edac_mc.c revision 49c0dab7e6000888b616bedcbbc8cd4710331610
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/kthread.h> 31#include <asm/uaccess.h> 32#include <asm/page.h> 33#include <asm/edac.h> 34#include "edac_mc.h" 35 36#define EDAC_MC_VERSION "Ver: 2.0.1 " __DATE__ 37 38 39#ifdef CONFIG_EDAC_DEBUG 40/* Values of 0 to 4 will generate output */ 41int edac_debug_level = 1; 42EXPORT_SYMBOL_GPL(edac_debug_level); 43#endif 44 45/* EDAC Controls, setable by module parameter, and sysfs */ 46static int log_ue = 1; 47static int log_ce = 1; 48static int panic_on_ue; 49static int poll_msec = 1000; 50 51/* lock to memory controller's control array */ 52static DECLARE_MUTEX(mem_ctls_mutex); 53static struct list_head mc_devices = LIST_HEAD_INIT(mc_devices); 54 55static struct task_struct *edac_thread; 56 57#ifdef CONFIG_PCI 58static int check_pci_parity = 0; /* default YES check PCI parity */ 59static int panic_on_pci_parity; /* default no panic on PCI Parity */ 60static atomic_t pci_parity_count = ATOMIC_INIT(0); 61 62static struct kobject edac_pci_kobj; /* /sys/devices/system/edac/pci */ 63static struct completion edac_pci_kobj_complete; 64#endif /* CONFIG_PCI */ 65 66/* START sysfs data and methods */ 67 68 69static const char *mem_types[] = { 70 [MEM_EMPTY] = "Empty", 71 [MEM_RESERVED] = "Reserved", 72 [MEM_UNKNOWN] = "Unknown", 73 [MEM_FPM] = "FPM", 74 [MEM_EDO] = "EDO", 75 [MEM_BEDO] = "BEDO", 76 [MEM_SDR] = "Unbuffered-SDR", 77 [MEM_RDR] = "Registered-SDR", 78 [MEM_DDR] = "Unbuffered-DDR", 79 [MEM_RDDR] = "Registered-DDR", 80 [MEM_RMBS] = "RMBS" 81}; 82 83static const char *dev_types[] = { 84 [DEV_UNKNOWN] = "Unknown", 85 [DEV_X1] = "x1", 86 [DEV_X2] = "x2", 87 [DEV_X4] = "x4", 88 [DEV_X8] = "x8", 89 [DEV_X16] = "x16", 90 [DEV_X32] = "x32", 91 [DEV_X64] = "x64" 92}; 93 94static const char *edac_caps[] = { 95 [EDAC_UNKNOWN] = "Unknown", 96 [EDAC_NONE] = "None", 97 [EDAC_RESERVED] = "Reserved", 98 [EDAC_PARITY] = "PARITY", 99 [EDAC_EC] = "EC", 100 [EDAC_SECDED] = "SECDED", 101 [EDAC_S2ECD2ED] = "S2ECD2ED", 102 [EDAC_S4ECD4ED] = "S4ECD4ED", 103 [EDAC_S8ECD8ED] = "S8ECD8ED", 104 [EDAC_S16ECD16ED] = "S16ECD16ED" 105}; 106 107/* sysfs object: /sys/devices/system/edac */ 108static struct sysdev_class edac_class = { 109 set_kset_name("edac"), 110}; 111 112/* sysfs object: 113 * /sys/devices/system/edac/mc 114 */ 115static struct kobject edac_memctrl_kobj; 116 117/* We use these to wait for the reference counts on edac_memctrl_kobj and 118 * edac_pci_kobj to reach 0. 119 */ 120static struct completion edac_memctrl_kobj_complete; 121 122/* 123 * /sys/devices/system/edac/mc; 124 * data structures and methods 125 */ 126static ssize_t memctrl_int_show(void *ptr, char *buffer) 127{ 128 int *value = (int*) ptr; 129 return sprintf(buffer, "%u\n", *value); 130} 131 132static ssize_t memctrl_int_store(void *ptr, const char *buffer, size_t count) 133{ 134 int *value = (int*) ptr; 135 136 if (isdigit(*buffer)) 137 *value = simple_strtoul(buffer, NULL, 0); 138 139 return count; 140} 141 142struct memctrl_dev_attribute { 143 struct attribute attr; 144 void *value; 145 ssize_t (*show)(void *,char *); 146 ssize_t (*store)(void *, const char *, size_t); 147}; 148 149/* Set of show/store abstract level functions for memory control object */ 150static ssize_t memctrl_dev_show(struct kobject *kobj, 151 struct attribute *attr, char *buffer) 152{ 153 struct memctrl_dev_attribute *memctrl_dev; 154 memctrl_dev = (struct memctrl_dev_attribute*)attr; 155 156 if (memctrl_dev->show) 157 return memctrl_dev->show(memctrl_dev->value, buffer); 158 159 return -EIO; 160} 161 162static ssize_t memctrl_dev_store(struct kobject *kobj, struct attribute *attr, 163 const char *buffer, size_t count) 164{ 165 struct memctrl_dev_attribute *memctrl_dev; 166 memctrl_dev = (struct memctrl_dev_attribute*)attr; 167 168 if (memctrl_dev->store) 169 return memctrl_dev->store(memctrl_dev->value, buffer, count); 170 171 return -EIO; 172} 173 174static struct sysfs_ops memctrlfs_ops = { 175 .show = memctrl_dev_show, 176 .store = memctrl_dev_store 177}; 178 179#define MEMCTRL_ATTR(_name,_mode,_show,_store) \ 180struct memctrl_dev_attribute attr_##_name = { \ 181 .attr = {.name = __stringify(_name), .mode = _mode }, \ 182 .value = &_name, \ 183 .show = _show, \ 184 .store = _store, \ 185}; 186 187#define MEMCTRL_STRING_ATTR(_name,_data,_mode,_show,_store) \ 188struct memctrl_dev_attribute attr_##_name = { \ 189 .attr = {.name = __stringify(_name), .mode = _mode }, \ 190 .value = _data, \ 191 .show = _show, \ 192 .store = _store, \ 193}; 194 195/* csrow<id> control files */ 196MEMCTRL_ATTR(panic_on_ue,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store); 197MEMCTRL_ATTR(log_ue,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store); 198MEMCTRL_ATTR(log_ce,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store); 199MEMCTRL_ATTR(poll_msec,S_IRUGO|S_IWUSR,memctrl_int_show,memctrl_int_store); 200 201/* Base Attributes of the memory ECC object */ 202static struct memctrl_dev_attribute *memctrl_attr[] = { 203 &attr_panic_on_ue, 204 &attr_log_ue, 205 &attr_log_ce, 206 &attr_poll_msec, 207 NULL, 208}; 209 210/* Main MC kobject release() function */ 211static void edac_memctrl_master_release(struct kobject *kobj) 212{ 213 debugf1("%s()\n", __func__); 214 complete(&edac_memctrl_kobj_complete); 215} 216 217static struct kobj_type ktype_memctrl = { 218 .release = edac_memctrl_master_release, 219 .sysfs_ops = &memctrlfs_ops, 220 .default_attrs = (struct attribute **) memctrl_attr, 221}; 222 223/* Initialize the main sysfs entries for edac: 224 * /sys/devices/system/edac 225 * 226 * and children 227 * 228 * Return: 0 SUCCESS 229 * !0 FAILURE 230 */ 231static int edac_sysfs_memctrl_setup(void) 232{ 233 int err=0; 234 235 debugf1("%s()\n", __func__); 236 237 /* create the /sys/devices/system/edac directory */ 238 err = sysdev_class_register(&edac_class); 239 240 if (!err) { 241 /* Init the MC's kobject */ 242 memset(&edac_memctrl_kobj, 0, sizeof (edac_memctrl_kobj)); 243 edac_memctrl_kobj.parent = &edac_class.kset.kobj; 244 edac_memctrl_kobj.ktype = &ktype_memctrl; 245 246 /* generate sysfs "..../edac/mc" */ 247 err = kobject_set_name(&edac_memctrl_kobj,"mc"); 248 249 if (!err) { 250 /* FIXME: maybe new sysdev_create_subdir() */ 251 err = kobject_register(&edac_memctrl_kobj); 252 253 if (err) 254 debugf1("Failed to register '.../edac/mc'\n"); 255 else 256 debugf1("Registered '.../edac/mc' kobject\n"); 257 } 258 } else 259 debugf1("%s() error=%d\n", __func__, err); 260 261 return err; 262} 263 264/* 265 * MC teardown: 266 * the '..../edac/mc' kobject followed by '..../edac' itself 267 */ 268static void edac_sysfs_memctrl_teardown(void) 269{ 270 debugf0("MC: " __FILE__ ": %s()\n", __func__); 271 272 /* Unregister the MC's kobject and wait for reference count to reach 273 * 0. 274 */ 275 init_completion(&edac_memctrl_kobj_complete); 276 kobject_unregister(&edac_memctrl_kobj); 277 wait_for_completion(&edac_memctrl_kobj_complete); 278 279 /* Unregister the 'edac' object */ 280 sysdev_class_unregister(&edac_class); 281} 282 283#ifdef CONFIG_PCI 284static ssize_t edac_pci_int_show(void *ptr, char *buffer) 285{ 286 int *value = ptr; 287 return sprintf(buffer,"%d\n",*value); 288} 289 290static ssize_t edac_pci_int_store(void *ptr, const char *buffer, size_t count) 291{ 292 int *value = ptr; 293 294 if (isdigit(*buffer)) 295 *value = simple_strtoul(buffer,NULL,0); 296 297 return count; 298} 299 300struct edac_pci_dev_attribute { 301 struct attribute attr; 302 void *value; 303 ssize_t (*show)(void *,char *); 304 ssize_t (*store)(void *, const char *,size_t); 305}; 306 307/* Set of show/store abstract level functions for PCI Parity object */ 308static ssize_t edac_pci_dev_show(struct kobject *kobj, struct attribute *attr, 309 char *buffer) 310{ 311 struct edac_pci_dev_attribute *edac_pci_dev; 312 edac_pci_dev= (struct edac_pci_dev_attribute*)attr; 313 314 if (edac_pci_dev->show) 315 return edac_pci_dev->show(edac_pci_dev->value, buffer); 316 return -EIO; 317} 318 319static ssize_t edac_pci_dev_store(struct kobject *kobj, 320 struct attribute *attr, const char *buffer, size_t count) 321{ 322 struct edac_pci_dev_attribute *edac_pci_dev; 323 edac_pci_dev= (struct edac_pci_dev_attribute*)attr; 324 325 if (edac_pci_dev->show) 326 return edac_pci_dev->store(edac_pci_dev->value, buffer, count); 327 return -EIO; 328} 329 330static struct sysfs_ops edac_pci_sysfs_ops = { 331 .show = edac_pci_dev_show, 332 .store = edac_pci_dev_store 333}; 334 335#define EDAC_PCI_ATTR(_name,_mode,_show,_store) \ 336struct edac_pci_dev_attribute edac_pci_attr_##_name = { \ 337 .attr = {.name = __stringify(_name), .mode = _mode }, \ 338 .value = &_name, \ 339 .show = _show, \ 340 .store = _store, \ 341}; 342 343#define EDAC_PCI_STRING_ATTR(_name,_data,_mode,_show,_store) \ 344struct edac_pci_dev_attribute edac_pci_attr_##_name = { \ 345 .attr = {.name = __stringify(_name), .mode = _mode }, \ 346 .value = _data, \ 347 .show = _show, \ 348 .store = _store, \ 349}; 350 351/* PCI Parity control files */ 352EDAC_PCI_ATTR(check_pci_parity, S_IRUGO|S_IWUSR, edac_pci_int_show, 353 edac_pci_int_store); 354EDAC_PCI_ATTR(panic_on_pci_parity, S_IRUGO|S_IWUSR, edac_pci_int_show, 355 edac_pci_int_store); 356EDAC_PCI_ATTR(pci_parity_count, S_IRUGO, edac_pci_int_show, NULL); 357 358/* Base Attributes of the memory ECC object */ 359static struct edac_pci_dev_attribute *edac_pci_attr[] = { 360 &edac_pci_attr_check_pci_parity, 361 &edac_pci_attr_panic_on_pci_parity, 362 &edac_pci_attr_pci_parity_count, 363 NULL, 364}; 365 366/* No memory to release */ 367static void edac_pci_release(struct kobject *kobj) 368{ 369 debugf1("%s()\n", __func__); 370 complete(&edac_pci_kobj_complete); 371} 372 373static struct kobj_type ktype_edac_pci = { 374 .release = edac_pci_release, 375 .sysfs_ops = &edac_pci_sysfs_ops, 376 .default_attrs = (struct attribute **) edac_pci_attr, 377}; 378 379/** 380 * edac_sysfs_pci_setup() 381 * 382 */ 383static int edac_sysfs_pci_setup(void) 384{ 385 int err; 386 387 debugf1("%s()\n", __func__); 388 389 memset(&edac_pci_kobj, 0, sizeof(edac_pci_kobj)); 390 edac_pci_kobj.parent = &edac_class.kset.kobj; 391 edac_pci_kobj.ktype = &ktype_edac_pci; 392 err = kobject_set_name(&edac_pci_kobj, "pci"); 393 394 if (!err) { 395 /* Instanstiate the csrow object */ 396 /* FIXME: maybe new sysdev_create_subdir() */ 397 err = kobject_register(&edac_pci_kobj); 398 399 if (err) 400 debugf1("Failed to register '.../edac/pci'\n"); 401 else 402 debugf1("Registered '.../edac/pci' kobject\n"); 403 } 404 405 return err; 406} 407 408static void edac_sysfs_pci_teardown(void) 409{ 410 debugf0("%s()\n", __func__); 411 init_completion(&edac_pci_kobj_complete); 412 kobject_unregister(&edac_pci_kobj); 413 wait_for_completion(&edac_pci_kobj_complete); 414} 415 416 417static u16 get_pci_parity_status(struct pci_dev *dev, int secondary) 418{ 419 int where; 420 u16 status; 421 422 where = secondary ? PCI_SEC_STATUS : PCI_STATUS; 423 pci_read_config_word(dev, where, &status); 424 425 /* If we get back 0xFFFF then we must suspect that the card has been 426 * pulled but the Linux PCI layer has not yet finished cleaning up. 427 * We don't want to report on such devices 428 */ 429 430 if (status == 0xFFFF) { 431 u32 sanity; 432 433 pci_read_config_dword(dev, 0, &sanity); 434 435 if (sanity == 0xFFFFFFFF) 436 return 0; 437 } 438 439 status &= PCI_STATUS_DETECTED_PARITY | PCI_STATUS_SIG_SYSTEM_ERROR | 440 PCI_STATUS_PARITY; 441 442 if (status) 443 /* reset only the bits we are interested in */ 444 pci_write_config_word(dev, where, status); 445 446 return status; 447} 448 449typedef void (*pci_parity_check_fn_t) (struct pci_dev *dev); 450 451/* Clear any PCI parity errors logged by this device. */ 452static void edac_pci_dev_parity_clear(struct pci_dev *dev) 453{ 454 u8 header_type; 455 456 get_pci_parity_status(dev, 0); 457 458 /* read the device TYPE, looking for bridges */ 459 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type); 460 461 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) 462 get_pci_parity_status(dev, 1); 463} 464 465/* 466 * PCI Parity polling 467 * 468 */ 469static void edac_pci_dev_parity_test(struct pci_dev *dev) 470{ 471 u16 status; 472 u8 header_type; 473 474 /* read the STATUS register on this device 475 */ 476 status = get_pci_parity_status(dev, 0); 477 478 debugf2("PCI STATUS= 0x%04x %s\n", status, dev->dev.bus_id ); 479 480 /* check the status reg for errors */ 481 if (status) { 482 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) 483 edac_printk(KERN_CRIT, EDAC_PCI, 484 "Signaled System Error on %s\n", 485 pci_name(dev)); 486 487 if (status & (PCI_STATUS_PARITY)) { 488 edac_printk(KERN_CRIT, EDAC_PCI, 489 "Master Data Parity Error on %s\n", 490 pci_name(dev)); 491 492 atomic_inc(&pci_parity_count); 493 } 494 495 if (status & (PCI_STATUS_DETECTED_PARITY)) { 496 edac_printk(KERN_CRIT, EDAC_PCI, 497 "Detected Parity Error on %s\n", 498 pci_name(dev)); 499 500 atomic_inc(&pci_parity_count); 501 } 502 } 503 504 /* read the device TYPE, looking for bridges */ 505 pci_read_config_byte(dev, PCI_HEADER_TYPE, &header_type); 506 507 debugf2("PCI HEADER TYPE= 0x%02x %s\n", header_type, dev->dev.bus_id ); 508 509 if ((header_type & 0x7F) == PCI_HEADER_TYPE_BRIDGE) { 510 /* On bridges, need to examine secondary status register */ 511 status = get_pci_parity_status(dev, 1); 512 513 debugf2("PCI SEC_STATUS= 0x%04x %s\n", 514 status, dev->dev.bus_id ); 515 516 /* check the secondary status reg for errors */ 517 if (status) { 518 if (status & (PCI_STATUS_SIG_SYSTEM_ERROR)) 519 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge " 520 "Signaled System Error on %s\n", 521 pci_name(dev)); 522 523 if (status & (PCI_STATUS_PARITY)) { 524 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge " 525 "Master Data Parity Error on " 526 "%s\n", pci_name(dev)); 527 528 atomic_inc(&pci_parity_count); 529 } 530 531 if (status & (PCI_STATUS_DETECTED_PARITY)) { 532 edac_printk(KERN_CRIT, EDAC_PCI, "Bridge " 533 "Detected Parity Error on %s\n", 534 pci_name(dev)); 535 536 atomic_inc(&pci_parity_count); 537 } 538 } 539 } 540} 541 542/* 543 * pci_dev parity list iterator 544 * Scan the PCI device list for one iteration, looking for SERRORs 545 * Master Parity ERRORS or Parity ERRORs on primary or secondary devices 546 */ 547static inline void edac_pci_dev_parity_iterator(pci_parity_check_fn_t fn) 548{ 549 struct pci_dev *dev = NULL; 550 551 /* request for kernel access to the next PCI device, if any, 552 * and while we are looking at it have its reference count 553 * bumped until we are done with it 554 */ 555 while((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) { 556 fn(dev); 557 } 558} 559 560static void do_pci_parity_check(void) 561{ 562 unsigned long flags; 563 int before_count; 564 565 debugf3("%s()\n", __func__); 566 567 if (!check_pci_parity) 568 return; 569 570 before_count = atomic_read(&pci_parity_count); 571 572 /* scan all PCI devices looking for a Parity Error on devices and 573 * bridges 574 */ 575 local_irq_save(flags); 576 edac_pci_dev_parity_iterator(edac_pci_dev_parity_test); 577 local_irq_restore(flags); 578 579 /* Only if operator has selected panic on PCI Error */ 580 if (panic_on_pci_parity) { 581 /* If the count is different 'after' from 'before' */ 582 if (before_count != atomic_read(&pci_parity_count)) 583 panic("EDAC: PCI Parity Error"); 584 } 585} 586 587static inline void clear_pci_parity_errors(void) 588{ 589 /* Clear any PCI bus parity errors that devices initially have logged 590 * in their registers. 591 */ 592 edac_pci_dev_parity_iterator(edac_pci_dev_parity_clear); 593} 594 595#else /* CONFIG_PCI */ 596 597/* pre-process these away */ 598#define do_pci_parity_check() 599#define clear_pci_parity_errors() 600#define edac_sysfs_pci_teardown() 601#define edac_sysfs_pci_setup() (0) 602 603#endif /* CONFIG_PCI */ 604 605/* EDAC sysfs CSROW data structures and methods 606 */ 607 608/* Set of more default csrow<id> attribute show/store functions */ 609static ssize_t csrow_ue_count_show(struct csrow_info *csrow, char *data, int private) 610{ 611 return sprintf(data,"%u\n", csrow->ue_count); 612} 613 614static ssize_t csrow_ce_count_show(struct csrow_info *csrow, char *data, int private) 615{ 616 return sprintf(data,"%u\n", csrow->ce_count); 617} 618 619static ssize_t csrow_size_show(struct csrow_info *csrow, char *data, int private) 620{ 621 return sprintf(data,"%u\n", PAGES_TO_MiB(csrow->nr_pages)); 622} 623 624static ssize_t csrow_mem_type_show(struct csrow_info *csrow, char *data, int private) 625{ 626 return sprintf(data,"%s\n", mem_types[csrow->mtype]); 627} 628 629static ssize_t csrow_dev_type_show(struct csrow_info *csrow, char *data, int private) 630{ 631 return sprintf(data,"%s\n", dev_types[csrow->dtype]); 632} 633 634static ssize_t csrow_edac_mode_show(struct csrow_info *csrow, char *data, int private) 635{ 636 return sprintf(data,"%s\n", edac_caps[csrow->edac_mode]); 637} 638 639/* show/store functions for DIMM Label attributes */ 640static ssize_t channel_dimm_label_show(struct csrow_info *csrow, 641 char *data, int channel) 642{ 643 return snprintf(data, EDAC_MC_LABEL_LEN,"%s", 644 csrow->channels[channel].label); 645} 646 647static ssize_t channel_dimm_label_store(struct csrow_info *csrow, 648 const char *data, 649 size_t count, 650 int channel) 651{ 652 ssize_t max_size = 0; 653 654 max_size = min((ssize_t)count,(ssize_t)EDAC_MC_LABEL_LEN-1); 655 strncpy(csrow->channels[channel].label, data, max_size); 656 csrow->channels[channel].label[max_size] = '\0'; 657 658 return max_size; 659} 660 661/* show function for dynamic chX_ce_count attribute */ 662static ssize_t channel_ce_count_show(struct csrow_info *csrow, 663 char *data, 664 int channel) 665{ 666 return sprintf(data, "%u\n", csrow->channels[channel].ce_count); 667} 668 669/* csrow specific attribute structure */ 670struct csrowdev_attribute { 671 struct attribute attr; 672 ssize_t (*show)(struct csrow_info *,char *,int); 673 ssize_t (*store)(struct csrow_info *, const char *,size_t,int); 674 int private; 675}; 676 677#define to_csrow(k) container_of(k, struct csrow_info, kobj) 678#define to_csrowdev_attr(a) container_of(a, struct csrowdev_attribute, attr) 679 680/* Set of show/store higher level functions for default csrow attributes */ 681static ssize_t csrowdev_show(struct kobject *kobj, 682 struct attribute *attr, 683 char *buffer) 684{ 685 struct csrow_info *csrow = to_csrow(kobj); 686 struct csrowdev_attribute *csrowdev_attr = to_csrowdev_attr(attr); 687 688 if (csrowdev_attr->show) 689 return csrowdev_attr->show(csrow, 690 buffer, 691 csrowdev_attr->private); 692 return -EIO; 693} 694 695static ssize_t csrowdev_store(struct kobject *kobj, struct attribute *attr, 696 const char *buffer, size_t count) 697{ 698 struct csrow_info *csrow = to_csrow(kobj); 699 struct csrowdev_attribute * csrowdev_attr = to_csrowdev_attr(attr); 700 701 if (csrowdev_attr->store) 702 return csrowdev_attr->store(csrow, 703 buffer, 704 count, 705 csrowdev_attr->private); 706 return -EIO; 707} 708 709static struct sysfs_ops csrowfs_ops = { 710 .show = csrowdev_show, 711 .store = csrowdev_store 712}; 713 714#define CSROWDEV_ATTR(_name,_mode,_show,_store,_private) \ 715struct csrowdev_attribute attr_##_name = { \ 716 .attr = {.name = __stringify(_name), .mode = _mode }, \ 717 .show = _show, \ 718 .store = _store, \ 719 .private = _private, \ 720}; 721 722/* default cwrow<id>/attribute files */ 723CSROWDEV_ATTR(size_mb,S_IRUGO,csrow_size_show,NULL,0); 724CSROWDEV_ATTR(dev_type,S_IRUGO,csrow_dev_type_show,NULL,0); 725CSROWDEV_ATTR(mem_type,S_IRUGO,csrow_mem_type_show,NULL,0); 726CSROWDEV_ATTR(edac_mode,S_IRUGO,csrow_edac_mode_show,NULL,0); 727CSROWDEV_ATTR(ue_count,S_IRUGO,csrow_ue_count_show,NULL,0); 728CSROWDEV_ATTR(ce_count,S_IRUGO,csrow_ce_count_show,NULL,0); 729 730/* default attributes of the CSROW<id> object */ 731static struct csrowdev_attribute *default_csrow_attr[] = { 732 &attr_dev_type, 733 &attr_mem_type, 734 &attr_edac_mode, 735 &attr_size_mb, 736 &attr_ue_count, 737 &attr_ce_count, 738 NULL, 739}; 740 741 742/* possible dynamic channel DIMM Label attribute files */ 743CSROWDEV_ATTR(ch0_dimm_label,S_IRUGO|S_IWUSR, 744 channel_dimm_label_show, 745 channel_dimm_label_store, 746 0 ); 747CSROWDEV_ATTR(ch1_dimm_label,S_IRUGO|S_IWUSR, 748 channel_dimm_label_show, 749 channel_dimm_label_store, 750 1 ); 751CSROWDEV_ATTR(ch2_dimm_label,S_IRUGO|S_IWUSR, 752 channel_dimm_label_show, 753 channel_dimm_label_store, 754 2 ); 755CSROWDEV_ATTR(ch3_dimm_label,S_IRUGO|S_IWUSR, 756 channel_dimm_label_show, 757 channel_dimm_label_store, 758 3 ); 759CSROWDEV_ATTR(ch4_dimm_label,S_IRUGO|S_IWUSR, 760 channel_dimm_label_show, 761 channel_dimm_label_store, 762 4 ); 763CSROWDEV_ATTR(ch5_dimm_label,S_IRUGO|S_IWUSR, 764 channel_dimm_label_show, 765 channel_dimm_label_store, 766 5 ); 767 768/* Total possible dynamic DIMM Label attribute file table */ 769static struct csrowdev_attribute *dynamic_csrow_dimm_attr[] = { 770 &attr_ch0_dimm_label, 771 &attr_ch1_dimm_label, 772 &attr_ch2_dimm_label, 773 &attr_ch3_dimm_label, 774 &attr_ch4_dimm_label, 775 &attr_ch5_dimm_label 776}; 777 778/* possible dynamic channel ce_count attribute files */ 779CSROWDEV_ATTR(ch0_ce_count,S_IRUGO|S_IWUSR, 780 channel_ce_count_show, 781 NULL, 782 0 ); 783CSROWDEV_ATTR(ch1_ce_count,S_IRUGO|S_IWUSR, 784 channel_ce_count_show, 785 NULL, 786 1 ); 787CSROWDEV_ATTR(ch2_ce_count,S_IRUGO|S_IWUSR, 788 channel_ce_count_show, 789 NULL, 790 2 ); 791CSROWDEV_ATTR(ch3_ce_count,S_IRUGO|S_IWUSR, 792 channel_ce_count_show, 793 NULL, 794 3 ); 795CSROWDEV_ATTR(ch4_ce_count,S_IRUGO|S_IWUSR, 796 channel_ce_count_show, 797 NULL, 798 4 ); 799CSROWDEV_ATTR(ch5_ce_count,S_IRUGO|S_IWUSR, 800 channel_ce_count_show, 801 NULL, 802 5 ); 803 804/* Total possible dynamic ce_count attribute file table */ 805static struct csrowdev_attribute *dynamic_csrow_ce_count_attr[] = { 806 &attr_ch0_ce_count, 807 &attr_ch1_ce_count, 808 &attr_ch2_ce_count, 809 &attr_ch3_ce_count, 810 &attr_ch4_ce_count, 811 &attr_ch5_ce_count 812}; 813 814 815#define EDAC_NR_CHANNELS 6 816 817/* Create dynamic CHANNEL files, indexed by 'chan', under specifed CSROW */ 818static int edac_create_channel_files(struct kobject *kobj, int chan) 819{ 820 int err=-ENODEV; 821 822 if (chan >= EDAC_NR_CHANNELS) 823 return err; 824 825 /* create the DIMM label attribute file */ 826 err = sysfs_create_file(kobj, 827 (struct attribute *) dynamic_csrow_dimm_attr[chan]); 828 829 if (!err) { 830 /* create the CE Count attribute file */ 831 err = sysfs_create_file(kobj, 832 (struct attribute *) dynamic_csrow_ce_count_attr[chan]); 833 } else { 834 debugf1("%s() dimm labels and ce_count files created", __func__); 835 } 836 837 return err; 838} 839 840/* No memory to release for this kobj */ 841static void edac_csrow_instance_release(struct kobject *kobj) 842{ 843 struct csrow_info *cs; 844 845 cs = container_of(kobj, struct csrow_info, kobj); 846 complete(&cs->kobj_complete); 847} 848 849/* the kobj_type instance for a CSROW */ 850static struct kobj_type ktype_csrow = { 851 .release = edac_csrow_instance_release, 852 .sysfs_ops = &csrowfs_ops, 853 .default_attrs = (struct attribute **) default_csrow_attr, 854}; 855 856/* Create a CSROW object under specifed edac_mc_device */ 857static int edac_create_csrow_object( 858 struct kobject *edac_mci_kobj, 859 struct csrow_info *csrow, 860 int index) 861{ 862 int err = 0; 863 int chan; 864 865 memset(&csrow->kobj, 0, sizeof(csrow->kobj)); 866 867 /* generate ..../edac/mc/mc<id>/csrow<index> */ 868 869 csrow->kobj.parent = edac_mci_kobj; 870 csrow->kobj.ktype = &ktype_csrow; 871 872 /* name this instance of csrow<id> */ 873 err = kobject_set_name(&csrow->kobj,"csrow%d",index); 874 if (err) 875 goto error_exit; 876 877 /* Instanstiate the csrow object */ 878 err = kobject_register(&csrow->kobj); 879 if (!err) { 880 /* Create the dyanmic attribute files on this csrow, 881 * namely, the DIMM labels and the channel ce_count 882 */ 883 for (chan = 0; chan < csrow->nr_channels; chan++) { 884 err = edac_create_channel_files(&csrow->kobj,chan); 885 if (err) 886 break; 887 } 888 } 889 890error_exit: 891 return err; 892} 893 894/* default sysfs methods and data structures for the main MCI kobject */ 895 896static ssize_t mci_reset_counters_store(struct mem_ctl_info *mci, 897 const char *data, size_t count) 898{ 899 int row, chan; 900 901 mci->ue_noinfo_count = 0; 902 mci->ce_noinfo_count = 0; 903 mci->ue_count = 0; 904 mci->ce_count = 0; 905 906 for (row = 0; row < mci->nr_csrows; row++) { 907 struct csrow_info *ri = &mci->csrows[row]; 908 909 ri->ue_count = 0; 910 ri->ce_count = 0; 911 912 for (chan = 0; chan < ri->nr_channels; chan++) 913 ri->channels[chan].ce_count = 0; 914 } 915 916 mci->start_time = jiffies; 917 return count; 918} 919 920/* default attribute files for the MCI object */ 921static ssize_t mci_ue_count_show(struct mem_ctl_info *mci, char *data) 922{ 923 return sprintf(data,"%d\n", mci->ue_count); 924} 925 926static ssize_t mci_ce_count_show(struct mem_ctl_info *mci, char *data) 927{ 928 return sprintf(data,"%d\n", mci->ce_count); 929} 930 931static ssize_t mci_ce_noinfo_show(struct mem_ctl_info *mci, char *data) 932{ 933 return sprintf(data,"%d\n", mci->ce_noinfo_count); 934} 935 936static ssize_t mci_ue_noinfo_show(struct mem_ctl_info *mci, char *data) 937{ 938 return sprintf(data,"%d\n", mci->ue_noinfo_count); 939} 940 941static ssize_t mci_seconds_show(struct mem_ctl_info *mci, char *data) 942{ 943 return sprintf(data,"%ld\n", (jiffies - mci->start_time) / HZ); 944} 945 946static ssize_t mci_ctl_name_show(struct mem_ctl_info *mci, char *data) 947{ 948 return sprintf(data,"%s\n", mci->ctl_name); 949} 950 951static ssize_t mci_size_mb_show(struct mem_ctl_info *mci, char *data) 952{ 953 int total_pages, csrow_idx; 954 955 for (total_pages = csrow_idx = 0; csrow_idx < mci->nr_csrows; 956 csrow_idx++) { 957 struct csrow_info *csrow = &mci->csrows[csrow_idx]; 958 959 if (!csrow->nr_pages) 960 continue; 961 962 total_pages += csrow->nr_pages; 963 } 964 965 return sprintf(data,"%u\n", PAGES_TO_MiB(total_pages)); 966} 967 968struct mcidev_attribute { 969 struct attribute attr; 970 ssize_t (*show)(struct mem_ctl_info *,char *); 971 ssize_t (*store)(struct mem_ctl_info *, const char *,size_t); 972}; 973 974#define to_mci(k) container_of(k, struct mem_ctl_info, edac_mci_kobj) 975#define to_mcidev_attr(a) container_of(a, struct mcidev_attribute, attr) 976 977/* MCI show/store functions for top most object */ 978static ssize_t mcidev_show(struct kobject *kobj, struct attribute *attr, 979 char *buffer) 980{ 981 struct mem_ctl_info *mem_ctl_info = to_mci(kobj); 982 struct mcidev_attribute * mcidev_attr = to_mcidev_attr(attr); 983 984 if (mcidev_attr->show) 985 return mcidev_attr->show(mem_ctl_info, buffer); 986 987 return -EIO; 988} 989 990static ssize_t mcidev_store(struct kobject *kobj, struct attribute *attr, 991 const char *buffer, size_t count) 992{ 993 struct mem_ctl_info *mem_ctl_info = to_mci(kobj); 994 struct mcidev_attribute * mcidev_attr = to_mcidev_attr(attr); 995 996 if (mcidev_attr->store) 997 return mcidev_attr->store(mem_ctl_info, buffer, count); 998 999 return -EIO; 1000} 1001 1002static struct sysfs_ops mci_ops = { 1003 .show = mcidev_show, 1004 .store = mcidev_store 1005}; 1006 1007#define MCIDEV_ATTR(_name,_mode,_show,_store) \ 1008struct mcidev_attribute mci_attr_##_name = { \ 1009 .attr = {.name = __stringify(_name), .mode = _mode }, \ 1010 .show = _show, \ 1011 .store = _store, \ 1012}; 1013 1014/* default Control file */ 1015MCIDEV_ATTR(reset_counters,S_IWUSR,NULL,mci_reset_counters_store); 1016 1017/* default Attribute files */ 1018MCIDEV_ATTR(mc_name,S_IRUGO,mci_ctl_name_show,NULL); 1019MCIDEV_ATTR(size_mb,S_IRUGO,mci_size_mb_show,NULL); 1020MCIDEV_ATTR(seconds_since_reset,S_IRUGO,mci_seconds_show,NULL); 1021MCIDEV_ATTR(ue_noinfo_count,S_IRUGO,mci_ue_noinfo_show,NULL); 1022MCIDEV_ATTR(ce_noinfo_count,S_IRUGO,mci_ce_noinfo_show,NULL); 1023MCIDEV_ATTR(ue_count,S_IRUGO,mci_ue_count_show,NULL); 1024MCIDEV_ATTR(ce_count,S_IRUGO,mci_ce_count_show,NULL); 1025 1026static struct mcidev_attribute *mci_attr[] = { 1027 &mci_attr_reset_counters, 1028 &mci_attr_mc_name, 1029 &mci_attr_size_mb, 1030 &mci_attr_seconds_since_reset, 1031 &mci_attr_ue_noinfo_count, 1032 &mci_attr_ce_noinfo_count, 1033 &mci_attr_ue_count, 1034 &mci_attr_ce_count, 1035 NULL 1036}; 1037 1038/* 1039 * Release of a MC controlling instance 1040 */ 1041static void edac_mci_instance_release(struct kobject *kobj) 1042{ 1043 struct mem_ctl_info *mci; 1044 1045 mci = to_mci(kobj); 1046 debugf0("%s() idx=%d\n", __func__, mci->mc_idx); 1047 complete(&mci->kobj_complete); 1048} 1049 1050static struct kobj_type ktype_mci = { 1051 .release = edac_mci_instance_release, 1052 .sysfs_ops = &mci_ops, 1053 .default_attrs = (struct attribute **) mci_attr, 1054}; 1055 1056 1057#define EDAC_DEVICE_SYMLINK "device" 1058 1059/* 1060 * Create a new Memory Controller kobject instance, 1061 * mc<id> under the 'mc' directory 1062 * 1063 * Return: 1064 * 0 Success 1065 * !0 Failure 1066 */ 1067static int edac_create_sysfs_mci_device(struct mem_ctl_info *mci) 1068{ 1069 int i; 1070 int err; 1071 struct csrow_info *csrow; 1072 struct kobject *edac_mci_kobj=&mci->edac_mci_kobj; 1073 1074 debugf0("%s() idx=%d\n", __func__, mci->mc_idx); 1075 memset(edac_mci_kobj, 0, sizeof(*edac_mci_kobj)); 1076 1077 /* set the name of the mc<id> object */ 1078 err = kobject_set_name(edac_mci_kobj,"mc%d",mci->mc_idx); 1079 if (err) 1080 return err; 1081 1082 /* link to our parent the '..../edac/mc' object */ 1083 edac_mci_kobj->parent = &edac_memctrl_kobj; 1084 edac_mci_kobj->ktype = &ktype_mci; 1085 1086 /* register the mc<id> kobject */ 1087 err = kobject_register(edac_mci_kobj); 1088 if (err) 1089 return err; 1090 1091 /* create a symlink for the device */ 1092 err = sysfs_create_link(edac_mci_kobj, &mci->dev->kobj, 1093 EDAC_DEVICE_SYMLINK); 1094 if (err) 1095 goto fail0; 1096 1097 /* Make directories for each CSROW object 1098 * under the mc<id> kobject 1099 */ 1100 for (i = 0; i < mci->nr_csrows; i++) { 1101 csrow = &mci->csrows[i]; 1102 1103 /* Only expose populated CSROWs */ 1104 if (csrow->nr_pages > 0) { 1105 err = edac_create_csrow_object(edac_mci_kobj,csrow,i); 1106 if (err) 1107 goto fail1; 1108 } 1109 } 1110 1111 return 0; 1112 1113 /* CSROW error: backout what has already been registered, */ 1114fail1: 1115 for ( i--; i >= 0; i--) { 1116 if (csrow->nr_pages > 0) { 1117 init_completion(&csrow->kobj_complete); 1118 kobject_unregister(&mci->csrows[i].kobj); 1119 wait_for_completion(&csrow->kobj_complete); 1120 } 1121 } 1122 1123fail0: 1124 init_completion(&mci->kobj_complete); 1125 kobject_unregister(edac_mci_kobj); 1126 wait_for_completion(&mci->kobj_complete); 1127 return err; 1128} 1129 1130/* 1131 * remove a Memory Controller instance 1132 */ 1133static void edac_remove_sysfs_mci_device(struct mem_ctl_info *mci) 1134{ 1135 int i; 1136 1137 debugf0("%s()\n", __func__); 1138 1139 /* remove all csrow kobjects */ 1140 for (i = 0; i < mci->nr_csrows; i++) { 1141 if (mci->csrows[i].nr_pages > 0) { 1142 init_completion(&mci->csrows[i].kobj_complete); 1143 kobject_unregister(&mci->csrows[i].kobj); 1144 wait_for_completion(&mci->csrows[i].kobj_complete); 1145 } 1146 } 1147 1148 sysfs_remove_link(&mci->edac_mci_kobj, EDAC_DEVICE_SYMLINK); 1149 init_completion(&mci->kobj_complete); 1150 kobject_unregister(&mci->edac_mci_kobj); 1151 wait_for_completion(&mci->kobj_complete); 1152} 1153 1154/* END OF sysfs data and methods */ 1155 1156#ifdef CONFIG_EDAC_DEBUG 1157 1158void edac_mc_dump_channel(struct channel_info *chan) 1159{ 1160 debugf4("\tchannel = %p\n", chan); 1161 debugf4("\tchannel->chan_idx = %d\n", chan->chan_idx); 1162 debugf4("\tchannel->ce_count = %d\n", chan->ce_count); 1163 debugf4("\tchannel->label = '%s'\n", chan->label); 1164 debugf4("\tchannel->csrow = %p\n\n", chan->csrow); 1165} 1166EXPORT_SYMBOL_GPL(edac_mc_dump_channel); 1167 1168void edac_mc_dump_csrow(struct csrow_info *csrow) 1169{ 1170 debugf4("\tcsrow = %p\n", csrow); 1171 debugf4("\tcsrow->csrow_idx = %d\n", csrow->csrow_idx); 1172 debugf4("\tcsrow->first_page = 0x%lx\n", 1173 csrow->first_page); 1174 debugf4("\tcsrow->last_page = 0x%lx\n", csrow->last_page); 1175 debugf4("\tcsrow->page_mask = 0x%lx\n", csrow->page_mask); 1176 debugf4("\tcsrow->nr_pages = 0x%x\n", csrow->nr_pages); 1177 debugf4("\tcsrow->nr_channels = %d\n", 1178 csrow->nr_channels); 1179 debugf4("\tcsrow->channels = %p\n", csrow->channels); 1180 debugf4("\tcsrow->mci = %p\n\n", csrow->mci); 1181} 1182EXPORT_SYMBOL_GPL(edac_mc_dump_csrow); 1183 1184void edac_mc_dump_mci(struct mem_ctl_info *mci) 1185{ 1186 debugf3("\tmci = %p\n", mci); 1187 debugf3("\tmci->mtype_cap = %lx\n", mci->mtype_cap); 1188 debugf3("\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap); 1189 debugf3("\tmci->edac_cap = %lx\n", mci->edac_cap); 1190 debugf4("\tmci->edac_check = %p\n", mci->edac_check); 1191 debugf3("\tmci->nr_csrows = %d, csrows = %p\n", 1192 mci->nr_csrows, mci->csrows); 1193 debugf3("\tdev = %p\n", mci->dev); 1194 debugf3("\tmod_name:ctl_name = %s:%s\n", 1195 mci->mod_name, mci->ctl_name); 1196 debugf3("\tpvt_info = %p\n\n", mci->pvt_info); 1197} 1198EXPORT_SYMBOL_GPL(edac_mc_dump_mci); 1199 1200#endif /* CONFIG_EDAC_DEBUG */ 1201 1202/* 'ptr' points to a possibly unaligned item X such that sizeof(X) is 'size'. 1203 * Adjust 'ptr' so that its alignment is at least as stringent as what the 1204 * compiler would provide for X and return the aligned result. 1205 * 1206 * If 'size' is a constant, the compiler will optimize this whole function 1207 * down to either a no-op or the addition of a constant to the value of 'ptr'. 1208 */ 1209static inline char * align_ptr(void *ptr, unsigned size) 1210{ 1211 unsigned align, r; 1212 1213 /* Here we assume that the alignment of a "long long" is the most 1214 * stringent alignment that the compiler will ever provide by default. 1215 * As far as I know, this is a reasonable assumption. 1216 */ 1217 if (size > sizeof(long)) 1218 align = sizeof(long long); 1219 else if (size > sizeof(int)) 1220 align = sizeof(long); 1221 else if (size > sizeof(short)) 1222 align = sizeof(int); 1223 else if (size > sizeof(char)) 1224 align = sizeof(short); 1225 else 1226 return (char *) ptr; 1227 1228 r = size % align; 1229 1230 if (r == 0) 1231 return (char *) ptr; 1232 1233 return (char *) (((unsigned long) ptr) + align - r); 1234} 1235 1236/** 1237 * edac_mc_alloc: Allocate a struct mem_ctl_info structure 1238 * @size_pvt: size of private storage needed 1239 * @nr_csrows: Number of CWROWS needed for this MC 1240 * @nr_chans: Number of channels for the MC 1241 * 1242 * Everything is kmalloc'ed as one big chunk - more efficient. 1243 * Only can be used if all structures have the same lifetime - otherwise 1244 * you have to allocate and initialize your own structures. 1245 * 1246 * Use edac_mc_free() to free mc structures allocated by this function. 1247 * 1248 * Returns: 1249 * NULL allocation failed 1250 * struct mem_ctl_info pointer 1251 */ 1252struct mem_ctl_info *edac_mc_alloc(unsigned sz_pvt, unsigned nr_csrows, 1253 unsigned nr_chans) 1254{ 1255 struct mem_ctl_info *mci; 1256 struct csrow_info *csi, *csrow; 1257 struct channel_info *chi, *chp, *chan; 1258 void *pvt; 1259 unsigned size; 1260 int row, chn; 1261 1262 /* Figure out the offsets of the various items from the start of an mc 1263 * structure. We want the alignment of each item to be at least as 1264 * stringent as what the compiler would provide if we could simply 1265 * hardcode everything into a single struct. 1266 */ 1267 mci = (struct mem_ctl_info *) 0; 1268 csi = (struct csrow_info *)align_ptr(&mci[1], sizeof(*csi)); 1269 chi = (struct channel_info *) 1270 align_ptr(&csi[nr_csrows], sizeof(*chi)); 1271 pvt = align_ptr(&chi[nr_chans * nr_csrows], sz_pvt); 1272 size = ((unsigned long) pvt) + sz_pvt; 1273 1274 if ((mci = kmalloc(size, GFP_KERNEL)) == NULL) 1275 return NULL; 1276 1277 /* Adjust pointers so they point within the memory we just allocated 1278 * rather than an imaginary chunk of memory located at address 0. 1279 */ 1280 csi = (struct csrow_info *) (((char *) mci) + ((unsigned long) csi)); 1281 chi = (struct channel_info *) (((char *) mci) + ((unsigned long) chi)); 1282 pvt = sz_pvt ? (((char *) mci) + ((unsigned long) pvt)) : NULL; 1283 1284 memset(mci, 0, size); /* clear all fields */ 1285 mci->csrows = csi; 1286 mci->pvt_info = pvt; 1287 mci->nr_csrows = nr_csrows; 1288 1289 for (row = 0; row < nr_csrows; row++) { 1290 csrow = &csi[row]; 1291 csrow->csrow_idx = row; 1292 csrow->mci = mci; 1293 csrow->nr_channels = nr_chans; 1294 chp = &chi[row * nr_chans]; 1295 csrow->channels = chp; 1296 1297 for (chn = 0; chn < nr_chans; chn++) { 1298 chan = &chp[chn]; 1299 chan->chan_idx = chn; 1300 chan->csrow = csrow; 1301 } 1302 } 1303 1304 return mci; 1305} 1306EXPORT_SYMBOL_GPL(edac_mc_alloc); 1307 1308/** 1309 * edac_mc_free: Free a previously allocated 'mci' structure 1310 * @mci: pointer to a struct mem_ctl_info structure 1311 */ 1312void edac_mc_free(struct mem_ctl_info *mci) 1313{ 1314 kfree(mci); 1315} 1316EXPORT_SYMBOL_GPL(edac_mc_free); 1317 1318static struct mem_ctl_info *find_mci_by_dev(struct device *dev) 1319{ 1320 struct mem_ctl_info *mci; 1321 struct list_head *item; 1322 1323 debugf3("%s()\n", __func__); 1324 1325 list_for_each(item, &mc_devices) { 1326 mci = list_entry(item, struct mem_ctl_info, link); 1327 1328 if (mci->dev == dev) 1329 return mci; 1330 } 1331 1332 return NULL; 1333} 1334 1335/* Return 0 on success, 1 on failure. 1336 * Before calling this function, caller must 1337 * assign a unique value to mci->mc_idx. 1338 */ 1339static int add_mc_to_global_list (struct mem_ctl_info *mci) 1340{ 1341 struct list_head *item, *insert_before; 1342 struct mem_ctl_info *p; 1343 1344 insert_before = &mc_devices; 1345 1346 if (unlikely((p = find_mci_by_dev(mci->dev)) != NULL)) 1347 goto fail0; 1348 1349 list_for_each(item, &mc_devices) { 1350 p = list_entry(item, struct mem_ctl_info, link); 1351 1352 if (p->mc_idx >= mci->mc_idx) { 1353 if (unlikely(p->mc_idx == mci->mc_idx)) 1354 goto fail1; 1355 1356 insert_before = item; 1357 break; 1358 } 1359 } 1360 1361 list_add_tail_rcu(&mci->link, insert_before); 1362 return 0; 1363 1364fail0: 1365 edac_printk(KERN_WARNING, EDAC_MC, 1366 "%s (%s) %s %s already assigned %d\n", p->dev->bus_id, 1367 dev_name(p->dev), p->mod_name, p->ctl_name, p->mc_idx); 1368 return 1; 1369 1370fail1: 1371 edac_printk(KERN_WARNING, EDAC_MC, 1372 "bug in low-level driver: attempt to assign\n" 1373 " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__); 1374 return 1; 1375} 1376 1377static void complete_mc_list_del(struct rcu_head *head) 1378{ 1379 struct mem_ctl_info *mci; 1380 1381 mci = container_of(head, struct mem_ctl_info, rcu); 1382 INIT_LIST_HEAD(&mci->link); 1383 complete(&mci->complete); 1384} 1385 1386static void del_mc_from_global_list(struct mem_ctl_info *mci) 1387{ 1388 list_del_rcu(&mci->link); 1389 init_completion(&mci->complete); 1390 call_rcu(&mci->rcu, complete_mc_list_del); 1391 wait_for_completion(&mci->complete); 1392} 1393 1394/** 1395 * edac_mc_add_mc: Insert the 'mci' structure into the mci global list and 1396 * create sysfs entries associated with mci structure 1397 * @mci: pointer to the mci structure to be added to the list 1398 * @mc_idx: A unique numeric identifier to be assigned to the 'mci' structure. 1399 * 1400 * Return: 1401 * 0 Success 1402 * !0 Failure 1403 */ 1404 1405/* FIXME - should a warning be printed if no error detection? correction? */ 1406int edac_mc_add_mc(struct mem_ctl_info *mci, int mc_idx) 1407{ 1408 debugf0("%s()\n", __func__); 1409 mci->mc_idx = mc_idx; 1410#ifdef CONFIG_EDAC_DEBUG 1411 if (edac_debug_level >= 3) 1412 edac_mc_dump_mci(mci); 1413 1414 if (edac_debug_level >= 4) { 1415 int i; 1416 1417 for (i = 0; i < mci->nr_csrows; i++) { 1418 int j; 1419 1420 edac_mc_dump_csrow(&mci->csrows[i]); 1421 for (j = 0; j < mci->csrows[i].nr_channels; j++) 1422 edac_mc_dump_channel( 1423 &mci->csrows[i].channels[j]); 1424 } 1425 } 1426#endif 1427 down(&mem_ctls_mutex); 1428 1429 if (add_mc_to_global_list(mci)) 1430 goto fail0; 1431 1432 /* set load time so that error rate can be tracked */ 1433 mci->start_time = jiffies; 1434 1435 if (edac_create_sysfs_mci_device(mci)) { 1436 edac_mc_printk(mci, KERN_WARNING, 1437 "failed to create sysfs device\n"); 1438 goto fail1; 1439 } 1440 1441 /* Report action taken */ 1442 edac_mc_printk(mci, KERN_INFO, "Giving out device to %s %s: DEV %s\n", 1443 mci->mod_name, mci->ctl_name, dev_name(mci->dev)); 1444 1445 up(&mem_ctls_mutex); 1446 return 0; 1447 1448fail1: 1449 del_mc_from_global_list(mci); 1450 1451fail0: 1452 up(&mem_ctls_mutex); 1453 return 1; 1454} 1455EXPORT_SYMBOL_GPL(edac_mc_add_mc); 1456 1457/** 1458 * edac_mc_del_mc: Remove sysfs entries for specified mci structure and 1459 * remove mci structure from global list 1460 * @pdev: Pointer to 'struct device' representing mci structure to remove. 1461 * 1462 * Return pointer to removed mci structure, or NULL if device not found. 1463 */ 1464struct mem_ctl_info * edac_mc_del_mc(struct device *dev) 1465{ 1466 struct mem_ctl_info *mci; 1467 1468 debugf0("MC: %s()\n", __func__); 1469 down(&mem_ctls_mutex); 1470 1471 if ((mci = find_mci_by_dev(dev)) == NULL) { 1472 up(&mem_ctls_mutex); 1473 return NULL; 1474 } 1475 1476 edac_remove_sysfs_mci_device(mci); 1477 del_mc_from_global_list(mci); 1478 up(&mem_ctls_mutex); 1479 edac_printk(KERN_INFO, EDAC_MC, 1480 "Removed device %d for %s %s: DEV %s\n", mci->mc_idx, 1481 mci->mod_name, mci->ctl_name, dev_name(mci->dev)); 1482 return mci; 1483} 1484EXPORT_SYMBOL_GPL(edac_mc_del_mc); 1485 1486void edac_mc_scrub_block(unsigned long page, unsigned long offset, u32 size) 1487{ 1488 struct page *pg; 1489 void *virt_addr; 1490 unsigned long flags = 0; 1491 1492 debugf3("%s()\n", __func__); 1493 1494 /* ECC error page was not in our memory. Ignore it. */ 1495 if(!pfn_valid(page)) 1496 return; 1497 1498 /* Find the actual page structure then map it and fix */ 1499 pg = pfn_to_page(page); 1500 1501 if (PageHighMem(pg)) 1502 local_irq_save(flags); 1503 1504 virt_addr = kmap_atomic(pg, KM_BOUNCE_READ); 1505 1506 /* Perform architecture specific atomic scrub operation */ 1507 atomic_scrub(virt_addr + offset, size); 1508 1509 /* Unmap and complete */ 1510 kunmap_atomic(virt_addr, KM_BOUNCE_READ); 1511 1512 if (PageHighMem(pg)) 1513 local_irq_restore(flags); 1514} 1515EXPORT_SYMBOL_GPL(edac_mc_scrub_block); 1516 1517/* FIXME - should return -1 */ 1518int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page) 1519{ 1520 struct csrow_info *csrows = mci->csrows; 1521 int row, i; 1522 1523 debugf1("MC%d: %s(): 0x%lx\n", mci->mc_idx, __func__, page); 1524 row = -1; 1525 1526 for (i = 0; i < mci->nr_csrows; i++) { 1527 struct csrow_info *csrow = &csrows[i]; 1528 1529 if (csrow->nr_pages == 0) 1530 continue; 1531 1532 debugf3("MC%d: %s(): first(0x%lx) page(0x%lx) last(0x%lx) " 1533 "mask(0x%lx)\n", mci->mc_idx, __func__, 1534 csrow->first_page, page, csrow->last_page, 1535 csrow->page_mask); 1536 1537 if ((page >= csrow->first_page) && 1538 (page <= csrow->last_page) && 1539 ((page & csrow->page_mask) == 1540 (csrow->first_page & csrow->page_mask))) { 1541 row = i; 1542 break; 1543 } 1544 } 1545 1546 if (row == -1) 1547 edac_mc_printk(mci, KERN_ERR, 1548 "could not look up page error address %lx\n", 1549 (unsigned long) page); 1550 1551 return row; 1552} 1553EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page); 1554 1555/* FIXME - setable log (warning/emerg) levels */ 1556/* FIXME - integrate with evlog: http://evlog.sourceforge.net/ */ 1557void edac_mc_handle_ce(struct mem_ctl_info *mci, 1558 unsigned long page_frame_number, unsigned long offset_in_page, 1559 unsigned long syndrome, int row, int channel, const char *msg) 1560{ 1561 unsigned long remapped_page; 1562 1563 debugf3("MC%d: %s()\n", mci->mc_idx, __func__); 1564 1565 /* FIXME - maybe make panic on INTERNAL ERROR an option */ 1566 if (row >= mci->nr_csrows || row < 0) { 1567 /* something is wrong */ 1568 edac_mc_printk(mci, KERN_ERR, 1569 "INTERNAL ERROR: row out of range " 1570 "(%d >= %d)\n", row, mci->nr_csrows); 1571 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); 1572 return; 1573 } 1574 1575 if (channel >= mci->csrows[row].nr_channels || channel < 0) { 1576 /* something is wrong */ 1577 edac_mc_printk(mci, KERN_ERR, 1578 "INTERNAL ERROR: channel out of range " 1579 "(%d >= %d)\n", channel, 1580 mci->csrows[row].nr_channels); 1581 edac_mc_handle_ce_no_info(mci, "INTERNAL ERROR"); 1582 return; 1583 } 1584 1585 if (log_ce) 1586 /* FIXME - put in DIMM location */ 1587 edac_mc_printk(mci, KERN_WARNING, 1588 "CE page 0x%lx, offset 0x%lx, grain %d, syndrome " 1589 "0x%lx, row %d, channel %d, label \"%s\": %s\n", 1590 page_frame_number, offset_in_page, 1591 mci->csrows[row].grain, syndrome, row, channel, 1592 mci->csrows[row].channels[channel].label, msg); 1593 1594 mci->ce_count++; 1595 mci->csrows[row].ce_count++; 1596 mci->csrows[row].channels[channel].ce_count++; 1597 1598 if (mci->scrub_mode & SCRUB_SW_SRC) { 1599 /* 1600 * Some MC's can remap memory so that it is still available 1601 * at a different address when PCI devices map into memory. 1602 * MC's that can't do this lose the memory where PCI devices 1603 * are mapped. This mapping is MC dependant and so we call 1604 * back into the MC driver for it to map the MC page to 1605 * a physical (CPU) page which can then be mapped to a virtual 1606 * page - which can then be scrubbed. 1607 */ 1608 remapped_page = mci->ctl_page_to_phys ? 1609 mci->ctl_page_to_phys(mci, page_frame_number) : 1610 page_frame_number; 1611 1612 edac_mc_scrub_block(remapped_page, offset_in_page, 1613 mci->csrows[row].grain); 1614 } 1615} 1616EXPORT_SYMBOL_GPL(edac_mc_handle_ce); 1617 1618void edac_mc_handle_ce_no_info(struct mem_ctl_info *mci, const char *msg) 1619{ 1620 if (log_ce) 1621 edac_mc_printk(mci, KERN_WARNING, 1622 "CE - no information available: %s\n", msg); 1623 1624 mci->ce_noinfo_count++; 1625 mci->ce_count++; 1626} 1627EXPORT_SYMBOL_GPL(edac_mc_handle_ce_no_info); 1628 1629void edac_mc_handle_ue(struct mem_ctl_info *mci, 1630 unsigned long page_frame_number, unsigned long offset_in_page, 1631 int row, const char *msg) 1632{ 1633 int len = EDAC_MC_LABEL_LEN * 4; 1634 char labels[len + 1]; 1635 char *pos = labels; 1636 int chan; 1637 int chars; 1638 1639 debugf3("MC%d: %s()\n", mci->mc_idx, __func__); 1640 1641 /* FIXME - maybe make panic on INTERNAL ERROR an option */ 1642 if (row >= mci->nr_csrows || row < 0) { 1643 /* something is wrong */ 1644 edac_mc_printk(mci, KERN_ERR, 1645 "INTERNAL ERROR: row out of range " 1646 "(%d >= %d)\n", row, mci->nr_csrows); 1647 edac_mc_handle_ue_no_info(mci, "INTERNAL ERROR"); 1648 return; 1649 } 1650 1651 chars = snprintf(pos, len + 1, "%s", 1652 mci->csrows[row].channels[0].label); 1653 len -= chars; 1654 pos += chars; 1655 1656 for (chan = 1; (chan < mci->csrows[row].nr_channels) && (len > 0); 1657 chan++) { 1658 chars = snprintf(pos, len + 1, ":%s", 1659 mci->csrows[row].channels[chan].label); 1660 len -= chars; 1661 pos += chars; 1662 } 1663 1664 if (log_ue) 1665 edac_mc_printk(mci, KERN_EMERG, 1666 "UE page 0x%lx, offset 0x%lx, grain %d, row %d, " 1667 "labels \"%s\": %s\n", page_frame_number, 1668 offset_in_page, mci->csrows[row].grain, row, labels, 1669 msg); 1670 1671 if (panic_on_ue) 1672 panic("EDAC MC%d: UE page 0x%lx, offset 0x%lx, grain %d, " 1673 "row %d, labels \"%s\": %s\n", mci->mc_idx, 1674 page_frame_number, offset_in_page, 1675 mci->csrows[row].grain, row, labels, msg); 1676 1677 mci->ue_count++; 1678 mci->csrows[row].ue_count++; 1679} 1680EXPORT_SYMBOL_GPL(edac_mc_handle_ue); 1681 1682void edac_mc_handle_ue_no_info(struct mem_ctl_info *mci, const char *msg) 1683{ 1684 if (panic_on_ue) 1685 panic("EDAC MC%d: Uncorrected Error", mci->mc_idx); 1686 1687 if (log_ue) 1688 edac_mc_printk(mci, KERN_WARNING, 1689 "UE - no information available: %s\n", msg); 1690 mci->ue_noinfo_count++; 1691 mci->ue_count++; 1692} 1693EXPORT_SYMBOL_GPL(edac_mc_handle_ue_no_info); 1694 1695 1696/* 1697 * Iterate over all MC instances and check for ECC, et al, errors 1698 */ 1699static inline void check_mc_devices(void) 1700{ 1701 struct list_head *item; 1702 struct mem_ctl_info *mci; 1703 1704 debugf3("%s()\n", __func__); 1705 down(&mem_ctls_mutex); 1706 1707 list_for_each(item, &mc_devices) { 1708 mci = list_entry(item, struct mem_ctl_info, link); 1709 1710 if (mci->edac_check != NULL) 1711 mci->edac_check(mci); 1712 } 1713 1714 up(&mem_ctls_mutex); 1715} 1716 1717/* 1718 * Check MC status every poll_msec. 1719 * Check PCI status every poll_msec as well. 1720 * 1721 * This where the work gets done for edac. 1722 * 1723 * SMP safe, doesn't use NMI, and auto-rate-limits. 1724 */ 1725static void do_edac_check(void) 1726{ 1727 debugf3("%s()\n", __func__); 1728 check_mc_devices(); 1729 do_pci_parity_check(); 1730} 1731 1732static int edac_kernel_thread(void *arg) 1733{ 1734 while (!kthread_should_stop()) { 1735 do_edac_check(); 1736 1737 /* goto sleep for the interval */ 1738 schedule_timeout_interruptible((HZ * poll_msec) / 1000); 1739 try_to_freeze(); 1740 } 1741 1742 return 0; 1743} 1744 1745/* 1746 * edac_mc_init 1747 * module initialization entry point 1748 */ 1749static int __init edac_mc_init(void) 1750{ 1751 edac_printk(KERN_INFO, EDAC_MC, EDAC_MC_VERSION "\n"); 1752 1753 /* 1754 * Harvest and clear any boot/initialization PCI parity errors 1755 * 1756 * FIXME: This only clears errors logged by devices present at time of 1757 * module initialization. We should also do an initial clear 1758 * of each newly hotplugged device. 1759 */ 1760 clear_pci_parity_errors(); 1761 1762 /* Create the MC sysfs entries */ 1763 if (edac_sysfs_memctrl_setup()) { 1764 edac_printk(KERN_ERR, EDAC_MC, 1765 "Error initializing sysfs code\n"); 1766 return -ENODEV; 1767 } 1768 1769 /* Create the PCI parity sysfs entries */ 1770 if (edac_sysfs_pci_setup()) { 1771 edac_sysfs_memctrl_teardown(); 1772 edac_printk(KERN_ERR, EDAC_MC, 1773 "EDAC PCI: Error initializing sysfs code\n"); 1774 return -ENODEV; 1775 } 1776 1777 /* create our kernel thread */ 1778 edac_thread = kthread_run(edac_kernel_thread, NULL, "kedac"); 1779 1780 if (IS_ERR(edac_thread)) { 1781 /* remove the sysfs entries */ 1782 edac_sysfs_memctrl_teardown(); 1783 edac_sysfs_pci_teardown(); 1784 return PTR_ERR(edac_thread); 1785 } 1786 1787 return 0; 1788} 1789 1790/* 1791 * edac_mc_exit() 1792 * module exit/termination functioni 1793 */ 1794static void __exit edac_mc_exit(void) 1795{ 1796 debugf0("%s()\n", __func__); 1797 kthread_stop(edac_thread); 1798 1799 /* tear down the sysfs device */ 1800 edac_sysfs_memctrl_teardown(); 1801 edac_sysfs_pci_teardown(); 1802} 1803 1804module_init(edac_mc_init); 1805module_exit(edac_mc_exit); 1806 1807MODULE_LICENSE("GPL"); 1808MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh et al\n" 1809 "Based on work by Dan Hollis et al"); 1810MODULE_DESCRIPTION("Core library routines for MC reporting"); 1811 1812module_param(panic_on_ue, int, 0644); 1813MODULE_PARM_DESC(panic_on_ue, "Panic on uncorrected error: 0=off 1=on"); 1814#ifdef CONFIG_PCI 1815module_param(check_pci_parity, int, 0644); 1816MODULE_PARM_DESC(check_pci_parity, "Check for PCI bus parity errors: 0=off 1=on"); 1817module_param(panic_on_pci_parity, int, 0644); 1818MODULE_PARM_DESC(panic_on_pci_parity, "Panic on PCI Bus Parity error: 0=off 1=on"); 1819#endif 1820module_param(log_ue, int, 0644); 1821MODULE_PARM_DESC(log_ue, "Log uncorrectable error to console: 0=off 1=on"); 1822module_param(log_ce, int, 0644); 1823MODULE_PARM_DESC(log_ce, "Log correctable error to console: 0=off 1=on"); 1824module_param(poll_msec, int, 0644); 1825MODULE_PARM_DESC(poll_msec, "Polling period in milliseconds"); 1826#ifdef CONFIG_EDAC_DEBUG 1827module_param(edac_debug_level, int, 0644); 1828MODULE_PARM_DESC(edac_debug_level, "Debug level"); 1829#endif 1830