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