blob: b955c58672e2e83ae5184a98ce97df40aa6b2911 [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) \
Douglas Thompsone27e3da2007-07-19 01:49:36 -070078 edac_printk(KERN_EMERG, 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
97#define BIT(x) (1 << (x))
98
99#define PCI_VEND_DEV(vend, dev) PCI_VENDOR_ID_ ## vend, \
100 PCI_DEVICE_ID_ ## vend ## _ ## dev
101
Dave Jiangc4192702007-07-19 01:49:47 -0700102#define dev_name(dev) (dev)->dev_name
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700103
104/* memory devices */
105enum dev_type {
106 DEV_UNKNOWN = 0,
107 DEV_X1,
108 DEV_X2,
109 DEV_X4,
110 DEV_X8,
111 DEV_X16,
112 DEV_X32, /* Do these parts exist? */
113 DEV_X64 /* Do these parts exist? */
114};
115
116#define DEV_FLAG_UNKNOWN BIT(DEV_UNKNOWN)
117#define DEV_FLAG_X1 BIT(DEV_X1)
118#define DEV_FLAG_X2 BIT(DEV_X2)
119#define DEV_FLAG_X4 BIT(DEV_X4)
120#define DEV_FLAG_X8 BIT(DEV_X8)
121#define DEV_FLAG_X16 BIT(DEV_X16)
122#define DEV_FLAG_X32 BIT(DEV_X32)
123#define DEV_FLAG_X64 BIT(DEV_X64)
124
125/* memory types */
126enum mem_type {
127 MEM_EMPTY = 0, /* Empty csrow */
128 MEM_RESERVED, /* Reserved csrow type */
129 MEM_UNKNOWN, /* Unknown csrow type */
130 MEM_FPM, /* Fast page mode */
131 MEM_EDO, /* Extended data out */
132 MEM_BEDO, /* Burst Extended data out */
133 MEM_SDR, /* Single data rate SDRAM */
134 MEM_RDR, /* Registered single data rate SDRAM */
135 MEM_DDR, /* Double data rate SDRAM */
136 MEM_RDDR, /* Registered Double data rate SDRAM */
137 MEM_RMBS, /* Rambus DRAM */
Douglas Thompson079708b2007-07-19 01:49:58 -0700138 MEM_DDR2, /* DDR2 RAM */
139 MEM_FB_DDR2, /* fully buffered DDR2 */
140 MEM_RDDR2, /* Registered DDR2 RAM */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700141};
142
143#define MEM_FLAG_EMPTY BIT(MEM_EMPTY)
144#define MEM_FLAG_RESERVED BIT(MEM_RESERVED)
145#define MEM_FLAG_UNKNOWN BIT(MEM_UNKNOWN)
146#define MEM_FLAG_FPM BIT(MEM_FPM)
147#define MEM_FLAG_EDO BIT(MEM_EDO)
148#define MEM_FLAG_BEDO BIT(MEM_BEDO)
149#define MEM_FLAG_SDR BIT(MEM_SDR)
150#define MEM_FLAG_RDR BIT(MEM_RDR)
151#define MEM_FLAG_DDR BIT(MEM_DDR)
152#define MEM_FLAG_RDDR BIT(MEM_RDDR)
153#define MEM_FLAG_RMBS BIT(MEM_RMBS)
154#define MEM_FLAG_DDR2 BIT(MEM_DDR2)
155#define MEM_FLAG_FB_DDR2 BIT(MEM_FB_DDR2)
156#define MEM_FLAG_RDDR2 BIT(MEM_RDDR2)
157
158/* chipset Error Detection and Correction capabilities and mode */
159enum edac_type {
160 EDAC_UNKNOWN = 0, /* Unknown if ECC is available */
161 EDAC_NONE, /* Doesnt support ECC */
162 EDAC_RESERVED, /* Reserved ECC type */
163 EDAC_PARITY, /* Detects parity errors */
164 EDAC_EC, /* Error Checking - no correction */
165 EDAC_SECDED, /* Single bit error correction, Double detection */
166 EDAC_S2ECD2ED, /* Chipkill x2 devices - do these exist? */
167 EDAC_S4ECD4ED, /* Chipkill x4 devices */
168 EDAC_S8ECD8ED, /* Chipkill x8 devices */
169 EDAC_S16ECD16ED, /* Chipkill x16 devices */
170};
171
172#define EDAC_FLAG_UNKNOWN BIT(EDAC_UNKNOWN)
173#define EDAC_FLAG_NONE BIT(EDAC_NONE)
174#define EDAC_FLAG_PARITY BIT(EDAC_PARITY)
175#define EDAC_FLAG_EC BIT(EDAC_EC)
176#define EDAC_FLAG_SECDED BIT(EDAC_SECDED)
177#define EDAC_FLAG_S2ECD2ED BIT(EDAC_S2ECD2ED)
178#define EDAC_FLAG_S4ECD4ED BIT(EDAC_S4ECD4ED)
179#define EDAC_FLAG_S8ECD8ED BIT(EDAC_S8ECD8ED)
180#define EDAC_FLAG_S16ECD16ED BIT(EDAC_S16ECD16ED)
181
182/* scrubbing capabilities */
183enum scrub_type {
184 SCRUB_UNKNOWN = 0, /* Unknown if scrubber is available */
185 SCRUB_NONE, /* No scrubber */
186 SCRUB_SW_PROG, /* SW progressive (sequential) scrubbing */
187 SCRUB_SW_SRC, /* Software scrub only errors */
188 SCRUB_SW_PROG_SRC, /* Progressive software scrub from an error */
189 SCRUB_SW_TUNABLE, /* Software scrub frequency is tunable */
190 SCRUB_HW_PROG, /* HW progressive (sequential) scrubbing */
191 SCRUB_HW_SRC, /* Hardware scrub only errors */
192 SCRUB_HW_PROG_SRC, /* Progressive hardware scrub from an error */
193 SCRUB_HW_TUNABLE /* Hardware scrub frequency is tunable */
194};
195
196#define SCRUB_FLAG_SW_PROG BIT(SCRUB_SW_PROG)
Douglas Thompson522a94b2007-07-19 01:49:41 -0700197#define SCRUB_FLAG_SW_SRC BIT(SCRUB_SW_SRC)
198#define SCRUB_FLAG_SW_PROG_SRC BIT(SCRUB_SW_PROG_SRC)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700199#define SCRUB_FLAG_SW_TUN BIT(SCRUB_SW_SCRUB_TUNABLE)
200#define SCRUB_FLAG_HW_PROG BIT(SCRUB_HW_PROG)
Douglas Thompson522a94b2007-07-19 01:49:41 -0700201#define SCRUB_FLAG_HW_SRC BIT(SCRUB_HW_SRC)
202#define SCRUB_FLAG_HW_PROG_SRC BIT(SCRUB_HW_PROG_SRC)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700203#define SCRUB_FLAG_HW_TUN BIT(SCRUB_HW_TUNABLE)
204
205/* FIXME - should have notify capabilities: NMI, LOG, PROC, etc */
206
Dave Jiang91b99042007-07-19 01:49:52 -0700207/* EDAC internal operation states */
208#define OP_ALLOC 0x100
209#define OP_RUNNING_POLL 0x201
210#define OP_RUNNING_INTERRUPT 0x202
211#define OP_RUNNING_POLL_INTR 0x203
212#define OP_OFFLINE 0x300
213
Douglas Thompson079708b2007-07-19 01:49:58 -0700214extern char *edac_align_ptr(void *ptr, unsigned size);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700215
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700216/*
217 * There are several things to be aware of that aren't at all obvious:
218 *
219 *
220 * SOCKETS, SOCKET SETS, BANKS, ROWS, CHIP-SELECT ROWS, CHANNELS, etc..
221 *
222 * These are some of the many terms that are thrown about that don't always
223 * mean what people think they mean (Inconceivable!). In the interest of
224 * creating a common ground for discussion, terms and their definitions
225 * will be established.
226 *
227 * Memory devices: The individual chip on a memory stick. These devices
228 * commonly output 4 and 8 bits each. Grouping several
229 * of these in parallel provides 64 bits which is common
230 * for a memory stick.
231 *
232 * Memory Stick: A printed circuit board that agregates multiple
233 * memory devices in parallel. This is the atomic
234 * memory component that is purchaseable by Joe consumer
235 * and loaded into a memory socket.
236 *
237 * Socket: A physical connector on the motherboard that accepts
238 * a single memory stick.
239 *
240 * Channel: Set of memory devices on a memory stick that must be
241 * grouped in parallel with one or more additional
242 * channels from other memory sticks. This parallel
243 * grouping of the output from multiple channels are
244 * necessary for the smallest granularity of memory access.
245 * Some memory controllers are capable of single channel -
246 * which means that memory sticks can be loaded
247 * individually. Other memory controllers are only
248 * capable of dual channel - which means that memory
249 * sticks must be loaded as pairs (see "socket set").
250 *
251 * Chip-select row: All of the memory devices that are selected together.
252 * for a single, minimum grain of memory access.
253 * This selects all of the parallel memory devices across
254 * all of the parallel channels. Common chip-select rows
255 * for single channel are 64 bits, for dual channel 128
256 * bits.
257 *
258 * Single-Ranked stick: A Single-ranked stick has 1 chip-select row of memmory.
259 * Motherboards commonly drive two chip-select pins to
260 * a memory stick. A single-ranked stick, will occupy
261 * only one of those rows. The other will be unused.
262 *
263 * Double-Ranked stick: A double-ranked stick has two chip-select rows which
264 * access different sets of memory devices. The two
265 * rows cannot be accessed concurrently.
266 *
267 * Double-sided stick: DEPRECATED TERM, see Double-Ranked stick.
268 * A double-sided stick has two chip-select rows which
269 * access different sets of memory devices. The two
270 * rows cannot be accessed concurrently. "Double-sided"
271 * is irrespective of the memory devices being mounted
272 * on both sides of the memory stick.
273 *
274 * Socket set: All of the memory sticks that are required for for
275 * a single memory access or all of the memory sticks
276 * spanned by a chip-select row. A single socket set
277 * has two chip-select rows and if double-sided sticks
278 * are used these will occupy those chip-select rows.
279 *
280 * Bank: This term is avoided because it is unclear when
281 * needing to distinguish between chip-select rows and
282 * socket sets.
283 *
284 * Controller pages:
285 *
286 * Physical pages:
287 *
288 * Virtual pages:
289 *
290 *
291 * STRUCTURE ORGANIZATION AND CHOICES
292 *
293 *
294 *
295 * PS - I enjoyed writing all that about as much as you enjoyed reading it.
296 */
297
298struct channel_info {
299 int chan_idx; /* channel index */
300 u32 ce_count; /* Correctable Errors for this CHANNEL */
Douglas Thompson079708b2007-07-19 01:49:58 -0700301 char label[EDAC_MC_LABEL_LEN + 1]; /* DIMM label on motherboard */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700302 struct csrow_info *csrow; /* the parent */
303};
304
305struct csrow_info {
306 unsigned long first_page; /* first page number in dimm */
307 unsigned long last_page; /* last page number in dimm */
308 unsigned long page_mask; /* used for interleaving -
309 * 0UL for non intlv
310 */
311 u32 nr_pages; /* number of pages in csrow */
312 u32 grain; /* granularity of reported error in bytes */
313 int csrow_idx; /* the chip-select row */
314 enum dev_type dtype; /* memory device type */
315 u32 ue_count; /* Uncorrectable Errors for this csrow */
316 u32 ce_count; /* Correctable Errors for this csrow */
317 enum mem_type mtype; /* memory csrow type */
318 enum edac_type edac_mode; /* EDAC mode for this csrow */
319 struct mem_ctl_info *mci; /* the parent */
320
321 struct kobject kobj; /* sysfs kobject for this csrow */
322 struct completion kobj_complete;
323
324 /* FIXME the number of CHANNELs might need to become dynamic */
325 u32 nr_channels;
326 struct channel_info *channels;
327};
328
329struct mem_ctl_info {
Douglas Thompson079708b2007-07-19 01:49:58 -0700330 struct list_head link; /* for global list of mem_ctl_info structs */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700331 unsigned long mtype_cap; /* memory types supported by mc */
332 unsigned long edac_ctl_cap; /* Mem controller EDAC capabilities */
333 unsigned long edac_cap; /* configuration capabilities - this is
334 * closely related to edac_ctl_cap. The
335 * difference is that the controller may be
336 * capable of s4ecd4ed which would be listed
337 * in edac_ctl_cap, but if channels aren't
338 * capable of s4ecd4ed then the edac_cap would
339 * not have that capability.
340 */
341 unsigned long scrub_cap; /* chipset scrub capabilities */
342 enum scrub_type scrub_mode; /* current scrub mode */
343
344 /* Translates sdram memory scrub rate given in bytes/sec to the
345 internal representation and configures whatever else needs
346 to be configured.
Douglas Thompson079708b2007-07-19 01:49:58 -0700347 */
348 int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 * bw);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700349
350 /* Get the current sdram memory scrub rate from the internal
351 representation and converts it to the closest matching
352 bandwith in bytes/sec.
Douglas Thompson079708b2007-07-19 01:49:58 -0700353 */
354 int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 * bw);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700355
356 /* pointer to edac checking routine */
357 void (*edac_check) (struct mem_ctl_info * mci);
358
359 /*
360 * Remaps memory pages: controller pages to physical pages.
361 * For most MC's, this will be NULL.
362 */
363 /* FIXME - why not send the phys page to begin with? */
364 unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700365 unsigned long page);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700366 int mc_idx;
367 int nr_csrows;
368 struct csrow_info *csrows;
369 /*
370 * FIXME - what about controllers on other busses? - IDs must be
371 * unique. dev pointer should be sufficiently unique, but
372 * BUS:SLOT.FUNC numbers may not be unique.
373 */
374 struct device *dev;
375 const char *mod_name;
376 const char *mod_ver;
377 const char *ctl_name;
Dave Jiangc4192702007-07-19 01:49:47 -0700378 const char *dev_name;
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700379 char proc_name[MC_PROC_NAME_MAX_LEN + 1];
380 void *pvt_info;
381 u32 ue_noinfo_count; /* Uncorrectable Errors w/o info */
382 u32 ce_noinfo_count; /* Correctable Errors w/o info */
383 u32 ue_count; /* Total Uncorrectable Errors for this MC */
384 u32 ce_count; /* Total Correctable Errors for this MC */
385 unsigned long start_time; /* mci load start time (in jiffies) */
386
387 /* this stuff is for safe removal of mc devices from global list while
388 * NMI handlers may be traversing list
389 */
390 struct rcu_head rcu;
391 struct completion complete;
392
393 /* edac sysfs device control */
394 struct kobject edac_mci_kobj;
395 struct completion kobj_complete;
Dave Jiang81d87cb2007-07-19 01:49:52 -0700396
397 /* work struct for this MC */
Dave Jiang81d87cb2007-07-19 01:49:52 -0700398 struct delayed_work work;
Douglas Thompson86aa8cb2007-07-19 01:50:01 -0700399
Dave Jiang81d87cb2007-07-19 01:49:52 -0700400 /* the internal state of this controller instance */
401 int op_state;
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700402};
403
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700404/*
405 * The following are the structures to provide for a generice
406 * or abstract 'edac_device'. This set of structures and the
407 * code that implements the APIs for the same, provide for
408 * registering EDAC type devices which are NOT standard memory.
409 *
410 * CPU caches (L1 and L2)
411 * DMA engines
412 * Core CPU swithces
413 * Fabric switch units
414 * PCIe interface controllers
415 * other EDAC/ECC type devices that can be monitored for
416 * errors, etc.
417 *
418 * It allows for a 2 level set of hiearchry. For example:
419 *
420 * cache could be composed of L1, L2 and L3 levels of cache.
421 * Each CPU core would have its own L1 cache, while sharing
422 * L2 and maybe L3 caches.
423 *
424 * View them arranged, via the sysfs presentation:
425 * /sys/devices/system/edac/..
426 *
427 * mc/ <existing memory device directory>
428 * cpu/cpu0/.. <L1 and L2 block directory>
429 * /L1-cache/ce_count
430 * /ue_count
431 * /L2-cache/ce_count
432 * /ue_count
433 * cpu/cpu1/.. <L1 and L2 block directory>
434 * /L1-cache/ce_count
435 * /ue_count
436 * /L2-cache/ce_count
437 * /ue_count
438 * ...
439 *
440 * the L1 and L2 directories would be "edac_device_block's"
441 */
442
443struct edac_device_counter {
Douglas Thompson079708b2007-07-19 01:49:58 -0700444 u32 ue_count;
445 u32 ce_count;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700446};
447
448#define INC_COUNTER(cnt) (cnt++)
449
450/*
451 * An array of these is passed to the alloc() function
452 * to specify attributes of the edac_block
453 */
454struct edac_attrib_spec {
Douglas Thompson079708b2007-07-19 01:49:58 -0700455 char name[EDAC_DEVICE_NAME_LEN + 1];
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700456
457 int type;
458#define EDAC_ATTR_INT 0x01
459#define EDAC_ATTR_CHAR 0x02
460};
461
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700462/* Attribute control structure
463 * In this structure is a pointer to the driver's edac_attrib_spec
464 * The life of this pointer is inclusive in the life of the driver's
465 * life cycle.
466 */
467struct edac_attrib {
468 struct edac_device_block *block; /* Up Pointer */
469
Douglas Thompson079708b2007-07-19 01:49:58 -0700470 struct edac_attrib_spec *spec; /* ptr to module spec entry */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700471
Douglas Thompson079708b2007-07-19 01:49:58 -0700472 union { /* actual value */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700473 int edac_attrib_int_value;
474 char edac_attrib_char_value[EDAC_ATTRIB_VALUE_LEN + 1];
475 } edac_attrib_value;
476};
477
478/* device block control structure */
479struct edac_device_block {
480 struct edac_device_instance *instance; /* Up Pointer */
Douglas Thompson079708b2007-07-19 01:49:58 -0700481 char name[EDAC_DEVICE_NAME_LEN + 1];
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700482
483 struct edac_device_counter counters; /* basic UE and CE counters */
484
Douglas Thompson079708b2007-07-19 01:49:58 -0700485 int nr_attribs; /* how many attributes */
486 struct edac_attrib *attribs; /* this block's attributes */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700487
488 /* edac sysfs device control */
489 struct kobject kobj;
490 struct completion kobj_complete;
491};
492
493/* device instance control structure */
494struct edac_device_instance {
495 struct edac_device_ctl_info *ctl; /* Up pointer */
496 char name[EDAC_DEVICE_NAME_LEN + 4];
497
498 struct edac_device_counter counters; /* instance counters */
499
Douglas Thompson079708b2007-07-19 01:49:58 -0700500 u32 nr_blocks; /* how many blocks */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700501 struct edac_device_block *blocks; /* block array */
502
503 /* edac sysfs device control */
504 struct kobject kobj;
505 struct completion kobj_complete;
506};
507
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700508/*
509 * Abstract edac_device control info structure
510 *
511 */
512struct edac_device_ctl_info {
513 /* for global list of edac_device_ctl_info structs */
514 struct list_head link;
515
516 int dev_idx;
517
518 /* Per instance controls for this edac_device */
519 int log_ue; /* boolean for logging UEs */
520 int log_ce; /* boolean for logging CEs */
521 int panic_on_ue; /* boolean for panic'ing on an UE */
522 unsigned poll_msec; /* number of milliseconds to poll interval */
523 unsigned long delay; /* number of jiffies for poll_msec */
524
525 struct sysdev_class *edac_class; /* pointer to class */
526
527 /* the internal state of this controller instance */
528 int op_state;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700529 /* work struct for this instance */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700530 struct delayed_work work;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700531
532 /* pointer to edac polling checking routine:
Douglas Thompson079708b2007-07-19 01:49:58 -0700533 * If NOT NULL: points to polling check routine
534 * If NULL: Then assumes INTERRUPT operation, where
535 * MC driver will receive events
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700536 */
537 void (*edac_check) (struct edac_device_ctl_info * edac_dev);
538
539 struct device *dev; /* pointer to device structure */
540
541 const char *mod_name; /* module name */
542 const char *ctl_name; /* edac controller name */
Dave Jiangc4192702007-07-19 01:49:47 -0700543 const char *dev_name; /* pci/platform/etc... name */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700544
545 void *pvt_info; /* pointer to 'private driver' info */
546
Douglas Thompson079708b2007-07-19 01:49:58 -0700547 unsigned long start_time; /* edac_device load start time (jiffies) */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700548
549 /* these are for safe removal of mc devices from global list while
550 * NMI handlers may be traversing list
551 */
552 struct rcu_head rcu;
553 struct completion complete;
554
555 /* sysfs top name under 'edac' directory
556 * and instance name:
Douglas Thompson079708b2007-07-19 01:49:58 -0700557 * cpu/cpu0/...
558 * cpu/cpu1/...
559 * cpu/cpu2/...
560 * ...
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700561 */
562 char name[EDAC_DEVICE_NAME_LEN + 1];
563
564 /* Number of instances supported on this control structure
565 * and the array of those instances
566 */
567 u32 nr_instances;
568 struct edac_device_instance *instances;
569
570 /* Event counters for the this whole EDAC Device */
571 struct edac_device_counter counters;
572
573 /* edac sysfs device control for the 'name'
574 * device this structure controls
575 */
576 struct kobject kobj;
577 struct completion kobj_complete;
578};
579
580/* To get from the instance's wq to the beginning of the ctl structure */
Dave Jiang81d87cb2007-07-19 01:49:52 -0700581#define to_edac_mem_ctl_work(w) \
582 container_of(w, struct mem_ctl_info, work)
583
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700584#define to_edac_device_ctl_work(w) \
585 container_of(w,struct edac_device_ctl_info,work)
586
587/* Function to calc the number of delay jiffies from poll_msec */
Douglas Thompson079708b2007-07-19 01:49:58 -0700588static inline void edac_device_calc_delay(struct edac_device_ctl_info *edac_dev)
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700589{
590 /* convert from msec to jiffies */
591 edac_dev->delay = edac_dev->poll_msec * HZ / 1000;
592}
593
Dave Jiang81d87cb2007-07-19 01:49:52 -0700594#define edac_calc_delay(dev) dev->delay = dev->poll_msec * HZ / 1000;
595
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700596/*
597 * The alloc() and free() functions for the 'edac_device' control info
598 * structure. A MC driver will allocate one of these for each edac_device
599 * it is going to control/register with the EDAC CORE.
600 */
601extern struct edac_device_ctl_info *edac_device_alloc_ctl_info(
Douglas Thompson079708b2007-07-19 01:49:58 -0700602 unsigned sizeof_private,
603 char *edac_device_name,
604 unsigned nr_instances,
605 char *edac_block_name,
606 unsigned nr_blocks,
607 unsigned offset_value,
608 struct edac_attrib_spec *attrib_spec,
609 unsigned nr_attribs);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700610
611/* The offset value can be:
612 * -1 indicating no offset value
613 * 0 for zero-based block numbers
614 * 1 for 1-based block number
615 * other for other-based block number
616 */
617#define BLOCK_OFFSET_VALUE_OFF ((unsigned) -1)
618
Douglas Thompson079708b2007-07-19 01:49:58 -0700619extern void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700620
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700621#ifdef CONFIG_PCI
622
Dave Jiang91b99042007-07-19 01:49:52 -0700623struct edac_pci_counter {
Douglas Thompson079708b2007-07-19 01:49:58 -0700624 atomic_t pe_count;
625 atomic_t npe_count;
Dave Jiang91b99042007-07-19 01:49:52 -0700626};
627
628/*
629 * Abstract edac_pci control info structure
630 *
631 */
632struct edac_pci_ctl_info {
633 /* for global list of edac_pci_ctl_info structs */
634 struct list_head link;
635
636 int pci_idx;
637
Dave Jiang91b99042007-07-19 01:49:52 -0700638 struct sysdev_class *edac_class; /* pointer to class */
639
640 /* the internal state of this controller instance */
641 int op_state;
642 /* work struct for this instance */
Dave Jiang91b99042007-07-19 01:49:52 -0700643 struct delayed_work work;
Dave Jiang91b99042007-07-19 01:49:52 -0700644
645 /* pointer to edac polling checking routine:
Douglas Thompson079708b2007-07-19 01:49:58 -0700646 * If NOT NULL: points to polling check routine
647 * If NULL: Then assumes INTERRUPT operation, where
648 * MC driver will receive events
Dave Jiang91b99042007-07-19 01:49:52 -0700649 */
650 void (*edac_check) (struct edac_pci_ctl_info * edac_dev);
651
652 struct device *dev; /* pointer to device structure */
653
654 const char *mod_name; /* module name */
655 const char *ctl_name; /* edac controller name */
656 const char *dev_name; /* pci/platform/etc... name */
657
658 void *pvt_info; /* pointer to 'private driver' info */
659
Douglas Thompson079708b2007-07-19 01:49:58 -0700660 unsigned long start_time; /* edac_pci load start time (jiffies) */
Dave Jiang91b99042007-07-19 01:49:52 -0700661
662 /* these are for safe removal of devices from global list while
663 * NMI handlers may be traversing list
664 */
665 struct rcu_head rcu;
666 struct completion complete;
667
668 /* sysfs top name under 'edac' directory
669 * and instance name:
Douglas Thompson079708b2007-07-19 01:49:58 -0700670 * cpu/cpu0/...
671 * cpu/cpu1/...
672 * cpu/cpu2/...
673 * ...
Dave Jiang91b99042007-07-19 01:49:52 -0700674 */
675 char name[EDAC_DEVICE_NAME_LEN + 1];
676
677 /* Event counters for the this whole EDAC Device */
678 struct edac_pci_counter counters;
679
680 /* edac sysfs device control for the 'name'
681 * device this structure controls
682 */
683 struct kobject kobj;
684 struct completion kobj_complete;
685};
686
687#define to_edac_pci_ctl_work(w) \
688 container_of(w, struct edac_pci_ctl_info,work)
689
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700690/* write all or some bits in a byte-register*/
691static inline void pci_write_bits8(struct pci_dev *pdev, int offset, u8 value,
Douglas Thompson079708b2007-07-19 01:49:58 -0700692 u8 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700693{
694 if (mask != 0xff) {
695 u8 buf;
696
697 pci_read_config_byte(pdev, offset, &buf);
698 value &= mask;
699 buf &= ~mask;
700 value |= buf;
701 }
702
703 pci_write_config_byte(pdev, offset, value);
704}
705
706/* write all or some bits in a word-register*/
707static inline void pci_write_bits16(struct pci_dev *pdev, int offset,
Douglas Thompson079708b2007-07-19 01:49:58 -0700708 u16 value, u16 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700709{
710 if (mask != 0xffff) {
711 u16 buf;
712
713 pci_read_config_word(pdev, offset, &buf);
714 value &= mask;
715 buf &= ~mask;
716 value |= buf;
717 }
718
719 pci_write_config_word(pdev, offset, value);
720}
721
722/* write all or some bits in a dword-register*/
723static inline void pci_write_bits32(struct pci_dev *pdev, int offset,
Douglas Thompson079708b2007-07-19 01:49:58 -0700724 u32 value, u32 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700725{
726 if (mask != 0xffff) {
727 u32 buf;
728
729 pci_read_config_dword(pdev, offset, &buf);
730 value &= mask;
731 buf &= ~mask;
732 value |= buf;
733 }
734
735 pci_write_config_dword(pdev, offset, value);
736}
737
Douglas Thompson079708b2007-07-19 01:49:58 -0700738#endif /* CONFIG_PCI */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700739
Douglas Thompson079708b2007-07-19 01:49:58 -0700740extern struct mem_ctl_info *edac_mc_find(int idx);
741extern int edac_mc_add_mc(struct mem_ctl_info *mci, int mc_idx);
742extern struct mem_ctl_info *edac_mc_del_mc(struct device *dev);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700743extern int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700744 unsigned long page);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700745
746/*
747 * The no info errors are used when error overflows are reported.
748 * There are a limited number of error logging registers that can
749 * be exausted. When all registers are exhausted and an additional
750 * error occurs then an error overflow register records that an
751 * error occured and the type of error, but doesn't have any
752 * further information. The ce/ue versions make for cleaner
753 * reporting logic and function interface - reduces conditional
754 * statement clutter and extra function arguments.
755 */
756extern void edac_mc_handle_ce(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700757 unsigned long page_frame_number,
758 unsigned long offset_in_page,
759 unsigned long syndrome, int row, int channel,
760 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700761extern void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700762 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700763extern void edac_mc_handle_ue(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700764 unsigned long page_frame_number,
765 unsigned long offset_in_page, int row,
766 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700767extern void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700768 const char *msg);
769extern void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci, unsigned int csrow,
770 unsigned int channel0, unsigned int channel1,
771 char *msg);
772extern void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci, unsigned int csrow,
773 unsigned int channel, char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700774
775/*
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700776 * edac_device APIs
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700777 */
778extern struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
Douglas Thompson079708b2007-07-19 01:49:58 -0700779 unsigned nr_chans);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700780extern void edac_mc_free(struct mem_ctl_info *mci);
Douglas Thompson079708b2007-07-19 01:49:58 -0700781extern int edac_device_add_device(struct edac_device_ctl_info *edac_dev,
782 int edac_idx);
783extern struct edac_device_ctl_info *edac_device_del_device(struct device *dev);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700784extern void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
Douglas Thompson079708b2007-07-19 01:49:58 -0700785 int inst_nr, int block_nr, const char *msg);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700786extern void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
Douglas Thompson079708b2007-07-19 01:49:58 -0700787 int inst_nr, int block_nr, const char *msg);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700788
Dave Jiang91b99042007-07-19 01:49:52 -0700789/*
790 * edac_pci APIs
791 */
Douglas Thompson079708b2007-07-19 01:49:58 -0700792extern struct edac_pci_ctl_info *edac_pci_alloc_ctl_info(unsigned int sz_pvt, const char
793 *edac_pci_name);
Dave Jiang91b99042007-07-19 01:49:52 -0700794
795extern void edac_pci_free_ctl_info(struct edac_pci_ctl_info *pci);
796
797extern void
798edac_pci_reset_delay_period(struct edac_pci_ctl_info *pci, unsigned long value);
799
800extern int edac_pci_add_device(struct edac_pci_ctl_info *pci, int edac_idx);
Douglas Thompson079708b2007-07-19 01:49:58 -0700801extern struct edac_pci_ctl_info *edac_pci_del_device(struct device *dev);
Dave Jiang91b99042007-07-19 01:49:52 -0700802
Douglas Thompson079708b2007-07-19 01:49:58 -0700803extern struct edac_pci_ctl_info *edac_pci_create_generic_ctl(struct device *dev, const char
804 *mod_name);
Dave Jiang91b99042007-07-19 01:49:52 -0700805
806extern void edac_pci_release_generic_ctl(struct edac_pci_ctl_info *pci);
807extern int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci);
808extern void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci);
809
810/*
811 * edac misc APIs
812 */
Douglas Thompson079708b2007-07-19 01:49:58 -0700813extern char *edac_op_state_toString(int op_state);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700814
815#endif /* _EDAC_CORE_H_ */