include/linux/edac.h

   1 /*
   2  * Generic EDAC defs
   3  *
   4  * Author: Dave Jiang <djiang@mvista.com>
   5  *
   6  * 2006-2008 (c) MontaVista Software, Inc. This file is licensed under
   7  * the terms of the GNU General Public License version 2. This program
   8  * is licensed "as is" without any warranty of any kind, whether express
   9  * or implied.
  10  *
  11  */
  12 #ifndef _LINUX_EDAC_H_
  13 #define _LINUX_EDAC_H_
  14
  15 #include <linux/atomic.h>
  16 #include <linux/device.h>
  17 #include <linux/completion.h>
  18 #include <linux/workqueue.h>
  19 #include <linux/debugfs.h>
  20 #include <linux/numa.h>
  21
  22 #define EDAC_DEVICE_NAME_LEN    31
  23
  24 struct device;
  25
  26 #define EDAC_OPSTATE_INVAL      -1
  27 #define EDAC_OPSTATE_POLL       0
  28 #define EDAC_OPSTATE_NMI        1
  29 #define EDAC_OPSTATE_INT        2
  30
  31 extern int edac_op_state;
  32
  33 struct bus_type *edac_get_sysfs_subsys(void);
  34 int edac_get_report_status(void);
  35 void edac_set_report_status(int new);
  36
  37 enum {
  38         EDAC_REPORTING_ENABLED,
  39         EDAC_REPORTING_DISABLED,
  40         EDAC_REPORTING_FORCE
  41 };
  42
  43 static inline void opstate_init(void)
  44 {
  45         switch (edac_op_state) {
  46         case EDAC_OPSTATE_POLL:
  47         case EDAC_OPSTATE_NMI:
  48                 break;
  49         default:
  50                 edac_op_state = EDAC_OPSTATE_POLL;
  51         }
  52         return;
  53 }
  54
  55 /* Max length of a DIMM label*/
  56 #define EDAC_MC_LABEL_LEN       31
  57
  58 /* Maximum size of the location string */
  59 #define LOCATION_SIZE 256
  60
  61 /* Defines the maximum number of labels that can be reported */
  62 #define EDAC_MAX_LABELS         8
  63
  64 /* String used to join two or more labels */
  65 #define OTHER_LABEL " or "
  66
  67 /**
  68  * enum dev_type - describe the type of memory DRAM chips used at the stick
  69  * @DEV_UNKNOWN:        Can't be determined, or MC doesn't support detect it
  70  * @DEV_X1:             1 bit for data
  71  * @DEV_X2:             2 bits for data
  72  * @DEV_X4:             4 bits for data
  73  * @DEV_X8:             8 bits for data
  74  * @DEV_X16:            16 bits for data
  75  * @DEV_X32:            32 bits for data
  76  * @DEV_X64:            64 bits for data
  77  *
  78  * Typical values are x4 and x8.
  79  */
  80 enum dev_type {
  81         DEV_UNKNOWN = 0,
  82         DEV_X1,
  83         DEV_X2,
  84         DEV_X4,
  85         DEV_X8,
  86         DEV_X16,
  87         DEV_X32,                /* Do these parts exist? */
  88         DEV_X64                 /* Do these parts exist? */
  89 };
  90
  91 #define DEV_FLAG_UNKNOWN        BIT(DEV_UNKNOWN)
  92 #define DEV_FLAG_X1             BIT(DEV_X1)
  93 #define DEV_FLAG_X2             BIT(DEV_X2)
  94 #define DEV_FLAG_X4             BIT(DEV_X4)
  95 #define DEV_FLAG_X8             BIT(DEV_X8)
  96 #define DEV_FLAG_X16            BIT(DEV_X16)
  97 #define DEV_FLAG_X32            BIT(DEV_X32)
  98 #define DEV_FLAG_X64            BIT(DEV_X64)
  99
 100 /**
 101  * enum hw_event_mc_err_type - type of the detected error
 102  *
 103  * @HW_EVENT_ERR_CORRECTED:     Corrected Error - Indicates that an ECC
 104  *                              corrected error was detected
 105  * @HW_EVENT_ERR_UNCORRECTED:   Uncorrected Error - Indicates an error that
 106  *                              can't be corrected by ECC, but it is not
 107  *                              fatal (maybe it is on an unused memory area,
 108  *                              or the memory controller could recover from
 109  *                              it for example, by re-trying the operation).
 110  * @HW_EVENT_ERR_DEFERRED:      Deferred Error - Indicates an uncorrectable
 111  *                              error whose handling is not urgent. This could
 112  *                              be due to hardware data poisoning where the
 113  *                              system can continue operation until the poisoned
 114  *                              data is consumed. Preemptive measures may also
 115  *                              be taken, e.g. offlining pages, etc.
 116  * @HW_EVENT_ERR_FATAL:         Fatal Error - Uncorrected error that could not
 117  *                              be recovered.
 118  * @HW_EVENT_ERR_INFO:          Informational - The CPER spec defines a forth
 119  *                              type of error: informational logs.
 120  */
 121 enum hw_event_mc_err_type {
 122         HW_EVENT_ERR_CORRECTED,
 123         HW_EVENT_ERR_UNCORRECTED,
 124         HW_EVENT_ERR_DEFERRED,
 125         HW_EVENT_ERR_FATAL,
 126         HW_EVENT_ERR_INFO,
 127 };
 128
 129 static inline char *mc_event_error_type(const unsigned int err_type)
 130 {
 131         switch (err_type) {
 132         case HW_EVENT_ERR_CORRECTED:
 133                 return "Corrected";
 134         case HW_EVENT_ERR_UNCORRECTED:
 135                 return "Uncorrected";
 136         case HW_EVENT_ERR_DEFERRED:
 137                 return "Deferred";
 138         case HW_EVENT_ERR_FATAL:
 139                 return "Fatal";
 140         default:
 141         case HW_EVENT_ERR_INFO:
 142                 return "Info";
 143         }
 144 }
 145
 146 /**
 147  * enum mem_type - memory types. For a more detailed reference, please see
 148  *                      http://en.wikipedia.org/wiki/DRAM
 149  *
 150  * @MEM_EMPTY:          Empty csrow
 151  * @MEM_RESERVED:       Reserved csrow type
 152  * @MEM_UNKNOWN:        Unknown csrow type
 153  * @MEM_FPM:            FPM - Fast Page Mode, used on systems up to 1995.
 154  * @MEM_EDO:            EDO - Extended data out, used on systems up to 1998.
 155  * @MEM_BEDO:           BEDO - Burst Extended data out, an EDO variant.
 156  * @MEM_SDR:            SDR - Single data rate SDRAM
 157  *                      http://en.wikipedia.org/wiki/Synchronous_dynamic_random-access_memory
 158  *                      They use 3 pins for chip select: Pins 0 and 2 are
 159  *                      for rank 0; pins 1 and 3 are for rank 1, if the memory
 160  *                      is dual-rank.
 161  * @MEM_RDR:            Registered SDR SDRAM
 162  * @MEM_DDR:            Double data rate SDRAM
 163  *                      http://en.wikipedia.org/wiki/DDR_SDRAM
 164  * @MEM_RDDR:           Registered Double data rate SDRAM
 165  *                      This is a variant of the DDR memories.
 166  *                      A registered memory has a buffer inside it, hiding
 167  *                      part of the memory details to the memory controller.
 168  * @MEM_RMBS:           Rambus DRAM, used on a few Pentium III/IV controllers.
 169  * @MEM_DDR2:           DDR2 RAM, as described at JEDEC JESD79-2F.
 170  *                      Those memories are labeled as "PC2-" instead of "PC" to
 171  *                      differentiate from DDR.
 172  * @MEM_FB_DDR2:        Fully-Buffered DDR2, as described at JEDEC Std No. 205
 173  *                      and JESD206.
 174  *                      Those memories are accessed per DIMM slot, and not by
 175  *                      a chip select signal.
 176  * @MEM_RDDR2:          Registered DDR2 RAM
 177  *                      This is a variant of the DDR2 memories.
 178  * @MEM_XDR:            Rambus XDR
 179  *                      It is an evolution of the original RAMBUS memories,
 180  *                      created to compete with DDR2. Weren't used on any
 181  *                      x86 arch, but cell_edac PPC memory controller uses it.
 182  * @MEM_DDR3:           DDR3 RAM
 183  * @MEM_RDDR3:          Registered DDR3 RAM
 184  *                      This is a variant of the DDR3 memories.
 185  * @MEM_LRDDR3:         Load-Reduced DDR3 memory.
 186  * @MEM_DDR4:           Unbuffered DDR4 RAM
 187  * @MEM_RDDR4:          Registered DDR4 RAM
 188  *                      This is a variant of the DDR4 memories.
 189  * @MEM_LRDDR4:         Load-Reduced DDR4 memory.
 190  * @MEM_NVDIMM:         Non-volatile RAM
 191  */
 192 enum mem_type {
 193         MEM_EMPTY = 0,
 194         MEM_RESERVED,
 195         MEM_UNKNOWN,
 196         MEM_FPM,
 197         MEM_EDO,
 198         MEM_BEDO,
 199         MEM_SDR,
 200         MEM_RDR,
 201         MEM_DDR,
 202         MEM_RDDR,
 203         MEM_RMBS,
 204         MEM_DDR2,
 205         MEM_FB_DDR2,
 206         MEM_RDDR2,
 207         MEM_XDR,
 208         MEM_DDR3,
 209         MEM_RDDR3,
 210         MEM_LRDDR3,
 211         MEM_DDR4,
 212         MEM_RDDR4,
 213         MEM_LRDDR4,
 214         MEM_NVDIMM,
 215 };
 216
 217 #define MEM_FLAG_EMPTY          BIT(MEM_EMPTY)
 218 #define MEM_FLAG_RESERVED       BIT(MEM_RESERVED)
 219 #define MEM_FLAG_UNKNOWN        BIT(MEM_UNKNOWN)
 220 #define MEM_FLAG_FPM            BIT(MEM_FPM)
 221 #define MEM_FLAG_EDO            BIT(MEM_EDO)
 222 #define MEM_FLAG_BEDO           BIT(MEM_BEDO)
 223 #define MEM_FLAG_SDR            BIT(MEM_SDR)
 224 #define MEM_FLAG_RDR            BIT(MEM_RDR)
 225 #define MEM_FLAG_DDR            BIT(MEM_DDR)
 226 #define MEM_FLAG_RDDR           BIT(MEM_RDDR)
 227 #define MEM_FLAG_RMBS           BIT(MEM_RMBS)
 228 #define MEM_FLAG_DDR2           BIT(MEM_DDR2)
 229 #define MEM_FLAG_FB_DDR2        BIT(MEM_FB_DDR2)
 230 #define MEM_FLAG_RDDR2          BIT(MEM_RDDR2)
 231 #define MEM_FLAG_XDR            BIT(MEM_XDR)
 232 #define MEM_FLAG_DDR3           BIT(MEM_DDR3)
 233 #define MEM_FLAG_RDDR3          BIT(MEM_RDDR3)
 234 #define MEM_FLAG_DDR4           BIT(MEM_DDR4)
 235 #define MEM_FLAG_RDDR4          BIT(MEM_RDDR4)
 236 #define MEM_FLAG_LRDDR4         BIT(MEM_LRDDR4)
 237 #define MEM_FLAG_NVDIMM         BIT(MEM_NVDIMM)
 238
 239 /**
 240  * enum edac-type - Error Detection and Correction capabilities and mode
 241  * @EDAC_UNKNOWN:       Unknown if ECC is available
 242  * @EDAC_NONE:          Doesn't support ECC
 243  * @EDAC_RESERVED:      Reserved ECC type
 244  * @EDAC_PARITY:        Detects parity errors
 245  * @EDAC_EC:            Error Checking - no correction
 246  * @EDAC_SECDED:        Single bit error correction, Double detection
 247  * @EDAC_S2ECD2ED:      Chipkill x2 devices - do these exist?
 248  * @EDAC_S4ECD4ED:      Chipkill x4 devices
 249  * @EDAC_S8ECD8ED:      Chipkill x8 devices
 250  * @EDAC_S16ECD16ED:    Chipkill x16 devices
 251  */
 252 enum edac_type {
 253         EDAC_UNKNOWN =  0,
 254         EDAC_NONE,
 255         EDAC_RESERVED,
 256         EDAC_PARITY,
 257         EDAC_EC,
 258         EDAC_SECDED,
 259         EDAC_S2ECD2ED,
 260         EDAC_S4ECD4ED,
 261         EDAC_S8ECD8ED,
 262         EDAC_S16ECD16ED,
 263 };
 264
 265 #define EDAC_FLAG_UNKNOWN       BIT(EDAC_UNKNOWN)
 266 #define EDAC_FLAG_NONE          BIT(EDAC_NONE)
 267 #define EDAC_FLAG_PARITY        BIT(EDAC_PARITY)
 268 #define EDAC_FLAG_EC            BIT(EDAC_EC)
 269 #define EDAC_FLAG_SECDED        BIT(EDAC_SECDED)
 270 #define EDAC_FLAG_S2ECD2ED      BIT(EDAC_S2ECD2ED)
 271 #define EDAC_FLAG_S4ECD4ED      BIT(EDAC_S4ECD4ED)
 272 #define EDAC_FLAG_S8ECD8ED      BIT(EDAC_S8ECD8ED)
 273 #define EDAC_FLAG_S16ECD16ED    BIT(EDAC_S16ECD16ED)
 274
 275 /**
 276  * enum scrub_type - scrubbing capabilities
 277  * @SCRUB_UNKNOWN:              Unknown if scrubber is available
 278  * @SCRUB_NONE:                 No scrubber
 279  * @SCRUB_SW_PROG:              SW progressive (sequential) scrubbing
 280  * @SCRUB_SW_SRC:               Software scrub only errors
 281  * @SCRUB_SW_PROG_SRC:          Progressive software scrub from an error
 282  * @SCRUB_SW_TUNABLE:           Software scrub frequency is tunable
 283  * @SCRUB_HW_PROG:              HW progressive (sequential) scrubbing
 284  * @SCRUB_HW_SRC:               Hardware scrub only errors
 285  * @SCRUB_HW_PROG_SRC:          Progressive hardware scrub from an error
 286  * @SCRUB_HW_TUNABLE:           Hardware scrub frequency is tunable
 287  */
 288 enum scrub_type {
 289         SCRUB_UNKNOWN = 0,
 290         SCRUB_NONE,
 291         SCRUB_SW_PROG,
 292         SCRUB_SW_SRC,
 293         SCRUB_SW_PROG_SRC,
 294         SCRUB_SW_TUNABLE,
 295         SCRUB_HW_PROG,
 296         SCRUB_HW_SRC,
 297         SCRUB_HW_PROG_SRC,
 298         SCRUB_HW_TUNABLE
 299 };
 300
 301 #define SCRUB_FLAG_SW_PROG      BIT(SCRUB_SW_PROG)
 302 #define SCRUB_FLAG_SW_SRC       BIT(SCRUB_SW_SRC)
 303 #define SCRUB_FLAG_SW_PROG_SRC  BIT(SCRUB_SW_PROG_SRC)
 304 #define SCRUB_FLAG_SW_TUN       BIT(SCRUB_SW_SCRUB_TUNABLE)
 305 #define SCRUB_FLAG_HW_PROG      BIT(SCRUB_HW_PROG)
 306 #define SCRUB_FLAG_HW_SRC       BIT(SCRUB_HW_SRC)
 307 #define SCRUB_FLAG_HW_PROG_SRC  BIT(SCRUB_HW_PROG_SRC)
 308 #define SCRUB_FLAG_HW_TUN       BIT(SCRUB_HW_TUNABLE)
 309
 310 /* FIXME - should have notify capabilities: NMI, LOG, PROC, etc */
 311
 312 /* EDAC internal operation states */
 313 #define OP_ALLOC                0x100
 314 #define OP_RUNNING_POLL         0x201
 315 #define OP_RUNNING_INTERRUPT    0x202
 316 #define OP_RUNNING_POLL_INTR    0x203
 317 #define OP_OFFLINE              0x300
 318
 319 /**
 320  * enum edac_mc_layer - memory controller hierarchy layer
 321  *
 322  * @EDAC_MC_LAYER_BRANCH:       memory layer is named "branch"
 323  * @EDAC_MC_LAYER_CHANNEL:      memory layer is named "channel"
 324  * @EDAC_MC_LAYER_SLOT:         memory layer is named "slot"
 325  * @EDAC_MC_LAYER_CHIP_SELECT:  memory layer is named "chip select"
 326  * @EDAC_MC_LAYER_ALL_MEM:      memory layout is unknown. All memory is mapped
 327  *                              as a single memory area. This is used when
 328  *                              retrieving errors from a firmware driven driver.
 329  *
 330  * This enum is used by the drivers to tell edac_mc_sysfs what name should
 331  * be used when describing a memory stick location.
 332  */
 333 enum edac_mc_layer_type {
 334         EDAC_MC_LAYER_BRANCH,
 335         EDAC_MC_LAYER_CHANNEL,
 336         EDAC_MC_LAYER_SLOT,
 337         EDAC_MC_LAYER_CHIP_SELECT,
 338         EDAC_MC_LAYER_ALL_MEM,
 339 };
 340
 341 /**
 342  * struct edac_mc_layer - describes the memory controller hierarchy
 343  * @type:               layer type
 344  * @size:               number of components per layer. For example,
 345  *                      if the channel layer has two channels, size = 2
 346  * @is_virt_csrow:      This layer is part of the "csrow" when old API
 347  *                      compatibility mode is enabled. Otherwise, it is
 348  *                      a channel
 349  */
 350 struct edac_mc_layer {
 351         enum edac_mc_layer_type type;
 352         unsigned                size;
 353         bool                    is_virt_csrow;
 354 };
 355
 356 /*
 357  * Maximum number of layers used by the memory controller to uniquely
 358  * identify a single memory stick.
 359  * NOTE: Changing this constant requires not only to change the constant
 360  * below, but also to change the existing code at the core, as there are
 361  * some code there that are optimized for 3 layers.
 362  */
 363 #define EDAC_MAX_LAYERS         3
 364
 365 /**
 366  * EDAC_DIMM_OFF - Macro responsible to get a pointer offset inside a pointer
 367  *                 array for the element given by [layer0,layer1,layer2]
 368  *                 position
 369  *
 370  * @layers:     a struct edac_mc_layer array, describing how many elements
 371  *              were allocated for each layer
 372  * @nlayers:    Number of layers at the @layers array
 373  * @layer0:     layer0 position
 374  * @layer1:     layer1 position. Unused if n_layers < 2
 375  * @layer2:     layer2 position. Unused if n_layers < 3
 376  *
 377  * For 1 layer, this macro returns "var[layer0] - var";
 378  *
 379  * For 2 layers, this macro is similar to allocate a bi-dimensional array
 380  * and to return "var[layer0][layer1] - var";
 381  *
 382  * For 3 layers, this macro is similar to allocate a tri-dimensional array
 383  * and to return "var[layer0][layer1][layer2] - var".
 384  *
 385  * A loop could be used here to make it more generic, but, as we only have
 386  * 3 layers, this is a little faster.
 387  *
 388  * By design, layers can never be 0 or more than 3. If that ever happens,
 389  * a NULL is returned, causing an OOPS during the memory allocation routine,
 390  * with would point to the developer that he's doing something wrong.
 391  */
 392 #define EDAC_DIMM_OFF(layers, nlayers, layer0, layer1, layer2) ({               \
 393         int __i;                                                        \
 394         if ((nlayers) == 1)                                             \
 395                 __i = layer0;                                           \
 396         else if ((nlayers) == 2)                                        \
 397                 __i = (layer1) + ((layers[1]).size * (layer0));         \
 398         else if ((nlayers) == 3)                                        \
 399                 __i = (layer2) + ((layers[2]).size * ((layer1) +        \
 400                             ((layers[1]).size * (layer0))));            \
 401         else                                                            \
 402                 __i = -EINVAL;                                          \
 403         __i;                                                            \
 404 })
 405
 406 /**
 407  * EDAC_DIMM_PTR - Macro responsible to get a pointer inside a pointer array
 408  *                 for the element given by [layer0,layer1,layer2] position
 409  *
 410  * @layers:     a struct edac_mc_layer array, describing how many elements
 411  *              were allocated for each layer
 412  * @var:        name of the var where we want to get the pointer
 413  *              (like mci->dimms)
 414  * @nlayers:    Number of layers at the @layers array
 415  * @layer0:     layer0 position
 416  * @layer1:     layer1 position. Unused if n_layers < 2
 417  * @layer2:     layer2 position. Unused if n_layers < 3
 418  *
 419  * For 1 layer, this macro returns "var[layer0]";
 420  *
 421  * For 2 layers, this macro is similar to allocate a bi-dimensional array
 422  * and to return "var[layer0][layer1]";
 423  *
 424  * For 3 layers, this macro is similar to allocate a tri-dimensional array
 425  * and to return "var[layer0][layer1][layer2]";
 426  */
 427 #define EDAC_DIMM_PTR(layers, var, nlayers, layer0, layer1, layer2) ({  \
 428         typeof(*var) __p;                                               \
 429         int ___i = EDAC_DIMM_OFF(layers, nlayers, layer0, layer1, layer2);      \
 430         if (___i < 0)                                                   \
 431                 __p = NULL;                                             \
 432         else                                                            \
 433                 __p = (var)[___i];                                      \
 434         __p;                                                            \
 435 })
 436
 437 struct dimm_info {
 438         struct device dev;
 439
 440         char label[EDAC_MC_LABEL_LEN + 1];      /* DIMM label on motherboard */
 441
 442         /* Memory location data */
 443         unsigned location[EDAC_MAX_LAYERS];
 444
 445         struct mem_ctl_info *mci;       /* the parent */
 446
 447         u32 grain;              /* granularity of reported error in bytes */
 448         enum dev_type dtype;    /* memory device type */
 449         enum mem_type mtype;    /* memory dimm type */
 450         enum edac_type edac_mode;       /* EDAC mode for this dimm */
 451
 452         u32 nr_pages;                   /* number of pages on this dimm */
 453
 454         unsigned csrow, cschannel;      /* Points to the old API data */
 455
 456         u16 smbios_handle;              /* Handle for SMBIOS type 17 */
 457 };
 458
 459 /**
 460  * struct rank_info - contains the information for one DIMM rank
 461  *
 462  * @chan_idx:   channel number where the rank is (typically, 0 or 1)
 463  * @ce_count:   number of correctable errors for this rank
 464  * @csrow:      A pointer to the chip select row structure (the parent
 465  *              structure). The location of the rank is given by
 466  *              the (csrow->csrow_idx, chan_idx) vector.
 467  * @dimm:       A pointer to the DIMM structure, where the DIMM label
 468  *              information is stored.
 469  *
 470  * FIXME: Currently, the EDAC core model will assume one DIMM per rank.
 471  *        This is a bad assumption, but it makes this patch easier. Later
 472  *        patches in this series will fix this issue.
 473  */
 474 struct rank_info {
 475         int chan_idx;
 476         struct csrow_info *csrow;
 477         struct dimm_info *dimm;
 478
 479         u32 ce_count;           /* Correctable Errors for this csrow */
 480 };
 481
 482 struct csrow_info {
 483         struct device dev;
 484
 485         /* Used only by edac_mc_find_csrow_by_page() */
 486         unsigned long first_page;       /* first page number in csrow */
 487         unsigned long last_page;        /* last page number in csrow */
 488         unsigned long page_mask;        /* used for interleaving -
 489                                          * 0UL for non intlv */
 490
 491         int csrow_idx;                  /* the chip-select row */
 492
 493         u32 ue_count;           /* Uncorrectable Errors for this csrow */
 494         u32 ce_count;           /* Correctable Errors for this csrow */
 495
 496         struct mem_ctl_info *mci;       /* the parent */
 497
 498         /* channel information for this csrow */
 499         u32 nr_channels;
 500         struct rank_info **channels;
 501 };
 502
 503 /*
 504  * struct errcount_attribute - used to store the several error counts
 505  */
 506 struct errcount_attribute_data {
 507         int n_layers;
 508         int pos[EDAC_MAX_LAYERS];
 509         int layer0, layer1, layer2;
 510 };
 511
 512 /**
 513  * struct edac_raw_error_desc - Raw error report structure
 514  * @grain:                      minimum granularity for an error report, in bytes
 515  * @error_count:                number of errors of the same type
 516  * @top_layer:                  top layer of the error (layer[0])
 517  * @mid_layer:                  middle layer of the error (layer[1])
 518  * @low_layer:                  low layer of the error (layer[2])
 519  * @page_frame_number:          page where the error happened
 520  * @offset_in_page:             page offset
 521  * @syndrome:                   syndrome of the error (or 0 if unknown or if
 522  *                              the syndrome is not applicable)
 523  * @msg:                        error message
 524  * @location:                   location of the error
 525  * @label:                      label of the affected DIMM(s)
 526  * @other_detail:               other driver-specific detail about the error
 527  * @enable_per_layer_report:    if false, the error affects all layers
 528  *                              (typically, a memory controller error)
 529  */
 530 struct edac_raw_error_desc {
 531         /*
 532          * NOTE: everything before grain won't be cleaned by
 533          * edac_raw_error_desc_clean()
 534          */
 535         char location[LOCATION_SIZE];
 536         char label[(EDAC_MC_LABEL_LEN + 1 + sizeof(OTHER_LABEL)) * EDAC_MAX_LABELS];
 537         long grain;
 538
 539         /* the vars below and grain will be cleaned on every new error report */
 540         u16 error_count;
 541         int top_layer;
 542         int mid_layer;
 543         int low_layer;
 544         unsigned long page_frame_number;
 545         unsigned long offset_in_page;
 546         unsigned long syndrome;
 547         const char *msg;
 548         const char *other_detail;
 549         bool enable_per_layer_report;
 550 };
 551
 552 /* MEMORY controller information structure
 553  */
 554 struct mem_ctl_info {
 555         struct device                   dev;
 556         struct bus_type                 *bus;
 557
 558         struct list_head link;  /* for global list of mem_ctl_info structs */
 559
 560         struct module *owner;   /* Module owner of this control struct */
 561
 562         unsigned long mtype_cap;        /* memory types supported by mc */
 563         unsigned long edac_ctl_cap;     /* Mem controller EDAC capabilities */
 564         unsigned long edac_cap; /* configuration capabilities - this is
 565                                  * closely related to edac_ctl_cap.  The
 566                                  * difference is that the controller may be
 567                                  * capable of s4ecd4ed which would be listed
 568                                  * in edac_ctl_cap, but if channels aren't
 569                                  * capable of s4ecd4ed then the edac_cap would
 570                                  * not have that capability.
 571                                  */
 572         unsigned long scrub_cap;        /* chipset scrub capabilities */
 573         enum scrub_type scrub_mode;     /* current scrub mode */
 574
 575         /* Translates sdram memory scrub rate given in bytes/sec to the
 576            internal representation and configures whatever else needs
 577            to be configured.
 578          */
 579         int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 bw);
 580
 581         /* Get the current sdram memory scrub rate from the internal
 582            representation and converts it to the closest matching
 583            bandwidth in bytes/sec.
 584          */
 585         int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci);
 586
 587
 588         /* pointer to edac checking routine */
 589         void (*edac_check) (struct mem_ctl_info * mci);
 590
 591         /*
 592          * Remaps memory pages: controller pages to physical pages.
 593          * For most MC's, this will be NULL.
 594          */
 595         /* FIXME - why not send the phys page to begin with? */
 596         unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci,
 597                                            unsigned long page);
 598         int mc_idx;
 599         struct csrow_info **csrows;
 600         unsigned nr_csrows, num_cschannel;
 601
 602         /*
 603          * Memory Controller hierarchy
 604          *
 605          * There are basically two types of memory controller: the ones that
 606          * sees memory sticks ("dimms"), and the ones that sees memory ranks.
 607          * All old memory controllers enumerate memories per rank, but most
 608          * of the recent drivers enumerate memories per DIMM, instead.
 609          * When the memory controller is per rank, csbased is true.
 610          */
 611         unsigned n_layers;
 612         struct edac_mc_layer *layers;
 613         bool csbased;
 614
 615         /*
 616          * DIMM info. Will eventually remove the entire csrows_info some day
 617          */
 618         unsigned tot_dimms;
 619         struct dimm_info **dimms;
 620
 621         /*
 622          * FIXME - what about controllers on other busses? - IDs must be
 623          * unique.  dev pointer should be sufficiently unique, but
 624          * BUS:SLOT.FUNC numbers may not be unique.
 625          */
 626         struct device *pdev;
 627         const char *mod_name;
 628         const char *ctl_name;
 629         const char *dev_name;
 630         void *pvt_info;
 631         unsigned long start_time;       /* mci load start time (in jiffies) */
 632
 633         /*
 634          * drivers shouldn't access those fields directly, as the core
 635          * already handles that.
 636          */
 637         u32 ce_noinfo_count, ue_noinfo_count;
 638         u32 ue_mc, ce_mc;
 639         u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS];
 640
 641         struct completion complete;
 642
 643         /* Additional top controller level attributes, but specified
 644          * by the low level driver.
 645          *
 646          * Set by the low level driver to provide attributes at the
 647          * controller level.
 648          * An array of structures, NULL terminated
 649          *
 650          * If attributes are desired, then set to array of attributes
 651          * If no attributes are desired, leave NULL
 652          */
 653         const struct mcidev_sysfs_attribute *mc_driver_sysfs_attributes;
 654
 655         /* work struct for this MC */
 656         struct delayed_work work;
 657
 658         /*
 659          * Used to report an error - by being at the global struct
 660          * makes the memory allocated by the EDAC core
 661          */
 662         struct edac_raw_error_desc error_desc;
 663
 664         /* the internal state of this controller instance */
 665         int op_state;
 666
 667         struct dentry *debugfs;
 668         u8 fake_inject_layer[EDAC_MAX_LAYERS];
 669         bool fake_inject_ue;
 670         u16 fake_inject_count;
 671 };
 672
 673 /*
 674  * Maximum number of memory controllers in the coherent fabric.
 675  */
 676 #define EDAC_MAX_MCS    2 * MAX_NUMNODES
 677
 678 #endif