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