (unsigned long)&memdev_dmi_entry->type;
1891                unsigned long speed_offset =
1892                        (unsigned long)&memdev_dmi_entry->speed -
1893                        (unsigned long)&memdev_dmi_entry->type;
1894
1895                /* Check that a DIMM is present */
1896                if (memdev_dmi_entry->size == 0)
1897                        return;
1898
1899                /*
1900                 * Pick the configured speed if it's available, otherwise
1901                 * pick the DIMM speed, or we don't have a speed.
1902                 */
1903                if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) {
1904                        dmi_mem_clk_speed =
1905                                memdev_dmi_entry->conf_mem_clk_speed;
1906                } else if (memdev_dmi_entry->length > speed_offset) {
1907                        dmi_mem_clk_speed = memdev_dmi_entry->speed;
1908                } else {
1909                        *dclk_freq = -1;
1910                        return;
1911                }
1912
1913                if (*dclk_freq == 0) {
1914                        /* First pass, speed was 0 */
1915                        if (dmi_mem_clk_speed > 0) {
1916                                /* Set speed if a valid speed is read */
1917                                *dclk_freq = dmi_mem_clk_speed;
1918                        } else {
1919                                /* Otherwise we don't have a valid speed */
1920                                *dclk_freq = -1;
1921                        }
1922                } else if (*dclk_freq > 0 &&
1923                           *dclk_freq != dmi_mem_clk_speed) {
1924                        /*
1925                         * If we have a speed, check that all DIMMS are the same
1926                         * speed, otherwise set the speed as invalid.
1927                         */
1928                        *dclk_freq = -1;
1929                }
1930        }
1931}
1932
1933/*
1934 * The default DCLK frequency is used as a fallback if we
1935 * fail to find anything reliable in the DMI. The value
1936 * is taken straight from the datasheet.
1937 */
1938#define DEFAULT_DCLK_FREQ 800
1939
1940static int get_dclk_freq(void)
1941{
1942        int dclk_freq = 0;
1943
1944        dmi_walk(decode_dclk, (void *)&dclk_freq);
1945
1946        if (dclk_freq < 1)
1947                return DEFAULT_DCLK_FREQ;
1948
1949        return dclk_freq;
1950}
1951
1952/*
1953 * set_sdram_scrub_rate         This routine sets byte/sec bandwidth scrub rate
1954 *                              to hardware according to SCRUBINTERVAL formula
1955 *                              found in datasheet.
1956 */
1957static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw)
1958{
1959        struct i7core_pvt *pvt = mci->pvt_info;
1960        struct pci_dev *pdev;
1961        u32 dw_scrub;
1962        u32 dw_ssr;
1963
1964        /* Get data from the MC register, function 2 */
1965        pdev = pvt->pci_mcr[2];
1966        if (!pdev)
1967                return -ENODEV;
1968
1969        pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub);
1970
1971        if (new_bw == 0) {
1972                /* Prepare to disable petrol scrub */
1973                dw_scrub &= ~STARTSCRUB;
1974                /* Stop the patrol scrub engine */
1975                write_and_test(pdev, MC_SCRUB_CONTROL,
1976                               dw_scrub & ~SCRUBINTERVAL_MASK);
1977
1978                /* Get current status of scrub rate and set bit to disable */
1979                pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
1980                dw_ssr &= ~SSR_MODE_MASK;
1981                dw_ssr |= SSR_MODE_DISABLE;
1982        } else {
1983                const int cache_line_size = 64;
1984                const u32 freq_dclk_mhz = pvt->dclk_freq;
1985                unsigned long long scrub_interval;
1986                /*
1987                 * Translate the desired scrub rate to a register value and
1988                 * program the corresponding register value.
1989                 */
1990                scrub_interval = (unsigned long long)freq_dclk_mhz *
1991                        cache_line_size * 1000000;
1992                do_div(scrub_interval, new_bw);
1993
1994                if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK)
1995                        return -EINVAL;
1996
1997                dw_scrub = SCRUBINTERVAL_MASK & scrub_interval;
1998
1999                /* Start the patrol scrub engine */
2000                pci_write_config_dword(pdev, MC_SCRUB_CONTROL,
2001                                       STARTSCRUB | dw_scrub);
2002
2003                /* Get current status of scrub rate and set bit to enable */
2004                pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr);
2005                dw_ssr &= ~SSR_MODE_MASK;
2006                dw_ssr |= SSR_MODE_ENABLE;
2007        }
2008        /* Disable or enable scrubbing */
2009        pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr);
2010
2011        return new_bw;
2012}
2013
2014/*
2015 * get_sdram_scrub_rate         This routine convert current scrub rate value
2016 *                              into byte/sec bandwidth according to
2017 *                              SCRUBINTERVAL formula found in datasheet.
2018 */
2019static int get_sdram_scrub_rate(struct mem_ctl_info *mci)
2020{
2021        struct i7core_pvt *pvt = mci->pvt_info;
2022        struct pci_dev *pdev;
2023        const u32 cache_line_size = 64;
2024        const u32 freq_dclk_mhz = pvt->dclk_freq;
2025        unsigned long long scrub_rate;
2026        u32 scrubval;
2027
2028        /* Get data from the MC register, function 2 */
2029        pdev = pvt->pci_mcr[2];
2030        if (!pdev)
2031                return -ENODEV;
2032
2033        /* Get current scrub control data */
2034        pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval);
2035
2036        /* Mask highest 8-bits to 0 */
2037        scrubval &=  SCRUBINTERVAL_MASK;
2038        if (!scrubval)
2039                return 0;
2040
2041        /* Calculate scrub rate value into byte/sec bandwidth */
2042        scrub_rate =  (unsigned long long)freq_dclk_mhz *
2043                1000000 * cache_line_size;
2044        do_div(scrub_rate, scrubval);
2045        return (int)scrub_rate;
2046}
2047
2048static void enable_sdram_scrub_setting(struct mem_ctl_info *mci)
2049{
2050        struct i7core_pvt *pvt = mci->pvt_info;
2051        u32 pci_lock;
2052
2053        /* Unlock writes to pci registers */
2054        pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2055        pci_lock &= ~0x3;
2056        pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2057                               pci_lock | MC_CFG_UNLOCK);
2058
2059        mci->set_sdram_scrub_rate = set_sdram_scrub_rate;
2060        mci->get_sdram_scrub_rate = get_sdram_scrub_rate;
2061}
2062
2063static void disable_sdram_scrub_setting(struct mem_ctl_info *mci)
2064{
2065        struct i7core_pvt *pvt = mci->pvt_info;
2066        u32 pci_lock;
2067
2068        /* Lock writes to pci registers */
2069        pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock);
2070        pci_lock &= ~0x3;
2071        pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL,
2072                               pci_lock | MC_CFG_LOCK);
2073}
2074
2075static void i7core_pci_ctl_create(struct i7core_pvt *pvt)
2076{
2077        pvt->i7core_pci = edac_pci_create_generic_ctl(
2078                                                &pvt->i7core_dev->pdev[0]->dev,
2079                                                EDAC_MOD_STR);
2080        if (unlikely(!pvt->i7core_pci))
2081                i7core_printk(KERN_WARNING,
2082                              "Unable to setup PCI error report via EDAC\n");
2083}
2084
2085static void i7core_pci_ctl_release(struct i7core_pvt *pvt)
2086{
2087        if (likely(pvt->i7core_pci))
2088                edac_pci_release_generic_ctl(pvt->i7core_pci);
2089        else
2090                i7core_printk(KERN_ERR,
2091                                "Couldn't find mem_ctl_info for socket %d\n",
2092                                pvt->i7core_dev->socket);
2093        pvt->i7core_pci = NULL;
2094}
2095
2096static void i7core_unregister_mci(struct i7core_dev *i7core_dev)
2097{
2098        struct mem_ctl_info *mci = i7core_dev->mci;
2099        struct i7core_pvt *pvt;
2100
2101        if (unlikely(!mci || !mci->pvt_info)) {
2102                edac_dbg(0, "MC: dev = %p\n", &i7core_dev->pdev[0]->dev);
2103
2104                i7core_printk(KERN_ERR, "Couldn't find mci handler\n");
2105                return;
2106        }
2107
2108        pvt = mci->pvt_info;
2109
2110        edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2111
2112        /* Disable scrubrate setting */
2113        if (pvt->enable_scrub)
2114                disable_sdram_scrub_setting(mci);
2115
2116        /* Disable EDAC polling */
2117        i7core_pci_ctl_release(pvt);
2118
2119        /* Remove MC sysfs nodes */
2120        i7core_delete_sysfs_devices(mci);
2121        edac_mc_del_mc(mci->pdev);
2122
2123        edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
2124        kfree(mci->ctl_name);
2125        edac_mc_free(mci);
2126        i7core_dev->mci = NULL;
2127}
2128
2129static int i7core_register_mci(struct i7core_dev *i7core_dev)
2130{
2131        struct mem_ctl_info *mci;
2132        struct i7core_pvt *pvt;
2133        int rc;
2134        struct edac_mc_layer layers[2];
2135
2136        /* allocate a new MC control structure */
2137
2138        layers[0].type = EDAC_MC_LAYER_CHANNEL;
2139        layers[0].size = NUM_CHANS;
2140        layers[0].is_virt_csrow = false;
2141        layers[1].type = EDAC_MC_LAYER_SLOT;
2142        layers[1].size = MAX_DIMMS;
2143        layers[1].is_virt_csrow = true;
2144        mci = edac_mc_alloc(i7core_dev->socket, ARRAY_SIZE(layers), layers,
2145                            sizeof(*pvt));
2146        if (unlikely(!mci))
2147                return -ENOMEM;
2148
2149        edac_dbg(0, "MC: mci = %p, dev = %p\n", mci, &i7core_dev->pdev[0]->dev);
2150
2151        pvt = mci->pvt_info;
2152        memset(pvt, 0, sizeof(*pvt));
2153
2154        /* Associates i7core_dev and mci for future usage */
2155        pvt->i7core_dev = i7core_dev;
2156        i7core_dev->mci = mci;
2157
2158        /*
2159         * FIXME: how to handle RDDR3 at MCI level? It is possible to have
2160         * Mixed RDDR3/UDDR3 with Nehalem, provided that they are on different
2161         * memory channels
2162         */
2163        mci->mtype_cap = MEM_FLAG_DDR3;
2164        mci->edac_ctl_cap = EDAC_FLAG_NONE;
2165        mci->edac_cap = EDAC_FLAG_NONE;
2166        mci->mod_name = "i7core_edac.c";
2167
2168        mci->ctl_name = kasprintf(GFP_KERNEL, "i7 core #%d", i7core_dev->socket);
2169        if (!mci->ctl_name) {
2170                rc = -ENOMEM;
2171                goto fail1;
2172        }
2173
2174        mci->dev_name = pci_name(i7core_dev->pdev[0]);
2175        mci->ctl_page_to_phys = NULL;
2176
2177        /* Store pci devices at mci for faster access */
2178        rc = mci_bind_devs(mci, i7core_dev);
2179        if (unlikely(rc < 0))
2180                goto fail0;
2181
2182
2183        /* Get dimm basic config */
2184        get_dimm_config(mci);
2185        /* record ptr to the generic device */
2186        mci->pdev = &i7core_dev->pdev[0]->dev;
2187
2188        /* Enable scrubrate setting */
2189        if (pvt->enable_scrub)
2190                enable_sdram_scrub_setting(mci);
2191
2192        /* add this new MC control structure to EDAC's list of MCs */
2193        if (unlikely(edac_mc_add_mc_with_groups(mci, i7core_dev_groups))) {
2194                edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
2195                /* FIXME: perhaps some code should go here that disables error
2196                 * reporting if we just enabled it
2197                 */
2198
2199                rc = -EINVAL;
2200                goto fail0;
2201        }
2202        if (i7core_create_sysfs_devices(mci)) {
2203                edac_dbg(0, "MC: failed to create sysfs nodes\n");
2204                edac_mc_del_mc(mci->pdev);
2205                rc = -EINVAL;
2206                goto fail0;
2207        }
2208
2209        /* Default error mask is any memory */
2210        pvt->inject.channel = 0;
2211        pvt->inject.dimm = -1;
2212        pvt->inject.rank = -1;
2213        pvt->inject.bank = -1;
2214        pvt->inject.page = -1;
2215        pvt->inject.col = -1;
2216
2217        /* allocating generic PCI control info */
2218        i7core_pci_ctl_create(pvt);
2219
2220        /* DCLK for scrub rate setting */
2221        pvt->dclk_freq = get_dclk_freq();
2222
2223        return 0;
2224
2225fail0:
2226        kfree(mci->ctl_name);
2227
2228fail1:
2229        edac_mc_free(mci);
2230        i7core_dev->mci = NULL;
2231        return rc;
2232}
2233
2234/*
2235 *      i7core_probe    Probe for ONE instance of device to see if it is
2236 *                      present.
2237 *      return:
2238 *              0 for FOUND a device
2239 *              < 0 for error code
2240 */
2241
2242static int i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2243{
2244        int rc, count = 0;
2245        struct i7core_dev *i7core_dev;
2246
2247        /* get the pci devices we want to reserve for our use */
2248        mutex_lock(&i7core_edac_lock);
2249
2250        /*
2251         * All memory controllers are allocated at the first pass.
2252         */
2253        if (unlikely(probed >= 1)) {
2254                mutex_unlock(&i7core_edac_lock);
2255                return -ENODEV;
2256        }
2257        probed++;
2258
2259        rc = i7core_get_all_devices();
2260        if (unlikely(rc < 0))
2261                goto fail0;
2262
2263        list_for_each_entry(i7core_dev, &i7core_edac_list, list) {
2264                count++;
2265                rc = i7core_register_mci(i7core_dev);
2266                if (unlikely(rc < 0))
2267                        goto fail1;
2268        }
2269
2270        /*
2271         * Nehalem-EX uses a different memory controller. However, as the
2272         * memory controller is not visible on some Nehalem/Nehalem-EP, we
2273         * need to indirectly probe via a X58 PCI device. The same devices
2274         * are found on (some) Nehalem-EX. So, on those machines, the
2275         * probe routine needs to return -ENODEV, as the actual Memory
2276         * Controller registers won't be detected.
2277         */
2278        if (!count) {
2279                rc = -ENODEV;
2280                goto fail1;
2281        }
2282
2283        i7core_printk(KERN_INFO,
2284                      "Driver loaded, %d memory controller(s) found.\n",
2285                      count);
2286
2287        mutex_unlock(&i7core_edac_lock);
2288        return 0;
2289
2290fail1:
2291        list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2292                i7core_unregister_mci(i7core_dev);
2293
2294        i7core_put_all_devices();
2295fail0:
2296        mutex_unlock(&i7core_edac_lock);
2297        return rc;
2298}
2299
2300/*
2301 *      i7core_remove   destructor for one instance of device
2302 *
2303 */
2304static void i7core_remove(struct pci_dev *pdev)
2305{
2306        struct i7core_dev *i7core_dev;
2307
2308        edac_dbg(0, "\n");
2309
2310        /*
2311         * we have a trouble here: pdev value for removal will be wrong, since
2312         * it will point to the X58 register used to detect that the machine
2313         * is a Nehalem or upper design. However, due to the way several PCI
2314         * devices are grouped together to provide MC functionality, we need
2315         * to use a different method for releasing the devices
2316         */
2317
2318        mutex_lock(&i7core_edac_lock);
2319
2320        if (unlikely(!probed)) {
2321                mutex_unlock(&i7core_edac_lock);
2322                return;
2323        }
2324
2325        list_for_each_entry(i7core_dev, &i7core_edac_list, list)
2326                i7core_unregister_mci(i7core_dev);
2327
2328        /* Release PCI resources */
2329        i7core_put_all_devices();
2330
2331        probed--;
2332
2333        mutex_unlock(&i7core_edac_lock);
2334}
2335
2336MODULE_DEVICE_TABLE(pci, i7core_pci_tbl);
2337
2338/*
2339 *      i7core_driver   pci_driver structure for this module
2340 *
2341 */
2342static struct pci_driver i7core_driver = {
2343        .name     = "i7core_edac",
2344        .probe    = i7core_probe,
2345        .remove   = i7core_remove,
2346        .id_table = i7core_pci_tbl,
2347};
2348
2349/*
2350 *      i7core_init             Module entry function
2351 *                      Try to initialize this module for its devices
2352 */
2353static int __init i7core_init(void)
2354{
2355        int pci_rc;
2356
2357        edac_dbg(2, "\n");
2358
2359        /* Ensure that the OPSTATE is set correctly for POLL or NMI */
2360        opstate_init();
2361
2362        if (use_pci_fixup)
2363                i7core_xeon_pci_fixup(pci_dev_table);
2364
2365        pci_rc = pci_register_driver(&i7core_driver);
2366
2367        if (pci_rc >= 0) {
2368                mce_register_decode_chain(&i7_mce_dec);
2369                return 0;
2370        }
2371
2372        i7core_printk(KERN_ERR, "Failed to register device with error %d.\n",
2373                      pci_rc);
2374
2375        return pci_rc;
2376}
2377
2378/*
2379 *      i7core_exit()   Module exit function
2380 *                      Unregister the driver
2381 */
2382static void __exit i7core_exit(void)
2383{
2384        edac_dbg(2, "\n");
2385        pci_unregister_driver(&i7core_driver);
2386        mce_unregister_decode_chain(&i7_mce_dec);
2387}
2388
2389module_init(i7core_init);
2390module_exit(i7core_exit);
2391
2392MODULE_LICENSE("GPL");
2393MODULE_AUTHOR("Mauro Carvalho Chehab");
2394MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
2395MODULE_DESCRIPTION("MC Driver for Intel i7 Core memory controllers - "
2396                   I7CORE_REVISION);
2397
2398module_param(edac_op_state, int, 0444);
2399MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
2400