blob: 8c114572b722a8ef68127f26cb317555a0279fbd [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
Douglas Thompson42a8e392007-07-19 01:50:10 -0700329/* mcidev_sysfs_attribute structure
330 * used for driver sysfs attributes and in mem_ctl_info
331 * sysfs top level entries
332 */
333struct mcidev_sysfs_attribute {
334 struct attribute attr;
335 ssize_t (*show)(struct mem_ctl_info *,char *);
336 ssize_t (*store)(struct mem_ctl_info *, const char *,size_t);
337};
338
339/* MEMORY controller information structure
340 */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700341struct mem_ctl_info {
Douglas Thompson079708b2007-07-19 01:49:58 -0700342 struct list_head link; /* for global list of mem_ctl_info structs */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700343 unsigned long mtype_cap; /* memory types supported by mc */
344 unsigned long edac_ctl_cap; /* Mem controller EDAC capabilities */
345 unsigned long edac_cap; /* configuration capabilities - this is
346 * closely related to edac_ctl_cap. The
347 * difference is that the controller may be
348 * capable of s4ecd4ed which would be listed
349 * in edac_ctl_cap, but if channels aren't
350 * capable of s4ecd4ed then the edac_cap would
351 * not have that capability.
352 */
353 unsigned long scrub_cap; /* chipset scrub capabilities */
354 enum scrub_type scrub_mode; /* current scrub mode */
355
356 /* Translates sdram memory scrub rate given in bytes/sec to the
357 internal representation and configures whatever else needs
358 to be configured.
Douglas Thompson079708b2007-07-19 01:49:58 -0700359 */
360 int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 * bw);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700361
362 /* Get the current sdram memory scrub rate from the internal
363 representation and converts it to the closest matching
364 bandwith in bytes/sec.
Douglas Thompson079708b2007-07-19 01:49:58 -0700365 */
366 int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 * bw);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700367
Douglas Thompson42a8e392007-07-19 01:50:10 -0700368
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700369 /* pointer to edac checking routine */
370 void (*edac_check) (struct mem_ctl_info * mci);
371
372 /*
373 * Remaps memory pages: controller pages to physical pages.
374 * For most MC's, this will be NULL.
375 */
376 /* FIXME - why not send the phys page to begin with? */
377 unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700378 unsigned long page);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700379 int mc_idx;
380 int nr_csrows;
381 struct csrow_info *csrows;
382 /*
383 * FIXME - what about controllers on other busses? - IDs must be
384 * unique. dev pointer should be sufficiently unique, but
385 * BUS:SLOT.FUNC numbers may not be unique.
386 */
387 struct device *dev;
388 const char *mod_name;
389 const char *mod_ver;
390 const char *ctl_name;
Dave Jiangc4192702007-07-19 01:49:47 -0700391 const char *dev_name;
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700392 char proc_name[MC_PROC_NAME_MAX_LEN + 1];
393 void *pvt_info;
394 u32 ue_noinfo_count; /* Uncorrectable Errors w/o info */
395 u32 ce_noinfo_count; /* Correctable Errors w/o info */
396 u32 ue_count; /* Total Uncorrectable Errors for this MC */
397 u32 ce_count; /* Total Correctable Errors for this MC */
398 unsigned long start_time; /* mci load start time (in jiffies) */
399
400 /* this stuff is for safe removal of mc devices from global list while
401 * NMI handlers may be traversing list
402 */
403 struct rcu_head rcu;
404 struct completion complete;
405
406 /* edac sysfs device control */
407 struct kobject edac_mci_kobj;
408 struct completion kobj_complete;
Dave Jiang81d87cb2007-07-19 01:49:52 -0700409
Douglas Thompson42a8e392007-07-19 01:50:10 -0700410 /* Additional top controller level attributes, but specified
411 * by the low level driver.
412 *
413 * Set by the low level driver to provide attributes at the
414 * controller level, same level as 'ue_count' and 'ce_count' above.
415 * An array of structures, NULL terminated
416 *
417 * If attributes are desired, then set to array of attributes
418 * If no attributes are desired, leave NULL
419 */
420 struct mcidev_sysfs_attribute *mc_driver_sysfs_attributes;
421
Dave Jiang81d87cb2007-07-19 01:49:52 -0700422 /* work struct for this MC */
Dave Jiang81d87cb2007-07-19 01:49:52 -0700423 struct delayed_work work;
Douglas Thompson86aa8cb2007-07-19 01:50:01 -0700424
Dave Jiang81d87cb2007-07-19 01:49:52 -0700425 /* the internal state of this controller instance */
426 int op_state;
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700427};
428
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700429/*
Douglas Thompson42a8e392007-07-19 01:50:10 -0700430 * The following are the structures to provide for a generic
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700431 * or abstract 'edac_device'. This set of structures and the
432 * code that implements the APIs for the same, provide for
433 * registering EDAC type devices which are NOT standard memory.
434 *
435 * CPU caches (L1 and L2)
436 * DMA engines
437 * Core CPU swithces
438 * Fabric switch units
439 * PCIe interface controllers
440 * other EDAC/ECC type devices that can be monitored for
441 * errors, etc.
442 *
443 * It allows for a 2 level set of hiearchry. For example:
444 *
445 * cache could be composed of L1, L2 and L3 levels of cache.
446 * Each CPU core would have its own L1 cache, while sharing
447 * L2 and maybe L3 caches.
448 *
449 * View them arranged, via the sysfs presentation:
450 * /sys/devices/system/edac/..
451 *
452 * mc/ <existing memory device directory>
453 * cpu/cpu0/.. <L1 and L2 block directory>
454 * /L1-cache/ce_count
455 * /ue_count
456 * /L2-cache/ce_count
457 * /ue_count
458 * cpu/cpu1/.. <L1 and L2 block directory>
459 * /L1-cache/ce_count
460 * /ue_count
461 * /L2-cache/ce_count
462 * /ue_count
463 * ...
464 *
465 * the L1 and L2 directories would be "edac_device_block's"
466 */
467
468struct edac_device_counter {
Douglas Thompson079708b2007-07-19 01:49:58 -0700469 u32 ue_count;
470 u32 ce_count;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700471};
472
473#define INC_COUNTER(cnt) (cnt++)
474
475/*
476 * An array of these is passed to the alloc() function
477 * to specify attributes of the edac_block
478 */
479struct edac_attrib_spec {
Douglas Thompson079708b2007-07-19 01:49:58 -0700480 char name[EDAC_DEVICE_NAME_LEN + 1];
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700481
482 int type;
483#define EDAC_ATTR_INT 0x01
484#define EDAC_ATTR_CHAR 0x02
485};
486
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700487/* Attribute control structure
488 * In this structure is a pointer to the driver's edac_attrib_spec
489 * The life of this pointer is inclusive in the life of the driver's
490 * life cycle.
491 */
492struct edac_attrib {
493 struct edac_device_block *block; /* Up Pointer */
494
Douglas Thompson079708b2007-07-19 01:49:58 -0700495 struct edac_attrib_spec *spec; /* ptr to module spec entry */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700496
Douglas Thompson079708b2007-07-19 01:49:58 -0700497 union { /* actual value */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700498 int edac_attrib_int_value;
499 char edac_attrib_char_value[EDAC_ATTRIB_VALUE_LEN + 1];
500 } edac_attrib_value;
501};
502
503/* device block control structure */
504struct edac_device_block {
505 struct edac_device_instance *instance; /* Up Pointer */
Douglas Thompson079708b2007-07-19 01:49:58 -0700506 char name[EDAC_DEVICE_NAME_LEN + 1];
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700507
508 struct edac_device_counter counters; /* basic UE and CE counters */
509
Douglas Thompson079708b2007-07-19 01:49:58 -0700510 int nr_attribs; /* how many attributes */
511 struct edac_attrib *attribs; /* this block's attributes */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700512
513 /* edac sysfs device control */
514 struct kobject kobj;
515 struct completion kobj_complete;
516};
517
518/* device instance control structure */
519struct edac_device_instance {
520 struct edac_device_ctl_info *ctl; /* Up pointer */
521 char name[EDAC_DEVICE_NAME_LEN + 4];
522
523 struct edac_device_counter counters; /* instance counters */
524
Douglas Thompson079708b2007-07-19 01:49:58 -0700525 u32 nr_blocks; /* how many blocks */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700526 struct edac_device_block *blocks; /* block array */
527
528 /* edac sysfs device control */
529 struct kobject kobj;
530 struct completion kobj_complete;
531};
532
Douglas Thompson42a8e392007-07-19 01:50:10 -0700533/* edac_dev_sysfs_attribute structure
534 * used for driver sysfs attributes and in mem_ctl_info
535 * sysfs top level entries
536 */
537struct edac_dev_sysfs_attribute {
538 struct attribute attr;
539 ssize_t (*show)(struct edac_device_ctl_info *,char *);
540 ssize_t (*store)(struct edac_device_ctl_info *, const char *,size_t);
541};
542
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700543/*
544 * Abstract edac_device control info structure
545 *
546 */
547struct edac_device_ctl_info {
548 /* for global list of edac_device_ctl_info structs */
549 struct list_head link;
550
551 int dev_idx;
552
553 /* Per instance controls for this edac_device */
554 int log_ue; /* boolean for logging UEs */
555 int log_ce; /* boolean for logging CEs */
556 int panic_on_ue; /* boolean for panic'ing on an UE */
557 unsigned poll_msec; /* number of milliseconds to poll interval */
558 unsigned long delay; /* number of jiffies for poll_msec */
559
Douglas Thompson42a8e392007-07-19 01:50:10 -0700560 /* Additional top controller level attributes, but specified
561 * by the low level driver.
562 *
563 * Set by the low level driver to provide attributes at the
564 * controller level, same level as 'ue_count' and 'ce_count' above.
565 * An array of structures, NULL terminated
566 *
567 * If attributes are desired, then set to array of attributes
568 * If no attributes are desired, leave NULL
569 */
570 struct edac_dev_sysfs_attribute *sysfs_attributes;
571
572 /* pointer to main 'edac' class in sysfs */
573 struct sysdev_class *edac_class;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700574
575 /* the internal state of this controller instance */
576 int op_state;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700577 /* work struct for this instance */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700578 struct delayed_work work;
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700579
580 /* pointer to edac polling checking routine:
Douglas Thompson079708b2007-07-19 01:49:58 -0700581 * If NOT NULL: points to polling check routine
582 * If NULL: Then assumes INTERRUPT operation, where
583 * MC driver will receive events
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700584 */
585 void (*edac_check) (struct edac_device_ctl_info * edac_dev);
586
587 struct device *dev; /* pointer to device structure */
588
589 const char *mod_name; /* module name */
590 const char *ctl_name; /* edac controller name */
Dave Jiangc4192702007-07-19 01:49:47 -0700591 const char *dev_name; /* pci/platform/etc... name */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700592
593 void *pvt_info; /* pointer to 'private driver' info */
594
Douglas Thompson079708b2007-07-19 01:49:58 -0700595 unsigned long start_time; /* edac_device load start time (jiffies) */
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700596
597 /* these are for safe removal of mc devices from global list while
598 * NMI handlers may be traversing list
599 */
600 struct rcu_head rcu;
601 struct completion complete;
602
603 /* sysfs top name under 'edac' directory
604 * and instance name:
Douglas Thompson079708b2007-07-19 01:49:58 -0700605 * cpu/cpu0/...
606 * cpu/cpu1/...
607 * cpu/cpu2/...
608 * ...
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700609 */
610 char name[EDAC_DEVICE_NAME_LEN + 1];
611
612 /* Number of instances supported on this control structure
613 * and the array of those instances
614 */
615 u32 nr_instances;
616 struct edac_device_instance *instances;
617
618 /* Event counters for the this whole EDAC Device */
619 struct edac_device_counter counters;
620
621 /* edac sysfs device control for the 'name'
622 * device this structure controls
623 */
624 struct kobject kobj;
625 struct completion kobj_complete;
626};
627
628/* To get from the instance's wq to the beginning of the ctl structure */
Dave Jiang81d87cb2007-07-19 01:49:52 -0700629#define to_edac_mem_ctl_work(w) \
630 container_of(w, struct mem_ctl_info, work)
631
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700632#define to_edac_device_ctl_work(w) \
633 container_of(w,struct edac_device_ctl_info,work)
634
635/* Function to calc the number of delay jiffies from poll_msec */
Douglas Thompson079708b2007-07-19 01:49:58 -0700636static inline void edac_device_calc_delay(struct edac_device_ctl_info *edac_dev)
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700637{
638 /* convert from msec to jiffies */
639 edac_dev->delay = edac_dev->poll_msec * HZ / 1000;
640}
641
Dave Jiang81d87cb2007-07-19 01:49:52 -0700642#define edac_calc_delay(dev) dev->delay = dev->poll_msec * HZ / 1000;
643
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700644/*
645 * The alloc() and free() functions for the 'edac_device' control info
646 * structure. A MC driver will allocate one of these for each edac_device
647 * it is going to control/register with the EDAC CORE.
648 */
649extern struct edac_device_ctl_info *edac_device_alloc_ctl_info(
Douglas Thompson079708b2007-07-19 01:49:58 -0700650 unsigned sizeof_private,
651 char *edac_device_name,
652 unsigned nr_instances,
653 char *edac_block_name,
654 unsigned nr_blocks,
655 unsigned offset_value,
656 struct edac_attrib_spec *attrib_spec,
657 unsigned nr_attribs);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700658
659/* The offset value can be:
660 * -1 indicating no offset value
661 * 0 for zero-based block numbers
662 * 1 for 1-based block number
663 * other for other-based block number
664 */
665#define BLOCK_OFFSET_VALUE_OFF ((unsigned) -1)
666
Douglas Thompson079708b2007-07-19 01:49:58 -0700667extern void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700668
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700669#ifdef CONFIG_PCI
670
Dave Jiang91b99042007-07-19 01:49:52 -0700671struct edac_pci_counter {
Douglas Thompson079708b2007-07-19 01:49:58 -0700672 atomic_t pe_count;
673 atomic_t npe_count;
Dave Jiang91b99042007-07-19 01:49:52 -0700674};
675
676/*
677 * Abstract edac_pci control info structure
678 *
679 */
680struct edac_pci_ctl_info {
681 /* for global list of edac_pci_ctl_info structs */
682 struct list_head link;
683
684 int pci_idx;
685
Dave Jiang91b99042007-07-19 01:49:52 -0700686 struct sysdev_class *edac_class; /* pointer to class */
687
688 /* the internal state of this controller instance */
689 int op_state;
690 /* work struct for this instance */
Dave Jiang91b99042007-07-19 01:49:52 -0700691 struct delayed_work work;
Dave Jiang91b99042007-07-19 01:49:52 -0700692
693 /* pointer to edac polling checking routine:
Douglas Thompson079708b2007-07-19 01:49:58 -0700694 * If NOT NULL: points to polling check routine
695 * If NULL: Then assumes INTERRUPT operation, where
696 * MC driver will receive events
Dave Jiang91b99042007-07-19 01:49:52 -0700697 */
698 void (*edac_check) (struct edac_pci_ctl_info * edac_dev);
699
700 struct device *dev; /* pointer to device structure */
701
702 const char *mod_name; /* module name */
703 const char *ctl_name; /* edac controller name */
704 const char *dev_name; /* pci/platform/etc... name */
705
706 void *pvt_info; /* pointer to 'private driver' info */
707
Douglas Thompson079708b2007-07-19 01:49:58 -0700708 unsigned long start_time; /* edac_pci load start time (jiffies) */
Dave Jiang91b99042007-07-19 01:49:52 -0700709
710 /* these are for safe removal of devices from global list while
711 * NMI handlers may be traversing list
712 */
713 struct rcu_head rcu;
714 struct completion complete;
715
716 /* sysfs top name under 'edac' directory
717 * and instance name:
Douglas Thompson079708b2007-07-19 01:49:58 -0700718 * cpu/cpu0/...
719 * cpu/cpu1/...
720 * cpu/cpu2/...
721 * ...
Dave Jiang91b99042007-07-19 01:49:52 -0700722 */
723 char name[EDAC_DEVICE_NAME_LEN + 1];
724
725 /* Event counters for the this whole EDAC Device */
726 struct edac_pci_counter counters;
727
728 /* edac sysfs device control for the 'name'
729 * device this structure controls
730 */
731 struct kobject kobj;
732 struct completion kobj_complete;
733};
734
735#define to_edac_pci_ctl_work(w) \
736 container_of(w, struct edac_pci_ctl_info,work)
737
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700738/* write all or some bits in a byte-register*/
739static inline void pci_write_bits8(struct pci_dev *pdev, int offset, u8 value,
Douglas Thompson079708b2007-07-19 01:49:58 -0700740 u8 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700741{
742 if (mask != 0xff) {
743 u8 buf;
744
745 pci_read_config_byte(pdev, offset, &buf);
746 value &= mask;
747 buf &= ~mask;
748 value |= buf;
749 }
750
751 pci_write_config_byte(pdev, offset, value);
752}
753
754/* write all or some bits in a word-register*/
755static inline void pci_write_bits16(struct pci_dev *pdev, int offset,
Douglas Thompson079708b2007-07-19 01:49:58 -0700756 u16 value, u16 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700757{
758 if (mask != 0xffff) {
759 u16 buf;
760
761 pci_read_config_word(pdev, offset, &buf);
762 value &= mask;
763 buf &= ~mask;
764 value |= buf;
765 }
766
767 pci_write_config_word(pdev, offset, value);
768}
769
770/* write all or some bits in a dword-register*/
771static inline void pci_write_bits32(struct pci_dev *pdev, int offset,
Douglas Thompson079708b2007-07-19 01:49:58 -0700772 u32 value, u32 mask)
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700773{
774 if (mask != 0xffff) {
775 u32 buf;
776
777 pci_read_config_dword(pdev, offset, &buf);
778 value &= mask;
779 buf &= ~mask;
780 value |= buf;
781 }
782
783 pci_write_config_dword(pdev, offset, value);
784}
785
Douglas Thompson079708b2007-07-19 01:49:58 -0700786#endif /* CONFIG_PCI */
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700787
Douglas Thompson079708b2007-07-19 01:49:58 -0700788extern struct mem_ctl_info *edac_mc_find(int idx);
789extern int edac_mc_add_mc(struct mem_ctl_info *mci, int mc_idx);
790extern struct mem_ctl_info *edac_mc_del_mc(struct device *dev);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700791extern int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700792 unsigned long page);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700793
794/*
795 * The no info errors are used when error overflows are reported.
796 * There are a limited number of error logging registers that can
797 * be exausted. When all registers are exhausted and an additional
798 * error occurs then an error overflow register records that an
799 * error occured and the type of error, but doesn't have any
800 * further information. The ce/ue versions make for cleaner
801 * reporting logic and function interface - reduces conditional
802 * statement clutter and extra function arguments.
803 */
804extern void edac_mc_handle_ce(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700805 unsigned long page_frame_number,
806 unsigned long offset_in_page,
807 unsigned long syndrome, int row, int channel,
808 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700809extern void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700810 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700811extern void edac_mc_handle_ue(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700812 unsigned long page_frame_number,
813 unsigned long offset_in_page, int row,
814 const char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700815extern void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci,
Douglas Thompson079708b2007-07-19 01:49:58 -0700816 const char *msg);
817extern void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci, unsigned int csrow,
818 unsigned int channel0, unsigned int channel1,
819 char *msg);
820extern void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci, unsigned int csrow,
821 unsigned int channel, char *msg);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700822
823/*
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700824 * edac_device APIs
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700825 */
826extern struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
Douglas Thompson079708b2007-07-19 01:49:58 -0700827 unsigned nr_chans);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700828extern void edac_mc_free(struct mem_ctl_info *mci);
Douglas Thompson079708b2007-07-19 01:49:58 -0700829extern int edac_device_add_device(struct edac_device_ctl_info *edac_dev,
830 int edac_idx);
831extern struct edac_device_ctl_info *edac_device_del_device(struct device *dev);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700832extern void edac_device_handle_ue(struct edac_device_ctl_info *edac_dev,
Douglas Thompson079708b2007-07-19 01:49:58 -0700833 int inst_nr, int block_nr, const char *msg);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700834extern void edac_device_handle_ce(struct edac_device_ctl_info *edac_dev,
Douglas Thompson079708b2007-07-19 01:49:58 -0700835 int inst_nr, int block_nr, const char *msg);
Douglas Thompsone27e3da2007-07-19 01:49:36 -0700836
Dave Jiang91b99042007-07-19 01:49:52 -0700837/*
838 * edac_pci APIs
839 */
Douglas Thompson079708b2007-07-19 01:49:58 -0700840extern struct edac_pci_ctl_info *edac_pci_alloc_ctl_info(unsigned int sz_pvt, const char
841 *edac_pci_name);
Dave Jiang91b99042007-07-19 01:49:52 -0700842
843extern void edac_pci_free_ctl_info(struct edac_pci_ctl_info *pci);
844
845extern void
846edac_pci_reset_delay_period(struct edac_pci_ctl_info *pci, unsigned long value);
847
848extern int edac_pci_add_device(struct edac_pci_ctl_info *pci, int edac_idx);
Douglas Thompson079708b2007-07-19 01:49:58 -0700849extern struct edac_pci_ctl_info *edac_pci_del_device(struct device *dev);
Dave Jiang91b99042007-07-19 01:49:52 -0700850
Douglas Thompson079708b2007-07-19 01:49:58 -0700851extern struct edac_pci_ctl_info *edac_pci_create_generic_ctl(struct device *dev, const char
852 *mod_name);
Dave Jiang91b99042007-07-19 01:49:52 -0700853
854extern void edac_pci_release_generic_ctl(struct edac_pci_ctl_info *pci);
855extern int edac_pci_create_sysfs(struct edac_pci_ctl_info *pci);
856extern void edac_pci_remove_sysfs(struct edac_pci_ctl_info *pci);
857
858/*
859 * edac misc APIs
860 */
Douglas Thompson079708b2007-07-19 01:49:58 -0700861extern char *edac_op_state_toString(int op_state);
Douglas Thompson7c9281d2007-07-19 01:49:33 -0700862
863#endif /* _EDAC_CORE_H_ */