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> | |
28 | #include <linux/sysdev.h> | |
29 | #include <linux/ctype.h> | |
c0d12172 | 30 | #include <linux/edac.h> |
da9bb1d2 AC |
31 | #include <asm/uaccess.h> |
32 | #include <asm/page.h> | |
33 | #include <asm/edac.h> | |
20bcb7a8 | 34 | #include "edac_core.h" |
7c9281d7 | 35 | #include "edac_module.h" |
da9bb1d2 | 36 | |
da9bb1d2 | 37 | /* lock to memory controller's control array */ |
63b7df91 | 38 | static DEFINE_MUTEX(mem_ctls_mutex); |
ff6ac2a6 | 39 | static LIST_HEAD(mc_devices); |
da9bb1d2 | 40 | |
da9bb1d2 AC |
41 | #ifdef CONFIG_EDAC_DEBUG |
42 | ||
2da1c119 | 43 | static void edac_mc_dump_channel(struct channel_info *chan) |
da9bb1d2 AC |
44 | { |
45 | debugf4("\tchannel = %p\n", chan); | |
46 | debugf4("\tchannel->chan_idx = %d\n", chan->chan_idx); | |
47 | debugf4("\tchannel->ce_count = %d\n", chan->ce_count); | |
48 | debugf4("\tchannel->label = '%s'\n", chan->label); | |
49 | debugf4("\tchannel->csrow = %p\n\n", chan->csrow); | |
50 | } | |
51 | ||
2da1c119 | 52 | static void edac_mc_dump_csrow(struct csrow_info *csrow) |
da9bb1d2 AC |
53 | { |
54 | debugf4("\tcsrow = %p\n", csrow); | |
55 | debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx); | |
079708b9 | 56 | debugf4("\tcsrow->first_page = 0x%lx\n", csrow->first_page); |
da9bb1d2 AC |
57 | debugf4("\tcsrow->last_page = 0x%lx\n", csrow->last_page); |
58 | debugf4("\tcsrow->page_mask = 0x%lx\n", csrow->page_mask); | |
59 | debugf4("\tcsrow->nr_pages = 0x%x\n", csrow->nr_pages); | |
079708b9 | 60 | debugf4("\tcsrow->nr_channels = %d\n", csrow->nr_channels); |
da9bb1d2 AC |
61 | debugf4("\tcsrow->channels = %p\n", csrow->channels); |
62 | debugf4("\tcsrow->mci = %p\n\n", csrow->mci); | |
63 | } | |
64 | ||
2da1c119 | 65 | static void edac_mc_dump_mci(struct mem_ctl_info *mci) |
da9bb1d2 AC |
66 | { |
67 | debugf3("\tmci = %p\n", mci); | |
68 | debugf3("\tmci->mtype_cap = %lx\n", mci->mtype_cap); | |
69 | debugf3("\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap); | |
70 | debugf3("\tmci->edac_cap = %lx\n", mci->edac_cap); | |
71 | debugf4("\tmci->edac_check = %p\n", mci->edac_check); | |
72 | debugf3("\tmci->nr_csrows = %d, csrows = %p\n", | |
73 | mci->nr_csrows, mci->csrows); | |
37f04581 | 74 | debugf3("\tdev = %p\n", mci->dev); |
079708b9 | 75 | debugf3("\tmod_name:ctl_name = %s:%s\n", mci->mod_name, mci->ctl_name); |
da9bb1d2 AC |
76 | debugf3("\tpvt_info = %p\n\n", mci->pvt_info); |
77 | } | |
78 | ||
239642fe BP |
79 | /* |
80 | * keep those in sync with the enum mem_type | |
81 | */ | |
82 | const char *edac_mem_types[] = { | |
83 | "Empty csrow", | |
84 | "Reserved csrow type", | |
85 | "Unknown csrow type", | |
86 | "Fast page mode RAM", | |
87 | "Extended data out RAM", | |
88 | "Burst Extended data out RAM", | |
89 | "Single data rate SDRAM", | |
90 | "Registered single data rate SDRAM", | |
91 | "Double data rate SDRAM", | |
92 | "Registered Double data rate SDRAM", | |
93 | "Rambus DRAM", | |
94 | "Unbuffered DDR2 RAM", | |
95 | "Fully buffered DDR2", | |
96 | "Registered DDR2 RAM", | |
97 | "Rambus XDR", | |
98 | "Unbuffered DDR3 RAM", | |
99 | "Registered DDR3 RAM", | |
100 | }; | |
101 | EXPORT_SYMBOL_GPL(edac_mem_types); | |
102 | ||
079708b9 | 103 | #endif /* CONFIG_EDAC_DEBUG */ |
da9bb1d2 AC |
104 | |
105 | /* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'. | |
106 | * Adjust 'ptr' so that its alignment is at least as stringent as what the | |
107 | * compiler would provide for X and return the aligned result. | |
108 | * | |
109 | * If 'size' is a constant, the compiler will optimize this whole function | |
110 | * down to either a no-op or the addition of a constant to the value of 'ptr'. | |
111 | */ | |
7391c6dc | 112 | void *edac_align_ptr(void *ptr, unsigned size) |
da9bb1d2 AC |
113 | { |
114 | unsigned align, r; | |
115 | ||
116 | /* Here we assume that the alignment of a "long long" is the most | |
117 | * stringent alignment that the compiler will ever provide by default. | |
118 | * As far as I know, this is a reasonable assumption. | |
119 | */ | |
120 | if (size > sizeof(long)) | |
121 | align = sizeof(long long); | |
122 | else if (size > sizeof(int)) | |
123 | align = sizeof(long); | |
124 | else if (size > sizeof(short)) | |
125 | align = sizeof(int); | |
126 | else if (size > sizeof(char)) | |
127 | align = sizeof(short); | |
128 | else | |
079708b9 | 129 | return (char *)ptr; |
da9bb1d2 AC |
130 | |
131 | r = size % align; | |
132 | ||
133 | if (r == 0) | |
079708b9 | 134 | return (char *)ptr; |
da9bb1d2 | 135 | |
7391c6dc | 136 | return (void *)(((unsigned long)ptr) + align - r); |
da9bb1d2 AC |
137 | } |
138 | ||
da9bb1d2 AC |
139 | /** |
140 | * edac_mc_alloc: Allocate a struct mem_ctl_info structure | |
141 | * @size_pvt: size of private storage needed | |
142 | * @nr_csrows: Number of CWROWS needed for this MC | |
143 | * @nr_chans: Number of channels for the MC | |
144 | * | |
145 | * Everything is kmalloc'ed as one big chunk - more efficient. | |
146 | * Only can be used if all structures have the same lifetime - otherwise | |
147 | * you have to allocate and initialize your own structures. | |
148 | * | |
149 | * Use edac_mc_free() to free mc structures allocated by this function. | |
150 | * | |
151 | * Returns: | |
152 | * NULL allocation failed | |
153 | * struct mem_ctl_info pointer | |
154 | */ | |
155 | struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows, | |
b8f6f975 | 156 | unsigned nr_chans, int edac_index) |
da9bb1d2 AC |
157 | { |
158 | struct mem_ctl_info *mci; | |
159 | struct csrow_info *csi, *csrow; | |
160 | struct channel_info *chi, *chp, *chan; | |
161 | void *pvt; | |
162 | unsigned size; | |
163 | int row, chn; | |
8096cfaf | 164 | int err; |
da9bb1d2 AC |
165 | |
166 | /* Figure out the offsets of the various items from the start of an mc | |
167 | * structure. We want the alignment of each item to be at least as | |
168 | * stringent as what the compiler would provide if we could simply | |
169 | * hardcode everything into a single struct. | |
170 | */ | |
079708b9 | 171 | mci = (struct mem_ctl_info *)0; |
7391c6dc DT |
172 | csi = edac_align_ptr(&mci[1], sizeof(*csi)); |
173 | chi = edac_align_ptr(&csi[nr_csrows], sizeof(*chi)); | |
e27e3dac | 174 | pvt = edac_align_ptr(&chi[nr_chans * nr_csrows], sz_pvt); |
079708b9 | 175 | size = ((unsigned long)pvt) + sz_pvt; |
da9bb1d2 | 176 | |
8096cfaf DT |
177 | mci = kzalloc(size, GFP_KERNEL); |
178 | if (mci == NULL) | |
da9bb1d2 AC |
179 | return NULL; |
180 | ||
181 | /* Adjust pointers so they point within the memory we just allocated | |
182 | * rather than an imaginary chunk of memory located at address 0. | |
183 | */ | |
079708b9 DT |
184 | csi = (struct csrow_info *)(((char *)mci) + ((unsigned long)csi)); |
185 | chi = (struct channel_info *)(((char *)mci) + ((unsigned long)chi)); | |
186 | pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL; | |
da9bb1d2 | 187 | |
b8f6f975 DT |
188 | /* setup index and various internal pointers */ |
189 | mci->mc_idx = edac_index; | |
da9bb1d2 AC |
190 | mci->csrows = csi; |
191 | mci->pvt_info = pvt; | |
192 | mci->nr_csrows = nr_csrows; | |
193 | ||
194 | for (row = 0; row < nr_csrows; row++) { | |
195 | csrow = &csi[row]; | |
196 | csrow->csrow_idx = row; | |
197 | csrow->mci = mci; | |
198 | csrow->nr_channels = nr_chans; | |
199 | chp = &chi[row * nr_chans]; | |
200 | csrow->channels = chp; | |
201 | ||
202 | for (chn = 0; chn < nr_chans; chn++) { | |
203 | chan = &chp[chn]; | |
204 | chan->chan_idx = chn; | |
205 | chan->csrow = csrow; | |
206 | } | |
207 | } | |
208 | ||
81d87cb1 | 209 | mci->op_state = OP_ALLOC; |
6fe1108f | 210 | INIT_LIST_HEAD(&mci->grp_kobj_list); |
81d87cb1 | 211 | |
8096cfaf DT |
212 | /* |
213 | * Initialize the 'root' kobj for the edac_mc controller | |
214 | */ | |
215 | err = edac_mc_register_sysfs_main_kobj(mci); | |
216 | if (err) { | |
217 | kfree(mci); | |
218 | return NULL; | |
219 | } | |
220 | ||
221 | /* at this point, the root kobj is valid, and in order to | |
222 | * 'free' the object, then the function: | |
223 | * edac_mc_unregister_sysfs_main_kobj() must be called | |
224 | * which will perform kobj unregistration and the actual free | |
225 | * will occur during the kobject callback operation | |
226 | */ | |
da9bb1d2 AC |
227 | return mci; |
228 | } | |
9110540f | 229 | EXPORT_SYMBOL_GPL(edac_mc_alloc); |
da9bb1d2 | 230 | |
da9bb1d2 | 231 | /** |
8096cfaf DT |
232 | * edac_mc_free |
233 | * 'Free' a previously allocated 'mci' structure | |
da9bb1d2 | 234 | * @mci: pointer to a struct mem_ctl_info structure |
da9bb1d2 AC |
235 | */ |
236 | void edac_mc_free(struct mem_ctl_info *mci) | |
237 | { | |
bbc560ae MCC |
238 | debugf1("%s()\n", __func__); |
239 | ||
8096cfaf | 240 | edac_mc_unregister_sysfs_main_kobj(mci); |
accf74ff MCC |
241 | |
242 | /* free the mci instance memory here */ | |
243 | kfree(mci); | |
da9bb1d2 | 244 | } |
9110540f | 245 | EXPORT_SYMBOL_GPL(edac_mc_free); |
da9bb1d2 | 246 | |
bce19683 | 247 | |
939747bd | 248 | /** |
bce19683 DT |
249 | * find_mci_by_dev |
250 | * | |
251 | * scan list of controllers looking for the one that manages | |
252 | * the 'dev' device | |
939747bd | 253 | * @dev: pointer to a struct device related with the MCI |
bce19683 | 254 | */ |
939747bd | 255 | struct mem_ctl_info *find_mci_by_dev(struct device *dev) |
da9bb1d2 AC |
256 | { |
257 | struct mem_ctl_info *mci; | |
258 | struct list_head *item; | |
259 | ||
537fba28 | 260 | debugf3("%s()\n", __func__); |
da9bb1d2 AC |
261 | |
262 | list_for_each(item, &mc_devices) { | |
263 | mci = list_entry(item, struct mem_ctl_info, link); | |
264 | ||
37f04581 | 265 | if (mci->dev == dev) |
da9bb1d2 AC |
266 | return mci; |
267 | } | |
268 | ||
269 | return NULL; | |
270 | } | |
939747bd | 271 | EXPORT_SYMBOL_GPL(find_mci_by_dev); |
da9bb1d2 | 272 | |
81d87cb1 DJ |
273 | /* |
274 | * handler for EDAC to check if NMI type handler has asserted interrupt | |
275 | */ | |
276 | static int edac_mc_assert_error_check_and_clear(void) | |
277 | { | |
66ee2f94 | 278 | int old_state; |
81d87cb1 | 279 | |
079708b9 | 280 | if (edac_op_state == EDAC_OPSTATE_POLL) |
81d87cb1 DJ |
281 | return 1; |
282 | ||
66ee2f94 DJ |
283 | old_state = edac_err_assert; |
284 | edac_err_assert = 0; | |
81d87cb1 | 285 | |
66ee2f94 | 286 | return old_state; |
81d87cb1 DJ |
287 | } |
288 | ||
289 | /* | |
290 | * edac_mc_workq_function | |
291 | * performs the operation scheduled by a workq request | |
292 | */ | |
81d87cb1 DJ |
293 | static void edac_mc_workq_function(struct work_struct *work_req) |
294 | { | |
fbeb4384 | 295 | struct delayed_work *d_work = to_delayed_work(work_req); |
81d87cb1 | 296 | struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work); |
81d87cb1 DJ |
297 | |
298 | mutex_lock(&mem_ctls_mutex); | |
299 | ||
bf52fa4a DT |
300 | /* if this control struct has movd to offline state, we are done */ |
301 | if (mci->op_state == OP_OFFLINE) { | |
302 | mutex_unlock(&mem_ctls_mutex); | |
303 | return; | |
304 | } | |
305 | ||
81d87cb1 DJ |
306 | /* Only poll controllers that are running polled and have a check */ |
307 | if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL)) | |
308 | mci->edac_check(mci); | |
309 | ||
81d87cb1 DJ |
310 | mutex_unlock(&mem_ctls_mutex); |
311 | ||
312 | /* Reschedule */ | |
4de78c68 | 313 | queue_delayed_work(edac_workqueue, &mci->work, |
052dfb45 | 314 | msecs_to_jiffies(edac_mc_get_poll_msec())); |
81d87cb1 DJ |
315 | } |
316 | ||
317 | /* | |
318 | * edac_mc_workq_setup | |
319 | * initialize a workq item for this mci | |
320 | * passing in the new delay period in msec | |
bf52fa4a DT |
321 | * |
322 | * locking model: | |
323 | * | |
324 | * called with the mem_ctls_mutex held | |
81d87cb1 | 325 | */ |
bf52fa4a | 326 | static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec) |
81d87cb1 DJ |
327 | { |
328 | debugf0("%s()\n", __func__); | |
329 | ||
bf52fa4a DT |
330 | /* if this instance is not in the POLL state, then simply return */ |
331 | if (mci->op_state != OP_RUNNING_POLL) | |
332 | return; | |
333 | ||
81d87cb1 | 334 | INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function); |
81d87cb1 DJ |
335 | queue_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec)); |
336 | } | |
337 | ||
338 | /* | |
339 | * edac_mc_workq_teardown | |
340 | * stop the workq processing on this mci | |
bf52fa4a DT |
341 | * |
342 | * locking model: | |
343 | * | |
344 | * called WITHOUT lock held | |
81d87cb1 | 345 | */ |
bf52fa4a | 346 | static void edac_mc_workq_teardown(struct mem_ctl_info *mci) |
81d87cb1 DJ |
347 | { |
348 | int status; | |
349 | ||
00740c58 BP |
350 | if (mci->op_state != OP_RUNNING_POLL) |
351 | return; | |
352 | ||
bce19683 DT |
353 | status = cancel_delayed_work(&mci->work); |
354 | if (status == 0) { | |
355 | debugf0("%s() not canceled, flush the queue\n", | |
356 | __func__); | |
bf52fa4a | 357 | |
bce19683 DT |
358 | /* workq instance might be running, wait for it */ |
359 | flush_workqueue(edac_workqueue); | |
81d87cb1 DJ |
360 | } |
361 | } | |
362 | ||
363 | /* | |
bce19683 DT |
364 | * edac_mc_reset_delay_period(unsigned long value) |
365 | * | |
366 | * user space has updated our poll period value, need to | |
367 | * reset our workq delays | |
81d87cb1 | 368 | */ |
bce19683 | 369 | void edac_mc_reset_delay_period(int value) |
81d87cb1 | 370 | { |
bce19683 DT |
371 | struct mem_ctl_info *mci; |
372 | struct list_head *item; | |
373 | ||
374 | mutex_lock(&mem_ctls_mutex); | |
375 | ||
376 | /* scan the list and turn off all workq timers, doing so under lock | |
377 | */ | |
378 | list_for_each(item, &mc_devices) { | |
379 | mci = list_entry(item, struct mem_ctl_info, link); | |
380 | ||
381 | if (mci->op_state == OP_RUNNING_POLL) | |
382 | cancel_delayed_work(&mci->work); | |
383 | } | |
384 | ||
385 | mutex_unlock(&mem_ctls_mutex); | |
81d87cb1 | 386 | |
bce19683 DT |
387 | |
388 | /* re-walk the list, and reset the poll delay */ | |
bf52fa4a DT |
389 | mutex_lock(&mem_ctls_mutex); |
390 | ||
bce19683 DT |
391 | list_for_each(item, &mc_devices) { |
392 | mci = list_entry(item, struct mem_ctl_info, link); | |
393 | ||
394 | edac_mc_workq_setup(mci, (unsigned long) value); | |
395 | } | |
81d87cb1 DJ |
396 | |
397 | mutex_unlock(&mem_ctls_mutex); | |
398 | } | |
399 | ||
bce19683 DT |
400 | |
401 | ||
2d7bbb91 DT |
402 | /* Return 0 on success, 1 on failure. |
403 | * Before calling this function, caller must | |
404 | * assign a unique value to mci->mc_idx. | |
bf52fa4a DT |
405 | * |
406 | * locking model: | |
407 | * | |
408 | * called with the mem_ctls_mutex lock held | |
2d7bbb91 | 409 | */ |
079708b9 | 410 | static int add_mc_to_global_list(struct mem_ctl_info *mci) |
da9bb1d2 AC |
411 | { |
412 | struct list_head *item, *insert_before; | |
413 | struct mem_ctl_info *p; | |
da9bb1d2 | 414 | |
2d7bbb91 | 415 | insert_before = &mc_devices; |
da9bb1d2 | 416 | |
bf52fa4a DT |
417 | p = find_mci_by_dev(mci->dev); |
418 | if (unlikely(p != NULL)) | |
2d7bbb91 | 419 | goto fail0; |
da9bb1d2 | 420 | |
2d7bbb91 DT |
421 | list_for_each(item, &mc_devices) { |
422 | p = list_entry(item, struct mem_ctl_info, link); | |
da9bb1d2 | 423 | |
2d7bbb91 DT |
424 | if (p->mc_idx >= mci->mc_idx) { |
425 | if (unlikely(p->mc_idx == mci->mc_idx)) | |
426 | goto fail1; | |
da9bb1d2 | 427 | |
2d7bbb91 DT |
428 | insert_before = item; |
429 | break; | |
da9bb1d2 | 430 | } |
da9bb1d2 AC |
431 | } |
432 | ||
433 | list_add_tail_rcu(&mci->link, insert_before); | |
c0d12172 | 434 | atomic_inc(&edac_handlers); |
da9bb1d2 | 435 | return 0; |
2d7bbb91 | 436 | |
052dfb45 | 437 | fail0: |
2d7bbb91 | 438 | edac_printk(KERN_WARNING, EDAC_MC, |
281efb17 | 439 | "%s (%s) %s %s already assigned %d\n", dev_name(p->dev), |
17aa7e03 | 440 | edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx); |
2d7bbb91 DT |
441 | return 1; |
442 | ||
052dfb45 | 443 | fail1: |
2d7bbb91 | 444 | edac_printk(KERN_WARNING, EDAC_MC, |
052dfb45 DT |
445 | "bug in low-level driver: attempt to assign\n" |
446 | " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__); | |
2d7bbb91 | 447 | return 1; |
da9bb1d2 AC |
448 | } |
449 | ||
e7ecd891 | 450 | static void complete_mc_list_del(struct rcu_head *head) |
a1d03fcc DP |
451 | { |
452 | struct mem_ctl_info *mci; | |
453 | ||
454 | mci = container_of(head, struct mem_ctl_info, rcu); | |
455 | INIT_LIST_HEAD(&mci->link); | |
a1d03fcc DP |
456 | } |
457 | ||
e7ecd891 | 458 | static void del_mc_from_global_list(struct mem_ctl_info *mci) |
a1d03fcc | 459 | { |
c0d12172 | 460 | atomic_dec(&edac_handlers); |
a1d03fcc | 461 | list_del_rcu(&mci->link); |
a1d03fcc | 462 | call_rcu(&mci->rcu, complete_mc_list_del); |
458e5ff1 | 463 | rcu_barrier(); |
a1d03fcc DP |
464 | } |
465 | ||
5da0831c DT |
466 | /** |
467 | * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'. | |
468 | * | |
469 | * If found, return a pointer to the structure. | |
470 | * Else return NULL. | |
471 | * | |
472 | * Caller must hold mem_ctls_mutex. | |
473 | */ | |
079708b9 | 474 | struct mem_ctl_info *edac_mc_find(int idx) |
5da0831c DT |
475 | { |
476 | struct list_head *item; | |
477 | struct mem_ctl_info *mci; | |
478 | ||
479 | list_for_each(item, &mc_devices) { | |
480 | mci = list_entry(item, struct mem_ctl_info, link); | |
481 | ||
482 | if (mci->mc_idx >= idx) { | |
483 | if (mci->mc_idx == idx) | |
484 | return mci; | |
485 | ||
486 | break; | |
487 | } | |
488 | } | |
489 | ||
490 | return NULL; | |
491 | } | |
492 | EXPORT_SYMBOL(edac_mc_find); | |
493 | ||
da9bb1d2 | 494 | /** |
472678eb DP |
495 | * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and |
496 | * create sysfs entries associated with mci structure | |
da9bb1d2 | 497 | * @mci: pointer to the mci structure to be added to the list |
2d7bbb91 | 498 | * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure. |
da9bb1d2 AC |
499 | * |
500 | * Return: | |
501 | * 0 Success | |
502 | * !0 Failure | |
503 | */ | |
504 | ||
505 | /* FIXME - should a warning be printed if no error detection? correction? */ | |
b8f6f975 | 506 | int edac_mc_add_mc(struct mem_ctl_info *mci) |
da9bb1d2 | 507 | { |
537fba28 | 508 | debugf0("%s()\n", __func__); |
b8f6f975 | 509 | |
da9bb1d2 AC |
510 | #ifdef CONFIG_EDAC_DEBUG |
511 | if (edac_debug_level >= 3) | |
512 | edac_mc_dump_mci(mci); | |
e7ecd891 | 513 | |
da9bb1d2 AC |
514 | if (edac_debug_level >= 4) { |
515 | int i; | |
516 | ||
517 | for (i = 0; i < mci->nr_csrows; i++) { | |
518 | int j; | |
e7ecd891 | 519 | |
da9bb1d2 AC |
520 | edac_mc_dump_csrow(&mci->csrows[i]); |
521 | for (j = 0; j < mci->csrows[i].nr_channels; j++) | |
079708b9 | 522 | edac_mc_dump_channel(&mci->csrows[i]. |
052dfb45 | 523 | channels[j]); |
da9bb1d2 AC |
524 | } |
525 | } | |
526 | #endif | |
63b7df91 | 527 | mutex_lock(&mem_ctls_mutex); |
da9bb1d2 AC |
528 | |
529 | if (add_mc_to_global_list(mci)) | |
028a7b6d | 530 | goto fail0; |
da9bb1d2 AC |
531 | |
532 | /* set load time so that error rate can be tracked */ | |
533 | mci->start_time = jiffies; | |
534 | ||
9794f33d | 535 | if (edac_create_sysfs_mci_device(mci)) { |
536 | edac_mc_printk(mci, KERN_WARNING, | |
052dfb45 | 537 | "failed to create sysfs device\n"); |
9794f33d | 538 | goto fail1; |
539 | } | |
da9bb1d2 | 540 | |
81d87cb1 DJ |
541 | /* If there IS a check routine, then we are running POLLED */ |
542 | if (mci->edac_check != NULL) { | |
543 | /* This instance is NOW RUNNING */ | |
544 | mci->op_state = OP_RUNNING_POLL; | |
545 | ||
546 | edac_mc_workq_setup(mci, edac_mc_get_poll_msec()); | |
547 | } else { | |
548 | mci->op_state = OP_RUNNING_INTERRUPT; | |
549 | } | |
550 | ||
da9bb1d2 | 551 | /* Report action taken */ |
bf52fa4a | 552 | edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':" |
17aa7e03 | 553 | " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci)); |
da9bb1d2 | 554 | |
63b7df91 | 555 | mutex_unlock(&mem_ctls_mutex); |
028a7b6d | 556 | return 0; |
da9bb1d2 | 557 | |
052dfb45 | 558 | fail1: |
028a7b6d DP |
559 | del_mc_from_global_list(mci); |
560 | ||
052dfb45 | 561 | fail0: |
63b7df91 | 562 | mutex_unlock(&mem_ctls_mutex); |
028a7b6d | 563 | return 1; |
da9bb1d2 | 564 | } |
9110540f | 565 | EXPORT_SYMBOL_GPL(edac_mc_add_mc); |
da9bb1d2 | 566 | |
da9bb1d2 | 567 | /** |
472678eb DP |
568 | * edac_mc_del_mc: Remove sysfs entries for specified mci structure and |
569 | * remove mci structure from global list | |
37f04581 | 570 | * @pdev: Pointer to 'struct device' representing mci structure to remove. |
da9bb1d2 | 571 | * |
18dbc337 | 572 | * Return pointer to removed mci structure, or NULL if device not found. |
da9bb1d2 | 573 | */ |
079708b9 | 574 | struct mem_ctl_info *edac_mc_del_mc(struct device *dev) |
da9bb1d2 | 575 | { |
18dbc337 | 576 | struct mem_ctl_info *mci; |
da9bb1d2 | 577 | |
bf52fa4a DT |
578 | debugf0("%s()\n", __func__); |
579 | ||
63b7df91 | 580 | mutex_lock(&mem_ctls_mutex); |
18dbc337 | 581 | |
bf52fa4a DT |
582 | /* find the requested mci struct in the global list */ |
583 | mci = find_mci_by_dev(dev); | |
584 | if (mci == NULL) { | |
63b7df91 | 585 | mutex_unlock(&mem_ctls_mutex); |
18dbc337 DP |
586 | return NULL; |
587 | } | |
588 | ||
81d87cb1 DJ |
589 | /* marking MCI offline */ |
590 | mci->op_state = OP_OFFLINE; | |
591 | ||
da9bb1d2 | 592 | del_mc_from_global_list(mci); |
63b7df91 | 593 | mutex_unlock(&mem_ctls_mutex); |
bf52fa4a DT |
594 | |
595 | /* flush workq processes and remove sysfs */ | |
596 | edac_mc_workq_teardown(mci); | |
597 | edac_remove_sysfs_mci_device(mci); | |
598 | ||
537fba28 | 599 | edac_printk(KERN_INFO, EDAC_MC, |
052dfb45 | 600 | "Removed device %d for %s %s: DEV %s\n", mci->mc_idx, |
17aa7e03 | 601 | mci->mod_name, mci->ctl_name, edac_dev_name(mci)); |
bf52fa4a | 602 | |
18dbc337 | 603 | return mci; |
da9bb1d2 | 604 | } |
9110540f | 605 | EXPORT_SYMBOL_GPL(edac_mc_del_mc); |
da9bb1d2 | 606 | |
2da1c119 AB |
607 | static void edac_mc_scrub_block(unsigned long page, unsigned long offset, |
608 | u32 size) | |
da9bb1d2 AC |
609 | { |
610 | struct page *pg; | |
611 | void *virt_addr; | |
612 | unsigned long flags = 0; | |
613 | ||
537fba28 | 614 | debugf3("%s()\n", __func__); |
da9bb1d2 AC |
615 | |
616 | /* ECC error page was not in our memory. Ignore it. */ | |
079708b9 | 617 | if (!pfn_valid(page)) |
da9bb1d2 AC |
618 | return; |
619 | ||
620 | /* Find the actual page structure then map it and fix */ | |
621 | pg = pfn_to_page(page); | |
622 | ||
623 | if (PageHighMem(pg)) | |
624 | local_irq_save(flags); | |
625 | ||
626 | virt_addr = kmap_atomic(pg, KM_BOUNCE_READ); | |
627 | ||
628 | /* Perform architecture specific atomic scrub operation */ | |
629 | atomic_scrub(virt_addr + offset, size); | |
630 | ||
631 | /* Unmap and complete */ | |
632 | kunmap_atomic(virt_addr, KM_BOUNCE_READ); | |
633 | ||
634 | if (PageHighMem(pg)) | |
635 | local_irq_restore(flags); | |
636 | } | |
637 | ||
da9bb1d2 | 638 | /* FIXME - should return -1 */ |
e7ecd891 | 639 | int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page) |
da9bb1d2 AC |
640 | { |
641 | struct csrow_info *csrows = mci->csrows; | |
642 | int row, i; | |
643 | ||
537fba28 | 644 | debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page); |
da9bb1d2 AC |
645 | row = -1; |
646 | ||
647 | for (i = 0; i < mci->nr_csrows; i++) { | |
648 | struct csrow_info *csrow = &csrows[i]; | |
649 | ||
650 | if (csrow->nr_pages == 0) | |
651 | continue; | |
652 | ||
537fba28 DP |
653 | debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) " |
654 | "mask(0x%lx)\n", mci->mc_idx, __func__, | |
655 | csrow->first_page, page, csrow->last_page, | |
656 | csrow->page_mask); | |
da9bb1d2 AC |
657 | |
658 | if ((page >= csrow->first_page) && | |
659 | (page <= csrow->last_page) && | |
660 | ((page & csrow->page_mask) == | |
661 | (csrow->first_page & csrow->page_mask))) { | |
662 | row = i; | |
663 | break; | |
664 | } | |
665 | } | |
666 | ||
667 | if (row == -1) | |
537fba28 | 668 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
669 | "could not look up page error address %lx\n", |
670 | (unsigned long)page); | |
da9bb1d2 AC |
671 | |
672 | return row; | |
673 | } | |
9110540f | 674 | EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page); |
da9bb1d2 | 675 | |
da9bb1d2 AC |
676 | /* FIXME - setable log (warning/emerg) levels */ |
677 | /* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */ | |
678 | void edac_mc_handle_ce(struct mem_ctl_info *mci, | |
052dfb45 DT |
679 | unsigned long page_frame_number, |
680 | unsigned long offset_in_page, unsigned long syndrome, | |
681 | int row, int channel, const char *msg) | |
da9bb1d2 AC |
682 | { |
683 | unsigned long remapped_page; | |
684 | ||
537fba28 | 685 | debugf3("MC%d: %s()\n", mci->mc_idx, __func__); |
da9bb1d2 AC |
686 | |
687 | /* FIXME - maybe make panic on INTERNAL ERROR an option */ | |
688 | if (row >= mci->nr_csrows || row < 0) { | |
689 | /* something is wrong */ | |
537fba28 | 690 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
691 | "INTERNAL ERROR: row out of range " |
692 | "(%d >= %d)\n", row, mci->nr_csrows); | |
da9bb1d2 AC |
693 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
694 | return; | |
695 | } | |
e7ecd891 | 696 | |
da9bb1d2 AC |
697 | if (channel >= mci->csrows[row].nr_channels || channel < 0) { |
698 | /* something is wrong */ | |
537fba28 | 699 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
700 | "INTERNAL ERROR: channel out of range " |
701 | "(%d >= %d)\n", channel, | |
702 | mci->csrows[row].nr_channels); | |
da9bb1d2 AC |
703 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
704 | return; | |
705 | } | |
706 | ||
4de78c68 | 707 | if (edac_mc_get_log_ce()) |
da9bb1d2 | 708 | /* FIXME - put in DIMM location */ |
537fba28 | 709 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 DT |
710 | "CE page 0x%lx, offset 0x%lx, grain %d, syndrome " |
711 | "0x%lx, row %d, channel %d, label \"%s\": %s\n", | |
712 | page_frame_number, offset_in_page, | |
713 | mci->csrows[row].grain, syndrome, row, channel, | |
714 | mci->csrows[row].channels[channel].label, msg); | |
da9bb1d2 AC |
715 | |
716 | mci->ce_count++; | |
717 | mci->csrows[row].ce_count++; | |
718 | mci->csrows[row].channels[channel].ce_count++; | |
719 | ||
720 | if (mci->scrub_mode & SCRUB_SW_SRC) { | |
721 | /* | |
722 | * Some MC's can remap memory so that it is still available | |
723 | * at a different address when PCI devices map into memory. | |
724 | * MC's that can't do this lose the memory where PCI devices | |
725 | * are mapped. This mapping is MC dependant and so we call | |
726 | * back into the MC driver for it to map the MC page to | |
727 | * a physical (CPU) page which can then be mapped to a virtual | |
728 | * page - which can then be scrubbed. | |
729 | */ | |
730 | remapped_page = mci->ctl_page_to_phys ? | |
052dfb45 DT |
731 | mci->ctl_page_to_phys(mci, page_frame_number) : |
732 | page_frame_number; | |
da9bb1d2 AC |
733 | |
734 | edac_mc_scrub_block(remapped_page, offset_in_page, | |
052dfb45 | 735 | mci->csrows[row].grain); |
da9bb1d2 AC |
736 | } |
737 | } | |
9110540f | 738 | EXPORT_SYMBOL_GPL(edac_mc_handle_ce); |
da9bb1d2 | 739 | |
e7ecd891 | 740 | void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg) |
da9bb1d2 | 741 | { |
4de78c68 | 742 | if (edac_mc_get_log_ce()) |
537fba28 | 743 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 | 744 | "CE - no information available: %s\n", msg); |
e7ecd891 | 745 | |
da9bb1d2 AC |
746 | mci->ce_noinfo_count++; |
747 | mci->ce_count++; | |
748 | } | |
9110540f | 749 | EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info); |
da9bb1d2 | 750 | |
da9bb1d2 | 751 | void edac_mc_handle_ue(struct mem_ctl_info *mci, |
052dfb45 DT |
752 | unsigned long page_frame_number, |
753 | unsigned long offset_in_page, int row, const char *msg) | |
da9bb1d2 AC |
754 | { |
755 | int len = EDAC_MC_LABEL_LEN * 4; | |
756 | char labels[len + 1]; | |
757 | char *pos = labels; | |
758 | int chan; | |
759 | int chars; | |
760 | ||
537fba28 | 761 | debugf3("MC%d: %s()\n", mci->mc_idx, __func__); |
da9bb1d2 AC |
762 | |
763 | /* FIXME - maybe make panic on INTERNAL ERROR an option */ | |
764 | if (row >= mci->nr_csrows || row < 0) { | |
765 | /* something is wrong */ | |
537fba28 | 766 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
767 | "INTERNAL ERROR: row out of range " |
768 | "(%d >= %d)\n", row, mci->nr_csrows); | |
da9bb1d2 AC |
769 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
770 | return; | |
771 | } | |
772 | ||
773 | chars = snprintf(pos, len + 1, "%s", | |
079708b9 | 774 | mci->csrows[row].channels[0].label); |
da9bb1d2 AC |
775 | len -= chars; |
776 | pos += chars; | |
e7ecd891 | 777 | |
da9bb1d2 | 778 | for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0); |
052dfb45 | 779 | chan++) { |
da9bb1d2 | 780 | chars = snprintf(pos, len + 1, ":%s", |
079708b9 | 781 | mci->csrows[row].channels[chan].label); |
da9bb1d2 AC |
782 | len -= chars; |
783 | pos += chars; | |
784 | } | |
785 | ||
4de78c68 | 786 | if (edac_mc_get_log_ue()) |
537fba28 | 787 | edac_mc_printk(mci, KERN_EMERG, |
052dfb45 DT |
788 | "UE page 0x%lx, offset 0x%lx, grain %d, row %d, " |
789 | "labels \"%s\": %s\n", page_frame_number, | |
790 | offset_in_page, mci->csrows[row].grain, row, | |
791 | labels, msg); | |
da9bb1d2 | 792 | |
4de78c68 | 793 | if (edac_mc_get_panic_on_ue()) |
e7ecd891 | 794 | panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, " |
052dfb45 DT |
795 | "row %d, labels \"%s\": %s\n", mci->mc_idx, |
796 | page_frame_number, offset_in_page, | |
797 | mci->csrows[row].grain, row, labels, msg); | |
da9bb1d2 AC |
798 | |
799 | mci->ue_count++; | |
800 | mci->csrows[row].ue_count++; | |
801 | } | |
9110540f | 802 | EXPORT_SYMBOL_GPL(edac_mc_handle_ue); |
da9bb1d2 | 803 | |
e7ecd891 | 804 | void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg) |
da9bb1d2 | 805 | { |
4de78c68 | 806 | if (edac_mc_get_panic_on_ue()) |
da9bb1d2 AC |
807 | panic("EDAC MC%d: Uncorrected Error", mci->mc_idx); |
808 | ||
4de78c68 | 809 | if (edac_mc_get_log_ue()) |
537fba28 | 810 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 | 811 | "UE - no information available: %s\n", msg); |
da9bb1d2 AC |
812 | mci->ue_noinfo_count++; |
813 | mci->ue_count++; | |
814 | } | |
079708b9 | 815 | EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info); |
da9bb1d2 | 816 | |
9794f33d | 817 | /************************************************************* |
818 | * On Fully Buffered DIMM modules, this help function is | |
819 | * called to process UE events | |
820 | */ | |
821 | void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci, | |
052dfb45 DT |
822 | unsigned int csrow, |
823 | unsigned int channela, | |
824 | unsigned int channelb, char *msg) | |
9794f33d | 825 | { |
826 | int len = EDAC_MC_LABEL_LEN * 4; | |
827 | char labels[len + 1]; | |
828 | char *pos = labels; | |
829 | int chars; | |
830 | ||
831 | if (csrow >= mci->nr_csrows) { | |
832 | /* something is wrong */ | |
833 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
834 | "INTERNAL ERROR: row out of range (%d >= %d)\n", |
835 | csrow, mci->nr_csrows); | |
9794f33d | 836 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
837 | return; | |
838 | } | |
839 | ||
840 | if (channela >= mci->csrows[csrow].nr_channels) { | |
841 | /* something is wrong */ | |
842 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
843 | "INTERNAL ERROR: channel-a out of range " |
844 | "(%d >= %d)\n", | |
845 | channela, mci->csrows[csrow].nr_channels); | |
9794f33d | 846 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
847 | return; | |
848 | } | |
849 | ||
850 | if (channelb >= mci->csrows[csrow].nr_channels) { | |
851 | /* something is wrong */ | |
852 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
853 | "INTERNAL ERROR: channel-b out of range " |
854 | "(%d >= %d)\n", | |
855 | channelb, mci->csrows[csrow].nr_channels); | |
9794f33d | 856 | edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); |
857 | return; | |
858 | } | |
859 | ||
860 | mci->ue_count++; | |
861 | mci->csrows[csrow].ue_count++; | |
862 | ||
863 | /* Generate the DIMM labels from the specified channels */ | |
864 | chars = snprintf(pos, len + 1, "%s", | |
865 | mci->csrows[csrow].channels[channela].label); | |
079708b9 DT |
866 | len -= chars; |
867 | pos += chars; | |
9794f33d | 868 | chars = snprintf(pos, len + 1, "-%s", |
869 | mci->csrows[csrow].channels[channelb].label); | |
870 | ||
4de78c68 | 871 | if (edac_mc_get_log_ue()) |
9794f33d | 872 | edac_mc_printk(mci, KERN_EMERG, |
052dfb45 DT |
873 | "UE row %d, channel-a= %d channel-b= %d " |
874 | "labels \"%s\": %s\n", csrow, channela, channelb, | |
875 | labels, msg); | |
9794f33d | 876 | |
4de78c68 | 877 | if (edac_mc_get_panic_on_ue()) |
9794f33d | 878 | panic("UE row %d, channel-a= %d channel-b= %d " |
052dfb45 DT |
879 | "labels \"%s\": %s\n", csrow, channela, |
880 | channelb, labels, msg); | |
9794f33d | 881 | } |
882 | EXPORT_SYMBOL(edac_mc_handle_fbd_ue); | |
883 | ||
884 | /************************************************************* | |
885 | * On Fully Buffered DIMM modules, this help function is | |
886 | * called to process CE events | |
887 | */ | |
888 | void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci, | |
052dfb45 | 889 | unsigned int csrow, unsigned int channel, char *msg) |
9794f33d | 890 | { |
891 | ||
892 | /* Ensure boundary values */ | |
893 | if (csrow >= mci->nr_csrows) { | |
894 | /* something is wrong */ | |
895 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
896 | "INTERNAL ERROR: row out of range (%d >= %d)\n", |
897 | csrow, mci->nr_csrows); | |
9794f33d | 898 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
899 | return; | |
900 | } | |
901 | if (channel >= mci->csrows[csrow].nr_channels) { | |
902 | /* something is wrong */ | |
903 | edac_mc_printk(mci, KERN_ERR, | |
052dfb45 DT |
904 | "INTERNAL ERROR: channel out of range (%d >= %d)\n", |
905 | channel, mci->csrows[csrow].nr_channels); | |
9794f33d | 906 | edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); |
907 | return; | |
908 | } | |
909 | ||
4de78c68 | 910 | if (edac_mc_get_log_ce()) |
9794f33d | 911 | /* FIXME - put in DIMM location */ |
912 | edac_mc_printk(mci, KERN_WARNING, | |
052dfb45 DT |
913 | "CE row %d, channel %d, label \"%s\": %s\n", |
914 | csrow, channel, | |
915 | mci->csrows[csrow].channels[channel].label, msg); | |
9794f33d | 916 | |
917 | mci->ce_count++; | |
918 | mci->csrows[csrow].ce_count++; | |
919 | mci->csrows[csrow].channels[channel].ce_count++; | |
920 | } | |
079708b9 | 921 | EXPORT_SYMBOL(edac_mc_handle_fbd_ce); |