Linux Audio

Check our new training course

Loading...
v6.8
   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/ctype.h>
  29#include <linux/edac.h>
  30#include <linux/bitops.h>
  31#include <linux/uaccess.h>
  32#include <asm/page.h>
  33#include "edac_mc.h"
  34#include "edac_module.h"
  35#include <ras/ras_event.h>
  36
  37#ifdef CONFIG_EDAC_ATOMIC_SCRUB
  38#include <asm/edac.h>
  39#else
  40#define edac_atomic_scrub(va, size) do { } while (0)
  41#endif
  42
  43int edac_op_state = EDAC_OPSTATE_INVAL;
  44EXPORT_SYMBOL_GPL(edac_op_state);
  45
 
 
  46/* lock to memory controller's control array */
  47static DEFINE_MUTEX(mem_ctls_mutex);
  48static LIST_HEAD(mc_devices);
  49
  50/*
  51 * Used to lock EDAC MC to just one module, avoiding two drivers e. g.
  52 *	apei/ghes and i7core_edac to be used at the same time.
  53 */
  54static const char *edac_mc_owner;
  55
  56static struct mem_ctl_info *error_desc_to_mci(struct edac_raw_error_desc *e)
 
 
 
 
 
 
  57{
  58	return container_of(e, struct mem_ctl_info, error_desc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  59}
  60
 
 
 
 
 
 
 
  61unsigned int edac_dimm_info_location(struct dimm_info *dimm, char *buf,
  62				     unsigned int len)
  63{
  64	struct mem_ctl_info *mci = dimm->mci;
  65	int i, n, count = 0;
  66	char *p = buf;
  67
  68	for (i = 0; i < mci->n_layers; i++) {
  69		n = scnprintf(p, len, "%s %d ",
  70			      edac_layer_name[mci->layers[i].type],
  71			      dimm->location[i]);
  72		p += n;
  73		len -= n;
  74		count += n;
 
 
  75	}
  76
  77	return count;
  78}
  79
  80#ifdef CONFIG_EDAC_DEBUG
  81
  82static void edac_mc_dump_channel(struct rank_info *chan)
  83{
  84	edac_dbg(4, "  channel->chan_idx = %d\n", chan->chan_idx);
  85	edac_dbg(4, "    channel = %p\n", chan);
  86	edac_dbg(4, "    channel->csrow = %p\n", chan->csrow);
  87	edac_dbg(4, "    channel->dimm = %p\n", chan->dimm);
  88}
  89
  90static void edac_mc_dump_dimm(struct dimm_info *dimm)
  91{
  92	char location[80];
  93
  94	if (!dimm->nr_pages)
  95		return;
  96
  97	edac_dimm_info_location(dimm, location, sizeof(location));
  98
  99	edac_dbg(4, "%s%i: %smapped as virtual row %d, chan %d\n",
 100		 dimm->mci->csbased ? "rank" : "dimm",
 101		 dimm->idx, location, dimm->csrow, dimm->cschannel);
 102	edac_dbg(4, "  dimm = %p\n", dimm);
 103	edac_dbg(4, "  dimm->label = '%s'\n", dimm->label);
 104	edac_dbg(4, "  dimm->nr_pages = 0x%x\n", dimm->nr_pages);
 105	edac_dbg(4, "  dimm->grain = %d\n", dimm->grain);
 
 106}
 107
 108static void edac_mc_dump_csrow(struct csrow_info *csrow)
 109{
 110	edac_dbg(4, "csrow->csrow_idx = %d\n", csrow->csrow_idx);
 111	edac_dbg(4, "  csrow = %p\n", csrow);
 112	edac_dbg(4, "  csrow->first_page = 0x%lx\n", csrow->first_page);
 113	edac_dbg(4, "  csrow->last_page = 0x%lx\n", csrow->last_page);
 114	edac_dbg(4, "  csrow->page_mask = 0x%lx\n", csrow->page_mask);
 115	edac_dbg(4, "  csrow->nr_channels = %d\n", csrow->nr_channels);
 116	edac_dbg(4, "  csrow->channels = %p\n", csrow->channels);
 117	edac_dbg(4, "  csrow->mci = %p\n", csrow->mci);
 118}
 119
 120static void edac_mc_dump_mci(struct mem_ctl_info *mci)
 121{
 122	edac_dbg(3, "\tmci = %p\n", mci);
 123	edac_dbg(3, "\tmci->mtype_cap = %lx\n", mci->mtype_cap);
 124	edac_dbg(3, "\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
 125	edac_dbg(3, "\tmci->edac_cap = %lx\n", mci->edac_cap);
 126	edac_dbg(4, "\tmci->edac_check = %p\n", mci->edac_check);
 127	edac_dbg(3, "\tmci->nr_csrows = %d, csrows = %p\n",
 128		 mci->nr_csrows, mci->csrows);
 129	edac_dbg(3, "\tmci->nr_dimms = %d, dimms = %p\n",
 130		 mci->tot_dimms, mci->dimms);
 131	edac_dbg(3, "\tdev = %p\n", mci->pdev);
 132	edac_dbg(3, "\tmod_name:ctl_name = %s:%s\n",
 133		 mci->mod_name, mci->ctl_name);
 134	edac_dbg(3, "\tpvt_info = %p\n\n", mci->pvt_info);
 135}
 136
 137#endif				/* CONFIG_EDAC_DEBUG */
 138
 139const char * const edac_mem_types[] = {
 140	[MEM_EMPTY]	= "Empty",
 141	[MEM_RESERVED]	= "Reserved",
 142	[MEM_UNKNOWN]	= "Unknown",
 143	[MEM_FPM]	= "FPM",
 144	[MEM_EDO]	= "EDO",
 145	[MEM_BEDO]	= "BEDO",
 146	[MEM_SDR]	= "Unbuffered-SDR",
 147	[MEM_RDR]	= "Registered-SDR",
 148	[MEM_DDR]	= "Unbuffered-DDR",
 149	[MEM_RDDR]	= "Registered-DDR",
 150	[MEM_RMBS]	= "RMBS",
 151	[MEM_DDR2]	= "Unbuffered-DDR2",
 152	[MEM_FB_DDR2]	= "FullyBuffered-DDR2",
 153	[MEM_RDDR2]	= "Registered-DDR2",
 154	[MEM_XDR]	= "XDR",
 155	[MEM_DDR3]	= "Unbuffered-DDR3",
 156	[MEM_RDDR3]	= "Registered-DDR3",
 157	[MEM_LRDDR3]	= "Load-Reduced-DDR3-RAM",
 158	[MEM_LPDDR3]	= "Low-Power-DDR3-RAM",
 159	[MEM_DDR4]	= "Unbuffered-DDR4",
 160	[MEM_RDDR4]	= "Registered-DDR4",
 161	[MEM_LPDDR4]	= "Low-Power-DDR4-RAM",
 162	[MEM_LRDDR4]	= "Load-Reduced-DDR4-RAM",
 163	[MEM_DDR5]	= "Unbuffered-DDR5",
 164	[MEM_RDDR5]	= "Registered-DDR5",
 165	[MEM_LRDDR5]	= "Load-Reduced-DDR5-RAM",
 166	[MEM_NVDIMM]	= "Non-volatile-RAM",
 167	[MEM_WIO2]	= "Wide-IO-2",
 168	[MEM_HBM2]	= "High-bandwidth-memory-Gen2",
 169	[MEM_HBM3]	= "High-bandwidth-memory-Gen3",
 170};
 171EXPORT_SYMBOL_GPL(edac_mem_types);
 172
 173static void _edac_mc_free(struct mem_ctl_info *mci)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 174{
 175	put_device(&mci->dev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 176}
 177
 178static void mci_release(struct device *dev)
 179{
 180	struct mem_ctl_info *mci = container_of(dev, struct mem_ctl_info, dev);
 181	struct csrow_info *csr;
 182	int i, chn, row;
 183
 184	if (mci->dimms) {
 185		for (i = 0; i < mci->tot_dimms; i++)
 186			kfree(mci->dimms[i]);
 187		kfree(mci->dimms);
 188	}
 189
 190	if (mci->csrows) {
 191		for (row = 0; row < mci->nr_csrows; row++) {
 192			csr = mci->csrows[row];
 193			if (!csr)
 194				continue;
 195
 196			if (csr->channels) {
 197				for (chn = 0; chn < mci->num_cschannel; chn++)
 198					kfree(csr->channels[chn]);
 199				kfree(csr->channels);
 200			}
 201			kfree(csr);
 202		}
 203		kfree(mci->csrows);
 204	}
 205	kfree(mci->pvt_info);
 206	kfree(mci->layers);
 207	kfree(mci);
 208}
 209
 210static int edac_mc_alloc_csrows(struct mem_ctl_info *mci)
 
 
 
 211{
 212	unsigned int tot_channels = mci->num_cschannel;
 213	unsigned int tot_csrows = mci->nr_csrows;
 214	unsigned int row, chn;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 215
 216	/*
 217	 * Alocate and fill the csrow/channels structs
 218	 */
 219	mci->csrows = kcalloc(tot_csrows, sizeof(*mci->csrows), GFP_KERNEL);
 220	if (!mci->csrows)
 221		return -ENOMEM;
 222
 223	for (row = 0; row < tot_csrows; row++) {
 224		struct csrow_info *csr;
 225
 226		csr = kzalloc(sizeof(**mci->csrows), GFP_KERNEL);
 227		if (!csr)
 228			return -ENOMEM;
 229
 230		mci->csrows[row] = csr;
 231		csr->csrow_idx = row;
 232		csr->mci = mci;
 233		csr->nr_channels = tot_channels;
 234		csr->channels = kcalloc(tot_channels, sizeof(*csr->channels),
 235					GFP_KERNEL);
 236		if (!csr->channels)
 237			return -ENOMEM;
 238
 239		for (chn = 0; chn < tot_channels; chn++) {
 240			struct rank_info *chan;
 241
 242			chan = kzalloc(sizeof(**csr->channels), GFP_KERNEL);
 243			if (!chan)
 244				return -ENOMEM;
 245
 246			csr->channels[chn] = chan;
 247			chan->chan_idx = chn;
 248			chan->csrow = csr;
 249		}
 250	}
 251
 252	return 0;
 253}
 254
 255static int edac_mc_alloc_dimms(struct mem_ctl_info *mci)
 256{
 257	unsigned int pos[EDAC_MAX_LAYERS];
 258	unsigned int row, chn, idx;
 259	int layer;
 260	void *p;
 261
 262	/*
 263	 * Allocate and fill the dimm structs
 264	 */
 265	mci->dimms  = kcalloc(mci->tot_dimms, sizeof(*mci->dimms), GFP_KERNEL);
 266	if (!mci->dimms)
 267		return -ENOMEM;
 268
 269	memset(&pos, 0, sizeof(pos));
 270	row = 0;
 271	chn = 0;
 272	for (idx = 0; idx < mci->tot_dimms; idx++) {
 273		struct dimm_info *dimm;
 274		struct rank_info *chan;
 275		int n, len;
 276
 277		chan = mci->csrows[row]->channels[chn];
 
 
 
 
 
 278
 279		dimm = kzalloc(sizeof(**mci->dimms), GFP_KERNEL);
 280		if (!dimm)
 281			return -ENOMEM;
 282		mci->dimms[idx] = dimm;
 283		dimm->mci = mci;
 284		dimm->idx = idx;
 285
 286		/*
 287		 * Copy DIMM location and initialize it.
 288		 */
 289		len = sizeof(dimm->label);
 290		p = dimm->label;
 291		n = scnprintf(p, len, "mc#%u", mci->mc_idx);
 292		p += n;
 293		len -= n;
 294		for (layer = 0; layer < mci->n_layers; layer++) {
 295			n = scnprintf(p, len, "%s#%u",
 296				      edac_layer_name[mci->layers[layer].type],
 297				      pos[layer]);
 298			p += n;
 299			len -= n;
 300			dimm->location[layer] = pos[layer];
 
 
 
 301		}
 302
 303		/* Link it to the csrows old API data */
 304		chan->dimm = dimm;
 305		dimm->csrow = row;
 306		dimm->cschannel = chn;
 307
 308		/* Increment csrow location */
 309		if (mci->layers[0].is_virt_csrow) {
 310			chn++;
 311			if (chn == mci->num_cschannel) {
 312				chn = 0;
 313				row++;
 314			}
 315		} else {
 316			row++;
 317			if (row == mci->nr_csrows) {
 318				row = 0;
 319				chn++;
 320			}
 321		}
 322
 323		/* Increment dimm location */
 324		for (layer = mci->n_layers - 1; layer >= 0; layer--) {
 325			pos[layer]++;
 326			if (pos[layer] < mci->layers[layer].size)
 327				break;
 328			pos[layer] = 0;
 329		}
 330	}
 331
 332	return 0;
 333}
 334
 335struct mem_ctl_info *edac_mc_alloc(unsigned int mc_num,
 336				   unsigned int n_layers,
 337				   struct edac_mc_layer *layers,
 338				   unsigned int sz_pvt)
 339{
 340	struct mem_ctl_info *mci;
 341	struct edac_mc_layer *layer;
 342	unsigned int idx, tot_dimms = 1;
 343	unsigned int tot_csrows = 1, tot_channels = 1;
 344	bool per_rank = false;
 345
 346	if (WARN_ON(n_layers > EDAC_MAX_LAYERS || n_layers == 0))
 347		return NULL;
 348
 349	/*
 350	 * Calculate the total amount of dimms and csrows/cschannels while
 351	 * in the old API emulation mode
 352	 */
 353	for (idx = 0; idx < n_layers; idx++) {
 354		tot_dimms *= layers[idx].size;
 355
 356		if (layers[idx].is_virt_csrow)
 357			tot_csrows *= layers[idx].size;
 358		else
 359			tot_channels *= layers[idx].size;
 360
 361		if (layers[idx].type == EDAC_MC_LAYER_CHIP_SELECT)
 362			per_rank = true;
 363	}
 364
 365	mci = kzalloc(sizeof(struct mem_ctl_info), GFP_KERNEL);
 366	if (!mci)
 367		return NULL;
 368
 369	mci->layers = kcalloc(n_layers, sizeof(struct edac_mc_layer), GFP_KERNEL);
 370	if (!mci->layers)
 371		goto error;
 372
 373	mci->pvt_info = kzalloc(sz_pvt, GFP_KERNEL);
 374	if (!mci->pvt_info)
 375		goto error;
 376
 377	mci->dev.release = mci_release;
 378	device_initialize(&mci->dev);
 379
 380	/* setup index and various internal pointers */
 381	mci->mc_idx = mc_num;
 382	mci->tot_dimms = tot_dimms;
 383	mci->n_layers = n_layers;
 384	memcpy(mci->layers, layers, sizeof(*layer) * n_layers);
 385	mci->nr_csrows = tot_csrows;
 386	mci->num_cschannel = tot_channels;
 387	mci->csbased = per_rank;
 388
 389	if (edac_mc_alloc_csrows(mci))
 390		goto error;
 391
 392	if (edac_mc_alloc_dimms(mci))
 393		goto error;
 394
 395	mci->op_state = OP_ALLOC;
 396
 397	return mci;
 398
 399error:
 400	_edac_mc_free(mci);
 401
 402	return NULL;
 403}
 404EXPORT_SYMBOL_GPL(edac_mc_alloc);
 405
 406void edac_mc_free(struct mem_ctl_info *mci)
 407{
 408	edac_dbg(1, "\n");
 409
 410	_edac_mc_free(mci);
 
 
 
 
 
 
 
 
 
 411}
 412EXPORT_SYMBOL_GPL(edac_mc_free);
 413
 414bool edac_has_mcs(void)
 415{
 416	bool ret;
 417
 418	mutex_lock(&mem_ctls_mutex);
 419
 420	ret = list_empty(&mc_devices);
 421
 422	mutex_unlock(&mem_ctls_mutex);
 423
 424	return !ret;
 425}
 426EXPORT_SYMBOL_GPL(edac_has_mcs);
 427
 428/* Caller must hold mem_ctls_mutex */
 429static struct mem_ctl_info *__find_mci_by_dev(struct device *dev)
 430{
 431	struct mem_ctl_info *mci;
 432	struct list_head *item;
 433
 434	edac_dbg(3, "\n");
 435
 436	list_for_each(item, &mc_devices) {
 437		mci = list_entry(item, struct mem_ctl_info, link);
 438
 439		if (mci->pdev == dev)
 440			return mci;
 441	}
 442
 443	return NULL;
 444}
 445
 446/**
 447 * find_mci_by_dev
 448 *
 449 *	scan list of controllers looking for the one that manages
 450 *	the 'dev' device
 451 * @dev: pointer to a struct device related with the MCI
 452 */
 453struct mem_ctl_info *find_mci_by_dev(struct device *dev)
 454{
 455	struct mem_ctl_info *ret;
 456
 457	mutex_lock(&mem_ctls_mutex);
 458	ret = __find_mci_by_dev(dev);
 459	mutex_unlock(&mem_ctls_mutex);
 460
 461	return ret;
 462}
 463EXPORT_SYMBOL_GPL(find_mci_by_dev);
 464
 465/*
 466 * edac_mc_workq_function
 467 *	performs the operation scheduled by a workq request
 468 */
 469static void edac_mc_workq_function(struct work_struct *work_req)
 470{
 471	struct delayed_work *d_work = to_delayed_work(work_req);
 472	struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
 473
 474	mutex_lock(&mem_ctls_mutex);
 475
 476	if (mci->op_state != OP_RUNNING_POLL) {
 477		mutex_unlock(&mem_ctls_mutex);
 478		return;
 479	}
 480
 481	if (edac_op_state == EDAC_OPSTATE_POLL)
 482		mci->edac_check(mci);
 483
 484	mutex_unlock(&mem_ctls_mutex);
 485
 486	/* Queue ourselves again. */
 487	edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
 488}
 489
 490/*
 491 * edac_mc_reset_delay_period(unsigned long value)
 492 *
 493 *	user space has updated our poll period value, need to
 494 *	reset our workq delays
 495 */
 496void edac_mc_reset_delay_period(unsigned long value)
 497{
 498	struct mem_ctl_info *mci;
 499	struct list_head *item;
 500
 501	mutex_lock(&mem_ctls_mutex);
 502
 503	list_for_each(item, &mc_devices) {
 504		mci = list_entry(item, struct mem_ctl_info, link);
 505
 506		if (mci->op_state == OP_RUNNING_POLL)
 507			edac_mod_work(&mci->work, value);
 508	}
 509	mutex_unlock(&mem_ctls_mutex);
 510}
 511
 512
 513
 514/* Return 0 on success, 1 on failure.
 515 * Before calling this function, caller must
 516 * assign a unique value to mci->mc_idx.
 517 *
 518 *	locking model:
 519 *
 520 *		called with the mem_ctls_mutex lock held
 521 */
 522static int add_mc_to_global_list(struct mem_ctl_info *mci)
 523{
 524	struct list_head *item, *insert_before;
 525	struct mem_ctl_info *p;
 526
 527	insert_before = &mc_devices;
 528
 529	p = __find_mci_by_dev(mci->pdev);
 530	if (unlikely(p != NULL))
 531		goto fail0;
 532
 533	list_for_each(item, &mc_devices) {
 534		p = list_entry(item, struct mem_ctl_info, link);
 535
 536		if (p->mc_idx >= mci->mc_idx) {
 537			if (unlikely(p->mc_idx == mci->mc_idx))
 538				goto fail1;
 539
 540			insert_before = item;
 541			break;
 542		}
 543	}
 544
 545	list_add_tail_rcu(&mci->link, insert_before);
 546	return 0;
 547
 548fail0:
 549	edac_printk(KERN_WARNING, EDAC_MC,
 550		"%s (%s) %s %s already assigned %d\n", dev_name(p->pdev),
 551		edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
 552	return 1;
 553
 554fail1:
 555	edac_printk(KERN_WARNING, EDAC_MC,
 556		"bug in low-level driver: attempt to assign\n"
 557		"    duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
 558	return 1;
 559}
 560
 561static int del_mc_from_global_list(struct mem_ctl_info *mci)
 562{
 563	list_del_rcu(&mci->link);
 564
 565	/* these are for safe removal of devices from global list while
 566	 * NMI handlers may be traversing list
 567	 */
 568	synchronize_rcu();
 569	INIT_LIST_HEAD(&mci->link);
 570
 571	return list_empty(&mc_devices);
 572}
 573
 574struct mem_ctl_info *edac_mc_find(int idx)
 575{
 576	struct mem_ctl_info *mci;
 577	struct list_head *item;
 578
 579	mutex_lock(&mem_ctls_mutex);
 580
 581	list_for_each(item, &mc_devices) {
 582		mci = list_entry(item, struct mem_ctl_info, link);
 583		if (mci->mc_idx == idx)
 584			goto unlock;
 585	}
 586
 587	mci = NULL;
 588unlock:
 589	mutex_unlock(&mem_ctls_mutex);
 590	return mci;
 591}
 592EXPORT_SYMBOL(edac_mc_find);
 593
 594const char *edac_get_owner(void)
 595{
 596	return edac_mc_owner;
 597}
 598EXPORT_SYMBOL_GPL(edac_get_owner);
 599
 600/* FIXME - should a warning be printed if no error detection? correction? */
 601int edac_mc_add_mc_with_groups(struct mem_ctl_info *mci,
 602			       const struct attribute_group **groups)
 603{
 604	int ret = -EINVAL;
 605	edac_dbg(0, "\n");
 606
 607#ifdef CONFIG_EDAC_DEBUG
 608	if (edac_debug_level >= 3)
 609		edac_mc_dump_mci(mci);
 610
 611	if (edac_debug_level >= 4) {
 612		struct dimm_info *dimm;
 613		int i;
 614
 615		for (i = 0; i < mci->nr_csrows; i++) {
 616			struct csrow_info *csrow = mci->csrows[i];
 617			u32 nr_pages = 0;
 618			int j;
 619
 620			for (j = 0; j < csrow->nr_channels; j++)
 621				nr_pages += csrow->channels[j]->dimm->nr_pages;
 622			if (!nr_pages)
 623				continue;
 624			edac_mc_dump_csrow(csrow);
 625			for (j = 0; j < csrow->nr_channels; j++)
 626				if (csrow->channels[j]->dimm->nr_pages)
 627					edac_mc_dump_channel(csrow->channels[j]);
 628		}
 629
 630		mci_for_each_dimm(mci, dimm)
 631			edac_mc_dump_dimm(dimm);
 632	}
 633#endif
 634	mutex_lock(&mem_ctls_mutex);
 635
 636	if (edac_mc_owner && edac_mc_owner != mci->mod_name) {
 637		ret = -EPERM;
 638		goto fail0;
 639	}
 640
 641	if (add_mc_to_global_list(mci))
 642		goto fail0;
 643
 644	/* set load time so that error rate can be tracked */
 645	mci->start_time = jiffies;
 646
 647	mci->bus = edac_get_sysfs_subsys();
 648
 649	if (edac_create_sysfs_mci_device(mci, groups)) {
 650		edac_mc_printk(mci, KERN_WARNING,
 651			"failed to create sysfs device\n");
 652		goto fail1;
 653	}
 654
 655	if (mci->edac_check) {
 656		mci->op_state = OP_RUNNING_POLL;
 657
 658		INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
 659		edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
 660
 661	} else {
 662		mci->op_state = OP_RUNNING_INTERRUPT;
 663	}
 664
 665	/* Report action taken */
 666	edac_mc_printk(mci, KERN_INFO,
 667		"Giving out device to module %s controller %s: DEV %s (%s)\n",
 668		mci->mod_name, mci->ctl_name, mci->dev_name,
 669		edac_op_state_to_string(mci->op_state));
 670
 671	edac_mc_owner = mci->mod_name;
 672
 673	mutex_unlock(&mem_ctls_mutex);
 674	return 0;
 675
 676fail1:
 677	del_mc_from_global_list(mci);
 678
 679fail0:
 680	mutex_unlock(&mem_ctls_mutex);
 681	return ret;
 682}
 683EXPORT_SYMBOL_GPL(edac_mc_add_mc_with_groups);
 684
 685struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
 686{
 687	struct mem_ctl_info *mci;
 688
 689	edac_dbg(0, "\n");
 690
 691	mutex_lock(&mem_ctls_mutex);
 692
 693	/* find the requested mci struct in the global list */
 694	mci = __find_mci_by_dev(dev);
 695	if (mci == NULL) {
 696		mutex_unlock(&mem_ctls_mutex);
 697		return NULL;
 698	}
 699
 700	/* mark MCI offline: */
 701	mci->op_state = OP_OFFLINE;
 702
 703	if (del_mc_from_global_list(mci))
 704		edac_mc_owner = NULL;
 705
 706	mutex_unlock(&mem_ctls_mutex);
 707
 708	if (mci->edac_check)
 709		edac_stop_work(&mci->work);
 710
 711	/* remove from sysfs */
 712	edac_remove_sysfs_mci_device(mci);
 713
 714	edac_printk(KERN_INFO, EDAC_MC,
 715		"Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
 716		mci->mod_name, mci->ctl_name, edac_dev_name(mci));
 717
 718	return mci;
 719}
 720EXPORT_SYMBOL_GPL(edac_mc_del_mc);
 721
 722static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
 723				u32 size)
 724{
 725	struct page *pg;
 726	void *virt_addr;
 727	unsigned long flags = 0;
 728
 729	edac_dbg(3, "\n");
 730
 731	/* ECC error page was not in our memory. Ignore it. */
 732	if (!pfn_valid(page))
 733		return;
 734
 735	/* Find the actual page structure then map it and fix */
 736	pg = pfn_to_page(page);
 737
 738	if (PageHighMem(pg))
 739		local_irq_save(flags);
 740
 741	virt_addr = kmap_atomic(pg);
 742
 743	/* Perform architecture specific atomic scrub operation */
 744	edac_atomic_scrub(virt_addr + offset, size);
 745
 746	/* Unmap and complete */
 747	kunmap_atomic(virt_addr);
 748
 749	if (PageHighMem(pg))
 750		local_irq_restore(flags);
 751}
 752
 753/* FIXME - should return -1 */
 754int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
 755{
 756	struct csrow_info **csrows = mci->csrows;
 757	int row, i, j, n;
 758
 759	edac_dbg(1, "MC%d: 0x%lx\n", mci->mc_idx, page);
 760	row = -1;
 761
 762	for (i = 0; i < mci->nr_csrows; i++) {
 763		struct csrow_info *csrow = csrows[i];
 764		n = 0;
 765		for (j = 0; j < csrow->nr_channels; j++) {
 766			struct dimm_info *dimm = csrow->channels[j]->dimm;
 767			n += dimm->nr_pages;
 768		}
 769		if (n == 0)
 770			continue;
 771
 772		edac_dbg(3, "MC%d: first(0x%lx) page(0x%lx) last(0x%lx) mask(0x%lx)\n",
 773			 mci->mc_idx,
 774			 csrow->first_page, page, csrow->last_page,
 775			 csrow->page_mask);
 776
 777		if ((page >= csrow->first_page) &&
 778		    (page <= csrow->last_page) &&
 779		    ((page & csrow->page_mask) ==
 780		     (csrow->first_page & csrow->page_mask))) {
 781			row = i;
 782			break;
 783		}
 784	}
 785
 786	if (row == -1)
 787		edac_mc_printk(mci, KERN_ERR,
 788			"could not look up page error address %lx\n",
 789			(unsigned long)page);
 790
 791	return row;
 792}
 793EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
 794
 795const char *edac_layer_name[] = {
 796	[EDAC_MC_LAYER_BRANCH] = "branch",
 797	[EDAC_MC_LAYER_CHANNEL] = "channel",
 798	[EDAC_MC_LAYER_SLOT] = "slot",
 799	[EDAC_MC_LAYER_CHIP_SELECT] = "csrow",
 800	[EDAC_MC_LAYER_ALL_MEM] = "memory",
 801};
 802EXPORT_SYMBOL_GPL(edac_layer_name);
 803
 804static void edac_inc_ce_error(struct edac_raw_error_desc *e)
 
 
 
 805{
 806	int pos[EDAC_MAX_LAYERS] = { e->top_layer, e->mid_layer, e->low_layer };
 807	struct mem_ctl_info *mci = error_desc_to_mci(e);
 808	struct dimm_info *dimm = edac_get_dimm(mci, pos[0], pos[1], pos[2]);
 809
 810	mci->ce_mc += e->error_count;
 811
 812	if (dimm)
 813		dimm->ce_count += e->error_count;
 814	else
 815		mci->ce_noinfo_count += e->error_count;
 
 
 
 
 
 
 
 
 
 
 816}
 817
 818static void edac_inc_ue_error(struct edac_raw_error_desc *e)
 
 
 
 819{
 820	int pos[EDAC_MAX_LAYERS] = { e->top_layer, e->mid_layer, e->low_layer };
 821	struct mem_ctl_info *mci = error_desc_to_mci(e);
 822	struct dimm_info *dimm = edac_get_dimm(mci, pos[0], pos[1], pos[2]);
 823
 824	mci->ue_mc += e->error_count;
 825
 826	if (dimm)
 827		dimm->ue_count += e->error_count;
 828	else
 829		mci->ue_noinfo_count += e->error_count;
 
 
 
 
 
 
 
 
 
 
 830}
 831
 832static void edac_ce_error(struct edac_raw_error_desc *e)
 
 
 
 
 
 
 
 
 
 
 
 833{
 834	struct mem_ctl_info *mci = error_desc_to_mci(e);
 835	unsigned long remapped_page;
 
 
 
 
 836
 837	if (edac_mc_get_log_ce()) {
 838		edac_mc_printk(mci, KERN_WARNING,
 839			"%d CE %s%son %s (%s page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx%s%s)\n",
 840			e->error_count, e->msg,
 841			*e->msg ? " " : "",
 842			e->label, e->location, e->page_frame_number, e->offset_in_page,
 843			e->grain, e->syndrome,
 844			*e->other_detail ? " - " : "",
 845			e->other_detail);
 
 
 846	}
 847
 848	edac_inc_ce_error(e);
 849
 850	if (mci->scrub_mode == SCRUB_SW_SRC) {
 851		/*
 852			* Some memory controllers (called MCs below) can remap
 853			* memory so that it is still available at a different
 854			* address when PCI devices map into memory.
 855			* MC's that can't do this, lose the memory where PCI
 856			* devices are mapped. This mapping is MC-dependent
 857			* and so we call back into the MC driver for it to
 858			* map the MC page to a physical (CPU) page which can
 859			* then be mapped to a virtual page - which can then
 860			* be scrubbed.
 861			*/
 862		remapped_page = mci->ctl_page_to_phys ?
 863			mci->ctl_page_to_phys(mci, e->page_frame_number) :
 864			e->page_frame_number;
 865
 866		edac_mc_scrub_block(remapped_page, e->offset_in_page, e->grain);
 
 867	}
 868}
 869
 870static void edac_ue_error(struct edac_raw_error_desc *e)
 
 
 
 
 
 
 
 
 871{
 872	struct mem_ctl_info *mci = error_desc_to_mci(e);
 
 
 
 873
 874	if (edac_mc_get_log_ue()) {
 875		edac_mc_printk(mci, KERN_WARNING,
 876			"%d UE %s%son %s (%s page:0x%lx offset:0x%lx grain:%ld%s%s)\n",
 877			e->error_count, e->msg,
 878			*e->msg ? " " : "",
 879			e->label, e->location, e->page_frame_number, e->offset_in_page,
 880			e->grain,
 881			*e->other_detail ? " - " : "",
 882			e->other_detail);
 
 
 883	}
 884
 885	edac_inc_ue_error(e);
 886
 887	if (edac_mc_get_panic_on_ue()) {
 888		panic("UE %s%son %s (%s page:0x%lx offset:0x%lx grain:%ld%s%s)\n",
 889			e->msg,
 890			*e->msg ? " " : "",
 891			e->label, e->location, e->page_frame_number, e->offset_in_page,
 892			e->grain,
 893			*e->other_detail ? " - " : "",
 894			e->other_detail);
 895	}
 
 
 896}
 897
 898static void edac_inc_csrow(struct edac_raw_error_desc *e, int row, int chan)
 
 
 899{
 900	struct mem_ctl_info *mci = error_desc_to_mci(e);
 901	enum hw_event_mc_err_type type = e->type;
 902	u16 count = e->error_count;
 903
 904	if (row < 0)
 905		return;
 906
 907	edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan);
 908
 
 909	if (type == HW_EVENT_ERR_CORRECTED) {
 910		mci->csrows[row]->ce_count += count;
 911		if (chan >= 0)
 912			mci->csrows[row]->channels[chan]->ce_count += count;
 
 
 
 
 913	} else {
 914		mci->csrows[row]->ue_count += count;
 915	}
 916}
 917
 918void edac_raw_mc_handle_error(struct edac_raw_error_desc *e)
 919{
 920	struct mem_ctl_info *mci = error_desc_to_mci(e);
 921	u8 grain_bits;
 922
 923	/* Sanity-check driver-supplied grain value. */
 924	if (WARN_ON_ONCE(!e->grain))
 925		e->grain = 1;
 926
 927	grain_bits = fls_long(e->grain - 1);
 928
 929	/* Report the error via the trace interface */
 930	if (IS_ENABLED(CONFIG_RAS))
 931		trace_mc_event(e->type, e->msg, e->label, e->error_count,
 932			       mci->mc_idx, e->top_layer, e->mid_layer,
 933			       e->low_layer,
 934			       (e->page_frame_number << PAGE_SHIFT) | e->offset_in_page,
 935			       grain_bits, e->syndrome, e->other_detail);
 936
 937	if (e->type == HW_EVENT_ERR_CORRECTED)
 938		edac_ce_error(e);
 939	else
 940		edac_ue_error(e);
 941}
 942EXPORT_SYMBOL_GPL(edac_raw_mc_handle_error);
 943
 944void edac_mc_handle_error(const enum hw_event_mc_err_type type,
 945			  struct mem_ctl_info *mci,
 946			  const u16 error_count,
 947			  const unsigned long page_frame_number,
 948			  const unsigned long offset_in_page,
 949			  const unsigned long syndrome,
 950			  const int top_layer,
 951			  const int mid_layer,
 952			  const int low_layer,
 953			  const char *msg,
 954			  const char *other_detail)
 955{
 956	struct dimm_info *dimm;
 957	char *p, *end;
 958	int row = -1, chan = -1;
 959	int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer };
 960	int i, n_labels = 0;
 
 961	struct edac_raw_error_desc *e = &mci->error_desc;
 962	bool any_memory = true;
 963	const char *prefix;
 964
 965	edac_dbg(3, "MC%d\n", mci->mc_idx);
 966
 967	/* Fills the error report buffer */
 968	memset(e, 0, sizeof (*e));
 969	e->error_count = error_count;
 970	e->type = type;
 971	e->top_layer = top_layer;
 972	e->mid_layer = mid_layer;
 973	e->low_layer = low_layer;
 974	e->page_frame_number = page_frame_number;
 975	e->offset_in_page = offset_in_page;
 976	e->syndrome = syndrome;
 977	/* need valid strings here for both: */
 978	e->msg = msg ?: "";
 979	e->other_detail = other_detail ?: "";
 980
 981	/*
 982	 * Check if the event report is consistent and if the memory location is
 983	 * known. If it is, the DIMM(s) label info will be filled and the DIMM's
 
 984	 * error counters will be incremented.
 985	 */
 986	for (i = 0; i < mci->n_layers; i++) {
 987		if (pos[i] >= (int)mci->layers[i].size) {
 988
 989			edac_mc_printk(mci, KERN_ERR,
 990				       "INTERNAL ERROR: %s value is out of range (%d >= %d)\n",
 991				       edac_layer_name[mci->layers[i].type],
 992				       pos[i], mci->layers[i].size);
 993			/*
 994			 * Instead of just returning it, let's use what's
 995			 * known about the error. The increment routines and
 996			 * the DIMM filter logic will do the right thing by
 997			 * pointing the likely damaged DIMMs.
 998			 */
 999			pos[i] = -1;
1000		}
1001		if (pos[i] >= 0)
1002			any_memory = false;
1003	}
1004
1005	/*
1006	 * Get the dimm label/grain that applies to the match criteria.
1007	 * As the error algorithm may not be able to point to just one memory
1008	 * stick, the logic here will get all possible labels that could
1009	 * pottentially be affected by the error.
1010	 * On FB-DIMM memory controllers, for uncorrected errors, it is common
1011	 * to have only the MC channel and the MC dimm (also called "branch")
1012	 * but the channel is not known, as the memory is arranged in pairs,
1013	 * where each memory belongs to a separate channel within the same
1014	 * branch.
1015	 */
1016	p = e->label;
1017	*p = '\0';
1018	end = p + sizeof(e->label);
1019	prefix = "";
1020
1021	mci_for_each_dimm(mci, dimm) {
 
 
1022		if (top_layer >= 0 && top_layer != dimm->location[0])
1023			continue;
1024		if (mid_layer >= 0 && mid_layer != dimm->location[1])
1025			continue;
1026		if (low_layer >= 0 && low_layer != dimm->location[2])
1027			continue;
1028
1029		/* get the max grain, over the error match range */
1030		if (dimm->grain > e->grain)
1031			e->grain = dimm->grain;
1032
1033		/*
1034		 * If the error is memory-controller wide, there's no need to
1035		 * seek for the affected DIMMs because the whole channel/memory
1036		 * controller/... may be affected. Also, don't show errors for
1037		 * empty DIMM slots.
1038		 */
1039		if (!dimm->nr_pages)
1040			continue;
1041
1042		n_labels++;
1043		if (n_labels > EDAC_MAX_LABELS) {
1044			p = e->label;
 
 
 
 
 
 
1045			*p = '\0';
1046		} else {
1047			p += scnprintf(p, end - p, "%s%s", prefix, dimm->label);
1048			prefix = OTHER_LABEL;
1049		}
1050
1051		/*
1052		 * get csrow/channel of the DIMM, in order to allow
1053		 * incrementing the compat API counters
1054		 */
1055		edac_dbg(4, "%s csrows map: (%d,%d)\n",
1056			mci->csbased ? "rank" : "dimm",
1057			dimm->csrow, dimm->cschannel);
1058		if (row == -1)
1059			row = dimm->csrow;
1060		else if (row >= 0 && row != dimm->csrow)
1061			row = -2;
1062
1063		if (chan == -1)
1064			chan = dimm->cschannel;
1065		else if (chan >= 0 && chan != dimm->cschannel)
1066			chan = -2;
1067	}
1068
1069	if (any_memory)
1070		strscpy(e->label, "any memory", sizeof(e->label));
1071	else if (!*e->label)
1072		strscpy(e->label, "unknown memory", sizeof(e->label));
1073
1074	edac_inc_csrow(e, row, chan);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1075
1076	/* Fill the RAM location data */
1077	p = e->location;
1078	end = p + sizeof(e->location);
1079	prefix = "";
1080
1081	for (i = 0; i < mci->n_layers; i++) {
1082		if (pos[i] < 0)
1083			continue;
1084
1085		p += scnprintf(p, end - p, "%s%s:%d", prefix,
1086			       edac_layer_name[mci->layers[i].type], pos[i]);
1087		prefix = " ";
1088	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1089
1090	edac_raw_mc_handle_error(e);
1091}
1092EXPORT_SYMBOL_GPL(edac_mc_handle_error);
v5.4
   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/ctype.h>
  29#include <linux/edac.h>
  30#include <linux/bitops.h>
  31#include <linux/uaccess.h>
  32#include <asm/page.h>
  33#include "edac_mc.h"
  34#include "edac_module.h"
  35#include <ras/ras_event.h>
  36
  37#ifdef CONFIG_EDAC_ATOMIC_SCRUB
  38#include <asm/edac.h>
  39#else
  40#define edac_atomic_scrub(va, size) do { } while (0)
  41#endif
  42
  43int edac_op_state = EDAC_OPSTATE_INVAL;
  44EXPORT_SYMBOL_GPL(edac_op_state);
  45
  46static int edac_report = EDAC_REPORTING_ENABLED;
  47
  48/* lock to memory controller's control array */
  49static DEFINE_MUTEX(mem_ctls_mutex);
  50static LIST_HEAD(mc_devices);
  51
  52/*
  53 * Used to lock EDAC MC to just one module, avoiding two drivers e. g.
  54 *	apei/ghes and i7core_edac to be used at the same time.
  55 */
  56static const char *edac_mc_owner;
  57
  58int edac_get_report_status(void)
  59{
  60	return edac_report;
  61}
  62EXPORT_SYMBOL_GPL(edac_get_report_status);
  63
  64void edac_set_report_status(int new)
  65{
  66	if (new == EDAC_REPORTING_ENABLED ||
  67	    new == EDAC_REPORTING_DISABLED ||
  68	    new == EDAC_REPORTING_FORCE)
  69		edac_report = new;
  70}
  71EXPORT_SYMBOL_GPL(edac_set_report_status);
  72
  73static int edac_report_set(const char *str, const struct kernel_param *kp)
  74{
  75	if (!str)
  76		return -EINVAL;
  77
  78	if (!strncmp(str, "on", 2))
  79		edac_report = EDAC_REPORTING_ENABLED;
  80	else if (!strncmp(str, "off", 3))
  81		edac_report = EDAC_REPORTING_DISABLED;
  82	else if (!strncmp(str, "force", 5))
  83		edac_report = EDAC_REPORTING_FORCE;
  84
  85	return 0;
  86}
  87
  88static int edac_report_get(char *buffer, const struct kernel_param *kp)
  89{
  90	int ret = 0;
  91
  92	switch (edac_report) {
  93	case EDAC_REPORTING_ENABLED:
  94		ret = sprintf(buffer, "on");
  95		break;
  96	case EDAC_REPORTING_DISABLED:
  97		ret = sprintf(buffer, "off");
  98		break;
  99	case EDAC_REPORTING_FORCE:
 100		ret = sprintf(buffer, "force");
 101		break;
 102	default:
 103		ret = -EINVAL;
 104		break;
 105	}
 106
 107	return ret;
 108}
 109
 110static const struct kernel_param_ops edac_report_ops = {
 111	.set = edac_report_set,
 112	.get = edac_report_get,
 113};
 114
 115module_param_cb(edac_report, &edac_report_ops, &edac_report, 0644);
 116
 117unsigned int edac_dimm_info_location(struct dimm_info *dimm, char *buf,
 118				     unsigned int len)
 119{
 120	struct mem_ctl_info *mci = dimm->mci;
 121	int i, n, count = 0;
 122	char *p = buf;
 123
 124	for (i = 0; i < mci->n_layers; i++) {
 125		n = snprintf(p, len, "%s %d ",
 126			      edac_layer_name[mci->layers[i].type],
 127			      dimm->location[i]);
 128		p += n;
 129		len -= n;
 130		count += n;
 131		if (!len)
 132			break;
 133	}
 134
 135	return count;
 136}
 137
 138#ifdef CONFIG_EDAC_DEBUG
 139
 140static void edac_mc_dump_channel(struct rank_info *chan)
 141{
 142	edac_dbg(4, "  channel->chan_idx = %d\n", chan->chan_idx);
 143	edac_dbg(4, "    channel = %p\n", chan);
 144	edac_dbg(4, "    channel->csrow = %p\n", chan->csrow);
 145	edac_dbg(4, "    channel->dimm = %p\n", chan->dimm);
 146}
 147
 148static void edac_mc_dump_dimm(struct dimm_info *dimm, int number)
 149{
 150	char location[80];
 151
 
 
 
 152	edac_dimm_info_location(dimm, location, sizeof(location));
 153
 154	edac_dbg(4, "%s%i: %smapped as virtual row %d, chan %d\n",
 155		 dimm->mci->csbased ? "rank" : "dimm",
 156		 number, location, dimm->csrow, dimm->cschannel);
 157	edac_dbg(4, "  dimm = %p\n", dimm);
 158	edac_dbg(4, "  dimm->label = '%s'\n", dimm->label);
 159	edac_dbg(4, "  dimm->nr_pages = 0x%x\n", dimm->nr_pages);
 160	edac_dbg(4, "  dimm->grain = %d\n", dimm->grain);
 161	edac_dbg(4, "  dimm->nr_pages = 0x%x\n", dimm->nr_pages);
 162}
 163
 164static void edac_mc_dump_csrow(struct csrow_info *csrow)
 165{
 166	edac_dbg(4, "csrow->csrow_idx = %d\n", csrow->csrow_idx);
 167	edac_dbg(4, "  csrow = %p\n", csrow);
 168	edac_dbg(4, "  csrow->first_page = 0x%lx\n", csrow->first_page);
 169	edac_dbg(4, "  csrow->last_page = 0x%lx\n", csrow->last_page);
 170	edac_dbg(4, "  csrow->page_mask = 0x%lx\n", csrow->page_mask);
 171	edac_dbg(4, "  csrow->nr_channels = %d\n", csrow->nr_channels);
 172	edac_dbg(4, "  csrow->channels = %p\n", csrow->channels);
 173	edac_dbg(4, "  csrow->mci = %p\n", csrow->mci);
 174}
 175
 176static void edac_mc_dump_mci(struct mem_ctl_info *mci)
 177{
 178	edac_dbg(3, "\tmci = %p\n", mci);
 179	edac_dbg(3, "\tmci->mtype_cap = %lx\n", mci->mtype_cap);
 180	edac_dbg(3, "\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap);
 181	edac_dbg(3, "\tmci->edac_cap = %lx\n", mci->edac_cap);
 182	edac_dbg(4, "\tmci->edac_check = %p\n", mci->edac_check);
 183	edac_dbg(3, "\tmci->nr_csrows = %d, csrows = %p\n",
 184		 mci->nr_csrows, mci->csrows);
 185	edac_dbg(3, "\tmci->nr_dimms = %d, dimms = %p\n",
 186		 mci->tot_dimms, mci->dimms);
 187	edac_dbg(3, "\tdev = %p\n", mci->pdev);
 188	edac_dbg(3, "\tmod_name:ctl_name = %s:%s\n",
 189		 mci->mod_name, mci->ctl_name);
 190	edac_dbg(3, "\tpvt_info = %p\n\n", mci->pvt_info);
 191}
 192
 193#endif				/* CONFIG_EDAC_DEBUG */
 194
 195const char * const edac_mem_types[] = {
 196	[MEM_EMPTY]	= "Empty",
 197	[MEM_RESERVED]	= "Reserved",
 198	[MEM_UNKNOWN]	= "Unknown",
 199	[MEM_FPM]	= "FPM",
 200	[MEM_EDO]	= "EDO",
 201	[MEM_BEDO]	= "BEDO",
 202	[MEM_SDR]	= "Unbuffered-SDR",
 203	[MEM_RDR]	= "Registered-SDR",
 204	[MEM_DDR]	= "Unbuffered-DDR",
 205	[MEM_RDDR]	= "Registered-DDR",
 206	[MEM_RMBS]	= "RMBS",
 207	[MEM_DDR2]	= "Unbuffered-DDR2",
 208	[MEM_FB_DDR2]	= "FullyBuffered-DDR2",
 209	[MEM_RDDR2]	= "Registered-DDR2",
 210	[MEM_XDR]	= "XDR",
 211	[MEM_DDR3]	= "Unbuffered-DDR3",
 212	[MEM_RDDR3]	= "Registered-DDR3",
 213	[MEM_LRDDR3]	= "Load-Reduced-DDR3-RAM",
 
 214	[MEM_DDR4]	= "Unbuffered-DDR4",
 215	[MEM_RDDR4]	= "Registered-DDR4",
 
 216	[MEM_LRDDR4]	= "Load-Reduced-DDR4-RAM",
 
 
 
 217	[MEM_NVDIMM]	= "Non-volatile-RAM",
 
 
 
 218};
 219EXPORT_SYMBOL_GPL(edac_mem_types);
 220
 221/**
 222 * edac_align_ptr - Prepares the pointer offsets for a single-shot allocation
 223 * @p:		pointer to a pointer with the memory offset to be used. At
 224 *		return, this will be incremented to point to the next offset
 225 * @size:	Size of the data structure to be reserved
 226 * @n_elems:	Number of elements that should be reserved
 227 *
 228 * If 'size' is a constant, the compiler will optimize this whole function
 229 * down to either a no-op or the addition of a constant to the value of '*p'.
 230 *
 231 * The 'p' pointer is absolutely needed to keep the proper advancing
 232 * further in memory to the proper offsets when allocating the struct along
 233 * with its embedded structs, as edac_device_alloc_ctl_info() does it
 234 * above, for example.
 235 *
 236 * At return, the pointer 'p' will be incremented to be used on a next call
 237 * to this function.
 238 */
 239void *edac_align_ptr(void **p, unsigned int size, int n_elems)
 240{
 241	unsigned int align, r;
 242	void *ptr = *p;
 243
 244	*p += size * n_elems;
 245
 246	/*
 247	 * 'p' can possibly be an unaligned item X such that sizeof(X) is
 248	 * 'size'.  Adjust 'p' so that its alignment is at least as
 249	 * stringent as what the compiler would provide for X and return
 250	 * the aligned result.
 251	 * Here we assume that the alignment of a "long long" is the most
 252	 * stringent alignment that the compiler will ever provide by default.
 253	 * As far as I know, this is a reasonable assumption.
 254	 */
 255	if (size > sizeof(long))
 256		align = sizeof(long long);
 257	else if (size > sizeof(int))
 258		align = sizeof(long);
 259	else if (size > sizeof(short))
 260		align = sizeof(int);
 261	else if (size > sizeof(char))
 262		align = sizeof(short);
 263	else
 264		return (char *)ptr;
 265
 266	r = (unsigned long)p % align;
 267
 268	if (r == 0)
 269		return (char *)ptr;
 270
 271	*p += align - r;
 272
 273	return (void *)(((unsigned long)ptr) + align - r);
 274}
 275
 276static void _edac_mc_free(struct mem_ctl_info *mci)
 277{
 
 278	struct csrow_info *csr;
 279	int i, chn, row;
 280
 281	if (mci->dimms) {
 282		for (i = 0; i < mci->tot_dimms; i++)
 283			kfree(mci->dimms[i]);
 284		kfree(mci->dimms);
 285	}
 286
 287	if (mci->csrows) {
 288		for (row = 0; row < mci->nr_csrows; row++) {
 289			csr = mci->csrows[row];
 290			if (!csr)
 291				continue;
 292
 293			if (csr->channels) {
 294				for (chn = 0; chn < mci->num_cschannel; chn++)
 295					kfree(csr->channels[chn]);
 296				kfree(csr->channels);
 297			}
 298			kfree(csr);
 299		}
 300		kfree(mci->csrows);
 301	}
 
 
 302	kfree(mci);
 303}
 304
 305struct mem_ctl_info *edac_mc_alloc(unsigned int mc_num,
 306				   unsigned int n_layers,
 307				   struct edac_mc_layer *layers,
 308				   unsigned int sz_pvt)
 309{
 310	struct mem_ctl_info *mci;
 311	struct edac_mc_layer *layer;
 312	struct csrow_info *csr;
 313	struct rank_info *chan;
 314	struct dimm_info *dimm;
 315	u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS];
 316	unsigned int pos[EDAC_MAX_LAYERS];
 317	unsigned int size, tot_dimms = 1, count = 1;
 318	unsigned int tot_csrows = 1, tot_channels = 1, tot_errcount = 0;
 319	void *pvt, *p, *ptr = NULL;
 320	int i, j, row, chn, n, len, off;
 321	bool per_rank = false;
 322
 323	BUG_ON(n_layers > EDAC_MAX_LAYERS || n_layers == 0);
 324	/*
 325	 * Calculate the total amount of dimms and csrows/cschannels while
 326	 * in the old API emulation mode
 327	 */
 328	for (i = 0; i < n_layers; i++) {
 329		tot_dimms *= layers[i].size;
 330		if (layers[i].is_virt_csrow)
 331			tot_csrows *= layers[i].size;
 332		else
 333			tot_channels *= layers[i].size;
 334
 335		if (layers[i].type == EDAC_MC_LAYER_CHIP_SELECT)
 336			per_rank = true;
 337	}
 338
 339	/* Figure out the offsets of the various items from the start of an mc
 340	 * structure.  We want the alignment of each item to be at least as
 341	 * stringent as what the compiler would provide if we could simply
 342	 * hardcode everything into a single struct.
 343	 */
 344	mci = edac_align_ptr(&ptr, sizeof(*mci), 1);
 345	layer = edac_align_ptr(&ptr, sizeof(*layer), n_layers);
 346	for (i = 0; i < n_layers; i++) {
 347		count *= layers[i].size;
 348		edac_dbg(4, "errcount layer %d size %d\n", i, count);
 349		ce_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
 350		ue_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count);
 351		tot_errcount += 2 * count;
 352	}
 353
 354	edac_dbg(4, "allocating %d error counters\n", tot_errcount);
 355	pvt = edac_align_ptr(&ptr, sz_pvt, 1);
 356	size = ((unsigned long)pvt) + sz_pvt;
 357
 358	edac_dbg(1, "allocating %u bytes for mci data (%d %s, %d csrows/channels)\n",
 359		 size,
 360		 tot_dimms,
 361		 per_rank ? "ranks" : "dimms",
 362		 tot_csrows * tot_channels);
 363
 364	mci = kzalloc(size, GFP_KERNEL);
 365	if (mci == NULL)
 366		return NULL;
 367
 368	/* Adjust pointers so they point within the memory we just allocated
 369	 * rather than an imaginary chunk of memory located at address 0.
 370	 */
 371	layer = (struct edac_mc_layer *)(((char *)mci) + ((unsigned long)layer));
 372	for (i = 0; i < n_layers; i++) {
 373		mci->ce_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ce_per_layer[i]));
 374		mci->ue_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ue_per_layer[i]));
 375	}
 376	pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL;
 377
 378	/* setup index and various internal pointers */
 379	mci->mc_idx = mc_num;
 380	mci->tot_dimms = tot_dimms;
 381	mci->pvt_info = pvt;
 382	mci->n_layers = n_layers;
 383	mci->layers = layer;
 384	memcpy(mci->layers, layers, sizeof(*layer) * n_layers);
 385	mci->nr_csrows = tot_csrows;
 386	mci->num_cschannel = tot_channels;
 387	mci->csbased = per_rank;
 388
 389	/*
 390	 * Alocate and fill the csrow/channels structs
 391	 */
 392	mci->csrows = kcalloc(tot_csrows, sizeof(*mci->csrows), GFP_KERNEL);
 393	if (!mci->csrows)
 394		goto error;
 
 395	for (row = 0; row < tot_csrows; row++) {
 
 
 396		csr = kzalloc(sizeof(**mci->csrows), GFP_KERNEL);
 397		if (!csr)
 398			goto error;
 
 399		mci->csrows[row] = csr;
 400		csr->csrow_idx = row;
 401		csr->mci = mci;
 402		csr->nr_channels = tot_channels;
 403		csr->channels = kcalloc(tot_channels, sizeof(*csr->channels),
 404					GFP_KERNEL);
 405		if (!csr->channels)
 406			goto error;
 407
 408		for (chn = 0; chn < tot_channels; chn++) {
 
 
 409			chan = kzalloc(sizeof(**csr->channels), GFP_KERNEL);
 410			if (!chan)
 411				goto error;
 
 412			csr->channels[chn] = chan;
 413			chan->chan_idx = chn;
 414			chan->csrow = csr;
 415		}
 416	}
 417
 
 
 
 
 
 
 
 
 
 
 418	/*
 419	 * Allocate and fill the dimm structs
 420	 */
 421	mci->dimms  = kcalloc(tot_dimms, sizeof(*mci->dimms), GFP_KERNEL);
 422	if (!mci->dimms)
 423		goto error;
 424
 425	memset(&pos, 0, sizeof(pos));
 426	row = 0;
 427	chn = 0;
 428	for (i = 0; i < tot_dimms; i++) {
 
 
 
 
 429		chan = mci->csrows[row]->channels[chn];
 430		off = EDAC_DIMM_OFF(layer, n_layers, pos[0], pos[1], pos[2]);
 431		if (off < 0 || off >= tot_dimms) {
 432			edac_mc_printk(mci, KERN_ERR, "EDAC core bug: EDAC_DIMM_OFF is trying to do an illegal data access\n");
 433			goto error;
 434		}
 435
 436		dimm = kzalloc(sizeof(**mci->dimms), GFP_KERNEL);
 437		if (!dimm)
 438			goto error;
 439		mci->dimms[off] = dimm;
 440		dimm->mci = mci;
 
 441
 442		/*
 443		 * Copy DIMM location and initialize it.
 444		 */
 445		len = sizeof(dimm->label);
 446		p = dimm->label;
 447		n = snprintf(p, len, "mc#%u", mc_num);
 448		p += n;
 449		len -= n;
 450		for (j = 0; j < n_layers; j++) {
 451			n = snprintf(p, len, "%s#%u",
 452				     edac_layer_name[layers[j].type],
 453				     pos[j]);
 454			p += n;
 455			len -= n;
 456			dimm->location[j] = pos[j];
 457
 458			if (len <= 0)
 459				break;
 460		}
 461
 462		/* Link it to the csrows old API data */
 463		chan->dimm = dimm;
 464		dimm->csrow = row;
 465		dimm->cschannel = chn;
 466
 467		/* Increment csrow location */
 468		if (layers[0].is_virt_csrow) {
 469			chn++;
 470			if (chn == tot_channels) {
 471				chn = 0;
 472				row++;
 473			}
 474		} else {
 475			row++;
 476			if (row == tot_csrows) {
 477				row = 0;
 478				chn++;
 479			}
 480		}
 481
 482		/* Increment dimm location */
 483		for (j = n_layers - 1; j >= 0; j--) {
 484			pos[j]++;
 485			if (pos[j] < layers[j].size)
 486				break;
 487			pos[j] = 0;
 488		}
 489	}
 490
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 491	mci->op_state = OP_ALLOC;
 492
 493	return mci;
 494
 495error:
 496	_edac_mc_free(mci);
 497
 498	return NULL;
 499}
 500EXPORT_SYMBOL_GPL(edac_mc_alloc);
 501
 502void edac_mc_free(struct mem_ctl_info *mci)
 503{
 504	edac_dbg(1, "\n");
 505
 506	/* If we're not yet registered with sysfs free only what was allocated
 507	 * in edac_mc_alloc().
 508	 */
 509	if (!device_is_registered(&mci->dev)) {
 510		_edac_mc_free(mci);
 511		return;
 512	}
 513
 514	/* the mci instance is freed here, when the sysfs object is dropped */
 515	edac_unregister_sysfs(mci);
 516}
 517EXPORT_SYMBOL_GPL(edac_mc_free);
 518
 519bool edac_has_mcs(void)
 520{
 521	bool ret;
 522
 523	mutex_lock(&mem_ctls_mutex);
 524
 525	ret = list_empty(&mc_devices);
 526
 527	mutex_unlock(&mem_ctls_mutex);
 528
 529	return !ret;
 530}
 531EXPORT_SYMBOL_GPL(edac_has_mcs);
 532
 533/* Caller must hold mem_ctls_mutex */
 534static struct mem_ctl_info *__find_mci_by_dev(struct device *dev)
 535{
 536	struct mem_ctl_info *mci;
 537	struct list_head *item;
 538
 539	edac_dbg(3, "\n");
 540
 541	list_for_each(item, &mc_devices) {
 542		mci = list_entry(item, struct mem_ctl_info, link);
 543
 544		if (mci->pdev == dev)
 545			return mci;
 546	}
 547
 548	return NULL;
 549}
 550
 551/**
 552 * find_mci_by_dev
 553 *
 554 *	scan list of controllers looking for the one that manages
 555 *	the 'dev' device
 556 * @dev: pointer to a struct device related with the MCI
 557 */
 558struct mem_ctl_info *find_mci_by_dev(struct device *dev)
 559{
 560	struct mem_ctl_info *ret;
 561
 562	mutex_lock(&mem_ctls_mutex);
 563	ret = __find_mci_by_dev(dev);
 564	mutex_unlock(&mem_ctls_mutex);
 565
 566	return ret;
 567}
 568EXPORT_SYMBOL_GPL(find_mci_by_dev);
 569
 570/*
 571 * edac_mc_workq_function
 572 *	performs the operation scheduled by a workq request
 573 */
 574static void edac_mc_workq_function(struct work_struct *work_req)
 575{
 576	struct delayed_work *d_work = to_delayed_work(work_req);
 577	struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work);
 578
 579	mutex_lock(&mem_ctls_mutex);
 580
 581	if (mci->op_state != OP_RUNNING_POLL) {
 582		mutex_unlock(&mem_ctls_mutex);
 583		return;
 584	}
 585
 586	if (edac_op_state == EDAC_OPSTATE_POLL)
 587		mci->edac_check(mci);
 588
 589	mutex_unlock(&mem_ctls_mutex);
 590
 591	/* Queue ourselves again. */
 592	edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
 593}
 594
 595/*
 596 * edac_mc_reset_delay_period(unsigned long value)
 597 *
 598 *	user space has updated our poll period value, need to
 599 *	reset our workq delays
 600 */
 601void edac_mc_reset_delay_period(unsigned long value)
 602{
 603	struct mem_ctl_info *mci;
 604	struct list_head *item;
 605
 606	mutex_lock(&mem_ctls_mutex);
 607
 608	list_for_each(item, &mc_devices) {
 609		mci = list_entry(item, struct mem_ctl_info, link);
 610
 611		if (mci->op_state == OP_RUNNING_POLL)
 612			edac_mod_work(&mci->work, value);
 613	}
 614	mutex_unlock(&mem_ctls_mutex);
 615}
 616
 617
 618
 619/* Return 0 on success, 1 on failure.
 620 * Before calling this function, caller must
 621 * assign a unique value to mci->mc_idx.
 622 *
 623 *	locking model:
 624 *
 625 *		called with the mem_ctls_mutex lock held
 626 */
 627static int add_mc_to_global_list(struct mem_ctl_info *mci)
 628{
 629	struct list_head *item, *insert_before;
 630	struct mem_ctl_info *p;
 631
 632	insert_before = &mc_devices;
 633
 634	p = __find_mci_by_dev(mci->pdev);
 635	if (unlikely(p != NULL))
 636		goto fail0;
 637
 638	list_for_each(item, &mc_devices) {
 639		p = list_entry(item, struct mem_ctl_info, link);
 640
 641		if (p->mc_idx >= mci->mc_idx) {
 642			if (unlikely(p->mc_idx == mci->mc_idx))
 643				goto fail1;
 644
 645			insert_before = item;
 646			break;
 647		}
 648	}
 649
 650	list_add_tail_rcu(&mci->link, insert_before);
 651	return 0;
 652
 653fail0:
 654	edac_printk(KERN_WARNING, EDAC_MC,
 655		"%s (%s) %s %s already assigned %d\n", dev_name(p->pdev),
 656		edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx);
 657	return 1;
 658
 659fail1:
 660	edac_printk(KERN_WARNING, EDAC_MC,
 661		"bug in low-level driver: attempt to assign\n"
 662		"    duplicate mc_idx %d in %s()\n", p->mc_idx, __func__);
 663	return 1;
 664}
 665
 666static int del_mc_from_global_list(struct mem_ctl_info *mci)
 667{
 668	list_del_rcu(&mci->link);
 669
 670	/* these are for safe removal of devices from global list while
 671	 * NMI handlers may be traversing list
 672	 */
 673	synchronize_rcu();
 674	INIT_LIST_HEAD(&mci->link);
 675
 676	return list_empty(&mc_devices);
 677}
 678
 679struct mem_ctl_info *edac_mc_find(int idx)
 680{
 681	struct mem_ctl_info *mci;
 682	struct list_head *item;
 683
 684	mutex_lock(&mem_ctls_mutex);
 685
 686	list_for_each(item, &mc_devices) {
 687		mci = list_entry(item, struct mem_ctl_info, link);
 688		if (mci->mc_idx == idx)
 689			goto unlock;
 690	}
 691
 692	mci = NULL;
 693unlock:
 694	mutex_unlock(&mem_ctls_mutex);
 695	return mci;
 696}
 697EXPORT_SYMBOL(edac_mc_find);
 698
 699const char *edac_get_owner(void)
 700{
 701	return edac_mc_owner;
 702}
 703EXPORT_SYMBOL_GPL(edac_get_owner);
 704
 705/* FIXME - should a warning be printed if no error detection? correction? */
 706int edac_mc_add_mc_with_groups(struct mem_ctl_info *mci,
 707			       const struct attribute_group **groups)
 708{
 709	int ret = -EINVAL;
 710	edac_dbg(0, "\n");
 711
 712#ifdef CONFIG_EDAC_DEBUG
 713	if (edac_debug_level >= 3)
 714		edac_mc_dump_mci(mci);
 715
 716	if (edac_debug_level >= 4) {
 
 717		int i;
 718
 719		for (i = 0; i < mci->nr_csrows; i++) {
 720			struct csrow_info *csrow = mci->csrows[i];
 721			u32 nr_pages = 0;
 722			int j;
 723
 724			for (j = 0; j < csrow->nr_channels; j++)
 725				nr_pages += csrow->channels[j]->dimm->nr_pages;
 726			if (!nr_pages)
 727				continue;
 728			edac_mc_dump_csrow(csrow);
 729			for (j = 0; j < csrow->nr_channels; j++)
 730				if (csrow->channels[j]->dimm->nr_pages)
 731					edac_mc_dump_channel(csrow->channels[j]);
 732		}
 733		for (i = 0; i < mci->tot_dimms; i++)
 734			if (mci->dimms[i]->nr_pages)
 735				edac_mc_dump_dimm(mci->dimms[i], i);
 736	}
 737#endif
 738	mutex_lock(&mem_ctls_mutex);
 739
 740	if (edac_mc_owner && edac_mc_owner != mci->mod_name) {
 741		ret = -EPERM;
 742		goto fail0;
 743	}
 744
 745	if (add_mc_to_global_list(mci))
 746		goto fail0;
 747
 748	/* set load time so that error rate can be tracked */
 749	mci->start_time = jiffies;
 750
 751	mci->bus = edac_get_sysfs_subsys();
 752
 753	if (edac_create_sysfs_mci_device(mci, groups)) {
 754		edac_mc_printk(mci, KERN_WARNING,
 755			"failed to create sysfs device\n");
 756		goto fail1;
 757	}
 758
 759	if (mci->edac_check) {
 760		mci->op_state = OP_RUNNING_POLL;
 761
 762		INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function);
 763		edac_queue_work(&mci->work, msecs_to_jiffies(edac_mc_get_poll_msec()));
 764
 765	} else {
 766		mci->op_state = OP_RUNNING_INTERRUPT;
 767	}
 768
 769	/* Report action taken */
 770	edac_mc_printk(mci, KERN_INFO,
 771		"Giving out device to module %s controller %s: DEV %s (%s)\n",
 772		mci->mod_name, mci->ctl_name, mci->dev_name,
 773		edac_op_state_to_string(mci->op_state));
 774
 775	edac_mc_owner = mci->mod_name;
 776
 777	mutex_unlock(&mem_ctls_mutex);
 778	return 0;
 779
 780fail1:
 781	del_mc_from_global_list(mci);
 782
 783fail0:
 784	mutex_unlock(&mem_ctls_mutex);
 785	return ret;
 786}
 787EXPORT_SYMBOL_GPL(edac_mc_add_mc_with_groups);
 788
 789struct mem_ctl_info *edac_mc_del_mc(struct device *dev)
 790{
 791	struct mem_ctl_info *mci;
 792
 793	edac_dbg(0, "\n");
 794
 795	mutex_lock(&mem_ctls_mutex);
 796
 797	/* find the requested mci struct in the global list */
 798	mci = __find_mci_by_dev(dev);
 799	if (mci == NULL) {
 800		mutex_unlock(&mem_ctls_mutex);
 801		return NULL;
 802	}
 803
 804	/* mark MCI offline: */
 805	mci->op_state = OP_OFFLINE;
 806
 807	if (del_mc_from_global_list(mci))
 808		edac_mc_owner = NULL;
 809
 810	mutex_unlock(&mem_ctls_mutex);
 811
 812	if (mci->edac_check)
 813		edac_stop_work(&mci->work);
 814
 815	/* remove from sysfs */
 816	edac_remove_sysfs_mci_device(mci);
 817
 818	edac_printk(KERN_INFO, EDAC_MC,
 819		"Removed device %d for %s %s: DEV %s\n", mci->mc_idx,
 820		mci->mod_name, mci->ctl_name, edac_dev_name(mci));
 821
 822	return mci;
 823}
 824EXPORT_SYMBOL_GPL(edac_mc_del_mc);
 825
 826static void edac_mc_scrub_block(unsigned long page, unsigned long offset,
 827				u32 size)
 828{
 829	struct page *pg;
 830	void *virt_addr;
 831	unsigned long flags = 0;
 832
 833	edac_dbg(3, "\n");
 834
 835	/* ECC error page was not in our memory. Ignore it. */
 836	if (!pfn_valid(page))
 837		return;
 838
 839	/* Find the actual page structure then map it and fix */
 840	pg = pfn_to_page(page);
 841
 842	if (PageHighMem(pg))
 843		local_irq_save(flags);
 844
 845	virt_addr = kmap_atomic(pg);
 846
 847	/* Perform architecture specific atomic scrub operation */
 848	edac_atomic_scrub(virt_addr + offset, size);
 849
 850	/* Unmap and complete */
 851	kunmap_atomic(virt_addr);
 852
 853	if (PageHighMem(pg))
 854		local_irq_restore(flags);
 855}
 856
 857/* FIXME - should return -1 */
 858int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page)
 859{
 860	struct csrow_info **csrows = mci->csrows;
 861	int row, i, j, n;
 862
 863	edac_dbg(1, "MC%d: 0x%lx\n", mci->mc_idx, page);
 864	row = -1;
 865
 866	for (i = 0; i < mci->nr_csrows; i++) {
 867		struct csrow_info *csrow = csrows[i];
 868		n = 0;
 869		for (j = 0; j < csrow->nr_channels; j++) {
 870			struct dimm_info *dimm = csrow->channels[j]->dimm;
 871			n += dimm->nr_pages;
 872		}
 873		if (n == 0)
 874			continue;
 875
 876		edac_dbg(3, "MC%d: first(0x%lx) page(0x%lx) last(0x%lx) mask(0x%lx)\n",
 877			 mci->mc_idx,
 878			 csrow->first_page, page, csrow->last_page,
 879			 csrow->page_mask);
 880
 881		if ((page >= csrow->first_page) &&
 882		    (page <= csrow->last_page) &&
 883		    ((page & csrow->page_mask) ==
 884		     (csrow->first_page & csrow->page_mask))) {
 885			row = i;
 886			break;
 887		}
 888	}
 889
 890	if (row == -1)
 891		edac_mc_printk(mci, KERN_ERR,
 892			"could not look up page error address %lx\n",
 893			(unsigned long)page);
 894
 895	return row;
 896}
 897EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page);
 898
 899const char *edac_layer_name[] = {
 900	[EDAC_MC_LAYER_BRANCH] = "branch",
 901	[EDAC_MC_LAYER_CHANNEL] = "channel",
 902	[EDAC_MC_LAYER_SLOT] = "slot",
 903	[EDAC_MC_LAYER_CHIP_SELECT] = "csrow",
 904	[EDAC_MC_LAYER_ALL_MEM] = "memory",
 905};
 906EXPORT_SYMBOL_GPL(edac_layer_name);
 907
 908static void edac_inc_ce_error(struct mem_ctl_info *mci,
 909			      bool enable_per_layer_report,
 910			      const int pos[EDAC_MAX_LAYERS],
 911			      const u16 count)
 912{
 913	int i, index = 0;
 
 
 914
 915	mci->ce_mc += count;
 916
 917	if (!enable_per_layer_report) {
 918		mci->ce_noinfo_count += count;
 919		return;
 920	}
 921
 922	for (i = 0; i < mci->n_layers; i++) {
 923		if (pos[i] < 0)
 924			break;
 925		index += pos[i];
 926		mci->ce_per_layer[i][index] += count;
 927
 928		if (i < mci->n_layers - 1)
 929			index *= mci->layers[i + 1].size;
 930	}
 931}
 932
 933static void edac_inc_ue_error(struct mem_ctl_info *mci,
 934				    bool enable_per_layer_report,
 935				    const int pos[EDAC_MAX_LAYERS],
 936				    const u16 count)
 937{
 938	int i, index = 0;
 
 
 939
 940	mci->ue_mc += count;
 941
 942	if (!enable_per_layer_report) {
 943		mci->ue_noinfo_count += count;
 944		return;
 945	}
 946
 947	for (i = 0; i < mci->n_layers; i++) {
 948		if (pos[i] < 0)
 949			break;
 950		index += pos[i];
 951		mci->ue_per_layer[i][index] += count;
 952
 953		if (i < mci->n_layers - 1)
 954			index *= mci->layers[i + 1].size;
 955	}
 956}
 957
 958static void edac_ce_error(struct mem_ctl_info *mci,
 959			  const u16 error_count,
 960			  const int pos[EDAC_MAX_LAYERS],
 961			  const char *msg,
 962			  const char *location,
 963			  const char *label,
 964			  const char *detail,
 965			  const char *other_detail,
 966			  const bool enable_per_layer_report,
 967			  const unsigned long page_frame_number,
 968			  const unsigned long offset_in_page,
 969			  long grain)
 970{
 
 971	unsigned long remapped_page;
 972	char *msg_aux = "";
 973
 974	if (*msg)
 975		msg_aux = " ";
 976
 977	if (edac_mc_get_log_ce()) {
 978		if (other_detail && *other_detail)
 979			edac_mc_printk(mci, KERN_WARNING,
 980				       "%d CE %s%son %s (%s %s - %s)\n",
 981				       error_count, msg, msg_aux, label,
 982				       location, detail, other_detail);
 983		else
 984			edac_mc_printk(mci, KERN_WARNING,
 985				       "%d CE %s%son %s (%s %s)\n",
 986				       error_count, msg, msg_aux, label,
 987				       location, detail);
 988	}
 989	edac_inc_ce_error(mci, enable_per_layer_report, pos, error_count);
 
 990
 991	if (mci->scrub_mode == SCRUB_SW_SRC) {
 992		/*
 993			* Some memory controllers (called MCs below) can remap
 994			* memory so that it is still available at a different
 995			* address when PCI devices map into memory.
 996			* MC's that can't do this, lose the memory where PCI
 997			* devices are mapped. This mapping is MC-dependent
 998			* and so we call back into the MC driver for it to
 999			* map the MC page to a physical (CPU) page which can
1000			* then be mapped to a virtual page - which can then
1001			* be scrubbed.
1002			*/
1003		remapped_page = mci->ctl_page_to_phys ?
1004			mci->ctl_page_to_phys(mci, page_frame_number) :
1005			page_frame_number;
1006
1007		edac_mc_scrub_block(remapped_page,
1008					offset_in_page, grain);
1009	}
1010}
1011
1012static void edac_ue_error(struct mem_ctl_info *mci,
1013			  const u16 error_count,
1014			  const int pos[EDAC_MAX_LAYERS],
1015			  const char *msg,
1016			  const char *location,
1017			  const char *label,
1018			  const char *detail,
1019			  const char *other_detail,
1020			  const bool enable_per_layer_report)
1021{
1022	char *msg_aux = "";
1023
1024	if (*msg)
1025		msg_aux = " ";
1026
1027	if (edac_mc_get_log_ue()) {
1028		if (other_detail && *other_detail)
1029			edac_mc_printk(mci, KERN_WARNING,
1030				       "%d UE %s%son %s (%s %s - %s)\n",
1031				       error_count, msg, msg_aux, label,
1032				       location, detail, other_detail);
1033		else
1034			edac_mc_printk(mci, KERN_WARNING,
1035				       "%d UE %s%son %s (%s %s)\n",
1036				       error_count, msg, msg_aux, label,
1037				       location, detail);
1038	}
1039
 
 
1040	if (edac_mc_get_panic_on_ue()) {
1041		if (other_detail && *other_detail)
1042			panic("UE %s%son %s (%s%s - %s)\n",
1043			      msg, msg_aux, label, location, detail, other_detail);
1044		else
1045			panic("UE %s%son %s (%s%s)\n",
1046			      msg, msg_aux, label, location, detail);
 
1047	}
1048
1049	edac_inc_ue_error(mci, enable_per_layer_report, pos, error_count);
1050}
1051
1052void edac_raw_mc_handle_error(const enum hw_event_mc_err_type type,
1053			      struct mem_ctl_info *mci,
1054			      struct edac_raw_error_desc *e)
1055{
1056	char detail[80];
1057	int pos[EDAC_MAX_LAYERS] = { e->top_layer, e->mid_layer, e->low_layer };
 
 
 
 
 
 
1058
1059	/* Memory type dependent details about the error */
1060	if (type == HW_EVENT_ERR_CORRECTED) {
1061		snprintf(detail, sizeof(detail),
1062			"page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx",
1063			e->page_frame_number, e->offset_in_page,
1064			e->grain, e->syndrome);
1065		edac_ce_error(mci, e->error_count, pos, e->msg, e->location, e->label,
1066			      detail, e->other_detail, e->enable_per_layer_report,
1067			      e->page_frame_number, e->offset_in_page, e->grain);
1068	} else {
1069		snprintf(detail, sizeof(detail),
1070			"page:0x%lx offset:0x%lx grain:%ld",
1071			e->page_frame_number, e->offset_in_page, e->grain);
1072
1073		edac_ue_error(mci, e->error_count, pos, e->msg, e->location, e->label,
1074			      detail, e->other_detail, e->enable_per_layer_report);
1075	}
 
 
 
 
 
 
 
1076
 
 
 
 
 
 
 
1077
 
 
 
 
1078}
1079EXPORT_SYMBOL_GPL(edac_raw_mc_handle_error);
1080
1081void edac_mc_handle_error(const enum hw_event_mc_err_type type,
1082			  struct mem_ctl_info *mci,
1083			  const u16 error_count,
1084			  const unsigned long page_frame_number,
1085			  const unsigned long offset_in_page,
1086			  const unsigned long syndrome,
1087			  const int top_layer,
1088			  const int mid_layer,
1089			  const int low_layer,
1090			  const char *msg,
1091			  const char *other_detail)
1092{
1093	char *p;
 
1094	int row = -1, chan = -1;
1095	int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer };
1096	int i, n_labels = 0;
1097	u8 grain_bits;
1098	struct edac_raw_error_desc *e = &mci->error_desc;
 
 
1099
1100	edac_dbg(3, "MC%d\n", mci->mc_idx);
1101
1102	/* Fills the error report buffer */
1103	memset(e, 0, sizeof (*e));
1104	e->error_count = error_count;
 
1105	e->top_layer = top_layer;
1106	e->mid_layer = mid_layer;
1107	e->low_layer = low_layer;
1108	e->page_frame_number = page_frame_number;
1109	e->offset_in_page = offset_in_page;
1110	e->syndrome = syndrome;
1111	e->msg = msg;
1112	e->other_detail = other_detail;
 
1113
1114	/*
1115	 * Check if the event report is consistent and if the memory
1116	 * location is known. If it is known, enable_per_layer_report will be
1117	 * true, the DIMM(s) label info will be filled and the per-layer
1118	 * error counters will be incremented.
1119	 */
1120	for (i = 0; i < mci->n_layers; i++) {
1121		if (pos[i] >= (int)mci->layers[i].size) {
1122
1123			edac_mc_printk(mci, KERN_ERR,
1124				       "INTERNAL ERROR: %s value is out of range (%d >= %d)\n",
1125				       edac_layer_name[mci->layers[i].type],
1126				       pos[i], mci->layers[i].size);
1127			/*
1128			 * Instead of just returning it, let's use what's
1129			 * known about the error. The increment routines and
1130			 * the DIMM filter logic will do the right thing by
1131			 * pointing the likely damaged DIMMs.
1132			 */
1133			pos[i] = -1;
1134		}
1135		if (pos[i] >= 0)
1136			e->enable_per_layer_report = true;
1137	}
1138
1139	/*
1140	 * Get the dimm label/grain that applies to the match criteria.
1141	 * As the error algorithm may not be able to point to just one memory
1142	 * stick, the logic here will get all possible labels that could
1143	 * pottentially be affected by the error.
1144	 * On FB-DIMM memory controllers, for uncorrected errors, it is common
1145	 * to have only the MC channel and the MC dimm (also called "branch")
1146	 * but the channel is not known, as the memory is arranged in pairs,
1147	 * where each memory belongs to a separate channel within the same
1148	 * branch.
1149	 */
1150	p = e->label;
1151	*p = '\0';
 
 
1152
1153	for (i = 0; i < mci->tot_dimms; i++) {
1154		struct dimm_info *dimm = mci->dimms[i];
1155
1156		if (top_layer >= 0 && top_layer != dimm->location[0])
1157			continue;
1158		if (mid_layer >= 0 && mid_layer != dimm->location[1])
1159			continue;
1160		if (low_layer >= 0 && low_layer != dimm->location[2])
1161			continue;
1162
1163		/* get the max grain, over the error match range */
1164		if (dimm->grain > e->grain)
1165			e->grain = dimm->grain;
1166
1167		/*
1168		 * If the error is memory-controller wide, there's no need to
1169		 * seek for the affected DIMMs because the whole
1170		 * channel/memory controller/...  may be affected.
1171		 * Also, don't show errors for empty DIMM slots.
1172		 */
1173		if (e->enable_per_layer_report && dimm->nr_pages) {
1174			if (n_labels >= EDAC_MAX_LABELS) {
1175				e->enable_per_layer_report = false;
1176				break;
1177			}
1178			n_labels++;
1179			if (p != e->label) {
1180				strcpy(p, OTHER_LABEL);
1181				p += strlen(OTHER_LABEL);
1182			}
1183			strcpy(p, dimm->label);
1184			p += strlen(p);
1185			*p = '\0';
 
 
 
 
1186
1187			/*
1188			 * get csrow/channel of the DIMM, in order to allow
1189			 * incrementing the compat API counters
1190			 */
1191			edac_dbg(4, "%s csrows map: (%d,%d)\n",
1192				 mci->csbased ? "rank" : "dimm",
1193				 dimm->csrow, dimm->cschannel);
1194			if (row == -1)
1195				row = dimm->csrow;
1196			else if (row >= 0 && row != dimm->csrow)
1197				row = -2;
1198
1199			if (chan == -1)
1200				chan = dimm->cschannel;
1201			else if (chan >= 0 && chan != dimm->cschannel)
1202				chan = -2;
1203		}
1204	}
 
 
 
 
1205
1206	if (!e->enable_per_layer_report) {
1207		strcpy(e->label, "any memory");
1208	} else {
1209		edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan);
1210		if (p == e->label)
1211			strcpy(e->label, "unknown memory");
1212		if (type == HW_EVENT_ERR_CORRECTED) {
1213			if (row >= 0) {
1214				mci->csrows[row]->ce_count += error_count;
1215				if (chan >= 0)
1216					mci->csrows[row]->channels[chan]->ce_count += error_count;
1217			}
1218		} else
1219			if (row >= 0)
1220				mci->csrows[row]->ue_count += error_count;
1221	}
1222
1223	/* Fill the RAM location data */
1224	p = e->location;
 
 
1225
1226	for (i = 0; i < mci->n_layers; i++) {
1227		if (pos[i] < 0)
1228			continue;
1229
1230		p += sprintf(p, "%s:%d ",
1231			     edac_layer_name[mci->layers[i].type],
1232			     pos[i]);
1233	}
1234	if (p > e->location)
1235		*(p - 1) = '\0';
1236
1237	/* Sanity-check driver-supplied grain value. */
1238	if (WARN_ON_ONCE(!e->grain))
1239		e->grain = 1;
1240
1241	grain_bits = fls_long(e->grain - 1);
1242
1243	/* Report the error via the trace interface */
1244	if (IS_ENABLED(CONFIG_RAS))
1245		trace_mc_event(type, e->msg, e->label, e->error_count,
1246			       mci->mc_idx, e->top_layer, e->mid_layer,
1247			       e->low_layer,
1248			       (e->page_frame_number << PAGE_SHIFT) | e->offset_in_page,
1249			       grain_bits, e->syndrome, e->other_detail);
1250
1251	edac_raw_mc_handle_error(type, mci, e);
1252}
1253EXPORT_SYMBOL_GPL(edac_mc_handle_error);