blob: 096b7fcdf484e292296ab3fd86216c026a188eb3 [file] [log] [blame]
Dave Jiangc0d12172007-07-19 01:49:46 -07001/*
2 * Generic EDAC defs
3 *
4 * Author: Dave Jiang <djiang@mvista.com>
5 *
Hitoshi Mitakec3c52bc2008-04-29 01:03:18 -07006 * 2006-2008 (c) MontaVista Software, Inc. This file is licensed under
Dave Jiangc0d12172007-07-19 01:49:46 -07007 * 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
Arun Sharma600634972011-07-26 16:09:06 -070015#include <linux/atomic.h>
Mauro Carvalho Chehab7a623c02012-04-16 16:41:11 -030016#include <linux/device.h>
Paul Gortmaker313162d2012-01-30 11:46:54 -050017#include <linux/completion.h>
18#include <linux/workqueue.h>
Mauro Carvalho Chehab452a6bf2012-03-26 09:35:11 -030019#include <linux/debugfs.h>
Paul Gortmaker313162d2012-01-30 11:46:54 -050020
21struct device;
Dave Jiangc0d12172007-07-19 01:49:46 -070022
23#define EDAC_OPSTATE_INVAL -1
24#define EDAC_OPSTATE_POLL 0
25#define EDAC_OPSTATE_NMI 1
26#define EDAC_OPSTATE_INT 2
27
28extern int edac_op_state;
Dave Jiang66ee2f92007-07-19 01:49:54 -070029extern int edac_err_assert;
Dave Jiangc0d12172007-07-19 01:49:46 -070030extern atomic_t edac_handlers;
Kay Sieversfe5ff8b2011-12-14 15:21:07 -080031extern struct bus_type edac_subsys;
Dave Jiangc0d12172007-07-19 01:49:46 -070032
33extern int edac_handler_set(void);
34extern void edac_atomic_assert_error(void);
Kay Sieversfe5ff8b2011-12-14 15:21:07 -080035extern struct bus_type *edac_get_sysfs_subsys(void);
36extern void edac_put_sysfs_subsys(void);
Dave Jiangc0d12172007-07-19 01:49:46 -070037
Hitoshi Mitakec3c52bc2008-04-29 01:03:18 -070038static inline void opstate_init(void)
39{
40 switch (edac_op_state) {
41 case EDAC_OPSTATE_POLL:
42 case EDAC_OPSTATE_NMI:
43 break;
44 default:
45 edac_op_state = EDAC_OPSTATE_POLL;
46 }
47 return;
48}
49
Mauro Carvalho Chehabc7ef7642013-02-21 13:36:45 -030050/* Max length of a DIMM label*/
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -030051#define EDAC_MC_LABEL_LEN 31
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -030052
Mauro Carvalho Chehabc7ef7642013-02-21 13:36:45 -030053/* Maximum size of the location string */
54#define LOCATION_SIZE 80
55
56/* Defines the maximum number of labels that can be reported */
57#define EDAC_MAX_LABELS 8
58
59/* String used to join two or more labels */
60#define OTHER_LABEL " or "
61
Mauro Carvalho Chehabb0610bb82012-03-21 16:21:07 -030062/**
63 * enum dev_type - describe the type of memory DRAM chips used at the stick
64 * @DEV_UNKNOWN: Can't be determined, or MC doesn't support detect it
65 * @DEV_X1: 1 bit for data
66 * @DEV_X2: 2 bits for data
67 * @DEV_X4: 4 bits for data
68 * @DEV_X8: 8 bits for data
69 * @DEV_X16: 16 bits for data
70 * @DEV_X32: 32 bits for data
71 * @DEV_X64: 64 bits for data
72 *
73 * Typical values are x4 and x8.
74 */
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -030075enum dev_type {
76 DEV_UNKNOWN = 0,
77 DEV_X1,
78 DEV_X2,
79 DEV_X4,
80 DEV_X8,
81 DEV_X16,
82 DEV_X32, /* Do these parts exist? */
83 DEV_X64 /* Do these parts exist? */
84};
85
86#define DEV_FLAG_UNKNOWN BIT(DEV_UNKNOWN)
87#define DEV_FLAG_X1 BIT(DEV_X1)
88#define DEV_FLAG_X2 BIT(DEV_X2)
89#define DEV_FLAG_X4 BIT(DEV_X4)
90#define DEV_FLAG_X8 BIT(DEV_X8)
91#define DEV_FLAG_X16 BIT(DEV_X16)
92#define DEV_FLAG_X32 BIT(DEV_X32)
93#define DEV_FLAG_X64 BIT(DEV_X64)
94
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -030095/**
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -030096 * enum hw_event_mc_err_type - type of the detected error
97 *
98 * @HW_EVENT_ERR_CORRECTED: Corrected Error - Indicates that an ECC
99 * corrected error was detected
100 * @HW_EVENT_ERR_UNCORRECTED: Uncorrected Error - Indicates an error that
101 * can't be corrected by ECC, but it is not
102 * fatal (maybe it is on an unused memory area,
103 * or the memory controller could recover from
104 * it for example, by re-trying the operation).
105 * @HW_EVENT_ERR_FATAL: Fatal Error - Uncorrected error that could not
106 * be recovered.
107 */
108enum hw_event_mc_err_type {
109 HW_EVENT_ERR_CORRECTED,
110 HW_EVENT_ERR_UNCORRECTED,
111 HW_EVENT_ERR_FATAL,
112};
113
114/**
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300115 * enum mem_type - memory types. For a more detailed reference, please see
116 * http://en.wikipedia.org/wiki/DRAM
117 *
118 * @MEM_EMPTY Empty csrow
119 * @MEM_RESERVED: Reserved csrow type
120 * @MEM_UNKNOWN: Unknown csrow type
121 * @MEM_FPM: FPM - Fast Page Mode, used on systems up to 1995.
122 * @MEM_EDO: EDO - Extended data out, used on systems up to 1998.
123 * @MEM_BEDO: BEDO - Burst Extended data out, an EDO variant.
124 * @MEM_SDR: SDR - Single data rate SDRAM
125 * http://en.wikipedia.org/wiki/Synchronous_dynamic_random-access_memory
126 * They use 3 pins for chip select: Pins 0 and 2 are
127 * for rank 0; pins 1 and 3 are for rank 1, if the memory
128 * is dual-rank.
129 * @MEM_RDR: Registered SDR SDRAM
130 * @MEM_DDR: Double data rate SDRAM
131 * http://en.wikipedia.org/wiki/DDR_SDRAM
132 * @MEM_RDDR: Registered Double data rate SDRAM
133 * This is a variant of the DDR memories.
134 * A registered memory has a buffer inside it, hiding
135 * part of the memory details to the memory controller.
136 * @MEM_RMBS: Rambus DRAM, used on a few Pentium III/IV controllers.
137 * @MEM_DDR2: DDR2 RAM, as described at JEDEC JESD79-2F.
138 * Those memories are labed as "PC2-" instead of "PC" to
139 * differenciate from DDR.
140 * @MEM_FB_DDR2: Fully-Buffered DDR2, as described at JEDEC Std No. 205
141 * and JESD206.
142 * Those memories are accessed per DIMM slot, and not by
143 * a chip select signal.
144 * @MEM_RDDR2: Registered DDR2 RAM
145 * This is a variant of the DDR2 memories.
146 * @MEM_XDR: Rambus XDR
147 * It is an evolution of the original RAMBUS memories,
148 * created to compete with DDR2. Weren't used on any
149 * x86 arch, but cell_edac PPC memory controller uses it.
150 * @MEM_DDR3: DDR3 RAM
151 * @MEM_RDDR3: Registered DDR3 RAM
152 * This is a variant of the DDR3 memories.
153 */
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300154enum mem_type {
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300155 MEM_EMPTY = 0,
156 MEM_RESERVED,
157 MEM_UNKNOWN,
158 MEM_FPM,
159 MEM_EDO,
160 MEM_BEDO,
161 MEM_SDR,
162 MEM_RDR,
163 MEM_DDR,
164 MEM_RDDR,
165 MEM_RMBS,
166 MEM_DDR2,
167 MEM_FB_DDR2,
168 MEM_RDDR2,
169 MEM_XDR,
170 MEM_DDR3,
171 MEM_RDDR3,
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300172};
173
174#define MEM_FLAG_EMPTY BIT(MEM_EMPTY)
175#define MEM_FLAG_RESERVED BIT(MEM_RESERVED)
176#define MEM_FLAG_UNKNOWN BIT(MEM_UNKNOWN)
177#define MEM_FLAG_FPM BIT(MEM_FPM)
178#define MEM_FLAG_EDO BIT(MEM_EDO)
179#define MEM_FLAG_BEDO BIT(MEM_BEDO)
180#define MEM_FLAG_SDR BIT(MEM_SDR)
181#define MEM_FLAG_RDR BIT(MEM_RDR)
182#define MEM_FLAG_DDR BIT(MEM_DDR)
183#define MEM_FLAG_RDDR BIT(MEM_RDDR)
184#define MEM_FLAG_RMBS BIT(MEM_RMBS)
185#define MEM_FLAG_DDR2 BIT(MEM_DDR2)
186#define MEM_FLAG_FB_DDR2 BIT(MEM_FB_DDR2)
187#define MEM_FLAG_RDDR2 BIT(MEM_RDDR2)
188#define MEM_FLAG_XDR BIT(MEM_XDR)
189#define MEM_FLAG_DDR3 BIT(MEM_DDR3)
190#define MEM_FLAG_RDDR3 BIT(MEM_RDDR3)
191
Mauro Carvalho Chehabb0610bb82012-03-21 16:21:07 -0300192/**
193 * enum edac-type - Error Detection and Correction capabilities and mode
194 * @EDAC_UNKNOWN: Unknown if ECC is available
195 * @EDAC_NONE: Doesn't support ECC
196 * @EDAC_RESERVED: Reserved ECC type
197 * @EDAC_PARITY: Detects parity errors
198 * @EDAC_EC: Error Checking - no correction
199 * @EDAC_SECDED: Single bit error correction, Double detection
200 * @EDAC_S2ECD2ED: Chipkill x2 devices - do these exist?
201 * @EDAC_S4ECD4ED: Chipkill x4 devices
202 * @EDAC_S8ECD8ED: Chipkill x8 devices
203 * @EDAC_S16ECD16ED: Chipkill x16 devices
204 */
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300205enum edac_type {
Mauro Carvalho Chehabb0610bb82012-03-21 16:21:07 -0300206 EDAC_UNKNOWN = 0,
207 EDAC_NONE,
208 EDAC_RESERVED,
209 EDAC_PARITY,
210 EDAC_EC,
211 EDAC_SECDED,
212 EDAC_S2ECD2ED,
213 EDAC_S4ECD4ED,
214 EDAC_S8ECD8ED,
215 EDAC_S16ECD16ED,
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300216};
217
218#define EDAC_FLAG_UNKNOWN BIT(EDAC_UNKNOWN)
219#define EDAC_FLAG_NONE BIT(EDAC_NONE)
220#define EDAC_FLAG_PARITY BIT(EDAC_PARITY)
221#define EDAC_FLAG_EC BIT(EDAC_EC)
222#define EDAC_FLAG_SECDED BIT(EDAC_SECDED)
223#define EDAC_FLAG_S2ECD2ED BIT(EDAC_S2ECD2ED)
224#define EDAC_FLAG_S4ECD4ED BIT(EDAC_S4ECD4ED)
225#define EDAC_FLAG_S8ECD8ED BIT(EDAC_S8ECD8ED)
226#define EDAC_FLAG_S16ECD16ED BIT(EDAC_S16ECD16ED)
227
Mauro Carvalho Chehabb0610bb82012-03-21 16:21:07 -0300228/**
229 * enum scrub_type - scrubbing capabilities
230 * @SCRUB_UNKNOWN Unknown if scrubber is available
231 * @SCRUB_NONE: No scrubber
232 * @SCRUB_SW_PROG: SW progressive (sequential) scrubbing
233 * @SCRUB_SW_SRC: Software scrub only errors
234 * @SCRUB_SW_PROG_SRC: Progressive software scrub from an error
235 * @SCRUB_SW_TUNABLE: Software scrub frequency is tunable
236 * @SCRUB_HW_PROG: HW progressive (sequential) scrubbing
237 * @SCRUB_HW_SRC: Hardware scrub only errors
238 * @SCRUB_HW_PROG_SRC: Progressive hardware scrub from an error
239 * SCRUB_HW_TUNABLE: Hardware scrub frequency is tunable
240 */
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300241enum scrub_type {
Mauro Carvalho Chehabb0610bb82012-03-21 16:21:07 -0300242 SCRUB_UNKNOWN = 0,
243 SCRUB_NONE,
244 SCRUB_SW_PROG,
245 SCRUB_SW_SRC,
246 SCRUB_SW_PROG_SRC,
247 SCRUB_SW_TUNABLE,
248 SCRUB_HW_PROG,
249 SCRUB_HW_SRC,
250 SCRUB_HW_PROG_SRC,
251 SCRUB_HW_TUNABLE
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300252};
253
254#define SCRUB_FLAG_SW_PROG BIT(SCRUB_SW_PROG)
255#define SCRUB_FLAG_SW_SRC BIT(SCRUB_SW_SRC)
256#define SCRUB_FLAG_SW_PROG_SRC BIT(SCRUB_SW_PROG_SRC)
257#define SCRUB_FLAG_SW_TUN BIT(SCRUB_SW_SCRUB_TUNABLE)
258#define SCRUB_FLAG_HW_PROG BIT(SCRUB_HW_PROG)
259#define SCRUB_FLAG_HW_SRC BIT(SCRUB_HW_SRC)
260#define SCRUB_FLAG_HW_PROG_SRC BIT(SCRUB_HW_PROG_SRC)
261#define SCRUB_FLAG_HW_TUN BIT(SCRUB_HW_TUNABLE)
262
263/* FIXME - should have notify capabilities: NMI, LOG, PROC, etc */
264
265/* EDAC internal operation states */
266#define OP_ALLOC 0x100
267#define OP_RUNNING_POLL 0x201
268#define OP_RUNNING_INTERRUPT 0x202
269#define OP_RUNNING_POLL_INTR 0x203
270#define OP_OFFLINE 0x300
271
272/*
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300273 * Concepts used at the EDAC subsystem
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300274 *
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300275 * There are several things to be aware of that aren't at all obvious:
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300276 *
277 * SOCKETS, SOCKET SETS, BANKS, ROWS, CHIP-SELECT ROWS, CHANNELS, etc..
278 *
279 * These are some of the many terms that are thrown about that don't always
280 * mean what people think they mean (Inconceivable!). In the interest of
281 * creating a common ground for discussion, terms and their definitions
282 * will be established.
283 *
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300284 * Memory devices: The individual DRAM chips on a memory stick. These
285 * devices commonly output 4 and 8 bits each (x4, x8).
286 * Grouping several of these in parallel provides the
287 * number of bits that the memory controller expects:
288 * typically 72 bits, in order to provide 64 bits +
289 * 8 bits of ECC data.
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300290 *
291 * Memory Stick: A printed circuit board that aggregates multiple
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300292 * memory devices in parallel. In general, this is the
293 * Field Replaceable Unit (FRU) which gets replaced, in
294 * the case of excessive errors. Most often it is also
295 * called DIMM (Dual Inline Memory Module).
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300296 *
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300297 * Memory Socket: A physical connector on the motherboard that accepts
298 * a single memory stick. Also called as "slot" on several
299 * datasheets.
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300300 *
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300301 * Channel: A memory controller channel, responsible to communicate
302 * with a group of DIMMs. Each channel has its own
303 * independent control (command) and data bus, and can
304 * be used independently or grouped with other channels.
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300305 *
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300306 * Branch: It is typically the highest hierarchy on a
307 * Fully-Buffered DIMM memory controller.
308 * Typically, it contains two channels.
309 * Two channels at the same branch can be used in single
310 * mode or in lockstep mode.
311 * When lockstep is enabled, the cacheline is doubled,
312 * but it generally brings some performance penalty.
313 * Also, it is generally not possible to point to just one
314 * memory stick when an error occurs, as the error
315 * correction code is calculated using two DIMMs instead
316 * of one. Due to that, it is capable of correcting more
317 * errors than on single mode.
318 *
319 * Single-channel: The data accessed by the memory controller is contained
320 * into one dimm only. E. g. if the data is 64 bits-wide,
321 * the data flows to the CPU using one 64 bits parallel
322 * access.
323 * Typically used with SDR, DDR, DDR2 and DDR3 memories.
324 * FB-DIMM and RAMBUS use a different concept for channel,
325 * so this concept doesn't apply there.
326 *
327 * Double-channel: The data size accessed by the memory controller is
328 * interlaced into two dimms, accessed at the same time.
329 * E. g. if the DIMM is 64 bits-wide (72 bits with ECC),
330 * the data flows to the CPU using a 128 bits parallel
331 * access.
332 *
333 * Chip-select row: This is the name of the DRAM signal used to select the
334 * DRAM ranks to be accessed. Common chip-select rows for
335 * single channel are 64 bits, for dual channel 128 bits.
336 * It may not be visible by the memory controller, as some
337 * DIMM types have a memory buffer that can hide direct
338 * access to it from the Memory Controller.
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300339 *
340 * Single-Ranked stick: A Single-ranked stick has 1 chip-select row of memory.
341 * Motherboards commonly drive two chip-select pins to
342 * a memory stick. A single-ranked stick, will occupy
343 * only one of those rows. The other will be unused.
344 *
345 * Double-Ranked stick: A double-ranked stick has two chip-select rows which
346 * access different sets of memory devices. The two
347 * rows cannot be accessed concurrently.
348 *
349 * Double-sided stick: DEPRECATED TERM, see Double-Ranked stick.
350 * A double-sided stick has two chip-select rows which
Mauro Carvalho Chehab01a6e282012-02-03 13:17:48 -0300351 * access different sets of memory devices. The two
352 * rows cannot be accessed concurrently. "Double-sided"
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300353 * is irrespective of the memory devices being mounted
354 * on both sides of the memory stick.
355 *
356 * Socket set: All of the memory sticks that are required for
357 * a single memory access or all of the memory sticks
358 * spanned by a chip-select row. A single socket set
359 * has two chip-select rows and if double-sided sticks
360 * are used these will occupy those chip-select rows.
361 *
362 * Bank: This term is avoided because it is unclear when
363 * needing to distinguish between chip-select rows and
364 * socket sets.
365 *
366 * Controller pages:
367 *
368 * Physical pages:
369 *
370 * Virtual pages:
371 *
372 *
373 * STRUCTURE ORGANIZATION AND CHOICES
374 *
375 *
376 *
377 * PS - I enjoyed writing all that about as much as you enjoyed reading it.
378 */
379
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300380/**
381 * enum edac_mc_layer - memory controller hierarchy layer
382 *
383 * @EDAC_MC_LAYER_BRANCH: memory layer is named "branch"
384 * @EDAC_MC_LAYER_CHANNEL: memory layer is named "channel"
385 * @EDAC_MC_LAYER_SLOT: memory layer is named "slot"
386 * @EDAC_MC_LAYER_CHIP_SELECT: memory layer is named "chip select"
Mauro Carvalho Chehabc66b5a72013-02-15 07:21:08 -0300387 * @EDAC_MC_LAYER_ALL_MEM: memory layout is unknown. All memory is mapped
388 * as a single memory area. This is used when
389 * retrieving errors from a firmware driven driver.
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300390 *
391 * This enum is used by the drivers to tell edac_mc_sysfs what name should
392 * be used when describing a memory stick location.
393 */
394enum edac_mc_layer_type {
395 EDAC_MC_LAYER_BRANCH,
396 EDAC_MC_LAYER_CHANNEL,
397 EDAC_MC_LAYER_SLOT,
398 EDAC_MC_LAYER_CHIP_SELECT,
Mauro Carvalho Chehabc66b5a72013-02-15 07:21:08 -0300399 EDAC_MC_LAYER_ALL_MEM,
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300400};
401
402/**
403 * struct edac_mc_layer - describes the memory controller hierarchy
404 * @layer: layer type
405 * @size: number of components per layer. For example,
406 * if the channel layer has two channels, size = 2
407 * @is_virt_csrow: This layer is part of the "csrow" when old API
408 * compatibility mode is enabled. Otherwise, it is
409 * a channel
410 */
411struct edac_mc_layer {
412 enum edac_mc_layer_type type;
413 unsigned size;
414 bool is_virt_csrow;
415};
416
417/*
418 * Maximum number of layers used by the memory controller to uniquely
419 * identify a single memory stick.
420 * NOTE: Changing this constant requires not only to change the constant
421 * below, but also to change the existing code at the core, as there are
422 * some code there that are optimized for 3 layers.
423 */
424#define EDAC_MAX_LAYERS 3
425
426/**
Mauro Carvalho Chehabde3910eb2012-04-24 15:05:43 -0300427 * EDAC_DIMM_OFF - Macro responsible to get a pointer offset inside a pointer array
428 * for the element given by [layer0,layer1,layer2] position
429 *
430 * @layers: a struct edac_mc_layer array, describing how many elements
431 * were allocated for each layer
432 * @n_layers: Number of layers at the @layers array
433 * @layer0: layer0 position
434 * @layer1: layer1 position. Unused if n_layers < 2
435 * @layer2: layer2 position. Unused if n_layers < 3
436 *
437 * For 1 layer, this macro returns &var[layer0] - &var
438 * For 2 layers, this macro is similar to allocate a bi-dimensional array
439 * and to return "&var[layer0][layer1] - &var"
440 * For 3 layers, this macro is similar to allocate a tri-dimensional array
441 * and to return "&var[layer0][layer1][layer2] - &var"
442 *
443 * A loop could be used here to make it more generic, but, as we only have
444 * 3 layers, this is a little faster.
445 * By design, layers can never be 0 or more than 3. If that ever happens,
446 * a NULL is returned, causing an OOPS during the memory allocation routine,
447 * with would point to the developer that he's doing something wrong.
448 */
449#define EDAC_DIMM_OFF(layers, nlayers, layer0, layer1, layer2) ({ \
450 int __i; \
451 if ((nlayers) == 1) \
452 __i = layer0; \
453 else if ((nlayers) == 2) \
454 __i = (layer1) + ((layers[1]).size * (layer0)); \
455 else if ((nlayers) == 3) \
456 __i = (layer2) + ((layers[2]).size * ((layer1) + \
457 ((layers[1]).size * (layer0)))); \
458 else \
459 __i = -EINVAL; \
460 __i; \
461})
462
463/**
464 * EDAC_DIMM_PTR - Macro responsible to get a pointer inside a pointer array
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300465 * for the element given by [layer0,layer1,layer2] position
466 *
467 * @layers: a struct edac_mc_layer array, describing how many elements
468 * were allocated for each layer
469 * @var: name of the var where we want to get the pointer
470 * (like mci->dimms)
471 * @n_layers: Number of layers at the @layers array
472 * @layer0: layer0 position
473 * @layer1: layer1 position. Unused if n_layers < 2
474 * @layer2: layer2 position. Unused if n_layers < 3
475 *
476 * For 1 layer, this macro returns &var[layer0]
477 * For 2 layers, this macro is similar to allocate a bi-dimensional array
478 * and to return "&var[layer0][layer1]"
479 * For 3 layers, this macro is similar to allocate a tri-dimensional array
480 * and to return "&var[layer0][layer1][layer2]"
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300481 */
482#define EDAC_DIMM_PTR(layers, var, nlayers, layer0, layer1, layer2) ({ \
Mauro Carvalho Chehabde3910eb2012-04-24 15:05:43 -0300483 typeof(*var) __p; \
484 int ___i = EDAC_DIMM_OFF(layers, nlayers, layer0, layer1, layer2); \
485 if (___i < 0) \
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300486 __p = NULL; \
Mauro Carvalho Chehabde3910eb2012-04-24 15:05:43 -0300487 else \
488 __p = (var)[___i]; \
Mauro Carvalho Chehab982216a2012-04-16 13:04:46 -0300489 __p; \
490})
491
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300492struct dimm_info {
Mauro Carvalho Chehab7a623c02012-04-16 16:41:11 -0300493 struct device dev;
494
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300495 char label[EDAC_MC_LABEL_LEN + 1]; /* DIMM label on motherboard */
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300496
497 /* Memory location data */
498 unsigned location[EDAC_MAX_LAYERS];
499
500 struct mem_ctl_info *mci; /* the parent */
Mauro Carvalho Chehab084a4fc2012-01-27 18:38:08 -0300501
502 u32 grain; /* granularity of reported error in bytes */
503 enum dev_type dtype; /* memory device type */
504 enum mem_type mtype; /* memory dimm type */
505 enum edac_type edac_mode; /* EDAC mode for this dimm */
506
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300507 u32 nr_pages; /* number of pages on this dimm */
Mauro Carvalho Chehaba895bf82012-01-28 09:09:38 -0300508
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300509 unsigned csrow, cschannel; /* Points to the old API data */
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300510};
511
Mauro Carvalho Chehaba4b4be32012-01-27 10:26:13 -0300512/**
513 * struct rank_info - contains the information for one DIMM rank
514 *
515 * @chan_idx: channel number where the rank is (typically, 0 or 1)
516 * @ce_count: number of correctable errors for this rank
Mauro Carvalho Chehaba4b4be32012-01-27 10:26:13 -0300517 * @csrow: A pointer to the chip select row structure (the parent
518 * structure). The location of the rank is given by
519 * the (csrow->csrow_idx, chan_idx) vector.
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300520 * @dimm: A pointer to the DIMM structure, where the DIMM label
521 * information is stored.
522 *
523 * FIXME: Currently, the EDAC core model will assume one DIMM per rank.
524 * This is a bad assumption, but it makes this patch easier. Later
525 * patches in this series will fix this issue.
Mauro Carvalho Chehaba4b4be32012-01-27 10:26:13 -0300526 */
527struct rank_info {
528 int chan_idx;
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300529 struct csrow_info *csrow;
530 struct dimm_info *dimm;
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300531
532 u32 ce_count; /* Correctable Errors for this csrow */
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300533};
534
535struct csrow_info {
Mauro Carvalho Chehab7a623c02012-04-16 16:41:11 -0300536 struct device dev;
537
Mauro Carvalho Chehaba895bf82012-01-28 09:09:38 -0300538 /* Used only by edac_mc_find_csrow_by_page() */
Mauro Carvalho Chehab084a4fc2012-01-27 18:38:08 -0300539 unsigned long first_page; /* first page number in csrow */
540 unsigned long last_page; /* last page number in csrow */
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300541 unsigned long page_mask; /* used for interleaving -
Mauro Carvalho Chehaba895bf82012-01-28 09:09:38 -0300542 * 0UL for non intlv */
543
Mauro Carvalho Chehab084a4fc2012-01-27 18:38:08 -0300544 int csrow_idx; /* the chip-select row */
545
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300546 u32 ue_count; /* Uncorrectable Errors for this csrow */
547 u32 ce_count; /* Correctable Errors for this csrow */
Borislav Petkov16a528ee2012-09-13 18:53:58 +0200548 u32 nr_pages; /* combined pages count of all channels */
Mauro Carvalho Chehab084a4fc2012-01-27 18:38:08 -0300549
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300550 struct mem_ctl_info *mci; /* the parent */
551
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300552 /* channel information for this csrow */
553 u32 nr_channels;
Mauro Carvalho Chehabde3910eb2012-04-24 15:05:43 -0300554 struct rank_info **channels;
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300555};
556
Mauro Carvalho Chehab7a623c02012-04-16 16:41:11 -0300557/*
558 * struct errcount_attribute - used to store the several error counts
559 */
560struct errcount_attribute_data {
561 int n_layers;
562 int pos[EDAC_MAX_LAYERS];
563 int layer0, layer1, layer2;
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300564};
565
Mauro Carvalho Chehabc7ef7642013-02-21 13:36:45 -0300566/**
567 * edac_raw_error_desc - Raw error report structure
568 * @grain: minimum granularity for an error report, in bytes
569 * @error_count: number of errors of the same type
570 * @top_layer: top layer of the error (layer[0])
571 * @mid_layer: middle layer of the error (layer[1])
572 * @low_layer: low layer of the error (layer[2])
573 * @page_frame_number: page where the error happened
574 * @offset_in_page: page offset
575 * @syndrome: syndrome of the error (or 0 if unknown or if
576 * the syndrome is not applicable)
577 * @msg: error message
578 * @location: location of the error
579 * @label: label of the affected DIMM(s)
580 * @other_detail: other driver-specific detail about the error
581 * @enable_per_layer_report: if false, the error affects all layers
582 * (typically, a memory controller error)
583 */
584struct edac_raw_error_desc {
585 /*
586 * NOTE: everything before grain won't be cleaned by
587 * edac_raw_error_desc_clean()
588 */
589 char location[LOCATION_SIZE];
590 char label[(EDAC_MC_LABEL_LEN + 1 + sizeof(OTHER_LABEL)) * EDAC_MAX_LABELS];
591 long grain;
592
593 /* the vars below and grain will be cleaned on every new error report */
594 u16 error_count;
595 int top_layer;
596 int mid_layer;
597 int low_layer;
598 unsigned long page_frame_number;
599 unsigned long offset_in_page;
600 unsigned long syndrome;
601 const char *msg;
602 const char *other_detail;
603 bool enable_per_layer_report;
604};
605
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300606/* MEMORY controller information structure
607 */
608struct mem_ctl_info {
Mauro Carvalho Chehab7a623c02012-04-16 16:41:11 -0300609 struct device dev;
610 struct bus_type bus;
611
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300612 struct list_head link; /* for global list of mem_ctl_info structs */
613
614 struct module *owner; /* Module owner of this control struct */
615
616 unsigned long mtype_cap; /* memory types supported by mc */
617 unsigned long edac_ctl_cap; /* Mem controller EDAC capabilities */
618 unsigned long edac_cap; /* configuration capabilities - this is
619 * closely related to edac_ctl_cap. The
620 * difference is that the controller may be
621 * capable of s4ecd4ed which would be listed
622 * in edac_ctl_cap, but if channels aren't
623 * capable of s4ecd4ed then the edac_cap would
624 * not have that capability.
625 */
626 unsigned long scrub_cap; /* chipset scrub capabilities */
627 enum scrub_type scrub_mode; /* current scrub mode */
628
629 /* Translates sdram memory scrub rate given in bytes/sec to the
630 internal representation and configures whatever else needs
631 to be configured.
632 */
633 int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 bw);
634
635 /* Get the current sdram memory scrub rate from the internal
636 representation and converts it to the closest matching
637 bandwidth in bytes/sec.
638 */
639 int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci);
640
641
642 /* pointer to edac checking routine */
643 void (*edac_check) (struct mem_ctl_info * mci);
644
645 /*
646 * Remaps memory pages: controller pages to physical pages.
647 * For most MC's, this will be NULL.
648 */
649 /* FIXME - why not send the phys page to begin with? */
650 unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci,
651 unsigned long page);
652 int mc_idx;
Mauro Carvalho Chehabde3910eb2012-04-24 15:05:43 -0300653 struct csrow_info **csrows;
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300654 unsigned nr_csrows, num_cschannel;
655
Mauro Carvalho Chehab7a623c02012-04-16 16:41:11 -0300656 /*
657 * Memory Controller hierarchy
658 *
659 * There are basically two types of memory controller: the ones that
660 * sees memory sticks ("dimms"), and the ones that sees memory ranks.
661 * All old memory controllers enumerate memories per rank, but most
662 * of the recent drivers enumerate memories per DIMM, instead.
663 * When the memory controller is per rank, mem_is_per_rank is true.
664 */
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300665 unsigned n_layers;
666 struct edac_mc_layer *layers;
667 bool mem_is_per_rank;
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300668
669 /*
670 * DIMM info. Will eventually remove the entire csrows_info some day
671 */
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300672 unsigned tot_dimms;
Mauro Carvalho Chehabde3910eb2012-04-24 15:05:43 -0300673 struct dimm_info **dimms;
Mauro Carvalho Chehaba7d7d2e2012-01-27 14:12:32 -0300674
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300675 /*
676 * FIXME - what about controllers on other busses? - IDs must be
677 * unique. dev pointer should be sufficiently unique, but
678 * BUS:SLOT.FUNC numbers may not be unique.
679 */
Mauro Carvalho Chehabfd687502012-03-16 07:44:18 -0300680 struct device *pdev;
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300681 const char *mod_name;
682 const char *mod_ver;
683 const char *ctl_name;
684 const char *dev_name;
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300685 void *pvt_info;
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300686 unsigned long start_time; /* mci load start time (in jiffies) */
687
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300688 /*
689 * drivers shouldn't access those fields directly, as the core
690 * already handles that.
691 */
692 u32 ce_noinfo_count, ue_noinfo_count;
Mauro Carvalho Chehab5926ff52012-02-09 11:05:20 -0300693 u32 ue_mc, ce_mc;
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300694 u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS];
695
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300696 struct completion complete;
697
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300698 /* Additional top controller level attributes, but specified
699 * by the low level driver.
700 *
701 * Set by the low level driver to provide attributes at the
Mauro Carvalho Chehab4275be62012-04-18 15:20:50 -0300702 * controller level.
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300703 * An array of structures, NULL terminated
704 *
705 * If attributes are desired, then set to array of attributes
706 * If no attributes are desired, leave NULL
707 */
708 const struct mcidev_sysfs_attribute *mc_driver_sysfs_attributes;
709
710 /* work struct for this MC */
711 struct delayed_work work;
712
Mauro Carvalho Chehabc7ef7642013-02-21 13:36:45 -0300713 /*
714 * Used to report an error - by being at the global struct
715 * makes the memory allocated by the EDAC core
716 */
717 struct edac_raw_error_desc error_desc;
718
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300719 /* the internal state of this controller instance */
720 int op_state;
Mauro Carvalho Chehab452a6bf2012-03-26 09:35:11 -0300721
722#ifdef CONFIG_EDAC_DEBUG
723 struct dentry *debugfs;
724 u8 fake_inject_layer[EDAC_MAX_LAYERS];
725 u32 fake_inject_ue;
Mauro Carvalho Chehab38ced282012-06-12 10:55:57 -0300726 u16 fake_inject_count;
Mauro Carvalho Chehab452a6bf2012-03-26 09:35:11 -0300727#endif
Borislav Petkov11652762012-09-13 17:19:40 +0200728 __u8 csbased : 1, /* csrow-based memory controller */
729 __resv : 7;
Mauro Carvalho Chehabddeb3542011-03-04 15:11:29 -0300730};
731
Dave Jiangc0d12172007-07-19 01:49:46 -0700732#endif