]>
Commit | Line | Data |
---|---|---|
da9bb1d2 AC |
1 | /* |
2 | * edac_mc kernel module | |
49c0dab7 | 3 | * (C) 2005, 2006 Linux Networx (http://lnxi.com) |
da9bb1d2 AC |
4 | * This file may be distributed under the terms of the |
5 | * GNU General Public License. | |
6 | * | |
7 | * Written by Thayne Harbaugh | |
8 | * Based on work by Dan Hollis <goemon at anime dot net> and others. | |
9 | * http://www.anime.net/~goemon/linux-ecc/ | |
10 | * | |
11 | * Modified by Dave Peterson and Doug Thompson | |
12 | * | |
13 | */ | |
14 | ||
da9bb1d2 AC |
15 | #include <linux/module.h> |
16 | #include <linux/proc_fs.h> | |
17 | #include <linux/kernel.h> | |
18 | #include <linux/types.h> | |
19 | #include <linux/smp.h> | |
20 | #include <linux/init.h> | |
21 | #include <linux/sysctl.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/timer.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/jiffies.h> | |
26 | #include <linux/spinlock.h> | |
27 | #include <linux/list.h> | |
da9bb1d2 | 28 | #include <linux/ctype.h> |
c0d12172 | 29 | #include <linux/edac.h> |
da9bb1d2 AC |
30 | #include <asm/uaccess.h> |
31 | #include <asm/page.h> | |
32 | #include <asm/edac.h> | |
20bcb7a8 | 33 | #include "edac_core.h" |
7c9281d7 | 34 | #include "edac_module.h" |
da9bb1d2 | 35 | |
da9bb1d2 | 36 | /* lock to memory controller's control array */ |
63b7df91 | 37 | static DEFINE_MUTEX(mem_ctls_mutex); |
ff6ac2a6 | 38 | static LIST_HEAD(mc_devices); |
da9bb1d2 | 39 | |
da9bb1d2 AC |
40 | #ifdef CONFIG_EDAC_DEBUG |
41 | ||
a4b4be3f | 42 | static void edac_mc_dump_channel(struct rank_info *chan) |
da9bb1d2 AC |
43 | { |
44 | debugf4("\tchannel = %p\n", chan); | |
45 | debugf4("\tchannel->chan_idx = %d\n", chan->chan_idx); | |
46 | debugf4("\tchannel->ce_count = %d\n", chan->ce_count); | |
47 | debugf4("\tchannel->label = '%s'\n", chan->label); | |
48 | debugf4("\tchannel->csrow = %p\n\n", chan->csrow); | |
49 | } | |
50 | ||
2da1c119 | 51 | static void edac_mc_dump_csrow(struct csrow_info *csrow) |
da9bb1d2 AC |
52 | { |
53 | debugf4("\tcsrow = %p\n", csrow); | |
54 | debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx); | |
079708b9 | 55 | debugf4("\tcsrow->first_page = 0x%lx\n", csrow->first_page); |
da9bb1d2 AC |
56 | debugf4("\tcsrow->last_page = 0x%lx\n", csrow->last_page); |
57 | debugf4("\tcsrow->page_mask = 0x%lx\n", csrow->page_mask); | |
58 | debugf4("\tcsrow->nr_pages = 0x%x\n", csrow->nr_pages); | |
079708b9 | 59 | debugf4("\tcsrow->nr_channels = %d\n", csrow->nr_channels); |
da9bb1d2 AC |
60 | debugf4("\tcsrow->channels = %p\n", csrow->channels); |
61 | debugf4("\tcsrow->mci = %p\n\n", csrow->mci); | |
62 | } | |
63 | ||
2da1c119 | 64 | static void edac_mc_dump_mci(struct mem_ctl_info *mci) |
da9bb1d2 AC |
65 | { |
66 | debugf3("\tmci = %p\n", mci); | |
67 | debugf3("\tmci->mtype_cap = %lx\n", mci->mtype_cap); | |
68 | debugf3("\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap); | |
69 | debugf3("\tmci->edac_cap = %lx\n", mci->edac_cap); | |
70 | debugf4("\tmci->edac_check = %p\n", mci->edac_check); | |
71 | debugf3("\tmci->nr_csrows = %d, csrows = %p\n", | |
72 | mci->nr_csrows, mci->csrows); | |
37f04581 | 73 | debugf3("\tdev = %p\n", mci->dev); |
079708b9 | 74 | debugf3("\tmod_name:ctl_name = %s:%s\n", mci->mod_name, mci->ctl_name); |
da9bb1d2 AC |
75 | debugf3("\tpvt_info = %p\n\n", mci->pvt_info); |
76 | } | |
77 | ||
24f9a7fe BP |
78 | #endif /* CONFIG_EDAC_DEBUG */ |
79 | ||
239642fe BP |
80 | /* |
81 | * keep those in sync with the enum mem_type | |
82 | */ | |
83 | const char *edac_mem_types[] = { | |
84 | "Empty csrow", | |
85 | "Reserved csrow type", | |
86 | "Unknown csrow type", | |
87 | "Fast page mode RAM", | |
88 | "Extended data out RAM", | |
89 | "Burst Extended data out RAM", | |
90 | "Single data rate SDRAM", | |
91 | "Registered single data rate SDRAM", | |
92 | "Double data rate SDRAM", | |
93 | "Registered Double data rate SDRAM", | |
94 | "Rambus DRAM", | |
95 | "Unbuffered DDR2 RAM", | |
96 | "Fully buffered DDR2", | |
97 | "Registered DDR2 RAM", | |
98 | "Rambus XDR", | |
99 | "Unbuffered DDR3 RAM", | |
100 | "Registered DDR3 RAM", | |
101 | }; | |
102 | EXPORT_SYMBOL_GPL(edac_mem_types); | |
103 | ||
da9bb1d2 AC |
104 | /* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'. |
105 | * Adjust 'ptr' so that its alignment is at least as stringent as what the | |
106 | * compiler would provide for X and return the aligned result. | |
107 | * | |
108 | * If 'size' is a constant, the compiler will optimize this whole function | |
109 | * down to either a no-op or the addition of a constant to the value of 'ptr'. | |
110 | */ | |
7391c6dc | 111 | void *edac_align_ptr(void *ptr, unsigned size) |
da9bb1d2 AC |
112 | { |
113 | unsigned align, r; | |
114 | ||
115 | /* Here we assume that the alignment of a "long long" is the most | |
116 | * stringent alignment that the compiler will ever provide by default. | |
117 | * As far as I know, this is a reasonable assumption. | |
118 | */ | |
119 | if (size > sizeof(long)) | |
120 | align = sizeof(long long); | |
121 | else if (size > sizeof(int)) | |
122 | align = sizeof(long); | |
123 | else if (size > sizeof(short)) | |
124 | align = sizeof(int); | |
125 | else if (size > sizeof(char)) | |
126 | align = sizeof(short); | |
127 | else | |
079708b9 | 128 | return (char *)ptr; |
da9bb1d2 AC |
129 | |
130 | r = size % align; | |
131 | ||
132 | if (r == 0) | |
079708b9 | 133 | return (char *)ptr; |
da9bb1d2 | 134 | |
7391c6dc | 135 | return (void *)(((unsigned long)ptr) + align - r); |
da9bb1d2 AC |
136 | } |
137 | ||
da9bb1d2 AC |
138 | /** |
139 | * edac_mc_alloc: Allocate a struct mem_ctl_info structure | |
140 | * @size_pvt: size of private storage needed | |
141 | * @nr_csrows: Number of CWROWS needed for this MC | |
142 | * @nr_chans: Number of channels for the MC | |
143 | * | |
144 | * Everything is kmalloc'ed as one big chunk - more efficient. | |
145 | * Only can be used if all structures have the same lifetime - otherwise | |
146 | * you have to allocate and initialize your own structures. | |
147 | * | |
148 | * Use edac_mc_free() to free mc structures allocated by this function. | |
149 | * | |
150 | * Returns: | |
151 | * NULL allocation failed | |
152 | * struct mem_ctl_info pointer | |
153 | */ | |
154 | struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows, | |
b8f6f975 | 155 | unsigned nr_chans, int edac_index) |
da9bb1d2 AC |
156 | { |
157 | struct mem_ctl_info *mci; | |
158 | struct csrow_info *csi, *csrow; | |
a4b4be3f | 159 | struct rank_info *chi, *chp, *chan; |
da9bb1d2 AC |
160 | void *pvt; |
161 | unsigned size; | |
162 | int row, chn; | |
8096cfaf | 163 | int err; |
da9bb1d2 AC |
164 | |
165 | /* Figure out the offsets of the various items from the start of an mc | |
166 | * structure. We want the alignment of each item to be at least as | |
167 | * stringent as what the compiler would provide if we could simply | |
168 | * hardcode everything into a single struct. | |
169 | */ | |
079708b9 | 170 | mci = (struct mem_ctl_info *)0; |
7391c6dc DT |
171 | csi = edac_align_ptr(&mci[1], sizeof(*csi)); |
172 | chi = edac_align_ptr(&csi[nr_csrows], sizeof(*chi)); | |
e27e3dac | 173 | pvt = edac_align_ptr(&chi[nr_chans * nr_csrows], sz_pvt); |
079708b9 | 174 | size = ((unsigned long)pvt) + sz_pvt; |
da9bb1d2 | 175 | |
8096cfaf DT |
176 | mci = kzalloc(size, GFP_KERNEL); |
177 | if (mci == NULL) | |
da9bb1d2 AC |
178 | return NULL; |
179 | ||
180 | /* Adjust pointers so they point within the memory we just allocated | |
181 | * rather than an imaginary chunk of memory located at address 0. | |
182 | */ | |
079708b9 | 183 | csi = (struct csrow_info *)(((char *)mci) + ((unsigned long)csi)); |
a4b4be3f | 184 | chi = (struct rank_info *)(((char *)mci) + ((unsigned long)chi)); |
079708b9 | 185 | pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL; |
da9bb1d2 | 186 | |
b8f6f975 DT |
187 | /* setup index and various internal pointers */ |
188 | mci->mc_idx = edac_index; | |
da9bb1d2 AC |
189 | mci->csrows = csi; |
190 | mci->pvt_info = pvt; | |
191 | mci->nr_csrows = nr_csrows; | |
192 | ||
193 | for (row = 0; row < nr_csrows; row++) { | |
194 | csrow = &csi[row]; | |
195 | csrow->csrow_idx = row; | |
196 | csrow->mci = mci; | |
197 | csrow->nr_channels = nr_chans; | |
198 | chp = &chi[row * nr_chans]; | |
199 | csrow->channels = chp; | |
200 | ||
201 | for (chn = 0; chn < nr_chans; chn++) { | |
202 | chan = &chp[chn]; | |
203 | chan->chan_idx = chn; | |
204 | chan->csrow = csrow; | |
205 | } | |
206 | } | |
207 | ||
81d87cb1 | 208 | mci->op_state = OP_ALLOC; |
6fe1108f | 209 | INIT_LIST_HEAD(&mci->grp_kobj_list); |
81d87cb1 | 210 | |
8096cfaf DT |
211 | /* |
212 | * Initialize the 'root' kobj for the edac_mc controller | |
213 | */ | |
214 | err = edac_mc_register_sysfs_main_kobj(mci); | |
215 | if (err) { | |
216 | kfree(mci); | |
217 | return NULL; | |
218 | } | |
219 | ||
220 | /* at this point, the root kobj is valid, and in order to | |
221 | * 'free' the object, then the function: | |
222 | * edac_mc_unregister_sysfs_main_kobj() must be called | |
223 | * which will perform kobj unregistration and the actual free | |
224 | * will occur during the kobject callback operation | |
225 | */ | |
da9bb1d2 AC |
226 | return mci; |
227 | } | |
9110540f | 228 | EXPORT_SYMBOL_GPL(edac_mc_alloc); |
da9bb1d2 | 229 | |
da9bb1d2 | 230 | /** |
8096cfaf DT |
231 | * edac_mc_free |
232 | * 'Free' a previously allocated 'mci' structure | |
da9bb1d2 | 233 | * @mci: pointer to a struct mem_ctl_info structure |
da9bb1d2 AC |
234 | */ |
235 | void edac_mc_free(struct mem_ctl_info *mci) | |
236 | { | |
bbc560ae MCC |
237 | debugf1("%s()\n", __func__); |
238 | ||
8096cfaf | 239 | edac_mc_unregister_sysfs_main_kobj(mci); |
accf74ff MCC |
240 | |
241 | /* free the mci instance memory here */ | |
242 | kfree(mci); | |
da9bb1d2 | 243 | } |
9110540f | 244 | EXPORT_SYMBOL_GPL(edac_mc_free); |
da9bb1d2 | 245 | |
bce19683 | 246 | |
939747bd | 247 | /** |
bce19683 DT |
248 | * find_mci_by_dev |
249 | * | |
250 | * scan list of controllers looking for the one that manages | |
251 | * the 'dev' device | |
939747bd | 252 | * @dev: pointer to a struct device related with the MCI |
bce19683 | 253 | */ |
939747bd | 254 | struct mem_ctl_info *find_mci_by_dev(struct device *dev) |
da9bb1d2 AC |
255 | { |
256 | struct mem_ctl_info *mci; | |
257 | struct list_head *item; | |
258 | ||
537fba28 | 259 | debugf3("%s()\n", __func__); |
da9bb1d2 AC |
260 | |
261 | list_for_each(item, &mc_devices) { | |
262 | mci = list_entry(item, struct mem_ctl_info, link); | |
263 | ||
37f04581 | 264 | if (mci->dev == dev) |
da9bb1d2 AC |
265 | return mci; |
266 | } | |
267 | ||
268 | return NULL; | |
269 | } | |
939747bd | 270 | EXPORT_SYMBOL_GPL(find_mci_by_dev); |
da9bb1d2 | 271 | |
81d87cb1 DJ |
272 | /* |
273 | * handler for EDAC to check if NMI type handler has asserted interrupt | |
274 | */ | |
275 | static int edac_mc_assert_error_check_and_clear(void) | |
276 | { | |
66ee2f94 | 277 | int old_state; |
81d87cb1 | 278 | |
079708b9 | 279 | if (edac_op_state == EDAC_OPSTATE_POLL) |
81d87cb1 DJ |
280 | return 1; |
281 | ||
66ee2f94 DJ |
282 | old_state = edac_err_assert; |
283 | edac_err_assert = 0; | |
81d87cb1 | 284 | |
66ee2f94 | 285 | return old_state; |
81d87cb1 DJ |
286 | } |
287 | ||
288 | /* | |
289 | * edac_mc_workq_function | |
290 | * performs the operation scheduled by a workq request | |
291 | */ | |
81d87cb1 DJ |
292 | static void edac_mc_workq_function(struct work_struct *work_req) |
293 | { | |
fbeb4384 | 294 | struct delayed_work *d_work = to_delayed_work(work_req); |
81d87cb1 | 295 | struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work); |
81d87cb1 DJ |
296 | |
297 | mutex_lock(&mem_ctls_mutex); | |
298 | ||
bf52fa4a DT |
299 | /* if this control struct has movd to offline state, we are done */ |
300 | if (mci->op_state == OP_OFFLINE) { | |
301 | mutex_unlock(&mem_ctls_mutex); | |
302 | return; | |
303 | } | |
304 | ||
81d87cb1 DJ |
305 | /* Only poll controllers that are running polled and have a check */ |
306 | if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL)) | |
307 | mci->edac_check(mci); | |
308 | ||
81d87cb1 DJ |
309 | mutex_unlock(&mem_ctls_mutex); |
310 | ||
311 | /* Reschedule */ | |
4de78c68 | 312 | queue_delayed_work(edac_workqueue, &mci->work, |
052dfb45 | 313 | msecs_to_jiffies(edac_mc_get_poll_msec())); |
81d87cb1 DJ |
314 | } |
315 | ||
316 | /* | |
317 | * edac_mc_workq_setup | |
318 | * initialize a workq item for this mci | |
319 | * passing in the new delay period in msec | |
bf52fa4a DT |
320 | * |
321 | * locking model: | |
322 | * | |
323 | * called with the mem_ctls_mutex held | |
81d87cb1 | 324 | */ |
bf52fa4a | 325 | static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec) |
81d87cb1 DJ |
326 | { |
327 | debugf0("%s()\n", __func__); | |
328 | ||
bf52fa4a DT |
329 | /* if this instance is not in the POLL state, then simply return */ |
330 | if (mci->op_state != OP_RUNNING_POLL) | |
331 | return; | |
332 | ||
81d87cb1 | 333 | INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function); |
81d87cb1 DJ |
334 | queue_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec)); |
335 | } | |
336 | ||
337 | /* | |
338 | * edac_mc_workq_teardown | |
339 | * stop the workq processing on this mci | |
bf52fa4a DT |
340 | * |
341 | * locking model: | |
342 | * | |
343 | * called WITHOUT lock held | |
81d87cb1 | 344 | */ |
bf52fa4a | 345 | static void edac_mc_workq_teardown(struct mem_ctl_info *mci) |
81d87cb1 DJ |
346 | { |
347 | int status; | |
348 | ||
00740c58 BP |
349 | if (mci->op_state != OP_RUNNING_POLL) |
350 | return; | |
351 | ||
bce19683 DT |
352 | status = cancel_delayed_work(&mci->work); |
353 | if (status == 0) { | |
354 | debugf0("%s() not canceled, flush the queue\n", | |
355 | __func__); | |
bf52fa4a | 356 | |
bce19683 DT |
357 | /* workq instance might be running, wait for it */ |
358 | flush_workqueue(edac_workqueue); | |
81d87cb1 DJ |
359 | } |
360 | } | |
361 | ||
362 | /* | |
bce19683 DT |
363 | * edac_mc_reset_delay_period(unsigned long value) |
364 | * | |
365 | * user space has updated our poll period value, need to | |
366 | * reset our workq delays | |
81d87cb1 | 367 | */ |
bce19683 | 368 | void edac_mc_reset_delay_period(int value) |
81d87cb1 | 369 | { |
bce19683 DT |
370 | struct mem_ctl_info *mci; |
371 | struct list_head *item; | |
372 | ||
373 | mutex_lock(&mem_ctls_mutex); | |
374 | ||
375 | /* scan the list and turn off all workq timers, doing so under lock | |
376 | */ | |
377 | list_for_each(item, &mc_devices) { | |
378 | mci = list_entry(item, struct mem_ctl_info, link); | |
379 | ||
380 | if (mci->op_state == OP_RUNNING_POLL) | |
381 | cancel_delayed_work(&mci->work); | |
382 | } | |
383 | ||
384 | mutex_unlock(&mem_ctls_mutex); | |
81d87cb1 | 385 | |
bce19683 DT |
386 | |
387 | /* re-walk the list, and reset the poll delay */ | |
bf52fa4a DT |
388 | mutex_lock(&mem_ctls_mutex); |
389 | ||
bce19683 DT |
390 | list_for_each(item, &mc_devices) { |
391 | mci = list_entry(item, struct mem_ctl_info, link); | |
392 | ||
393 | edac_mc_workq_setup(mci, (unsigned long) value); | |
394 | } | |
81d87cb1 DJ |
395 | |
396 | mutex_unlock(&mem_ctls_mutex); | |
397 | } | |
398 | ||
bce19683 DT |
399 | |
400 | ||
2d7bbb91 DT |
401 | /* Return 0 on success, 1 on failure. |
402 | * Before calling this function, caller must | |
403 | * assign a unique value to mci->mc_idx. | |
bf52fa4a DT |
404 | * |
405 | * locking model: | |
406 | * | |
407 | * called with the mem_ctls_mutex lock held | |
2d7bbb91 | 408 | */ |
079708b9 | 409 | static int add_mc_to_global_list(struct mem_ctl_info *mci) |
da9bb1d2 AC |
410 | { |
411 | struct list_head *item, *insert_before; | |
412 | struct mem_ctl_info *p; | |
da9bb1d2 | 413 | |
2d7bbb91 | 414 | insert_before = &mc_devices; |
da9bb1d2 | 415 | |
bf52fa4a DT |
416 | p = find_mci_by_dev(mci->dev); |
417 | if (unlikely(p != NULL)) | |
2d7bbb91 | 418 | goto fail0; |
da9bb1d2 | 419 | |
2d7bbb91 DT |
420 | list_for_each(item, &mc_devices) { |
421 | p = list_entry(item, struct mem_ctl_info, link); | |
da9bb1d2 | 422 | |
2d7bbb91 DT |
423 | if (p->mc_idx >= mci->mc_idx) { |
424 | if (unlikely(p->mc_idx == mci->mc_idx)) | |
425 | goto fail1; | |
da9bb1d2 | 426 | |
2d7bbb91 DT |
427 | insert_before = item; |
428 | break; | |
da9bb1d2 | 429 | } |
da9bb1d2 AC |
430 | } |
431 | ||
432 | list_add_tail_rcu(&mci->link, insert_before); | |
c0d12172 | 433 | atomic_inc(&edac_handlers); |
da9bb1d2 | 434 | return 0; |
2d7bbb91 | 435 | |
052dfb45 | 436 | fail0: |
2d7bbb91 | 437 | edac_printk(KERN_WARNING, EDAC_MC, |
281efb17 | 438 | "%s (%s) %s %s already assigned %d\n", dev_name(p->dev), |
17aa7e03 | 439 | edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx); |
2d7bbb91 DT |
440 | return 1; |
441 | ||
052dfb45 | 442 | fail1: |
2d7bbb91 | 443 | edac_printk(KERN_WARNING, EDAC_MC, |
052dfb45 DT |
444 | "bug in low-level driver: attempt to assign\n" |
445 | " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__); | |
2d7bbb91 | 446 | return 1; |
da9bb1d2 AC |
447 | } |
448 | ||
e7ecd891 | 449 | static void del_mc_from_global_list(struct mem_ctl_info *mci) |
a1d03fcc | 450 | { |
c0d12172 | 451 | atomic_dec(&edac_handlers); |
a1d03fcc | 452 | list_del_rcu(&mci->link); |
e2e77098 LJ |
453 | |
454 | /* these are for safe removal of devices from global list while | |
455 | * NMI handlers may be traversing list | |
456 | */ | |
457 | synchronize_rcu(); | |
458 | INIT_LIST_HEAD(&mci->link); | |
a1d03fcc DP |
459 | } |
460 | ||
5da0831c DT |
461 | /** |
462 | * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'. | |
463 | * | |
464 | * If found, return a pointer to the structure. | |
465 | * Else return NULL. | |
466 | * | |
467 | * Caller must hold mem_ctls_mutex. | |
468 | */ | |
079708b9 | 469 | struct mem_ctl_info *edac_mc_find(int idx) |
5da0831c DT |
470 | { |
471 | struct list_head *item; | |
472 | struct mem_ctl_info *mci; | |
473 | ||
474 | list_for_each(item, &mc_devices) { | |
475 | mci = list_entry(item, struct mem_ctl_info, link); | |
476 | ||
477 | if (mci->mc_idx >= idx) { | |
478 | if (mci->mc_idx == idx) | |
479 | return mci; | |
480 | ||
481 | break; | |
482 | } | |
483 | } | |
484 | ||
485 | return NULL; | |
486 | } | |
487 | EXPORT_SYMBOL(edac_mc_find); | |
488 | ||
da9bb1d2 | 489 | /** |
472678eb DP |
490 | * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and |
491 | * create sysfs entries associated with mci structure | |
da9bb1d2 | 492 | * @mci: pointer to the mci structure to be added to the list |
2d7bbb91 | 493 | * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure. |
da9bb1d2 AC |
494 | * |
495 | * Return: | |
496 | * 0 Success | |
497 | * !0 Failure | |
498 | */ | |
499 | ||
500 | /* FIXME - should a warning be printed if no error detection? correction? */ | |
b8f6f975 | 501 | int edac_mc_add_mc(struct mem_ctl_info *mci) |
da9bb1d2 | 502 | { |
537fba28 | 503 | debugf0("%s()\n", __func__); |
b8f6f975 | 504 | |
da9bb1d2 AC |
505 | #ifdef CONFIG_EDAC_DEBUG |
506 | if (edac_debug_level >= 3) | |
507 | edac_mc_dump_mci(mci); | |
e7ecd891 | 508 | |
da9bb1d2 AC |
509 | if (edac_debug_level >= 4) { |
510 | int i; | |
511 | ||
512 | for (i = 0; i < mci->nr_csrows; i++) { | |
513 | int j; | |
e7ecd891 | 514 | |
da9bb1d2 AC |
515 | edac_mc_dump_csrow(&mci->csrows[i]); |
516 | for (j = 0; j < mci->csrows[i].nr_channels; j++) | |
079708b9 | 517 | edac_mc_dump_channel(&mci->csrows[i]. |
052dfb45 | 518 | channels[j]); |
da9bb1d2 AC |
519 | } |
520 | } | |
521 | #endif | |
63b7df91 | 522 | mutex_lock(&mem_ctls_mutex); |
da9bb1d2 AC |
523 | |
524 | if (add_mc_to_global_list(mci)) | |
028a7b6d | 525 | goto fail0; |
da9bb1d2 AC |
526 | |
527 | /* set load time so that error rate can be tracked */ | |
528 | mci->start_time = jiffies; | |
529 | ||
9794f33d | 530 | if (edac_create_sysfs_mci_device(mci)) { |
531 | edac_mc_printk(mci, KERN_WARNING, | |
052dfb45 | 532 | "failed to create sysfs device\n"); |
9794f33d | 533 | goto fail1; |
534 | } | |
da9bb1d2 | 535 | |
81d87cb1 DJ |
536 | /* If there IS a check routine, then we are running POLLED */ |
537 | if (mci->edac_check != NULL) { | |
538 | /* This instance is NOW RUNNING */ | |
539 | mci->op_state = OP_RUNNING_POLL; | |
540 | ||
541 | edac_mc_workq_setup(mci, edac_mc_get_poll_msec()); | |
542 | } else { | |
543 | mci->op_state = OP_RUNNING_INTERRUPT; | |
544 | } | |
545 | ||
da9bb1d2 | 546 | /* Report action taken */ |
bf52fa4a | 547 | edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':" |
17aa7e03 | 548 | " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci)); |
da9bb1d2 | 549 | |
63b7df91 | 550 | mutex_unlock(&mem_ctls_mutex); |
028a7b6d | 551 | return 0; |
da9bb1d2 | 552 | |
052dfb45 | 553 | fail1: |
028a7b6d DP |
554 | del_mc_from_global_list(mci); |
555 | ||
052dfb45 | 556 | fail0: |
63b7df91 | 557 | mutex_unlock(&mem_ctls_mutex); |
028a7b6d | 558 | return 1; |
da9bb1d2 | 559 | } |
9110540f | 560 | EXPORT_SYMBOL_GPL(edac_mc_add_mc); |
da9bb1d2 | 561 | |
da9bb1d2 | 562 | /** |
472678eb DP |
563 | * edac_mc_del_mc: Remove sysfs entries for specified mci structure and |
564 | * remove mci structure from global list | |
37f04581 | 565 | * @pdev: Pointer to 'struct device' representing mci structure to remove. |
da9bb1d2 | 566 | * |
18dbc337 | 567 | * Return pointer to removed mci structure, or NULL if device not found. |
da9bb1d2 | 568 | */ |
079708b9 | 569 | struct mem_ctl_info *edac_mc_del_mc(struct device *dev) |
da9bb1d2 | 570 | { |
18dbc337 | 571 | struct mem_ctl_info *mci; |
da9bb1d2 | 572 | |
bf52fa4a DT |
573 | debugf0("%s()\n", __func__); |
574 | ||
63b7df91 | 575 | mutex_lock(&mem_ctls_mutex); |
18dbc337 | 576 | |
bf52fa4a DT |
577 | /* find the requested mci struct in the global list */ |
578 | mci = find_mci_by_dev(dev); | |
579 | if (mci == NULL) { | |
63b7df91 | 580 | mutex_unlock(&mem_ctls_mutex); |
18dbc337 DP |
581 | return NULL; |
582 | } | |
583 | ||
da9bb1d2 | 584 | del_mc_from_global_list(mci); |
63b7df91 | 585 | mutex_unlock(&mem_ctls_mutex); |
bf52fa4a | 586 | |
bb31b312 | 587 | /* flush workq processes */ |
bf52fa4a | 588 | edac_mc_workq_teardown(mci); |
bb31b312 BP |
589 | |
590 | /* marking MCI offline */ | |
591 | mci->op_state = OP_OFFLINE; | |
592 | ||
593 | /* remove from sysfs */ | |
bf52fa4a DT |
594 | edac_remove_sysfs_mci_device(mci); |
595 | ||
537fba28 | 596 | edac_printk(KERN_INFO, EDAC_MC, |
052dfb45 | 597 | "Removed device %d for %s %s: DEV %s\n", mci->mc_idx, |
17aa7e03 | 598 | mci->mod_name, mci->ctl_name, edac_dev_name(mci)); |
bf52fa4a | 599 | |
18dbc337 | 600 | return mci; |
da9bb1d2 | 601 | } |
9110540f | 602 | EXPORT_SYMBOL_GPL(edac_mc_del_mc); |
da9bb1d2 | 603 | |
2da1c119 AB |
604 | static void edac_mc_scrub_block(unsigned long page, unsigned long offset, |
605 | u32 size) | |
da9bb1d2 AC |
606 | { |
607 | struct page *pg; | |
608 | void *virt_addr; | |
609 | unsigned long flags = 0; | |
610 | ||
537fba28 | 611 | debugf3("%s()\n", __func__); |
da9bb1d2 AC |
612 | |
613 | /* ECC error page was not in our memory. Ignore it. */ | |
079708b9 | 614 | if (!pfn_valid(page)) |
da9bb1d2 AC |
615 | return; |
616 | ||
617 | /* Find the actual page structure then map it and fix */ | |
618 | pg = pfn_to_page(page); | |
619 | ||
620 | if (PageHighMem(pg)) | |
621 | local_irq_save(flags); | |
622 | ||
4e5df7ca | 623 | virt_addr = kmap_atomic(pg); |
da9bb1d2 AC |
624 | |
625 | /* Perform architecture specific atomic scrub operation */ | |
626 | atomic_scrub(virt_addr + offset, size); | |
627 | ||
628 | /* Unmap and complete */ | |
4e5df7ca | 629 | kunmap_atomic(virt_addr); |
da9bb1d2 AC |
630 | |
631 | if (PageHighMem(pg)) | |
632 | local_irq_restore(flags); | |
633 | } | |
634 | ||
da9bb1d2 | 635 | /* FIXME - should return -1 */ |
e7ecd891 | 636 | int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page) |
da9bb1d2 AC |
637 | { |
638 | struct csrow_info *csrows = mci->csrows; | |
639 | int row, i; | |
640 | ||
537fba28 | 641 | debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page); |
da9bb1d2 AC |
642 | row = -1; |
643 | ||
644 | for (i = 0; i < mci->nr_csrows; i++) { | |
645 | struct csrow_info *csrow = &csrows[i]; | |
646 | ||
647 | if (csrow->nr_pages == 0) | |
648 | continue; | |
649 | ||
537fba28 DP |
650 | debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) " |
651 | "mask(0x%lx)\n", mci->mc_idx, __func__, | |
652 | csrow->first_page, page, csrow->last_page, | |
653 | csrow->page_mask); | |
da9bb1d2 AC |
654 | |
655 | if ((page >= csrow->first_page) && | |
656 | (page <= csrow->last_page) && | |
657 | ((page & csrow->page_mask) == | |
658 | (csrow->first_page & csrow->page_mask))) { | |
659 | row = i; | |
660 | break; | |
661 | } | |
662 | } | |
663 | ||
664 | if (row == -1) | |
537fba28 | 665 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
666 | "could not look up page error address %lx\n", |
667 | (unsigned long)page); | |
da9bb1d2 AC |
668 | |
669 | return row; | |
670 | } | |
9110540f | 671 | EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page); |
da9bb1d2 | 672 | |
da9bb1d2 AC |
673 | /* FIXME - setable log (warning/emerg) levels */ |
674 | /* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */ | |
675 | void edac_mc_handle_ce(struct mem_ctl_info *mci, | |
052dfb45 DT |
676 | unsigned long page_frame_number, |
677 | unsigned long offset_in_page, unsigned long syndrome, | |
678 | int row, int channel, const char *msg) | |
da9bb1d2 AC |
679 | { |
680 | unsigned long remapped_page; | |
681 | ||
537fba28 | 682 | debugf3("MC%d: %s()\n", mci->mc_idx, __func__); |
da9bb1d2 AC |
683 | |
684 | /* FIXME - maybe make panic on INTERNAL ERROR an option */ | |
685 | if (row >= mci->nr_csrows || row < 0) { | |
686 | /* something is wrong */ | |
537fba28 | 687 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
688 | "INTERNAL ERROR: row out of range " |
689 | "(%d >= %d)\n", row, mci->nr_csrows); | |
da9bb1d2 AC |
690 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
691 | return; | |
692 | } | |
e7ecd891 | 693 | |
da9bb1d2 AC |
694 | if (channel >= mci->csrows[row].nr_channels || channel < 0) { |
695 | /* something is wrong */ | |
537fba28 | 696 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
697 | "INTERNAL ERROR: channel out of range " |
698 | "(%d >= %d)\n", channel, | |
699 | mci->csrows[row].nr_channels); | |
da9bb1d2 AC |
700 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
701 | return; | |
702 | } | |
703 | ||
4de78c68 | 704 | if (edac_mc_get_log_ce()) |
da9bb1d2 | 705 | /* FIXME - put in DIMM location */ |
537fba28 | 706 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 DT |
707 | "CE page 0x%lx, offset 0x%lx, grain %d, syndrome " |
708 | "0x%lx, row %d, channel %d, label \"%s\": %s\n", | |
709 | page_frame_number, offset_in_page, | |
710 | mci->csrows[row].grain, syndrome, row, channel, | |
711 | mci->csrows[row].channels[channel].label, msg); | |
da9bb1d2 AC |
712 | |
713 | mci->ce_count++; | |
714 | mci->csrows[row].ce_count++; | |
715 | mci->csrows[row].channels[channel].ce_count++; | |
716 | ||
717 | if (mci->scrub_mode & SCRUB_SW_SRC) { | |
718 | /* | |
719 | * Some MC's can remap memory so that it is still available | |
720 | * at a different address when PCI devices map into memory. | |
721 | * MC's that can't do this lose the memory where PCI devices | |
25985edc | 722 | * are mapped. This mapping is MC dependent and so we call |
da9bb1d2 AC |
723 | * back into the MC driver for it to map the MC page to |
724 | * a physical (CPU) page which can then be mapped to a virtual | |
725 | * page - which can then be scrubbed. | |
726 | */ | |
727 | remapped_page = mci->ctl_page_to_phys ? | |
052dfb45 DT |
728 | mci->ctl_page_to_phys(mci, page_frame_number) : |
729 | page_frame_number; | |
da9bb1d2 AC |
730 | |
731 | edac_mc_scrub_block(remapped_page, offset_in_page, | |
052dfb45 | 732 | mci->csrows[row].grain); |
da9bb1d2 AC |
733 | } |
734 | } | |
9110540f | 735 | EXPORT_SYMBOL_GPL(edac_mc_handle_ce); |
da9bb1d2 | 736 | |
e7ecd891 | 737 | void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg) |
da9bb1d2 | 738 | { |
4de78c68 | 739 | if (edac_mc_get_log_ce()) |
537fba28 | 740 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 | 741 | "CE - no information available: %s\n", msg); |
e7ecd891 | 742 | |
da9bb1d2 AC |
743 | mci->ce_noinfo_count++; |
744 | mci->ce_count++; | |
745 | } | |
9110540f | 746 | EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info); |
da9bb1d2 | 747 | |
da9bb1d2 | 748 | void edac_mc_handle_ue(struct mem_ctl_info *mci, |
052dfb45 DT |
749 | unsigned long page_frame_number, |
750 | unsigned long offset_in_page, int row, const char *msg) | |
da9bb1d2 AC |
751 | { |
752 | int len = EDAC_MC_LABEL_LEN * 4; | |
753 | char labels[len + 1]; | |
754 | char *pos = labels; | |
755 | int chan; | |
756 | int chars; | |
757 | ||
537fba28 | 758 | debugf3("MC%d: %s()\n", mci->mc_idx, __func__); |
da9bb1d2 AC |
759 | |
760 | /* FIXME - maybe make panic on INTERNAL ERROR an option */ | |
761 | if (row >= mci->nr_csrows || row < 0) { | |
762 | /* something is wrong */ | |
537fba28 | 763 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
764 | "INTERNAL ERROR: row out of range " |
765 | "(%d >= %d)\n", row, mci->nr_csrows); | |
da9bb1d2 AC |
766 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
767 | return; | |
768 | } | |
769 | ||
770 | chars = snprintf(pos, len + 1, "%s", | |
079708b9 | 771 | mci->csrows[row].channels[0].label); |
da9bb1d2 AC |
772 | len -= chars; |
773 | pos += chars; | |
e7ecd891 | 774 | |
da9bb1d2 | 775 | for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0); |
052dfb45 | 776 | chan++) { |
da9bb1d2 | 777 | chars = snprintf(pos, len + 1, ":%s", |
079708b9 | 778 | mci->csrows[row].channels[chan].label); |
da9bb1d2 AC |
779 | len -= chars; |
780 | pos += chars; | |
781 | } | |
782 | ||
4de78c68 | 783 | if (edac_mc_get_log_ue()) |
537fba28 | 784 | edac_mc_printk(mci, KERN_EMERG, |
052dfb45 DT |
785 | "UE page 0x%lx, offset 0x%lx, grain %d, row %d, " |
786 | "labels \"%s\": %s\n", page_frame_number, | |
787 | offset_in_page, mci->csrows[row].grain, row, | |
788 | labels, msg); | |
da9bb1d2 | 789 | |
4de78c68 | 790 | if (edac_mc_get_panic_on_ue()) |
e7ecd891 | 791 | panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, " |
052dfb45 DT |
792 | "row %d, labels \"%s\": %s\n", mci->mc_idx, |
793 | page_frame_number, offset_in_page, | |
794 | mci->csrows[row].grain, row, labels, msg); | |
da9bb1d2 AC |
795 | |
796 | mci->ue_count++; | |
797 | mci->csrows[row].ue_count++; | |
798 | } | |
9110540f | 799 | EXPORT_SYMBOL_GPL(edac_mc_handle_ue); |
da9bb1d2 | 800 | |
e7ecd891 | 801 | void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg) |
da9bb1d2 | 802 | { |
4de78c68 | 803 | if (edac_mc_get_panic_on_ue()) |
da9bb1d2 AC |
804 | panic("EDAC MC%d: Uncorrected Error", mci->mc_idx); |
805 | ||
4de78c68 | 806 | if (edac_mc_get_log_ue()) |
537fba28 | 807 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 | 808 | "UE - no information available: %s\n", msg); |
da9bb1d2 AC |
809 | mci->ue_noinfo_count++; |
810 | mci->ue_count++; | |
811 | } | |
079708b9 | 812 | EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info); |
da9bb1d2 | 813 | |
9794f33d | 814 | /************************************************************* |
815 | * On Fully Buffered DIMM modules, this help function is | |
816 | * called to process UE events | |
817 | */ | |
818 | void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci, | |
052dfb45 DT |
819 | unsigned int csrow, |
820 | unsigned int channela, | |
821 | unsigned int channelb, char *msg) | |
9794f33d | 822 | { |
823 | int len = EDAC_MC_LABEL_LEN * 4; | |
824 | char labels[len + 1]; | |
825 | char *pos = labels; | |
826 | int chars; | |
827 | ||
828 | if (csrow >= mci->nr_csrows) { | |
829 | /* something is wrong */ | |
830 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
831 | "INTERNAL ERROR: row out of range (%d >= %d)\n", |
832 | csrow, mci->nr_csrows); | |
9794f33d | 833 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
834 | return; | |
835 | } | |
836 | ||
837 | if (channela >= mci->csrows[csrow].nr_channels) { | |
838 | /* something is wrong */ | |
839 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
840 | "INTERNAL ERROR: channel-a out of range " |
841 | "(%d >= %d)\n", | |
842 | channela, mci->csrows[csrow].nr_channels); | |
9794f33d | 843 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
844 | return; | |
845 | } | |
846 | ||
847 | if (channelb >= mci->csrows[csrow].nr_channels) { | |
848 | /* something is wrong */ | |
849 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
850 | "INTERNAL ERROR: channel-b out of range " |
851 | "(%d >= %d)\n", | |
852 | channelb, mci->csrows[csrow].nr_channels); | |
9794f33d | 853 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
854 | return; | |
855 | } | |
856 | ||
857 | mci->ue_count++; | |
858 | mci->csrows[csrow].ue_count++; | |
859 | ||
860 | /* Generate the DIMM labels from the specified channels */ | |
861 | chars = snprintf(pos, len + 1, "%s", | |
862 | mci->csrows[csrow].channels[channela].label); | |
079708b9 DT |
863 | len -= chars; |
864 | pos += chars; | |
9794f33d | 865 | chars = snprintf(pos, len + 1, "-%s", |
866 | mci->csrows[csrow].channels[channelb].label); | |
867 | ||
4de78c68 | 868 | if (edac_mc_get_log_ue()) |
9794f33d | 869 | edac_mc_printk(mci, KERN_EMERG, |
052dfb45 DT |
870 | "UE row %d, channel-a= %d channel-b= %d " |
871 | "labels \"%s\": %s\n", csrow, channela, channelb, | |
872 | labels, msg); | |
9794f33d | 873 | |
4de78c68 | 874 | if (edac_mc_get_panic_on_ue()) |
9794f33d | 875 | panic("UE row %d, channel-a= %d channel-b= %d " |
052dfb45 DT |
876 | "labels \"%s\": %s\n", csrow, channela, |
877 | channelb, labels, msg); | |
9794f33d | 878 | } |
879 | EXPORT_SYMBOL(edac_mc_handle_fbd_ue); | |
880 | ||
881 | /************************************************************* | |
882 | * On Fully Buffered DIMM modules, this help function is | |
883 | * called to process CE events | |
884 | */ | |
885 | void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci, | |
052dfb45 | 886 | unsigned int csrow, unsigned int channel, char *msg) |
9794f33d | 887 | { |
888 | ||
889 | /* Ensure boundary values */ | |
890 | if (csrow >= mci->nr_csrows) { | |
891 | /* something is wrong */ | |
892 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
893 | "INTERNAL ERROR: row out of range (%d >= %d)\n", |
894 | csrow, mci->nr_csrows); | |
9794f33d | 895 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
896 | return; | |
897 | } | |
898 | if (channel >= mci->csrows[csrow].nr_channels) { | |
899 | /* something is wrong */ | |
900 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
901 | "INTERNAL ERROR: channel out of range (%d >= %d)\n", |
902 | channel, mci->csrows[csrow].nr_channels); | |
9794f33d | 903 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
904 | return; | |
905 | } | |
906 | ||
4de78c68 | 907 | if (edac_mc_get_log_ce()) |
9794f33d | 908 | /* FIXME - put in DIMM location */ |
909 | edac_mc_printk(mci, KERN_WARNING, | |
052dfb45 DT |
910 | "CE row %d, channel %d, label \"%s\": %s\n", |
911 | csrow, channel, | |
912 | mci->csrows[csrow].channels[channel].label, msg); | |
9794f33d | 913 | |
914 | mci->ce_count++; | |
915 | mci->csrows[csrow].ce_count++; | |
916 | mci->csrows[csrow].channels[channel].ce_count++; | |
917 | } | |
079708b9 | 918 | EXPORT_SYMBOL(edac_mc_handle_fbd_ce); |