blob: 2d23e304f5ec1083b9850e46fe817c74b4e78d00 [file] [log] [blame]
Douglas Thompson7c9281d2007-07-19 01:49:33 -07001/*
2 * Defines, structures, APIs for edac_core module
3 *
4 * (C) 2007 Linux Networx (http://lnxi.com)
5 * This file may be distributed under the terms of the
6 * GNU General Public License.
7 *
8 * Written by Thayne Harbaugh
9 * Based on work by Dan Hollis <goemon at anime dot net> and others.
10 * http://www.anime.net/~goemon/linux-ecc/
11 *
12 * NMI handling support added by
13 * Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>
14 *
15 * Refactored for multi-source files:
16 * Doug Thompson <norsk5@xmission.com>
17 *
18 */
19
20#ifndef _EDAC_CORE_H_
21#define _EDAC_CORE_H_
22
23#include <linux/kernel.h>
24#include <linux/types.h>
25#include <linux/module.h>
26#include <linux/spinlock.h>
27#include <linux/smp.h>
28#include <linux/pci.h>
29#include <linux/time.h>
30#include <linux/nmi.h>
31#include <linux/rcupdate.h>
32#include <linux/completion.h>
33#include <linux/kobject.h>
34#include <linux/platform_device.h>
Douglas Thompsone27e3da2007-07-19 01:49:36 -070035#include <linux/sysdev.h>
36#include <linux/workqueue.h>
37#include <linux/version.h>
Douglas Thompson7c9281d2007-07-19 01:49:33 -070038
39#define EDAC_MC_LABEL_LEN 31
Douglas Thompsone27e3da2007-07-19 01:49:36 -070040#define EDAC_DEVICE_NAME_LEN 31
41#define EDAC_ATTRIB_VALUE_LEN 15
42#define MC_PROC_NAME_MAX_LEN 7
Douglas Thompson7c9281d2007-07-19 01:49:33 -070043
44#if PAGE_SHIFT < 20
45#define PAGES_TO_MiB( pages ) ( ( pages ) >> ( 20 - PAGE_SHIFT ) )
46#else /* PAGE_SHIFT > 20 */
47#define PAGES_TO_MiB( pages ) ( ( pages ) << ( PAGE_SHIFT - 20 ) )
48#endif
49
50#define edac_printk(level, prefix, fmt, arg...) \
51 printk(level "EDAC " prefix ": " fmt, ##arg)
52
53#define edac_mc_printk(mci, level, fmt, arg...) \
54 printk(level "EDAC MC%d: " fmt, mci->mc_idx, ##arg)
55
56#define edac_mc_chipset_printk(mci, level, prefix, fmt, arg...) \
57 printk(level "EDAC " prefix " MC%d: " fmt, mci->mc_idx, ##arg)
58
Douglas Thompsone27e3da2007-07-19 01:49:36 -070059/* edac_device printk */
60#define edac_device_printk(ctl, level, fmt, arg...) \
61 printk(level "EDAC DEVICE%d: " fmt, ctl->dev_idx, ##arg)
62
Dave Jiang91b99042007-07-19 01:49:52 -070063/* edac_pci printk */
64#define edac_pci_printk(ctl, level, fmt, arg...) \
65 printk(level "EDAC PCI%d: " fmt, ctl->pci_idx, ##arg)
66
Douglas Thompson7c9281d2007-07-19 01:49:33 -070067/* prefixes for edac_printk() and edac_mc_printk() */
68#define EDAC_MC "MC"
69#define EDAC_PCI "PCI"
70#define EDAC_DEBUG "DEBUG"
71
72#ifdef CONFIG_EDAC_DEBUG
73extern int edac_debug_level;
74
75#define edac_debug_printk(level, fmt, arg...) \
76 do { \
77 if (level <= edac_debug_level) \
Doug Thompson3c8bb2c2007-09-11 15:23:31 -070078 edac_printk(KERN_DEBUG, EDAC_DEBUG, fmt, ##arg); \
Douglas Thompson7c9281d2007-07-19 01:49:33 -070079 } while(0)
80
81#define debugf0( ... ) edac_debug_printk(0, __VA_ARGS__ )
82#define debugf1( ... ) edac_debug_printk(1, __VA_ARGS__ )
83#define debugf2( ... ) edac_debug_printk(2, __VA_ARGS__ )
84#define debugf3( ... ) edac_debug_printk(3, __VA_ARGS__ )
85#define debugf4( ... ) edac_debug_printk(4, __VA_ARGS__ )
86
Douglas Thompson079708b2007-07-19 01:49:58 -070087#else /* !CONFIG_EDAC_DEBUG */
Douglas Thompson7c9281d2007-07-19 01:49:33 -070088
89#define debugf0( ... )
90#define debugf1( ... )
91#define debugf2( ... )
92#define debugf3( ... )
93#define debugf4( ... )
94
Douglas Thompson079708b2007-07-19 01:49:58 -070095#endif /* !CONFIG_EDAC_DEBUG */
Douglas Thompson7c9281d2007-07-19 01:49:33 -070096
Douglas Thompson7c9281d2007-07-19 01:49:33 -070097#define PCI_VEND_DEV(vend, dev) PCI_VENDOR_ID_ ## vend, \
98 PCI_DEVICE_ID_ ## vend ## _ ## dev
99
Dave Jiangc4192702007-07-19 01:49:47 -0700100#define dev_name(dev) (dev)->dev_name
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700101
102/* memory devices */
103enum dev_type {
104 DEV_UNKNOWN = 0,
105 DEV_X1,
106 DEV_X2,
107 DEV_X4,
108 DEV_X8,
109 DEV_X16,
110 DEV_X32, /* Do these parts exist? */
111 DEV_X64 /* Do these parts exist? */
112};
113
114#define DEV_FLAG_UNKNOWN BIT(DEV_UNKNOWN)
115#define DEV_FLAG_X1 BIT(DEV_X1)
116#define DEV_FLAG_X2 BIT(DEV_X2)
117#define DEV_FLAG_X4 BIT(DEV_X4)
118#define DEV_FLAG_X8 BIT(DEV_X8)
119#define DEV_FLAG_X16 BIT(DEV_X16)
120#define DEV_FLAG_X32 BIT(DEV_X32)
121#define DEV_FLAG_X64 BIT(DEV_X64)
122
123/* memory types */
124enum mem_type {
125 MEM_EMPTY = 0, /* Empty csrow */
126 MEM_RESERVED, /* Reserved csrow type */
127 MEM_UNKNOWN, /* Unknown csrow type */
128 MEM_FPM, /* Fast page mode */
129 MEM_EDO, /* Extended data out */
130 MEM_BEDO, /* Burst Extended data out */
131 MEM_SDR, /* Single data rate SDRAM */
132 MEM_RDR, /* Registered single data rate SDRAM */
133 MEM_DDR, /* Double data rate SDRAM */
134 MEM_RDDR, /* Registered Double data rate SDRAM */
135 MEM_RMBS, /* Rambus DRAM */
Douglas Thompson079708b2007-07-19 01:49:58 -0700136 MEM_DDR2, /* DDR2 RAM */
137 MEM_FB_DDR2, /* fully buffered DDR2 */
138 MEM_RDDR2, /* Registered DDR2 RAM */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700139};
140
141#define MEM_FLAG_EMPTY BIT(MEM_EMPTY)
142#define MEM_FLAG_RESERVED BIT(MEM_RESERVED)
143#define MEM_FLAG_UNKNOWN BIT(MEM_UNKNOWN)
144#define MEM_FLAG_FPM BIT(MEM_FPM)
145#define MEM_FLAG_EDO BIT(MEM_EDO)
146#define MEM_FLAG_BEDO BIT(MEM_BEDO)
147#define MEM_FLAG_SDR BIT(MEM_SDR)
148#define MEM_FLAG_RDR BIT(MEM_RDR)
149#define MEM_FLAG_DDR BIT(MEM_DDR)
150#define MEM_FLAG_RDDR BIT(MEM_RDDR)
151#define MEM_FLAG_RMBS BIT(MEM_RMBS)
152#define MEM_FLAG_DDR2 BIT(MEM_DDR2)
153#define MEM_FLAG_FB_DDR2 BIT(MEM_FB_DDR2)
154#define MEM_FLAG_RDDR2 BIT(MEM_RDDR2)
155
156/* chipset Error Detection and Correction capabilities and mode */
157enum edac_type {
158 EDAC_UNKNOWN = 0, /* Unknown if ECC is available */
159 EDAC_NONE, /* Doesnt support ECC */
160 EDAC_RESERVED, /* Reserved ECC type */
161 EDAC_PARITY, /* Detects parity errors */
162 EDAC_EC, /* Error Checking - no correction */
163 EDAC_SECDED, /* Single bit error correction, Double detection */
164 EDAC_S2ECD2ED, /* Chipkill x2 devices - do these exist? */
165 EDAC_S4ECD4ED, /* Chipkill x4 devices */
166 EDAC_S8ECD8ED, /* Chipkill x8 devices */
167 EDAC_S16ECD16ED, /* Chipkill x16 devices */
168};
169
170#define EDAC_FLAG_UNKNOWN BIT(EDAC_UNKNOWN)
171#define EDAC_FLAG_NONE BIT(EDAC_NONE)
172#define EDAC_FLAG_PARITY BIT(EDAC_PARITY)
173#define EDAC_FLAG_EC BIT(EDAC_EC)
174#define EDAC_FLAG_SECDED BIT(EDAC_SECDED)
175#define EDAC_FLAG_S2ECD2ED BIT(EDAC_S2ECD2ED)
176#define EDAC_FLAG_S4ECD4ED BIT(EDAC_S4ECD4ED)
177#define EDAC_FLAG_S8ECD8ED BIT(EDAC_S8ECD8ED)
178#define EDAC_FLAG_S16ECD16ED BIT(EDAC_S16ECD16ED)
179
180/* scrubbing capabilities */
181enum scrub_type {
182 SCRUB_UNKNOWN = 0, /* Unknown if scrubber is available */
183 SCRUB_NONE, /* No scrubber */
184 SCRUB_SW_PROG, /* SW progressive (sequential) scrubbing */
185 SCRUB_SW_SRC, /* Software scrub only errors */
186 SCRUB_SW_PROG_SRC, /* Progressive software scrub from an error */
187 SCRUB_SW_TUNABLE, /* Software scrub frequency is tunable */
188 SCRUB_HW_PROG, /* HW progressive (sequential) scrubbing */
189 SCRUB_HW_SRC, /* Hardware scrub only errors */
190 SCRUB_HW_PROG_SRC, /* Progressive hardware scrub from an error */
191 SCRUB_HW_TUNABLE /* Hardware scrub frequency is tunable */
192};
193
194#define SCRUB_FLAG_SW_PROG BIT(SCRUB_SW_PROG)
Douglas Thompson522a94b2007-07-19 01:49:41 -0700195#define SCRUB_FLAG_SW_SRC BIT(SCRUB_SW_SRC)
196#define SCRUB_FLAG_SW_PROG_SRC BIT(SCRUB_SW_PROG_SRC)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700197#define SCRUB_FLAG_SW_TUN BIT(SCRUB_SW_SCRUB_TUNABLE)
198#define SCRUB_FLAG_HW_PROG BIT(SCRUB_HW_PROG)
Douglas Thompson522a94b2007-07-19 01:49:41 -0700199#define SCRUB_FLAG_HW_SRC BIT(SCRUB_HW_SRC)
200#define SCRUB_FLAG_HW_PROG_SRC BIT(SCRUB_HW_PROG_SRC)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700201#define SCRUB_FLAG_HW_TUN BIT(SCRUB_HW_TUNABLE)
202
203/* FIXME - should have notify capabilities: NMI, LOG, PROC, etc */
204
Dave Jiang91b99042007-07-19 01:49:52 -0700205/* EDAC internal operation states */
206#define OP_ALLOC 0x100
207#define OP_RUNNING_POLL 0x201
208#define OP_RUNNING_INTERRUPT 0x202
209#define OP_RUNNING_POLL_INTR 0x203
210#define OP_OFFLINE 0x300
211
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700212/*
213 * There are several things to be aware of that aren't at all obvious:
214 *
215 *
216 * SOCKETS, SOCKET SETS, BANKS, ROWS, CHIP-SELECT ROWS, CHANNELS, etc..
217 *
218 * These are some of the many terms that are thrown about that don't always
219 * mean what people think they mean (Inconceivable!). In the interest of
220 * creating a common ground for discussion, terms and their definitions
221 * will be established.
222 *
223 * Memory devices: The individual chip on a memory stick. These devices
224 * commonly output 4 and 8 bits each. Grouping several
225 * of these in parallel provides 64 bits which is common
226 * for a memory stick.
227 *
228 * Memory Stick: A printed circuit board that agregates multiple
229 * memory devices in parallel. This is the atomic
230 * memory component that is purchaseable by Joe consumer
231 * and loaded into a memory socket.
232 *
233 * Socket: A physical connector on the motherboard that accepts
234 * a single memory stick.
235 *
236 * Channel: Set of memory devices on a memory stick that must be
237 * grouped in parallel with one or more additional
238 * channels from other memory sticks. This parallel
239 * grouping of the output from multiple channels are
240 * necessary for the smallest granularity of memory access.
241 * Some memory controllers are capable of single channel -
242 * which means that memory sticks can be loaded
243 * individually. Other memory controllers are only
244 * capable of dual channel - which means that memory
245 * sticks must be loaded as pairs (see "socket set").
246 *
247 * Chip-select row: All of the memory devices that are selected together.
248 * for a single, minimum grain of memory access.
249 * This selects all of the parallel memory devices across
250 * all of the parallel channels. Common chip-select rows
251 * for single channel are 64 bits, for dual channel 128
252 * bits.
253 *
254 * Single-Ranked stick: A Single-ranked stick has 1 chip-select row of memmory.
255 * Motherboards commonly drive two chip-select pins to
256 * a memory stick. A single-ranked stick, will occupy
257 * only one of those rows. The other will be unused.
258 *
259 * Double-Ranked stick: A double-ranked stick has two chip-select rows which
260 * access different sets of memory devices. The two
261 * rows cannot be accessed concurrently.
262 *
263 * Double-sided stick: DEPRECATED TERM, see Double-Ranked stick.
264 * A double-sided stick has two chip-select rows which
265 * access different sets of memory devices. The two
266 * rows cannot be accessed concurrently. "Double-sided"
267 * is irrespective of the memory devices being mounted
268 * on both sides of the memory stick.
269 *
270 * Socket set: All of the memory sticks that are required for for
271 * a single memory access or all of the memory sticks
272 * spanned by a chip-select row. A single socket set
273 * has two chip-select rows and if double-sided sticks
274 * are used these will occupy those chip-select rows.
275 *
276 * Bank: This term is avoided because it is unclear when
277 * needing to distinguish between chip-select rows and
278 * socket sets.
279 *
280 * Controller pages:
281 *
282 * Physical pages:
283 *
284 * Virtual pages:
285 *
286 *
287 * STRUCTURE ORGANIZATION AND CHOICES
288 *
289 *
290 *
291 * PS - I enjoyed writing all that about as much as you enjoyed reading it.
292 */
293
294struct channel_info {
295 int chan_idx; /* channel index */
296 u32 ce_count; /* Correctable Errors for this CHANNEL */
Douglas Thompson079708b2007-07-19 01:49:58 -0700297 char label[EDAC_MC_LABEL_LEN + 1]; /* DIMM label on motherboard */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700298 struct csrow_info *csrow; /* the parent */
299};
300
301struct csrow_info {
302 unsigned long first_page; /* first page number in dimm */
303 unsigned long last_page; /* last page number in dimm */
304 unsigned long page_mask; /* used for interleaving -
305 * 0UL for non intlv
306 */
307 u32 nr_pages; /* number of pages in csrow */
308 u32 grain; /* granularity of reported error in bytes */
309 int csrow_idx; /* the chip-select row */
310 enum dev_type dtype; /* memory device type */
311 u32 ue_count; /* Uncorrectable Errors for this csrow */
312 u32 ce_count; /* Correctable Errors for this csrow */
313 enum mem_type mtype; /* memory csrow type */
314 enum edac_type edac_mode; /* EDAC mode for this csrow */
315 struct mem_ctl_info *mci; /* the parent */
316
317 struct kobject kobj; /* sysfs kobject for this csrow */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700318
Doug Thompson8096cfa2007-07-19 01:50:27 -0700319 /* channel information for this csrow */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700320 u32 nr_channels;
321 struct channel_info *channels;
322};
323
Douglas Thompson42a8e392007-07-19 01:50:10 -0700324/* mcidev_sysfs_attribute structure
325 * used for driver sysfs attributes and in mem_ctl_info
326 * sysfs top level entries
327 */
328struct mcidev_sysfs_attribute {
329 struct attribute attr;
330 ssize_t (*show)(struct mem_ctl_info *,char *);
331 ssize_t (*store)(struct mem_ctl_info *, const char *,size_t);
332};
333
334/* MEMORY controller information structure
335 */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700336struct mem_ctl_info {
Douglas Thompson079708b2007-07-19 01:49:58 -0700337 struct list_head link; /* for global list of mem_ctl_info structs */
Douglas Thompson1c3631f2007-07-19 01:50:29 -0700338
339 struct module *owner; /* Module owner of this control struct */
340
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700341 unsigned long mtype_cap; /* memory types supported by mc */
342 unsigned long edac_ctl_cap; /* Mem controller EDAC capabilities */
343 unsigned long edac_cap; /* configuration capabilities - this is
344 * closely related to edac_ctl_cap. The
345 * difference is that the controller may be
346 * capable of s4ecd4ed which would be listed
347 * in edac_ctl_cap, but if channels aren't
348 * capable of s4ecd4ed then the edac_cap would
349 * not have that capability.
350 */
351 unsigned long scrub_cap; /* chipset scrub capabilities */
352 enum scrub_type scrub_mode; /* current scrub mode */
353
354 /* Translates sdram memory scrub rate given in bytes/sec to the
355 internal representation and configures whatever else needs
356 to be configured.
Douglas Thompson079708b2007-07-19 01:49:58 -0700357 */
358 int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 * bw);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700359
360 /* Get the current sdram memory scrub rate from the internal
361 representation and converts it to the closest matching
362 bandwith in bytes/sec.
Douglas Thompson079708b2007-07-19 01:49:58 -0700363 */
364 int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 * bw);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700365
Douglas Thompson42a8e392007-07-19 01:50:10 -0700366
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700367 /* pointer to edac checking routine */
368 void (*edac_check) (struct mem_ctl_info * mci);
369
370 /*
371 * Remaps memory pages: controller pages to physical pages.
372 * For most MC's, this will be NULL.
373 */
374 /* FIXME - why not send the phys page to begin with? */
375 unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700376 unsigned long page);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700377 int mc_idx;
378 int nr_csrows;
379 struct csrow_info *csrows;
380 /*
381 * FIXME - what about controllers on other busses? - IDs must be
382 * unique. dev pointer should be sufficiently unique, but
383 * BUS:SLOT.FUNC numbers may not be unique.
384 */
385 struct device *dev;
386 const char *mod_name;
387 const char *mod_ver;
388 const char *ctl_name;
Dave Jiangc4192702007-07-19 01:49:47 -0700389 const char *dev_name;
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700390 char proc_name[MC_PROC_NAME_MAX_LEN + 1];
391 void *pvt_info;
392 u32 ue_noinfo_count; /* Uncorrectable Errors w/o info */
393 u32 ce_noinfo_count; /* Correctable Errors w/o info */
394 u32 ue_count; /* Total Uncorrectable Errors for this MC */
395 u32 ce_count; /* Total Correctable Errors for this MC */
396 unsigned long start_time; /* mci load start time (in jiffies) */
397
398 /* this stuff is for safe removal of mc devices from global list while
399 * NMI handlers may be traversing list
400 */
401 struct rcu_head rcu;
402 struct completion complete;
403
404 /* edac sysfs device control */
405 struct kobject edac_mci_kobj;
Dave Jiang81d87cb2007-07-19 01:49:52 -0700406
Douglas Thompson42a8e392007-07-19 01:50:10 -0700407 /* Additional top controller level attributes, but specified
408 * by the low level driver.
409 *
410 * Set by the low level driver to provide attributes at the
411 * controller level, same level as 'ue_count' and 'ce_count' above.
412 * An array of structures, NULL terminated
413 *
414 * If attributes are desired, then set to array of attributes
415 * If no attributes are desired, leave NULL
416 */
417 struct mcidev_sysfs_attribute *mc_driver_sysfs_attributes;
418
Dave Jiang81d87cb2007-07-19 01:49:52 -0700419 /* work struct for this MC */
Dave Jiang81d87cb2007-07-19 01:49:52 -0700420 struct delayed_work work;
Douglas Thompson86aa8cb2007-07-19 01:50:01 -0700421
Dave Jiang81d87cb2007-07-19 01:49:52 -0700422 /* the internal state of this controller instance */
423 int op_state;
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700424};
425
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700426/*
Douglas Thompson42a8e392007-07-19 01:50:10 -0700427 * The following are the structures to provide for a generic
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700428 * or abstract 'edac_device'. This set of structures and the
429 * code that implements the APIs for the same, provide for
430 * registering EDAC type devices which are NOT standard memory.
431 *
432 * CPU caches (L1 and L2)
433 * DMA engines
434 * Core CPU swithces
435 * Fabric switch units
436 * PCIe interface controllers
437 * other EDAC/ECC type devices that can be monitored for
438 * errors, etc.
439 *
440 * It allows for a 2 level set of hiearchry. For example:
441 *
442 * cache could be composed of L1, L2 and L3 levels of cache.
443 * Each CPU core would have its own L1 cache, while sharing
444 * L2 and maybe L3 caches.
445 *
446 * View them arranged, via the sysfs presentation:
447 * /sys/devices/system/edac/..
448 *
449 * mc/ <existing memory device directory>
450 * cpu/cpu0/.. <L1 and L2 block directory>
451 * /L1-cache/ce_count
452 * /ue_count
453 * /L2-cache/ce_count
454 * /ue_count
455 * cpu/cpu1/.. <L1 and L2 block directory>
456 * /L1-cache/ce_count
457 * /ue_count
458 * /L2-cache/ce_count
459 * /ue_count
460 * ...
461 *
462 * the L1 and L2 directories would be "edac_device_block's"
463 */
464
465struct edac_device_counter {
Douglas Thompson079708b2007-07-19 01:49:58 -0700466 u32 ue_count;
467 u32 ce_count;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700468};
469
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700470/* forward reference */
471struct edac_device_ctl_info;
472struct edac_device_block;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700473
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700474/* edac_dev_sysfs_attribute structure
475 * used for driver sysfs attributes in mem_ctl_info
476 * for extra controls and attributes:
477 * like high level error Injection controls
478 */
479struct edac_dev_sysfs_attribute {
480 struct attribute attr;
481 ssize_t (*show)(struct edac_device_ctl_info *, char *);
482 ssize_t (*store)(struct edac_device_ctl_info *, const char *, size_t);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700483};
484
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700485/* edac_dev_sysfs_block_attribute structure
Doug Thompsonb2a4ac02007-07-19 01:50:33 -0700486 *
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700487 * used in leaf 'block' nodes for adding controls/attributes
Doug Thompsonb2a4ac02007-07-19 01:50:33 -0700488 *
489 * each block in each instance of the containing control structure
490 * can have an array of the following. The show and store functions
491 * will be filled in with the show/store function in the
492 * low level driver.
493 *
494 * The 'value' field will be the actual value field used for
495 * counting
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700496 */
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700497struct edac_dev_sysfs_block_attribute {
498 struct attribute attr;
499 ssize_t (*show)(struct kobject *, struct attribute *, char *);
500 ssize_t (*store)(struct kobject *, struct attribute *,
501 const char *, size_t);
502 struct edac_device_block *block;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700503
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700504 unsigned int value;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700505};
506
507/* device block control structure */
508struct edac_device_block {
509 struct edac_device_instance *instance; /* Up Pointer */
Douglas Thompson079708b2007-07-19 01:49:58 -0700510 char name[EDAC_DEVICE_NAME_LEN + 1];
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700511
512 struct edac_device_counter counters; /* basic UE and CE counters */
513
Douglas Thompson079708b2007-07-19 01:49:58 -0700514 int nr_attribs; /* how many attributes */
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700515
516 /* this block's attributes, could be NULL */
517 struct edac_dev_sysfs_block_attribute *block_attributes;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700518
519 /* edac sysfs device control */
520 struct kobject kobj;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700521};
522
523/* device instance control structure */
524struct edac_device_instance {
525 struct edac_device_ctl_info *ctl; /* Up pointer */
526 char name[EDAC_DEVICE_NAME_LEN + 4];
527
528 struct edac_device_counter counters; /* instance counters */
529
Douglas Thompson079708b2007-07-19 01:49:58 -0700530 u32 nr_blocks; /* how many blocks */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700531 struct edac_device_block *blocks; /* block array */
532
533 /* edac sysfs device control */
534 struct kobject kobj;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700535};
536
Douglas Thompson42a8e392007-07-19 01:50:10 -0700537
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700538/*
539 * Abstract edac_device control info structure
540 *
541 */
542struct edac_device_ctl_info {
543 /* for global list of edac_device_ctl_info structs */
544 struct list_head link;
545
Douglas Thompson1c3631f2007-07-19 01:50:29 -0700546 struct module *owner; /* Module owner of this control struct */
547
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700548 int dev_idx;
549
550 /* Per instance controls for this edac_device */
551 int log_ue; /* boolean for logging UEs */
552 int log_ce; /* boolean for logging CEs */
553 int panic_on_ue; /* boolean for panic'ing on an UE */
554 unsigned poll_msec; /* number of milliseconds to poll interval */
555 unsigned long delay; /* number of jiffies for poll_msec */
556
Douglas Thompson42a8e392007-07-19 01:50:10 -0700557 /* Additional top controller level attributes, but specified
558 * by the low level driver.
559 *
560 * Set by the low level driver to provide attributes at the
561 * controller level, same level as 'ue_count' and 'ce_count' above.
562 * An array of structures, NULL terminated
563 *
564 * If attributes are desired, then set to array of attributes
565 * If no attributes are desired, leave NULL
566 */
567 struct edac_dev_sysfs_attribute *sysfs_attributes;
568
569 /* pointer to main 'edac' class in sysfs */
570 struct sysdev_class *edac_class;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700571
572 /* the internal state of this controller instance */
573 int op_state;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700574 /* work struct for this instance */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700575 struct delayed_work work;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700576
577 /* pointer to edac polling checking routine:
Douglas Thompson079708b2007-07-19 01:49:58 -0700578 * If NOT NULL: points to polling check routine
579 * If NULL: Then assumes INTERRUPT operation, where
580 * MC driver will receive events
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700581 */
582 void (*edac_check) (struct edac_device_ctl_info * edac_dev);
583
584 struct device *dev; /* pointer to device structure */
585
586 const char *mod_name; /* module name */
587 const char *ctl_name; /* edac controller name */
Dave Jiangc4192702007-07-19 01:49:47 -0700588 const char *dev_name; /* pci/platform/etc... name */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700589
590 void *pvt_info; /* pointer to 'private driver' info */
591
Douglas Thompson079708b2007-07-19 01:49:58 -0700592 unsigned long start_time; /* edac_device load start time (jiffies) */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700593
594 /* these are for safe removal of mc devices from global list while
595 * NMI handlers may be traversing list
596 */
597 struct rcu_head rcu;
Douglas Thompson1c3631f2007-07-19 01:50:29 -0700598 struct completion removal_complete;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700599
600 /* sysfs top name under 'edac' directory
601 * and instance name:
Douglas Thompson079708b2007-07-19 01:49:58 -0700602 * cpu/cpu0/...
603 * cpu/cpu1/...
604 * cpu/cpu2/...
605 * ...
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700606 */
607 char name[EDAC_DEVICE_NAME_LEN + 1];
608
609 /* Number of instances supported on this control structure
610 * and the array of those instances
611 */
612 u32 nr_instances;
613 struct edac_device_instance *instances;
614
615 /* Event counters for the this whole EDAC Device */
616 struct edac_device_counter counters;
617
618 /* edac sysfs device control for the 'name'
619 * device this structure controls
620 */
621 struct kobject kobj;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700622};
623
624/* To get from the instance's wq to the beginning of the ctl structure */
Dave Jiang81d87cb2007-07-19 01:49:52 -0700625#define to_edac_mem_ctl_work(w) \
626 container_of(w, struct mem_ctl_info, work)
627
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700628#define to_edac_device_ctl_work(w) \
629 container_of(w,struct edac_device_ctl_info,work)
630
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700631/*
632 * The alloc() and free() functions for the 'edac_device' control info
633 * structure. A MC driver will allocate one of these for each edac_device
634 * it is going to control/register with the EDAC CORE.
635 */
636extern struct edac_device_ctl_info *edac_device_alloc_ctl_info(
Douglas Thompson079708b2007-07-19 01:49:58 -0700637 unsigned sizeof_private,
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700638 char *edac_device_name, unsigned nr_instances,
639 char *edac_block_name, unsigned nr_blocks,
Douglas Thompson079708b2007-07-19 01:49:58 -0700640 unsigned offset_value,
Douglas Thompsonfd309a9d2007-07-19 01:50:25 -0700641 struct edac_dev_sysfs_block_attribute *block_attributes,
Doug Thompsond45e7822007-07-19 01:50:27 -0700642 unsigned nr_attribs,
643 int device_index);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700644
645/* The offset value can be:
646 * -1 indicating no offset value
647 * 0 for zero-based block numbers
648 * 1 for 1-based block number
649 * other for other-based block number
650 */
651#define BLOCK_OFFSET_VALUE_OFF ((unsigned) -1)
652
Douglas Thompson079708b2007-07-19 01:49:58 -0700653extern void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700654
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700655#ifdef CONFIG_PCI
656
Dave Jiang91b99042007-07-19 01:49:52 -0700657struct edac_pci_counter {
Douglas Thompson079708b2007-07-19 01:49:58 -0700658 atomic_t pe_count;
659 atomic_t npe_count;
Dave Jiang91b99042007-07-19 01:49:52 -0700660};
661
662/*
663 * Abstract edac_pci control info structure
664 *
665 */
666struct edac_pci_ctl_info {
667 /* for global list of edac_pci_ctl_info structs */
668 struct list_head link;
669
670 int pci_idx;
671
Dave Jiang91b99042007-07-19 01:49:52 -0700672 struct sysdev_class *edac_class; /* pointer to class */
673
674 /* the internal state of this controller instance */
675 int op_state;
676 /* work struct for this instance */
Dave Jiang91b99042007-07-19 01:49:52 -0700677 struct delayed_work work;
Dave Jiang91b99042007-07-19 01:49:52 -0700678
679 /* pointer to edac polling checking routine:
Douglas Thompson079708b2007-07-19 01:49:58 -0700680 * If NOT NULL: points to polling check routine
681 * If NULL: Then assumes INTERRUPT operation, where
682 * MC driver will receive events
Dave Jiang91b99042007-07-19 01:49:52 -0700683 */
684 void (*edac_check) (struct edac_pci_ctl_info * edac_dev);
685
686 struct device *dev; /* pointer to device structure */
687
688 const char *mod_name; /* module name */
689 const char *ctl_name; /* edac controller name */
690 const char *dev_name; /* pci/platform/etc... name */
691
692 void *pvt_info; /* pointer to 'private driver' info */
693
Douglas Thompson079708b2007-07-19 01:49:58 -0700694 unsigned long start_time; /* edac_pci load start time (jiffies) */
Dave Jiang91b99042007-07-19 01:49:52 -0700695
696 /* these are for safe removal of devices from global list while
697 * NMI handlers may be traversing list
698 */
699 struct rcu_head rcu;
700 struct completion complete;
701
702 /* sysfs top name under 'edac' directory
703 * and instance name:
Douglas Thompson079708b2007-07-19 01:49:58 -0700704 * cpu/cpu0/...
705 * cpu/cpu1/...
706 * cpu/cpu2/...
707 * ...
Dave Jiang91b99042007-07-19 01:49:52 -0700708 */
709 char name[EDAC_DEVICE_NAME_LEN + 1];
710
711 /* Event counters for the this whole EDAC Device */
712 struct edac_pci_counter counters;
713
714 /* edac sysfs device control for the 'name'
715 * device this structure controls
716 */
717 struct kobject kobj;
718 struct completion kobj_complete;
719};
720
721#define to_edac_pci_ctl_work(w) \
722 container_of(w, struct edac_pci_ctl_info,work)
723
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700724/* write all or some bits in a byte-register*/
725static inline void pci_write_bits8(struct pci_dev *pdev, int offset, u8 value,
Douglas Thompson079708b2007-07-19 01:49:58 -0700726 u8 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700727{
728 if (mask != 0xff) {
729 u8 buf;
730
731 pci_read_config_byte(pdev, offset, &buf);
732 value &= mask;
733 buf &= ~mask;
734 value |= buf;
735 }
736
737 pci_write_config_byte(pdev, offset, value);
738}
739
740/* write all or some bits in a word-register*/
741static inline void pci_write_bits16(struct pci_dev *pdev, int offset,
Douglas Thompson079708b2007-07-19 01:49:58 -0700742 u16 value, u16 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700743{
744 if (mask != 0xffff) {
745 u16 buf;
746
747 pci_read_config_word(pdev, offset, &buf);
748 value &= mask;
749 buf &= ~mask;
750 value |= buf;
751 }
752
753 pci_write_config_word(pdev, offset, value);
754}
755
756/* write all or some bits in a dword-register*/
757static inline void pci_write_bits32(struct pci_dev *pdev, int offset,
Douglas Thompson079708b2007-07-19 01:49:58 -0700758 u32 value, u32 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700759{
760 if (mask != 0xffff) {
761 u32 buf;
762
763 pci_read_config_dword(pdev, offset, &buf);
764 value &= mask;
765 buf &= ~mask;
766 value |= buf;
767 }
768
769 pci_write_config_dword(pdev, offset, value);
770}
771
Douglas Thompson079708b2007-07-19 01:49:58 -0700772#endif /* CONFIG_PCI */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700773
Doug Thompsonb8f6f972007-07-19 01:50:26 -0700774extern struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
775 unsigned nr_chans, int edac_index);
776extern int edac_mc_add_mc(struct mem_ctl_info *mci);
777extern void edac_mc_free(struct mem_ctl_info *mci);
Douglas Thompson079708b2007-07-19 01:49:58 -0700778extern struct mem_ctl_info *edac_mc_find(int idx);
Douglas Thompson079708b2007-07-19 01:49:58 -0700779extern struct mem_ctl_info *edac_mc_del_mc(struct device *dev);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700780extern int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700781 unsigned long page);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700782
783/*
784 * The no info errors are used when error overflows are reported.
785 * There are a limited number of error logging registers that can
786 * be exausted. When all registers are exhausted and an additional
787 * error occurs then an error overflow register records that an
788 * error occured and the type of error, but doesn't have any
789 * further information. The ce/ue versions make for cleaner
790 * reporting logic and function interface - reduces conditional
791 * statement clutter and extra function arguments.
792 */
793extern void edac_mc_handle_ce(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700794 unsigned long page_frame_number,
795 unsigned long offset_in_page,
796 unsigned long syndrome, int row, int channel,
797 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700798extern void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700799 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700800extern void edac_mc_handle_ue(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700801 unsigned long page_frame_number,
802 unsigned long offset_in_page, int row,
803 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700804extern void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700805 const char *msg);
806extern void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci, unsigned int csrow,
807 unsigned int channel0, unsigned int channel1,
808 char *msg);
809extern void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci, unsigned int csrow,
810 unsigned int channel, char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700811
812/*
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700813 * edac_device APIs
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700814 */
Doug Thompsond45e7822007-07-19 01:50:27 -0700815extern int edac_device_add_device(struct edac_device_ctl_info *edac_dev);
Douglas Thompson079708b2007-07-19 01:49:58 -0700816extern struct edac_device_ctl_info *edac_device_del_device(struct device *dev);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700817extern void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
Doug Thompsonb8f6f972007-07-19 01:50:26 -0700818 int inst_nr, int block_nr, const char *msg);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700819extern void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
Doug Thompsonb8f6f972007-07-19 01:50:26 -0700820 int inst_nr, int block_nr, const char *msg);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700821
Dave Jiang91b99042007-07-19 01:49:52 -0700822/*
823 * edac_pci APIs
824 */
Doug Thompsonb8f6f972007-07-19 01:50:26 -0700825extern struct edac_pci_ctl_info *edac_pci_alloc_ctl_info(unsigned int sz_pvt,
826 const char *edac_pci_name);
Dave Jiang91b99042007-07-19 01:49:52 -0700827
828extern void edac_pci_free_ctl_info(struct edac_pci_ctl_info *pci);
829
Doug Thompsonb8f6f972007-07-19 01:50:26 -0700830extern void edac_pci_reset_delay_period(struct edac_pci_ctl_info *pci,
831 unsigned long value);
Dave Jiang91b99042007-07-19 01:49:52 -0700832
833extern int edac_pci_add_device(struct edac_pci_ctl_info *pci, int edac_idx);
Douglas Thompson079708b2007-07-19 01:49:58 -0700834extern struct edac_pci_ctl_info *edac_pci_del_device(struct device *dev);
Dave Jiang91b99042007-07-19 01:49:52 -0700835
Doug Thompsonb8f6f972007-07-19 01:50:26 -0700836extern struct edac_pci_ctl_info *edac_pci_create_generic_ctl(
837 struct device *dev,
838 const char *mod_name);
Dave Jiang91b99042007-07-19 01:49:52 -0700839
840extern void edac_pci_release_generic_ctl(struct edac_pci_ctl_info *pci);
841extern int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci);
842extern void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci);
843
844/*
845 * edac misc APIs
846 */
Douglas Thompson494d0d52007-07-19 01:50:21 -0700847extern char *edac_op_state_to_string(int op_state);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700848
849#endif /* _EDAC_CORE_H_ */