edac_mc.c 23 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886
  1. /*
  2. * edac_mc kernel module
  3. * (C) 2005, 2006 Linux Networx (http://lnxi.com)
  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. #include <linux/module.h>
  15. #include <linux/proc_fs.h>
  16. #include <linux/kernel.h>
  17. #include <linux/types.h>
  18. #include <linux/smp.h>
  19. #include <linux/init.h>
  20. #include <linux/sysctl.h>
  21. #include <linux/highmem.h>
  22. #include <linux/timer.h>
  23. #include <linux/slab.h>
  24. #include <linux/jiffies.h>
  25. #include <linux/spinlock.h>
  26. #include <linux/list.h>
  27. #include <linux/sysdev.h>
  28. #include <linux/ctype.h>
  29. #include <linux/edac.h>
  30. #include <asm/uaccess.h>
  31. #include <asm/page.h>
  32. #include <asm/edac.h>
  33. #include "edac_core.h"
  34. #include "edac_module.h"
  35. /* lock to memory controller's control array */
  36. static DEFINE_MUTEX(mem_ctls_mutex);
  37. static LIST_HEAD(mc_devices);
  38. #ifdef CONFIG_EDAC_DEBUG
  39. static void edac_mc_dump_channel(struct channel_info *chan)
  40. {
  41. debugf4("\tchannel = %p\n", chan);
  42. debugf4("\tchannel->chan_idx = %d\n", chan->chan_idx);
  43. debugf4("\tchannel->ce_count = %d\n", chan->ce_count);
  44. debugf4("\tchannel->label = '%s'\n", chan->label);
  45. debugf4("\tchannel->csrow = %p\n\n", chan->csrow);
  46. }
  47. static void edac_mc_dump_csrow(struct csrow_info *csrow)
  48. {
  49. debugf4("\tcsrow = %p\n", csrow);
  50. debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx);
  51. debugf4("\tcsrow->first_page = 0x%lx\n", csrow->first_page);
  52. debugf4("\tcsrow->last_page = 0x%lx\n", csrow->last_page);
  53. debugf4("\tcsrow->page_mask = 0x%lx\n", csrow->page_mask);
  54. debugf4("\tcsrow->nr_pages = 0x%x\n", csrow->nr_pages);
  55. debugf4("\tcsrow->nr_channels = %d\n", csrow->nr_channels);
  56. debugf4("\tcsrow->channels = %p\n", csrow->channels);
  57. debugf4("\tcsrow->mci = %p\n\n", csrow->mci);
  58. }
  59. static void edac_mc_dump_mci(struct mem_ctl_info *mci)
  60. {
  61. debugf3("\tmci = %p\n", mci);
  62. debugf3("\tmci->mtype_cap = %lx\n", mci->mtype_cap);
  63. debugf3("\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
  64. debugf3("\tmci->edac_cap = %lx\n", mci->edac_cap);
  65. debugf4("\tmci->edac_check = %p\n", mci->edac_check);
  66. debugf3("\tmci->nr_csrows = %d, csrows = %p\n",
  67. mci->nr_csrows, mci->csrows);
  68. debugf3("\tdev = %p\n", mci->dev);
  69. debugf3("\tmod_name:ctl_name = %s:%s\n", mci->mod_name, mci->ctl_name);
  70. debugf3("\tpvt_info = %p\n\n", mci->pvt_info);
  71. }
  72. #endif /* CONFIG_EDAC_DEBUG */
  73. /* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'.
  74. * Adjust 'ptr' so that its alignment is at least as stringent as what the
  75. * compiler would provide for X and return the aligned result.
  76. *
  77. * If 'size' is a constant, the compiler will optimize this whole function
  78. * down to either a no-op or the addition of a constant to the value of 'ptr'.
  79. */
  80. void *edac_align_ptr(void *ptr, unsigned size)
  81. {
  82. unsigned align, r;
  83. /* Here we assume that the alignment of a "long long" is the most
  84. * stringent alignment that the compiler will ever provide by default.
  85. * As far as I know, this is a reasonable assumption.
  86. */
  87. if (size > sizeof(long))
  88. align = sizeof(long long);
  89. else if (size > sizeof(int))
  90. align = sizeof(long);
  91. else if (size > sizeof(short))
  92. align = sizeof(int);
  93. else if (size > sizeof(char))
  94. align = sizeof(short);
  95. else
  96. return (char *)ptr;
  97. r = size % align;
  98. if (r == 0)
  99. return (char *)ptr;
  100. return (void *)(((unsigned long)ptr) + align - r);
  101. }
  102. /**
  103. * edac_mc_alloc: Allocate a struct mem_ctl_info structure
  104. * @size_pvt: size of private storage needed
  105. * @nr_csrows: Number of CWROWS needed for this MC
  106. * @nr_chans: Number of channels for the MC
  107. *
  108. * Everything is kmalloc'ed as one big chunk - more efficient.
  109. * Only can be used if all structures have the same lifetime - otherwise
  110. * you have to allocate and initialize your own structures.
  111. *
  112. * Use edac_mc_free() to free mc structures allocated by this function.
  113. *
  114. * Returns:
  115. * NULL allocation failed
  116. * struct mem_ctl_info pointer
  117. */
  118. struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows,
  119. unsigned nr_chans, int edac_index)
  120. {
  121. struct mem_ctl_info *mci;
  122. struct csrow_info *csi, *csrow;
  123. struct channel_info *chi, *chp, *chan;
  124. void *pvt;
  125. unsigned size;
  126. int row, chn;
  127. int err;
  128. /* Figure out the offsets of the various items from the start of an mc
  129. * structure. We want the alignment of each item to be at least as
  130. * stringent as what the compiler would provide if we could simply
  131. * hardcode everything into a single struct.
  132. */
  133. mci = (struct mem_ctl_info *)0;
  134. csi = edac_align_ptr(&mci[1], sizeof(*csi));
  135. chi = edac_align_ptr(&csi[nr_csrows], sizeof(*chi));
  136. pvt = edac_align_ptr(&chi[nr_chans * nr_csrows], sz_pvt);
  137. size = ((unsigned long)pvt) + sz_pvt;
  138. mci = kzalloc(size, GFP_KERNEL);
  139. if (mci == NULL)
  140. return NULL;
  141. /* Adjust pointers so they point within the memory we just allocated
  142. * rather than an imaginary chunk of memory located at address 0.
  143. */
  144. csi = (struct csrow_info *)(((char *)mci) + ((unsigned long)csi));
  145. chi = (struct channel_info *)(((char *)mci) + ((unsigned long)chi));
  146. pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
  147. /* setup index and various internal pointers */
  148. mci->mc_idx = edac_index;
  149. mci->csrows = csi;
  150. mci->pvt_info = pvt;
  151. mci->nr_csrows = nr_csrows;
  152. for (row = 0; row < nr_csrows; row++) {
  153. csrow = &csi[row];
  154. csrow->csrow_idx = row;
  155. csrow->mci = mci;
  156. csrow->nr_channels = nr_chans;
  157. chp = &chi[row * nr_chans];
  158. csrow->channels = chp;
  159. for (chn = 0; chn < nr_chans; chn++) {
  160. chan = &chp[chn];
  161. chan->chan_idx = chn;
  162. chan->csrow = csrow;
  163. }
  164. }
  165. mci->op_state = OP_ALLOC;
  166. /*
  167. * Initialize the 'root' kobj for the edac_mc controller
  168. */
  169. err = edac_mc_register_sysfs_main_kobj(mci);
  170. if (err) {
  171. kfree(mci);
  172. return NULL;
  173. }
  174. /* at this point, the root kobj is valid, and in order to
  175. * 'free' the object, then the function:
  176. * edac_mc_unregister_sysfs_main_kobj() must be called
  177. * which will perform kobj unregistration and the actual free
  178. * will occur during the kobject callback operation
  179. */
  180. return mci;
  181. }
  182. EXPORT_SYMBOL_GPL(edac_mc_alloc);
  183. /**
  184. * edac_mc_free
  185. * 'Free' a previously allocated 'mci' structure
  186. * @mci: pointer to a struct mem_ctl_info structure
  187. */
  188. void edac_mc_free(struct mem_ctl_info *mci)
  189. {
  190. edac_mc_unregister_sysfs_main_kobj(mci);
  191. }
  192. EXPORT_SYMBOL_GPL(edac_mc_free);
  193. /*
  194. * find_mci_by_dev
  195. *
  196. * scan list of controllers looking for the one that manages
  197. * the 'dev' device
  198. */
  199. static struct mem_ctl_info *find_mci_by_dev(struct device *dev)
  200. {
  201. struct mem_ctl_info *mci;
  202. struct list_head *item;
  203. debugf3("%s()\n", __func__);
  204. list_for_each(item, &mc_devices) {
  205. mci = list_entry(item, struct mem_ctl_info, link);
  206. if (mci->dev == dev)
  207. return mci;
  208. }
  209. return NULL;
  210. }
  211. /*
  212. * handler for EDAC to check if NMI type handler has asserted interrupt
  213. */
  214. static int edac_mc_assert_error_check_and_clear(void)
  215. {
  216. int old_state;
  217. if (edac_op_state == EDAC_OPSTATE_POLL)
  218. return 1;
  219. old_state = edac_err_assert;
  220. edac_err_assert = 0;
  221. return old_state;
  222. }
  223. /*
  224. * edac_mc_workq_function
  225. * performs the operation scheduled by a workq request
  226. */
  227. static void edac_mc_workq_function(struct work_struct *work_req)
  228. {
  229. struct delayed_work *d_work = to_delayed_work(work_req);
  230. struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
  231. mutex_lock(&mem_ctls_mutex);
  232. /* if this control struct has movd to offline state, we are done */
  233. if (mci->op_state == OP_OFFLINE) {
  234. mutex_unlock(&mem_ctls_mutex);
  235. return;
  236. }
  237. /* Only poll controllers that are running polled and have a check */
  238. if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL))
  239. mci->edac_check(mci);
  240. mutex_unlock(&mem_ctls_mutex);
  241. /* Reschedule */
  242. queue_delayed_work(edac_workqueue, &mci->work,
  243. msecs_to_jiffies(edac_mc_get_poll_msec()));
  244. }
  245. /*
  246. * edac_mc_workq_setup
  247. * initialize a workq item for this mci
  248. * passing in the new delay period in msec
  249. *
  250. * locking model:
  251. *
  252. * called with the mem_ctls_mutex held
  253. */
  254. static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec)
  255. {
  256. debugf0("%s()\n", __func__);
  257. /* if this instance is not in the POLL state, then simply return */
  258. if (mci->op_state != OP_RUNNING_POLL)
  259. return;
  260. INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
  261. queue_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec));
  262. }
  263. /*
  264. * edac_mc_workq_teardown
  265. * stop the workq processing on this mci
  266. *
  267. * locking model:
  268. *
  269. * called WITHOUT lock held
  270. */
  271. static void edac_mc_workq_teardown(struct mem_ctl_info *mci)
  272. {
  273. int status;
  274. status = cancel_delayed_work(&mci->work);
  275. if (status == 0) {
  276. debugf0("%s() not canceled, flush the queue\n",
  277. __func__);
  278. /* workq instance might be running, wait for it */
  279. flush_workqueue(edac_workqueue);
  280. }
  281. }
  282. /*
  283. * edac_mc_reset_delay_period(unsigned long value)
  284. *
  285. * user space has updated our poll period value, need to
  286. * reset our workq delays
  287. */
  288. void edac_mc_reset_delay_period(int value)
  289. {
  290. struct mem_ctl_info *mci;
  291. struct list_head *item;
  292. mutex_lock(&mem_ctls_mutex);
  293. /* scan the list and turn off all workq timers, doing so under lock
  294. */
  295. list_for_each(item, &mc_devices) {
  296. mci = list_entry(item, struct mem_ctl_info, link);
  297. if (mci->op_state == OP_RUNNING_POLL)
  298. cancel_delayed_work(&mci->work);
  299. }
  300. mutex_unlock(&mem_ctls_mutex);
  301. /* re-walk the list, and reset the poll delay */
  302. mutex_lock(&mem_ctls_mutex);
  303. list_for_each(item, &mc_devices) {
  304. mci = list_entry(item, struct mem_ctl_info, link);
  305. edac_mc_workq_setup(mci, (unsigned long) value);
  306. }
  307. mutex_unlock(&mem_ctls_mutex);
  308. }
  309. /* Return 0 on success, 1 on failure.
  310. * Before calling this function, caller must
  311. * assign a unique value to mci->mc_idx.
  312. *
  313. * locking model:
  314. *
  315. * called with the mem_ctls_mutex lock held
  316. */
  317. static int add_mc_to_global_list(struct mem_ctl_info *mci)
  318. {
  319. struct list_head *item, *insert_before;
  320. struct mem_ctl_info *p;
  321. insert_before = &mc_devices;
  322. p = find_mci_by_dev(mci->dev);
  323. if (unlikely(p != NULL))
  324. goto fail0;
  325. list_for_each(item, &mc_devices) {
  326. p = list_entry(item, struct mem_ctl_info, link);
  327. if (p->mc_idx >= mci->mc_idx) {
  328. if (unlikely(p->mc_idx == mci->mc_idx))
  329. goto fail1;
  330. insert_before = item;
  331. break;
  332. }
  333. }
  334. list_add_tail_rcu(&mci->link, insert_before);
  335. atomic_inc(&edac_handlers);
  336. return 0;
  337. fail0:
  338. edac_printk(KERN_WARNING, EDAC_MC,
  339. "%s (%s) %s %s already assigned %d\n", dev_name(p->dev),
  340. edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
  341. return 1;
  342. fail1:
  343. edac_printk(KERN_WARNING, EDAC_MC,
  344. "bug in low-level driver: attempt to assign\n"
  345. " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
  346. return 1;
  347. }
  348. static void complete_mc_list_del(struct rcu_head *head)
  349. {
  350. struct mem_ctl_info *mci;
  351. mci = container_of(head, struct mem_ctl_info, rcu);
  352. INIT_LIST_HEAD(&mci->link);
  353. }
  354. static void del_mc_from_global_list(struct mem_ctl_info *mci)
  355. {
  356. atomic_dec(&edac_handlers);
  357. list_del_rcu(&mci->link);
  358. call_rcu(&mci->rcu, complete_mc_list_del);
  359. rcu_barrier();
  360. }
  361. /**
  362. * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'.
  363. *
  364. * If found, return a pointer to the structure.
  365. * Else return NULL.
  366. *
  367. * Caller must hold mem_ctls_mutex.
  368. */
  369. struct mem_ctl_info *edac_mc_find(int idx)
  370. {
  371. struct list_head *item;
  372. struct mem_ctl_info *mci;
  373. list_for_each(item, &mc_devices) {
  374. mci = list_entry(item, struct mem_ctl_info, link);
  375. if (mci->mc_idx >= idx) {
  376. if (mci->mc_idx == idx)
  377. return mci;
  378. break;
  379. }
  380. }
  381. return NULL;
  382. }
  383. EXPORT_SYMBOL(edac_mc_find);
  384. /**
  385. * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and
  386. * create sysfs entries associated with mci structure
  387. * @mci: pointer to the mci structure to be added to the list
  388. * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure.
  389. *
  390. * Return:
  391. * 0 Success
  392. * !0 Failure
  393. */
  394. /* FIXME - should a warning be printed if no error detection? correction? */
  395. int edac_mc_add_mc(struct mem_ctl_info *mci)
  396. {
  397. debugf0("%s()\n", __func__);
  398. #ifdef CONFIG_EDAC_DEBUG
  399. if (edac_debug_level >= 3)
  400. edac_mc_dump_mci(mci);
  401. if (edac_debug_level >= 4) {
  402. int i;
  403. for (i = 0; i < mci->nr_csrows; i++) {
  404. int j;
  405. edac_mc_dump_csrow(&mci->csrows[i]);
  406. for (j = 0; j < mci->csrows[i].nr_channels; j++)
  407. edac_mc_dump_channel(&mci->csrows[i].
  408. channels[j]);
  409. }
  410. }
  411. #endif
  412. mutex_lock(&mem_ctls_mutex);
  413. if (add_mc_to_global_list(mci))
  414. goto fail0;
  415. /* set load time so that error rate can be tracked */
  416. mci->start_time = jiffies;
  417. if (edac_create_sysfs_mci_device(mci)) {
  418. edac_mc_printk(mci, KERN_WARNING,
  419. "failed to create sysfs device\n");
  420. goto fail1;
  421. }
  422. /* If there IS a check routine, then we are running POLLED */
  423. if (mci->edac_check != NULL) {
  424. /* This instance is NOW RUNNING */
  425. mci->op_state = OP_RUNNING_POLL;
  426. edac_mc_workq_setup(mci, edac_mc_get_poll_msec());
  427. } else {
  428. mci->op_state = OP_RUNNING_INTERRUPT;
  429. }
  430. /* Report action taken */
  431. edac_mc_printk(mci, KERN_INFO, "Giving out device to '%s' '%s':"
  432. " DEV %s\n", mci->mod_name, mci->ctl_name, edac_dev_name(mci));
  433. mutex_unlock(&mem_ctls_mutex);
  434. return 0;
  435. fail1:
  436. del_mc_from_global_list(mci);
  437. fail0:
  438. mutex_unlock(&mem_ctls_mutex);
  439. return 1;
  440. }
  441. EXPORT_SYMBOL_GPL(edac_mc_add_mc);
  442. /**
  443. * edac_mc_del_mc: Remove sysfs entries for specified mci structure and
  444. * remove mci structure from global list
  445. * @pdev: Pointer to 'struct device' representing mci structure to remove.
  446. *
  447. * Return pointer to removed mci structure, or NULL if device not found.
  448. */
  449. struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
  450. {
  451. struct mem_ctl_info *mci;
  452. debugf0("%s()\n", __func__);
  453. mutex_lock(&mem_ctls_mutex);
  454. /* find the requested mci struct in the global list */
  455. mci = find_mci_by_dev(dev);
  456. if (mci == NULL) {
  457. mutex_unlock(&mem_ctls_mutex);
  458. return NULL;
  459. }
  460. /* marking MCI offline */
  461. mci->op_state = OP_OFFLINE;
  462. del_mc_from_global_list(mci);
  463. mutex_unlock(&mem_ctls_mutex);
  464. /* flush workq processes and remove sysfs */
  465. edac_mc_workq_teardown(mci);
  466. edac_remove_sysfs_mci_device(mci);
  467. edac_printk(KERN_INFO, EDAC_MC,
  468. "Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
  469. mci->mod_name, mci->ctl_name, edac_dev_name(mci));
  470. return mci;
  471. }
  472. EXPORT_SYMBOL_GPL(edac_mc_del_mc);
  473. static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
  474. u32 size)
  475. {
  476. struct page *pg;
  477. void *virt_addr;
  478. unsigned long flags = 0;
  479. debugf3("%s()\n", __func__);
  480. /* ECC error page was not in our memory. Ignore it. */
  481. if (!pfn_valid(page))
  482. return;
  483. /* Find the actual page structure then map it and fix */
  484. pg = pfn_to_page(page);
  485. if (PageHighMem(pg))
  486. local_irq_save(flags);
  487. virt_addr = kmap_atomic(pg, KM_BOUNCE_READ);
  488. /* Perform architecture specific atomic scrub operation */
  489. atomic_scrub(virt_addr + offset, size);
  490. /* Unmap and complete */
  491. kunmap_atomic(virt_addr, KM_BOUNCE_READ);
  492. if (PageHighMem(pg))
  493. local_irq_restore(flags);
  494. }
  495. /* FIXME - should return -1 */
  496. int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
  497. {
  498. struct csrow_info *csrows = mci->csrows;
  499. int row, i;
  500. debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page);
  501. row = -1;
  502. for (i = 0; i < mci->nr_csrows; i++) {
  503. struct csrow_info *csrow = &csrows[i];
  504. if (csrow->nr_pages == 0)
  505. continue;
  506. debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) "
  507. "mask(0x%lx)\n", mci->mc_idx, __func__,
  508. csrow->first_page, page, csrow->last_page,
  509. csrow->page_mask);
  510. if ((page >= csrow->first_page) &&
  511. (page <= csrow->last_page) &&
  512. ((page & csrow->page_mask) ==
  513. (csrow->first_page & csrow->page_mask))) {
  514. row = i;
  515. break;
  516. }
  517. }
  518. if (row == -1)
  519. edac_mc_printk(mci, KERN_ERR,
  520. "could not look up page error address %lx\n",
  521. (unsigned long)page);
  522. return row;
  523. }
  524. EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
  525. /* FIXME - setable log (warning/emerg) levels */
  526. /* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */
  527. void edac_mc_handle_ce(struct mem_ctl_info *mci,
  528. unsigned long page_frame_number,
  529. unsigned long offset_in_page, unsigned long syndrome,
  530. int row, int channel, const char *msg)
  531. {
  532. unsigned long remapped_page;
  533. debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
  534. /* FIXME - maybe make panic on INTERNAL ERROR an option */
  535. if (row >= mci->nr_csrows || row < 0) {
  536. /* something is wrong */
  537. edac_mc_printk(mci, KERN_ERR,
  538. "INTERNAL ERROR: row out of range "
  539. "(%d >= %d)\n", row, mci->nr_csrows);
  540. edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
  541. return;
  542. }
  543. if (channel >= mci->csrows[row].nr_channels || channel < 0) {
  544. /* something is wrong */
  545. edac_mc_printk(mci, KERN_ERR,
  546. "INTERNAL ERROR: channel out of range "
  547. "(%d >= %d)\n", channel,
  548. mci->csrows[row].nr_channels);
  549. edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
  550. return;
  551. }
  552. if (edac_mc_get_log_ce())
  553. /* FIXME - put in DIMM location */
  554. edac_mc_printk(mci, KERN_WARNING,
  555. "CE page 0x%lx, offset 0x%lx, grain %d, syndrome "
  556. "0x%lx, row %d, channel %d, label \"%s\": %s\n",
  557. page_frame_number, offset_in_page,
  558. mci->csrows[row].grain, syndrome, row, channel,
  559. mci->csrows[row].channels[channel].label, msg);
  560. mci->ce_count++;
  561. mci->csrows[row].ce_count++;
  562. mci->csrows[row].channels[channel].ce_count++;
  563. if (mci->scrub_mode & SCRUB_SW_SRC) {
  564. /*
  565. * Some MC's can remap memory so that it is still available
  566. * at a different address when PCI devices map into memory.
  567. * MC's that can't do this lose the memory where PCI devices
  568. * are mapped. This mapping is MC dependant and so we call
  569. * back into the MC driver for it to map the MC page to
  570. * a physical (CPU) page which can then be mapped to a virtual
  571. * page - which can then be scrubbed.
  572. */
  573. remapped_page = mci->ctl_page_to_phys ?
  574. mci->ctl_page_to_phys(mci, page_frame_number) :
  575. page_frame_number;
  576. edac_mc_scrub_block(remapped_page, offset_in_page,
  577. mci->csrows[row].grain);
  578. }
  579. }
  580. EXPORT_SYMBOL_GPL(edac_mc_handle_ce);
  581. void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg)
  582. {
  583. if (edac_mc_get_log_ce())
  584. edac_mc_printk(mci, KERN_WARNING,
  585. "CE - no information available: %s\n", msg);
  586. mci->ce_noinfo_count++;
  587. mci->ce_count++;
  588. }
  589. EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info);
  590. void edac_mc_handle_ue(struct mem_ctl_info *mci,
  591. unsigned long page_frame_number,
  592. unsigned long offset_in_page, int row, const char *msg)
  593. {
  594. int len = EDAC_MC_LABEL_LEN * 4;
  595. char labels[len + 1];
  596. char *pos = labels;
  597. int chan;
  598. int chars;
  599. debugf3("MC%d: %s()\n", mci->mc_idx, __func__);
  600. /* FIXME - maybe make panic on INTERNAL ERROR an option */
  601. if (row >= mci->nr_csrows || row < 0) {
  602. /* something is wrong */
  603. edac_mc_printk(mci, KERN_ERR,
  604. "INTERNAL ERROR: row out of range "
  605. "(%d >= %d)\n", row, mci->nr_csrows);
  606. edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
  607. return;
  608. }
  609. chars = snprintf(pos, len + 1, "%s",
  610. mci->csrows[row].channels[0].label);
  611. len -= chars;
  612. pos += chars;
  613. for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0);
  614. chan++) {
  615. chars = snprintf(pos, len + 1, ":%s",
  616. mci->csrows[row].channels[chan].label);
  617. len -= chars;
  618. pos += chars;
  619. }
  620. if (edac_mc_get_log_ue())
  621. edac_mc_printk(mci, KERN_EMERG,
  622. "UE page 0x%lx, offset 0x%lx, grain %d, row %d, "
  623. "labels \"%s\": %s\n", page_frame_number,
  624. offset_in_page, mci->csrows[row].grain, row,
  625. labels, msg);
  626. if (edac_mc_get_panic_on_ue())
  627. panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, "
  628. "row %d, labels \"%s\": %s\n", mci->mc_idx,
  629. page_frame_number, offset_in_page,
  630. mci->csrows[row].grain, row, labels, msg);
  631. mci->ue_count++;
  632. mci->csrows[row].ue_count++;
  633. }
  634. EXPORT_SYMBOL_GPL(edac_mc_handle_ue);
  635. void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg)
  636. {
  637. if (edac_mc_get_panic_on_ue())
  638. panic("EDAC MC%d: Uncorrected Error", mci->mc_idx);
  639. if (edac_mc_get_log_ue())
  640. edac_mc_printk(mci, KERN_WARNING,
  641. "UE - no information available: %s\n", msg);
  642. mci->ue_noinfo_count++;
  643. mci->ue_count++;
  644. }
  645. EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info);
  646. /*************************************************************
  647. * On Fully Buffered DIMM modules, this help function is
  648. * called to process UE events
  649. */
  650. void edac_mc_handle_fbd_ue(struct mem_ctl_info *mci,
  651. unsigned int csrow,
  652. unsigned int channela,
  653. unsigned int channelb, char *msg)
  654. {
  655. int len = EDAC_MC_LABEL_LEN * 4;
  656. char labels[len + 1];
  657. char *pos = labels;
  658. int chars;
  659. if (csrow >= mci->nr_csrows) {
  660. /* something is wrong */
  661. edac_mc_printk(mci, KERN_ERR,
  662. "INTERNAL ERROR: row out of range (%d >= %d)\n",
  663. csrow, mci->nr_csrows);
  664. edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
  665. return;
  666. }
  667. if (channela >= mci->csrows[csrow].nr_channels) {
  668. /* something is wrong */
  669. edac_mc_printk(mci, KERN_ERR,
  670. "INTERNAL ERROR: channel-a out of range "
  671. "(%d >= %d)\n",
  672. channela, mci->csrows[csrow].nr_channels);
  673. edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
  674. return;
  675. }
  676. if (channelb >= mci->csrows[csrow].nr_channels) {
  677. /* something is wrong */
  678. edac_mc_printk(mci, KERN_ERR,
  679. "INTERNAL ERROR: channel-b out of range "
  680. "(%d >= %d)\n",
  681. channelb, mci->csrows[csrow].nr_channels);
  682. edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR");
  683. return;
  684. }
  685. mci->ue_count++;
  686. mci->csrows[csrow].ue_count++;
  687. /* Generate the DIMM labels from the specified channels */
  688. chars = snprintf(pos, len + 1, "%s",
  689. mci->csrows[csrow].channels[channela].label);
  690. len -= chars;
  691. pos += chars;
  692. chars = snprintf(pos, len + 1, "-%s",
  693. mci->csrows[csrow].channels[channelb].label);
  694. if (edac_mc_get_log_ue())
  695. edac_mc_printk(mci, KERN_EMERG,
  696. "UE row %d, channel-a= %d channel-b= %d "
  697. "labels \"%s\": %s\n", csrow, channela, channelb,
  698. labels, msg);
  699. if (edac_mc_get_panic_on_ue())
  700. panic("UE row %d, channel-a= %d channel-b= %d "
  701. "labels \"%s\": %s\n", csrow, channela,
  702. channelb, labels, msg);
  703. }
  704. EXPORT_SYMBOL(edac_mc_handle_fbd_ue);
  705. /*************************************************************
  706. * On Fully Buffered DIMM modules, this help function is
  707. * called to process CE events
  708. */
  709. void edac_mc_handle_fbd_ce(struct mem_ctl_info *mci,
  710. unsigned int csrow, unsigned int channel, char *msg)
  711. {
  712. /* Ensure boundary values */
  713. if (csrow >= mci->nr_csrows) {
  714. /* something is wrong */
  715. edac_mc_printk(mci, KERN_ERR,
  716. "INTERNAL ERROR: row out of range (%d >= %d)\n",
  717. csrow, mci->nr_csrows);
  718. edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
  719. return;
  720. }
  721. if (channel >= mci->csrows[csrow].nr_channels) {
  722. /* something is wrong */
  723. edac_mc_printk(mci, KERN_ERR,
  724. "INTERNAL ERROR: channel out of range (%d >= %d)\n",
  725. channel, mci->csrows[csrow].nr_channels);
  726. edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR");
  727. return;
  728. }
  729. if (edac_mc_get_log_ce())
  730. /* FIXME - put in DIMM location */
  731. edac_mc_printk(mci, KERN_WARNING,
  732. "CE row %d, channel %d, label \"%s\": %s\n",
  733. csrow, channel,
  734. mci->csrows[csrow].channels[channel].label, msg);
  735. mci->ce_count++;
  736. mci->csrows[csrow].ce_count++;
  737. mci->csrows[csrow].channels[channel].ce_count++;
  738. }
  739. EXPORT_SYMBOL(edac_mc_handle_fbd_ce);