Linux 4.16.11
[linux/fpc-iii.git] / drivers / nvdimm / region_devs.c
blob1593e1806b16c6b413ea1e5555987b88c6740286
1 /*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/scatterlist.h>
14 #include <linux/highmem.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/hash.h>
18 #include <linux/sort.h>
19 #include <linux/io.h>
20 #include <linux/nd.h>
21 #include "nd-core.h"
22 #include "nd.h"
25 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
26 * irrelevant.
28 #include <linux/io-64-nonatomic-hi-lo.h>
30 static DEFINE_IDA(region_ida);
31 static DEFINE_PER_CPU(int, flush_idx);
33 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
34 struct nd_region_data *ndrd)
36 int i, j;
38 dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
39 nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
40 for (i = 0; i < (1 << ndrd->hints_shift); i++) {
41 struct resource *res = &nvdimm->flush_wpq[i];
42 unsigned long pfn = PHYS_PFN(res->start);
43 void __iomem *flush_page;
45 /* check if flush hints share a page */
46 for (j = 0; j < i; j++) {
47 struct resource *res_j = &nvdimm->flush_wpq[j];
48 unsigned long pfn_j = PHYS_PFN(res_j->start);
50 if (pfn == pfn_j)
51 break;
54 if (j < i)
55 flush_page = (void __iomem *) ((unsigned long)
56 ndrd_get_flush_wpq(ndrd, dimm, j)
57 & PAGE_MASK);
58 else
59 flush_page = devm_nvdimm_ioremap(dev,
60 PFN_PHYS(pfn), PAGE_SIZE);
61 if (!flush_page)
62 return -ENXIO;
63 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
64 + (res->start & ~PAGE_MASK));
67 return 0;
70 int nd_region_activate(struct nd_region *nd_region)
72 int i, j, num_flush = 0;
73 struct nd_region_data *ndrd;
74 struct device *dev = &nd_region->dev;
75 size_t flush_data_size = sizeof(void *);
77 nvdimm_bus_lock(&nd_region->dev);
78 for (i = 0; i < nd_region->ndr_mappings; i++) {
79 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
80 struct nvdimm *nvdimm = nd_mapping->nvdimm;
82 /* at least one null hint slot per-dimm for the "no-hint" case */
83 flush_data_size += sizeof(void *);
84 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
85 if (!nvdimm->num_flush)
86 continue;
87 flush_data_size += nvdimm->num_flush * sizeof(void *);
89 nvdimm_bus_unlock(&nd_region->dev);
91 ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
92 if (!ndrd)
93 return -ENOMEM;
94 dev_set_drvdata(dev, ndrd);
96 if (!num_flush)
97 return 0;
99 ndrd->hints_shift = ilog2(num_flush);
100 for (i = 0; i < nd_region->ndr_mappings; i++) {
101 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
102 struct nvdimm *nvdimm = nd_mapping->nvdimm;
103 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
105 if (rc)
106 return rc;
110 * Clear out entries that are duplicates. This should prevent the
111 * extra flushings.
113 for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
114 /* ignore if NULL already */
115 if (!ndrd_get_flush_wpq(ndrd, i, 0))
116 continue;
118 for (j = i + 1; j < nd_region->ndr_mappings; j++)
119 if (ndrd_get_flush_wpq(ndrd, i, 0) ==
120 ndrd_get_flush_wpq(ndrd, j, 0))
121 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
124 return 0;
127 static void nd_region_release(struct device *dev)
129 struct nd_region *nd_region = to_nd_region(dev);
130 u16 i;
132 for (i = 0; i < nd_region->ndr_mappings; i++) {
133 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
134 struct nvdimm *nvdimm = nd_mapping->nvdimm;
136 put_device(&nvdimm->dev);
138 free_percpu(nd_region->lane);
139 ida_simple_remove(&region_ida, nd_region->id);
140 if (is_nd_blk(dev))
141 kfree(to_nd_blk_region(dev));
142 else
143 kfree(nd_region);
146 static struct device_type nd_blk_device_type = {
147 .name = "nd_blk",
148 .release = nd_region_release,
151 static struct device_type nd_pmem_device_type = {
152 .name = "nd_pmem",
153 .release = nd_region_release,
156 static struct device_type nd_volatile_device_type = {
157 .name = "nd_volatile",
158 .release = nd_region_release,
161 bool is_nd_pmem(struct device *dev)
163 return dev ? dev->type == &nd_pmem_device_type : false;
166 bool is_nd_blk(struct device *dev)
168 return dev ? dev->type == &nd_blk_device_type : false;
171 bool is_nd_volatile(struct device *dev)
173 return dev ? dev->type == &nd_volatile_device_type : false;
176 struct nd_region *to_nd_region(struct device *dev)
178 struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
180 WARN_ON(dev->type->release != nd_region_release);
181 return nd_region;
183 EXPORT_SYMBOL_GPL(to_nd_region);
185 struct nd_blk_region *to_nd_blk_region(struct device *dev)
187 struct nd_region *nd_region = to_nd_region(dev);
189 WARN_ON(!is_nd_blk(dev));
190 return container_of(nd_region, struct nd_blk_region, nd_region);
192 EXPORT_SYMBOL_GPL(to_nd_blk_region);
194 void *nd_region_provider_data(struct nd_region *nd_region)
196 return nd_region->provider_data;
198 EXPORT_SYMBOL_GPL(nd_region_provider_data);
200 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
202 return ndbr->blk_provider_data;
204 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
206 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
208 ndbr->blk_provider_data = data;
210 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
213 * nd_region_to_nstype() - region to an integer namespace type
214 * @nd_region: region-device to interrogate
216 * This is the 'nstype' attribute of a region as well, an input to the
217 * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
218 * namespace devices with namespace drivers.
220 int nd_region_to_nstype(struct nd_region *nd_region)
222 if (is_memory(&nd_region->dev)) {
223 u16 i, alias;
225 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
226 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
227 struct nvdimm *nvdimm = nd_mapping->nvdimm;
229 if (test_bit(NDD_ALIASING, &nvdimm->flags))
230 alias++;
232 if (alias)
233 return ND_DEVICE_NAMESPACE_PMEM;
234 else
235 return ND_DEVICE_NAMESPACE_IO;
236 } else if (is_nd_blk(&nd_region->dev)) {
237 return ND_DEVICE_NAMESPACE_BLK;
240 return 0;
242 EXPORT_SYMBOL(nd_region_to_nstype);
244 static ssize_t size_show(struct device *dev,
245 struct device_attribute *attr, char *buf)
247 struct nd_region *nd_region = to_nd_region(dev);
248 unsigned long long size = 0;
250 if (is_memory(dev)) {
251 size = nd_region->ndr_size;
252 } else if (nd_region->ndr_mappings == 1) {
253 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
255 size = nd_mapping->size;
258 return sprintf(buf, "%llu\n", size);
260 static DEVICE_ATTR_RO(size);
262 static ssize_t deep_flush_show(struct device *dev,
263 struct device_attribute *attr, char *buf)
265 struct nd_region *nd_region = to_nd_region(dev);
268 * NOTE: in the nvdimm_has_flush() error case this attribute is
269 * not visible.
271 return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
274 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
275 const char *buf, size_t len)
277 bool flush;
278 int rc = strtobool(buf, &flush);
279 struct nd_region *nd_region = to_nd_region(dev);
281 if (rc)
282 return rc;
283 if (!flush)
284 return -EINVAL;
285 nvdimm_flush(nd_region);
287 return len;
289 static DEVICE_ATTR_RW(deep_flush);
291 static ssize_t mappings_show(struct device *dev,
292 struct device_attribute *attr, char *buf)
294 struct nd_region *nd_region = to_nd_region(dev);
296 return sprintf(buf, "%d\n", nd_region->ndr_mappings);
298 static DEVICE_ATTR_RO(mappings);
300 static ssize_t nstype_show(struct device *dev,
301 struct device_attribute *attr, char *buf)
303 struct nd_region *nd_region = to_nd_region(dev);
305 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
307 static DEVICE_ATTR_RO(nstype);
309 static ssize_t set_cookie_show(struct device *dev,
310 struct device_attribute *attr, char *buf)
312 struct nd_region *nd_region = to_nd_region(dev);
313 struct nd_interleave_set *nd_set = nd_region->nd_set;
314 ssize_t rc = 0;
316 if (is_memory(dev) && nd_set)
317 /* pass, should be precluded by region_visible */;
318 else
319 return -ENXIO;
322 * The cookie to show depends on which specification of the
323 * labels we are using. If there are not labels then default to
324 * the v1.1 namespace label cookie definition. To read all this
325 * data we need to wait for probing to settle.
327 device_lock(dev);
328 nvdimm_bus_lock(dev);
329 wait_nvdimm_bus_probe_idle(dev);
330 if (nd_region->ndr_mappings) {
331 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
332 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
334 if (ndd) {
335 struct nd_namespace_index *nsindex;
337 nsindex = to_namespace_index(ndd, ndd->ns_current);
338 rc = sprintf(buf, "%#llx\n",
339 nd_region_interleave_set_cookie(nd_region,
340 nsindex));
343 nvdimm_bus_unlock(dev);
344 device_unlock(dev);
346 if (rc)
347 return rc;
348 return sprintf(buf, "%#llx\n", nd_set->cookie1);
350 static DEVICE_ATTR_RO(set_cookie);
352 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
354 resource_size_t blk_max_overlap = 0, available, overlap;
355 int i;
357 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
359 retry:
360 available = 0;
361 overlap = blk_max_overlap;
362 for (i = 0; i < nd_region->ndr_mappings; i++) {
363 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
364 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
366 /* if a dimm is disabled the available capacity is zero */
367 if (!ndd)
368 return 0;
370 if (is_memory(&nd_region->dev)) {
371 available += nd_pmem_available_dpa(nd_region,
372 nd_mapping, &overlap);
373 if (overlap > blk_max_overlap) {
374 blk_max_overlap = overlap;
375 goto retry;
377 } else if (is_nd_blk(&nd_region->dev))
378 available += nd_blk_available_dpa(nd_region);
381 return available;
384 static ssize_t available_size_show(struct device *dev,
385 struct device_attribute *attr, char *buf)
387 struct nd_region *nd_region = to_nd_region(dev);
388 unsigned long long available = 0;
391 * Flush in-flight updates and grab a snapshot of the available
392 * size. Of course, this value is potentially invalidated the
393 * memory nvdimm_bus_lock() is dropped, but that's userspace's
394 * problem to not race itself.
396 nvdimm_bus_lock(dev);
397 wait_nvdimm_bus_probe_idle(dev);
398 available = nd_region_available_dpa(nd_region);
399 nvdimm_bus_unlock(dev);
401 return sprintf(buf, "%llu\n", available);
403 static DEVICE_ATTR_RO(available_size);
405 static ssize_t init_namespaces_show(struct device *dev,
406 struct device_attribute *attr, char *buf)
408 struct nd_region_data *ndrd = dev_get_drvdata(dev);
409 ssize_t rc;
411 nvdimm_bus_lock(dev);
412 if (ndrd)
413 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
414 else
415 rc = -ENXIO;
416 nvdimm_bus_unlock(dev);
418 return rc;
420 static DEVICE_ATTR_RO(init_namespaces);
422 static ssize_t namespace_seed_show(struct device *dev,
423 struct device_attribute *attr, char *buf)
425 struct nd_region *nd_region = to_nd_region(dev);
426 ssize_t rc;
428 nvdimm_bus_lock(dev);
429 if (nd_region->ns_seed)
430 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
431 else
432 rc = sprintf(buf, "\n");
433 nvdimm_bus_unlock(dev);
434 return rc;
436 static DEVICE_ATTR_RO(namespace_seed);
438 static ssize_t btt_seed_show(struct device *dev,
439 struct device_attribute *attr, char *buf)
441 struct nd_region *nd_region = to_nd_region(dev);
442 ssize_t rc;
444 nvdimm_bus_lock(dev);
445 if (nd_region->btt_seed)
446 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
447 else
448 rc = sprintf(buf, "\n");
449 nvdimm_bus_unlock(dev);
451 return rc;
453 static DEVICE_ATTR_RO(btt_seed);
455 static ssize_t pfn_seed_show(struct device *dev,
456 struct device_attribute *attr, char *buf)
458 struct nd_region *nd_region = to_nd_region(dev);
459 ssize_t rc;
461 nvdimm_bus_lock(dev);
462 if (nd_region->pfn_seed)
463 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
464 else
465 rc = sprintf(buf, "\n");
466 nvdimm_bus_unlock(dev);
468 return rc;
470 static DEVICE_ATTR_RO(pfn_seed);
472 static ssize_t dax_seed_show(struct device *dev,
473 struct device_attribute *attr, char *buf)
475 struct nd_region *nd_region = to_nd_region(dev);
476 ssize_t rc;
478 nvdimm_bus_lock(dev);
479 if (nd_region->dax_seed)
480 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
481 else
482 rc = sprintf(buf, "\n");
483 nvdimm_bus_unlock(dev);
485 return rc;
487 static DEVICE_ATTR_RO(dax_seed);
489 static ssize_t read_only_show(struct device *dev,
490 struct device_attribute *attr, char *buf)
492 struct nd_region *nd_region = to_nd_region(dev);
494 return sprintf(buf, "%d\n", nd_region->ro);
497 static ssize_t read_only_store(struct device *dev,
498 struct device_attribute *attr, const char *buf, size_t len)
500 bool ro;
501 int rc = strtobool(buf, &ro);
502 struct nd_region *nd_region = to_nd_region(dev);
504 if (rc)
505 return rc;
507 nd_region->ro = ro;
508 return len;
510 static DEVICE_ATTR_RW(read_only);
512 static ssize_t region_badblocks_show(struct device *dev,
513 struct device_attribute *attr, char *buf)
515 struct nd_region *nd_region = to_nd_region(dev);
517 return badblocks_show(&nd_region->bb, buf, 0);
520 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
522 static ssize_t resource_show(struct device *dev,
523 struct device_attribute *attr, char *buf)
525 struct nd_region *nd_region = to_nd_region(dev);
527 return sprintf(buf, "%#llx\n", nd_region->ndr_start);
529 static DEVICE_ATTR_RO(resource);
531 static ssize_t persistence_domain_show(struct device *dev,
532 struct device_attribute *attr, char *buf)
534 struct nd_region *nd_region = to_nd_region(dev);
536 if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
537 return sprintf(buf, "cpu_cache\n");
538 else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
539 return sprintf(buf, "memory_controller\n");
540 else
541 return sprintf(buf, "\n");
543 static DEVICE_ATTR_RO(persistence_domain);
545 static struct attribute *nd_region_attributes[] = {
546 &dev_attr_size.attr,
547 &dev_attr_nstype.attr,
548 &dev_attr_mappings.attr,
549 &dev_attr_btt_seed.attr,
550 &dev_attr_pfn_seed.attr,
551 &dev_attr_dax_seed.attr,
552 &dev_attr_deep_flush.attr,
553 &dev_attr_read_only.attr,
554 &dev_attr_set_cookie.attr,
555 &dev_attr_available_size.attr,
556 &dev_attr_namespace_seed.attr,
557 &dev_attr_init_namespaces.attr,
558 &dev_attr_badblocks.attr,
559 &dev_attr_resource.attr,
560 &dev_attr_persistence_domain.attr,
561 NULL,
564 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
566 struct device *dev = container_of(kobj, typeof(*dev), kobj);
567 struct nd_region *nd_region = to_nd_region(dev);
568 struct nd_interleave_set *nd_set = nd_region->nd_set;
569 int type = nd_region_to_nstype(nd_region);
571 if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
572 return 0;
574 if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
575 return 0;
577 if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
578 return 0;
580 if (a == &dev_attr_resource.attr) {
581 if (is_nd_pmem(dev))
582 return 0400;
583 else
584 return 0;
587 if (a == &dev_attr_deep_flush.attr) {
588 int has_flush = nvdimm_has_flush(nd_region);
590 if (has_flush == 1)
591 return a->mode;
592 else if (has_flush == 0)
593 return 0444;
594 else
595 return 0;
598 if (a == &dev_attr_persistence_domain.attr) {
599 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
600 | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
601 return 0;
602 return a->mode;
605 if (a != &dev_attr_set_cookie.attr
606 && a != &dev_attr_available_size.attr)
607 return a->mode;
609 if ((type == ND_DEVICE_NAMESPACE_PMEM
610 || type == ND_DEVICE_NAMESPACE_BLK)
611 && a == &dev_attr_available_size.attr)
612 return a->mode;
613 else if (is_memory(dev) && nd_set)
614 return a->mode;
616 return 0;
619 struct attribute_group nd_region_attribute_group = {
620 .attrs = nd_region_attributes,
621 .is_visible = region_visible,
623 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
625 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
626 struct nd_namespace_index *nsindex)
628 struct nd_interleave_set *nd_set = nd_region->nd_set;
630 if (!nd_set)
631 return 0;
633 if (nsindex && __le16_to_cpu(nsindex->major) == 1
634 && __le16_to_cpu(nsindex->minor) == 1)
635 return nd_set->cookie1;
636 return nd_set->cookie2;
639 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
641 struct nd_interleave_set *nd_set = nd_region->nd_set;
643 if (nd_set)
644 return nd_set->altcookie;
645 return 0;
648 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
650 struct nd_label_ent *label_ent, *e;
652 lockdep_assert_held(&nd_mapping->lock);
653 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
654 list_del(&label_ent->list);
655 kfree(label_ent);
660 * Upon successful probe/remove, take/release a reference on the
661 * associated interleave set (if present), and plant new btt + namespace
662 * seeds. Also, on the removal of a BLK region, notify the provider to
663 * disable the region.
665 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
666 struct device *dev, bool probe)
668 struct nd_region *nd_region;
670 if (!probe && is_nd_region(dev)) {
671 int i;
673 nd_region = to_nd_region(dev);
674 for (i = 0; i < nd_region->ndr_mappings; i++) {
675 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
676 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
677 struct nvdimm *nvdimm = nd_mapping->nvdimm;
679 mutex_lock(&nd_mapping->lock);
680 nd_mapping_free_labels(nd_mapping);
681 mutex_unlock(&nd_mapping->lock);
683 put_ndd(ndd);
684 nd_mapping->ndd = NULL;
685 if (ndd)
686 atomic_dec(&nvdimm->busy);
689 if (dev->parent && is_nd_region(dev->parent) && probe) {
690 nd_region = to_nd_region(dev->parent);
691 nvdimm_bus_lock(dev);
692 if (nd_region->ns_seed == dev)
693 nd_region_create_ns_seed(nd_region);
694 nvdimm_bus_unlock(dev);
696 if (is_nd_btt(dev) && probe) {
697 struct nd_btt *nd_btt = to_nd_btt(dev);
699 nd_region = to_nd_region(dev->parent);
700 nvdimm_bus_lock(dev);
701 if (nd_region->btt_seed == dev)
702 nd_region_create_btt_seed(nd_region);
703 if (nd_region->ns_seed == &nd_btt->ndns->dev)
704 nd_region_create_ns_seed(nd_region);
705 nvdimm_bus_unlock(dev);
707 if (is_nd_pfn(dev) && probe) {
708 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
710 nd_region = to_nd_region(dev->parent);
711 nvdimm_bus_lock(dev);
712 if (nd_region->pfn_seed == dev)
713 nd_region_create_pfn_seed(nd_region);
714 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
715 nd_region_create_ns_seed(nd_region);
716 nvdimm_bus_unlock(dev);
718 if (is_nd_dax(dev) && probe) {
719 struct nd_dax *nd_dax = to_nd_dax(dev);
721 nd_region = to_nd_region(dev->parent);
722 nvdimm_bus_lock(dev);
723 if (nd_region->dax_seed == dev)
724 nd_region_create_dax_seed(nd_region);
725 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
726 nd_region_create_ns_seed(nd_region);
727 nvdimm_bus_unlock(dev);
731 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
733 nd_region_notify_driver_action(nvdimm_bus, dev, true);
736 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
738 nd_region_notify_driver_action(nvdimm_bus, dev, false);
741 static ssize_t mappingN(struct device *dev, char *buf, int n)
743 struct nd_region *nd_region = to_nd_region(dev);
744 struct nd_mapping *nd_mapping;
745 struct nvdimm *nvdimm;
747 if (n >= nd_region->ndr_mappings)
748 return -ENXIO;
749 nd_mapping = &nd_region->mapping[n];
750 nvdimm = nd_mapping->nvdimm;
752 return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
753 nd_mapping->start, nd_mapping->size,
754 nd_mapping->position);
757 #define REGION_MAPPING(idx) \
758 static ssize_t mapping##idx##_show(struct device *dev, \
759 struct device_attribute *attr, char *buf) \
761 return mappingN(dev, buf, idx); \
763 static DEVICE_ATTR_RO(mapping##idx)
766 * 32 should be enough for a while, even in the presence of socket
767 * interleave a 32-way interleave set is a degenerate case.
769 REGION_MAPPING(0);
770 REGION_MAPPING(1);
771 REGION_MAPPING(2);
772 REGION_MAPPING(3);
773 REGION_MAPPING(4);
774 REGION_MAPPING(5);
775 REGION_MAPPING(6);
776 REGION_MAPPING(7);
777 REGION_MAPPING(8);
778 REGION_MAPPING(9);
779 REGION_MAPPING(10);
780 REGION_MAPPING(11);
781 REGION_MAPPING(12);
782 REGION_MAPPING(13);
783 REGION_MAPPING(14);
784 REGION_MAPPING(15);
785 REGION_MAPPING(16);
786 REGION_MAPPING(17);
787 REGION_MAPPING(18);
788 REGION_MAPPING(19);
789 REGION_MAPPING(20);
790 REGION_MAPPING(21);
791 REGION_MAPPING(22);
792 REGION_MAPPING(23);
793 REGION_MAPPING(24);
794 REGION_MAPPING(25);
795 REGION_MAPPING(26);
796 REGION_MAPPING(27);
797 REGION_MAPPING(28);
798 REGION_MAPPING(29);
799 REGION_MAPPING(30);
800 REGION_MAPPING(31);
802 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
804 struct device *dev = container_of(kobj, struct device, kobj);
805 struct nd_region *nd_region = to_nd_region(dev);
807 if (n < nd_region->ndr_mappings)
808 return a->mode;
809 return 0;
812 static struct attribute *mapping_attributes[] = {
813 &dev_attr_mapping0.attr,
814 &dev_attr_mapping1.attr,
815 &dev_attr_mapping2.attr,
816 &dev_attr_mapping3.attr,
817 &dev_attr_mapping4.attr,
818 &dev_attr_mapping5.attr,
819 &dev_attr_mapping6.attr,
820 &dev_attr_mapping7.attr,
821 &dev_attr_mapping8.attr,
822 &dev_attr_mapping9.attr,
823 &dev_attr_mapping10.attr,
824 &dev_attr_mapping11.attr,
825 &dev_attr_mapping12.attr,
826 &dev_attr_mapping13.attr,
827 &dev_attr_mapping14.attr,
828 &dev_attr_mapping15.attr,
829 &dev_attr_mapping16.attr,
830 &dev_attr_mapping17.attr,
831 &dev_attr_mapping18.attr,
832 &dev_attr_mapping19.attr,
833 &dev_attr_mapping20.attr,
834 &dev_attr_mapping21.attr,
835 &dev_attr_mapping22.attr,
836 &dev_attr_mapping23.attr,
837 &dev_attr_mapping24.attr,
838 &dev_attr_mapping25.attr,
839 &dev_attr_mapping26.attr,
840 &dev_attr_mapping27.attr,
841 &dev_attr_mapping28.attr,
842 &dev_attr_mapping29.attr,
843 &dev_attr_mapping30.attr,
844 &dev_attr_mapping31.attr,
845 NULL,
848 struct attribute_group nd_mapping_attribute_group = {
849 .is_visible = mapping_visible,
850 .attrs = mapping_attributes,
852 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
854 int nd_blk_region_init(struct nd_region *nd_region)
856 struct device *dev = &nd_region->dev;
857 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
859 if (!is_nd_blk(dev))
860 return 0;
862 if (nd_region->ndr_mappings < 1) {
863 dev_dbg(dev, "invalid BLK region\n");
864 return -ENXIO;
867 return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
871 * nd_region_acquire_lane - allocate and lock a lane
872 * @nd_region: region id and number of lanes possible
874 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
875 * We optimize for the common case where there are 256 lanes, one
876 * per-cpu. For larger systems we need to lock to share lanes. For now
877 * this implementation assumes the cost of maintaining an allocator for
878 * free lanes is on the order of the lock hold time, so it implements a
879 * static lane = cpu % num_lanes mapping.
881 * In the case of a BTT instance on top of a BLK namespace a lane may be
882 * acquired recursively. We lock on the first instance.
884 * In the case of a BTT instance on top of PMEM, we only acquire a lane
885 * for the BTT metadata updates.
887 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
889 unsigned int cpu, lane;
891 cpu = get_cpu();
892 if (nd_region->num_lanes < nr_cpu_ids) {
893 struct nd_percpu_lane *ndl_lock, *ndl_count;
895 lane = cpu % nd_region->num_lanes;
896 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
897 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
898 if (ndl_count->count++ == 0)
899 spin_lock(&ndl_lock->lock);
900 } else
901 lane = cpu;
903 return lane;
905 EXPORT_SYMBOL(nd_region_acquire_lane);
907 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
909 if (nd_region->num_lanes < nr_cpu_ids) {
910 unsigned int cpu = get_cpu();
911 struct nd_percpu_lane *ndl_lock, *ndl_count;
913 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
914 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
915 if (--ndl_count->count == 0)
916 spin_unlock(&ndl_lock->lock);
917 put_cpu();
919 put_cpu();
921 EXPORT_SYMBOL(nd_region_release_lane);
923 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
924 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
925 const char *caller)
927 struct nd_region *nd_region;
928 struct device *dev;
929 void *region_buf;
930 unsigned int i;
931 int ro = 0;
933 for (i = 0; i < ndr_desc->num_mappings; i++) {
934 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
935 struct nvdimm *nvdimm = mapping->nvdimm;
937 if ((mapping->start | mapping->size) % SZ_4K) {
938 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
939 caller, dev_name(&nvdimm->dev), i);
941 return NULL;
944 if (test_bit(NDD_UNARMED, &nvdimm->flags))
945 ro = 1;
948 if (dev_type == &nd_blk_device_type) {
949 struct nd_blk_region_desc *ndbr_desc;
950 struct nd_blk_region *ndbr;
952 ndbr_desc = to_blk_region_desc(ndr_desc);
953 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
954 * ndr_desc->num_mappings,
955 GFP_KERNEL);
956 if (ndbr) {
957 nd_region = &ndbr->nd_region;
958 ndbr->enable = ndbr_desc->enable;
959 ndbr->do_io = ndbr_desc->do_io;
961 region_buf = ndbr;
962 } else {
963 nd_region = kzalloc(sizeof(struct nd_region)
964 + sizeof(struct nd_mapping)
965 * ndr_desc->num_mappings,
966 GFP_KERNEL);
967 region_buf = nd_region;
970 if (!region_buf)
971 return NULL;
972 nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
973 if (nd_region->id < 0)
974 goto err_id;
976 nd_region->lane = alloc_percpu(struct nd_percpu_lane);
977 if (!nd_region->lane)
978 goto err_percpu;
980 for (i = 0; i < nr_cpu_ids; i++) {
981 struct nd_percpu_lane *ndl;
983 ndl = per_cpu_ptr(nd_region->lane, i);
984 spin_lock_init(&ndl->lock);
985 ndl->count = 0;
988 for (i = 0; i < ndr_desc->num_mappings; i++) {
989 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
990 struct nvdimm *nvdimm = mapping->nvdimm;
992 nd_region->mapping[i].nvdimm = nvdimm;
993 nd_region->mapping[i].start = mapping->start;
994 nd_region->mapping[i].size = mapping->size;
995 nd_region->mapping[i].position = mapping->position;
996 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
997 mutex_init(&nd_region->mapping[i].lock);
999 get_device(&nvdimm->dev);
1001 nd_region->ndr_mappings = ndr_desc->num_mappings;
1002 nd_region->provider_data = ndr_desc->provider_data;
1003 nd_region->nd_set = ndr_desc->nd_set;
1004 nd_region->num_lanes = ndr_desc->num_lanes;
1005 nd_region->flags = ndr_desc->flags;
1006 nd_region->ro = ro;
1007 nd_region->numa_node = ndr_desc->numa_node;
1008 ida_init(&nd_region->ns_ida);
1009 ida_init(&nd_region->btt_ida);
1010 ida_init(&nd_region->pfn_ida);
1011 ida_init(&nd_region->dax_ida);
1012 dev = &nd_region->dev;
1013 dev_set_name(dev, "region%d", nd_region->id);
1014 dev->parent = &nvdimm_bus->dev;
1015 dev->type = dev_type;
1016 dev->groups = ndr_desc->attr_groups;
1017 nd_region->ndr_size = resource_size(ndr_desc->res);
1018 nd_region->ndr_start = ndr_desc->res->start;
1019 nd_device_register(dev);
1021 return nd_region;
1023 err_percpu:
1024 ida_simple_remove(&region_ida, nd_region->id);
1025 err_id:
1026 kfree(region_buf);
1027 return NULL;
1030 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1031 struct nd_region_desc *ndr_desc)
1033 ndr_desc->num_lanes = ND_MAX_LANES;
1034 return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1035 __func__);
1037 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1039 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1040 struct nd_region_desc *ndr_desc)
1042 if (ndr_desc->num_mappings > 1)
1043 return NULL;
1044 ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1045 return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1046 __func__);
1048 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1050 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1051 struct nd_region_desc *ndr_desc)
1053 ndr_desc->num_lanes = ND_MAX_LANES;
1054 return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1055 __func__);
1057 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1060 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1061 * @nd_region: blk or interleaved pmem region
1063 void nvdimm_flush(struct nd_region *nd_region)
1065 struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1066 int i, idx;
1069 * Try to encourage some diversity in flush hint addresses
1070 * across cpus assuming a limited number of flush hints.
1072 idx = this_cpu_read(flush_idx);
1073 idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1076 * The first wmb() is needed to 'sfence' all previous writes
1077 * such that they are architecturally visible for the platform
1078 * buffer flush. Note that we've already arranged for pmem
1079 * writes to avoid the cache via memcpy_flushcache(). The final
1080 * wmb() ensures ordering for the NVDIMM flush write.
1082 wmb();
1083 for (i = 0; i < nd_region->ndr_mappings; i++)
1084 if (ndrd_get_flush_wpq(ndrd, i, 0))
1085 writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1086 wmb();
1088 EXPORT_SYMBOL_GPL(nvdimm_flush);
1091 * nvdimm_has_flush - determine write flushing requirements
1092 * @nd_region: blk or interleaved pmem region
1094 * Returns 1 if writes require flushing
1095 * Returns 0 if writes do not require flushing
1096 * Returns -ENXIO if flushing capability can not be determined
1098 int nvdimm_has_flush(struct nd_region *nd_region)
1100 int i;
1102 /* no nvdimm or pmem api == flushing capability unknown */
1103 if (nd_region->ndr_mappings == 0
1104 || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1105 return -ENXIO;
1107 for (i = 0; i < nd_region->ndr_mappings; i++) {
1108 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1109 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1111 /* flush hints present / available */
1112 if (nvdimm->num_flush)
1113 return 1;
1117 * The platform defines dimm devices without hints, assume
1118 * platform persistence mechanism like ADR
1120 return 0;
1122 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1124 int nvdimm_has_cache(struct nd_region *nd_region)
1126 return is_nd_pmem(&nd_region->dev);
1128 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1130 void __exit nd_region_devs_exit(void)
1132 ida_destroy(&region_ida);