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>
25 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
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)
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);
55 flush_page = (void __iomem *) ((unsigned long)
56 ndrd_get_flush_wpq(ndrd, dimm, j)
59 flush_page = devm_nvdimm_ioremap(dev,
60 PFN_PHYS(pfn), PAGE_SIZE);
63 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
64 + (res->start & ~PAGE_MASK));
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)
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);
94 dev_set_drvdata(dev, ndrd);
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);
110 * Clear out entries that are duplicates. This should prevent the
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))
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);
127 static void nd_region_release(struct device *dev)
129 struct nd_region *nd_region = to_nd_region(dev);
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(®ion_ida, nd_region->id);
141 kfree(to_nd_blk_region(dev));
146 static struct device_type nd_blk_device_type = {
148 .release = nd_region_release,
151 static struct device_type nd_pmem_device_type = {
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);
183 EXPORT_SYMBOL_GPL(to_nd_region);
185 struct device *nd_region_dev(struct nd_region *nd_region)
189 return &nd_region->dev;
191 EXPORT_SYMBOL_GPL(nd_region_dev);
193 struct nd_blk_region *to_nd_blk_region(struct device *dev)
195 struct nd_region *nd_region = to_nd_region(dev);
197 WARN_ON(!is_nd_blk(dev));
198 return container_of(nd_region, struct nd_blk_region, nd_region);
200 EXPORT_SYMBOL_GPL(to_nd_blk_region);
202 void *nd_region_provider_data(struct nd_region *nd_region)
204 return nd_region->provider_data;
206 EXPORT_SYMBOL_GPL(nd_region_provider_data);
208 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
210 return ndbr->blk_provider_data;
212 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
214 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
216 ndbr->blk_provider_data = data;
218 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
221 * nd_region_to_nstype() - region to an integer namespace type
222 * @nd_region: region-device to interrogate
224 * This is the 'nstype' attribute of a region as well, an input to the
225 * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
226 * namespace devices with namespace drivers.
228 int nd_region_to_nstype(struct nd_region *nd_region)
230 if (is_memory(&nd_region->dev)) {
233 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
234 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
235 struct nvdimm *nvdimm = nd_mapping->nvdimm;
237 if (test_bit(NDD_ALIASING, &nvdimm->flags))
241 return ND_DEVICE_NAMESPACE_PMEM;
243 return ND_DEVICE_NAMESPACE_IO;
244 } else if (is_nd_blk(&nd_region->dev)) {
245 return ND_DEVICE_NAMESPACE_BLK;
250 EXPORT_SYMBOL(nd_region_to_nstype);
252 static ssize_t size_show(struct device *dev,
253 struct device_attribute *attr, char *buf)
255 struct nd_region *nd_region = to_nd_region(dev);
256 unsigned long long size = 0;
258 if (is_memory(dev)) {
259 size = nd_region->ndr_size;
260 } else if (nd_region->ndr_mappings == 1) {
261 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
263 size = nd_mapping->size;
266 return sprintf(buf, "%llu\n", size);
268 static DEVICE_ATTR_RO(size);
270 static ssize_t deep_flush_show(struct device *dev,
271 struct device_attribute *attr, char *buf)
273 struct nd_region *nd_region = to_nd_region(dev);
276 * NOTE: in the nvdimm_has_flush() error case this attribute is
279 return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
282 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
283 const char *buf, size_t len)
286 int rc = strtobool(buf, &flush);
287 struct nd_region *nd_region = to_nd_region(dev);
293 nvdimm_flush(nd_region);
297 static DEVICE_ATTR_RW(deep_flush);
299 static ssize_t mappings_show(struct device *dev,
300 struct device_attribute *attr, char *buf)
302 struct nd_region *nd_region = to_nd_region(dev);
304 return sprintf(buf, "%d\n", nd_region->ndr_mappings);
306 static DEVICE_ATTR_RO(mappings);
308 static ssize_t nstype_show(struct device *dev,
309 struct device_attribute *attr, char *buf)
311 struct nd_region *nd_region = to_nd_region(dev);
313 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
315 static DEVICE_ATTR_RO(nstype);
317 static ssize_t set_cookie_show(struct device *dev,
318 struct device_attribute *attr, char *buf)
320 struct nd_region *nd_region = to_nd_region(dev);
321 struct nd_interleave_set *nd_set = nd_region->nd_set;
324 if (is_memory(dev) && nd_set)
325 /* pass, should be precluded by region_visible */;
330 * The cookie to show depends on which specification of the
331 * labels we are using. If there are not labels then default to
332 * the v1.1 namespace label cookie definition. To read all this
333 * data we need to wait for probing to settle.
336 nvdimm_bus_lock(dev);
337 wait_nvdimm_bus_probe_idle(dev);
338 if (nd_region->ndr_mappings) {
339 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
340 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
343 struct nd_namespace_index *nsindex;
345 nsindex = to_namespace_index(ndd, ndd->ns_current);
346 rc = sprintf(buf, "%#llx\n",
347 nd_region_interleave_set_cookie(nd_region,
351 nvdimm_bus_unlock(dev);
356 return sprintf(buf, "%#llx\n", nd_set->cookie1);
358 static DEVICE_ATTR_RO(set_cookie);
360 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
362 resource_size_t blk_max_overlap = 0, available, overlap;
365 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
369 overlap = blk_max_overlap;
370 for (i = 0; i < nd_region->ndr_mappings; i++) {
371 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
372 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
374 /* if a dimm is disabled the available capacity is zero */
378 if (is_memory(&nd_region->dev)) {
379 available += nd_pmem_available_dpa(nd_region,
380 nd_mapping, &overlap);
381 if (overlap > blk_max_overlap) {
382 blk_max_overlap = overlap;
385 } else if (is_nd_blk(&nd_region->dev))
386 available += nd_blk_available_dpa(nd_region);
392 static ssize_t available_size_show(struct device *dev,
393 struct device_attribute *attr, char *buf)
395 struct nd_region *nd_region = to_nd_region(dev);
396 unsigned long long available = 0;
399 * Flush in-flight updates and grab a snapshot of the available
400 * size. Of course, this value is potentially invalidated the
401 * memory nvdimm_bus_lock() is dropped, but that's userspace's
402 * problem to not race itself.
404 nvdimm_bus_lock(dev);
405 wait_nvdimm_bus_probe_idle(dev);
406 available = nd_region_available_dpa(nd_region);
407 nvdimm_bus_unlock(dev);
409 return sprintf(buf, "%llu\n", available);
411 static DEVICE_ATTR_RO(available_size);
413 static ssize_t init_namespaces_show(struct device *dev,
414 struct device_attribute *attr, char *buf)
416 struct nd_region_data *ndrd = dev_get_drvdata(dev);
419 nvdimm_bus_lock(dev);
421 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
424 nvdimm_bus_unlock(dev);
428 static DEVICE_ATTR_RO(init_namespaces);
430 static ssize_t namespace_seed_show(struct device *dev,
431 struct device_attribute *attr, char *buf)
433 struct nd_region *nd_region = to_nd_region(dev);
436 nvdimm_bus_lock(dev);
437 if (nd_region->ns_seed)
438 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
440 rc = sprintf(buf, "\n");
441 nvdimm_bus_unlock(dev);
444 static DEVICE_ATTR_RO(namespace_seed);
446 static ssize_t btt_seed_show(struct device *dev,
447 struct device_attribute *attr, char *buf)
449 struct nd_region *nd_region = to_nd_region(dev);
452 nvdimm_bus_lock(dev);
453 if (nd_region->btt_seed)
454 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
456 rc = sprintf(buf, "\n");
457 nvdimm_bus_unlock(dev);
461 static DEVICE_ATTR_RO(btt_seed);
463 static ssize_t pfn_seed_show(struct device *dev,
464 struct device_attribute *attr, char *buf)
466 struct nd_region *nd_region = to_nd_region(dev);
469 nvdimm_bus_lock(dev);
470 if (nd_region->pfn_seed)
471 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
473 rc = sprintf(buf, "\n");
474 nvdimm_bus_unlock(dev);
478 static DEVICE_ATTR_RO(pfn_seed);
480 static ssize_t dax_seed_show(struct device *dev,
481 struct device_attribute *attr, char *buf)
483 struct nd_region *nd_region = to_nd_region(dev);
486 nvdimm_bus_lock(dev);
487 if (nd_region->dax_seed)
488 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
490 rc = sprintf(buf, "\n");
491 nvdimm_bus_unlock(dev);
495 static DEVICE_ATTR_RO(dax_seed);
497 static ssize_t read_only_show(struct device *dev,
498 struct device_attribute *attr, char *buf)
500 struct nd_region *nd_region = to_nd_region(dev);
502 return sprintf(buf, "%d\n", nd_region->ro);
505 static ssize_t read_only_store(struct device *dev,
506 struct device_attribute *attr, const char *buf, size_t len)
509 int rc = strtobool(buf, &ro);
510 struct nd_region *nd_region = to_nd_region(dev);
518 static DEVICE_ATTR_RW(read_only);
520 static ssize_t region_badblocks_show(struct device *dev,
521 struct device_attribute *attr, char *buf)
523 struct nd_region *nd_region = to_nd_region(dev);
525 return badblocks_show(&nd_region->bb, buf, 0);
528 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
530 static ssize_t resource_show(struct device *dev,
531 struct device_attribute *attr, char *buf)
533 struct nd_region *nd_region = to_nd_region(dev);
535 return sprintf(buf, "%#llx\n", nd_region->ndr_start);
537 static DEVICE_ATTR_RO(resource);
539 static ssize_t persistence_domain_show(struct device *dev,
540 struct device_attribute *attr, char *buf)
542 struct nd_region *nd_region = to_nd_region(dev);
544 if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
545 return sprintf(buf, "cpu_cache\n");
546 else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
547 return sprintf(buf, "memory_controller\n");
549 return sprintf(buf, "\n");
551 static DEVICE_ATTR_RO(persistence_domain);
553 static struct attribute *nd_region_attributes[] = {
555 &dev_attr_nstype.attr,
556 &dev_attr_mappings.attr,
557 &dev_attr_btt_seed.attr,
558 &dev_attr_pfn_seed.attr,
559 &dev_attr_dax_seed.attr,
560 &dev_attr_deep_flush.attr,
561 &dev_attr_read_only.attr,
562 &dev_attr_set_cookie.attr,
563 &dev_attr_available_size.attr,
564 &dev_attr_namespace_seed.attr,
565 &dev_attr_init_namespaces.attr,
566 &dev_attr_badblocks.attr,
567 &dev_attr_resource.attr,
568 &dev_attr_persistence_domain.attr,
572 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
574 struct device *dev = container_of(kobj, typeof(*dev), kobj);
575 struct nd_region *nd_region = to_nd_region(dev);
576 struct nd_interleave_set *nd_set = nd_region->nd_set;
577 int type = nd_region_to_nstype(nd_region);
579 if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
582 if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
585 if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
588 if (a == &dev_attr_resource.attr) {
595 if (a == &dev_attr_deep_flush.attr) {
596 int has_flush = nvdimm_has_flush(nd_region);
600 else if (has_flush == 0)
606 if (a == &dev_attr_persistence_domain.attr) {
607 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
608 | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
613 if (a != &dev_attr_set_cookie.attr
614 && a != &dev_attr_available_size.attr)
617 if ((type == ND_DEVICE_NAMESPACE_PMEM
618 || type == ND_DEVICE_NAMESPACE_BLK)
619 && a == &dev_attr_available_size.attr)
621 else if (is_memory(dev) && nd_set)
627 struct attribute_group nd_region_attribute_group = {
628 .attrs = nd_region_attributes,
629 .is_visible = region_visible,
631 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
633 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
634 struct nd_namespace_index *nsindex)
636 struct nd_interleave_set *nd_set = nd_region->nd_set;
641 if (nsindex && __le16_to_cpu(nsindex->major) == 1
642 && __le16_to_cpu(nsindex->minor) == 1)
643 return nd_set->cookie1;
644 return nd_set->cookie2;
647 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
649 struct nd_interleave_set *nd_set = nd_region->nd_set;
652 return nd_set->altcookie;
656 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
658 struct nd_label_ent *label_ent, *e;
660 lockdep_assert_held(&nd_mapping->lock);
661 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
662 list_del(&label_ent->list);
668 * Upon successful probe/remove, take/release a reference on the
669 * associated interleave set (if present), and plant new btt + namespace
670 * seeds. Also, on the removal of a BLK region, notify the provider to
671 * disable the region.
673 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
674 struct device *dev, bool probe)
676 struct nd_region *nd_region;
678 if (!probe && is_nd_region(dev)) {
681 nd_region = to_nd_region(dev);
682 for (i = 0; i < nd_region->ndr_mappings; i++) {
683 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
684 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
685 struct nvdimm *nvdimm = nd_mapping->nvdimm;
687 mutex_lock(&nd_mapping->lock);
688 nd_mapping_free_labels(nd_mapping);
689 mutex_unlock(&nd_mapping->lock);
692 nd_mapping->ndd = NULL;
694 atomic_dec(&nvdimm->busy);
697 if (dev->parent && is_nd_region(dev->parent) && probe) {
698 nd_region = to_nd_region(dev->parent);
699 nvdimm_bus_lock(dev);
700 if (nd_region->ns_seed == dev)
701 nd_region_create_ns_seed(nd_region);
702 nvdimm_bus_unlock(dev);
704 if (is_nd_btt(dev) && probe) {
705 struct nd_btt *nd_btt = to_nd_btt(dev);
707 nd_region = to_nd_region(dev->parent);
708 nvdimm_bus_lock(dev);
709 if (nd_region->btt_seed == dev)
710 nd_region_create_btt_seed(nd_region);
711 if (nd_region->ns_seed == &nd_btt->ndns->dev)
712 nd_region_create_ns_seed(nd_region);
713 nvdimm_bus_unlock(dev);
715 if (is_nd_pfn(dev) && probe) {
716 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
718 nd_region = to_nd_region(dev->parent);
719 nvdimm_bus_lock(dev);
720 if (nd_region->pfn_seed == dev)
721 nd_region_create_pfn_seed(nd_region);
722 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
723 nd_region_create_ns_seed(nd_region);
724 nvdimm_bus_unlock(dev);
726 if (is_nd_dax(dev) && probe) {
727 struct nd_dax *nd_dax = to_nd_dax(dev);
729 nd_region = to_nd_region(dev->parent);
730 nvdimm_bus_lock(dev);
731 if (nd_region->dax_seed == dev)
732 nd_region_create_dax_seed(nd_region);
733 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
734 nd_region_create_ns_seed(nd_region);
735 nvdimm_bus_unlock(dev);
739 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
741 nd_region_notify_driver_action(nvdimm_bus, dev, true);
744 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
746 nd_region_notify_driver_action(nvdimm_bus, dev, false);
749 static ssize_t mappingN(struct device *dev, char *buf, int n)
751 struct nd_region *nd_region = to_nd_region(dev);
752 struct nd_mapping *nd_mapping;
753 struct nvdimm *nvdimm;
755 if (n >= nd_region->ndr_mappings)
757 nd_mapping = &nd_region->mapping[n];
758 nvdimm = nd_mapping->nvdimm;
760 return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
761 nd_mapping->start, nd_mapping->size,
762 nd_mapping->position);
765 #define REGION_MAPPING(idx) \
766 static ssize_t mapping##idx##_show(struct device *dev, \
767 struct device_attribute *attr, char *buf) \
769 return mappingN(dev, buf, idx); \
771 static DEVICE_ATTR_RO(mapping##idx)
774 * 32 should be enough for a while, even in the presence of socket
775 * interleave a 32-way interleave set is a degenerate case.
810 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
812 struct device *dev = container_of(kobj, struct device, kobj);
813 struct nd_region *nd_region = to_nd_region(dev);
815 if (n < nd_region->ndr_mappings)
820 static struct attribute *mapping_attributes[] = {
821 &dev_attr_mapping0.attr,
822 &dev_attr_mapping1.attr,
823 &dev_attr_mapping2.attr,
824 &dev_attr_mapping3.attr,
825 &dev_attr_mapping4.attr,
826 &dev_attr_mapping5.attr,
827 &dev_attr_mapping6.attr,
828 &dev_attr_mapping7.attr,
829 &dev_attr_mapping8.attr,
830 &dev_attr_mapping9.attr,
831 &dev_attr_mapping10.attr,
832 &dev_attr_mapping11.attr,
833 &dev_attr_mapping12.attr,
834 &dev_attr_mapping13.attr,
835 &dev_attr_mapping14.attr,
836 &dev_attr_mapping15.attr,
837 &dev_attr_mapping16.attr,
838 &dev_attr_mapping17.attr,
839 &dev_attr_mapping18.attr,
840 &dev_attr_mapping19.attr,
841 &dev_attr_mapping20.attr,
842 &dev_attr_mapping21.attr,
843 &dev_attr_mapping22.attr,
844 &dev_attr_mapping23.attr,
845 &dev_attr_mapping24.attr,
846 &dev_attr_mapping25.attr,
847 &dev_attr_mapping26.attr,
848 &dev_attr_mapping27.attr,
849 &dev_attr_mapping28.attr,
850 &dev_attr_mapping29.attr,
851 &dev_attr_mapping30.attr,
852 &dev_attr_mapping31.attr,
856 struct attribute_group nd_mapping_attribute_group = {
857 .is_visible = mapping_visible,
858 .attrs = mapping_attributes,
860 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
862 int nd_blk_region_init(struct nd_region *nd_region)
864 struct device *dev = &nd_region->dev;
865 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
870 if (nd_region->ndr_mappings < 1) {
871 dev_dbg(dev, "invalid BLK region\n");
875 return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
879 * nd_region_acquire_lane - allocate and lock a lane
880 * @nd_region: region id and number of lanes possible
882 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
883 * We optimize for the common case where there are 256 lanes, one
884 * per-cpu. For larger systems we need to lock to share lanes. For now
885 * this implementation assumes the cost of maintaining an allocator for
886 * free lanes is on the order of the lock hold time, so it implements a
887 * static lane = cpu % num_lanes mapping.
889 * In the case of a BTT instance on top of a BLK namespace a lane may be
890 * acquired recursively. We lock on the first instance.
892 * In the case of a BTT instance on top of PMEM, we only acquire a lane
893 * for the BTT metadata updates.
895 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
897 unsigned int cpu, lane;
900 if (nd_region->num_lanes < nr_cpu_ids) {
901 struct nd_percpu_lane *ndl_lock, *ndl_count;
903 lane = cpu % nd_region->num_lanes;
904 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
905 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
906 if (ndl_count->count++ == 0)
907 spin_lock(&ndl_lock->lock);
913 EXPORT_SYMBOL(nd_region_acquire_lane);
915 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
917 if (nd_region->num_lanes < nr_cpu_ids) {
918 unsigned int cpu = get_cpu();
919 struct nd_percpu_lane *ndl_lock, *ndl_count;
921 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
922 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
923 if (--ndl_count->count == 0)
924 spin_unlock(&ndl_lock->lock);
929 EXPORT_SYMBOL(nd_region_release_lane);
931 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
932 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
935 struct nd_region *nd_region;
941 for (i = 0; i < ndr_desc->num_mappings; i++) {
942 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
943 struct nvdimm *nvdimm = mapping->nvdimm;
945 if ((mapping->start | mapping->size) % SZ_4K) {
946 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
947 caller, dev_name(&nvdimm->dev), i);
952 if (test_bit(NDD_UNARMED, &nvdimm->flags))
956 if (dev_type == &nd_blk_device_type) {
957 struct nd_blk_region_desc *ndbr_desc;
958 struct nd_blk_region *ndbr;
960 ndbr_desc = to_blk_region_desc(ndr_desc);
961 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
962 * ndr_desc->num_mappings,
965 nd_region = &ndbr->nd_region;
966 ndbr->enable = ndbr_desc->enable;
967 ndbr->do_io = ndbr_desc->do_io;
971 nd_region = kzalloc(sizeof(struct nd_region)
972 + sizeof(struct nd_mapping)
973 * ndr_desc->num_mappings,
975 region_buf = nd_region;
980 nd_region->id = ida_simple_get(®ion_ida, 0, 0, GFP_KERNEL);
981 if (nd_region->id < 0)
984 nd_region->lane = alloc_percpu(struct nd_percpu_lane);
985 if (!nd_region->lane)
988 for (i = 0; i < nr_cpu_ids; i++) {
989 struct nd_percpu_lane *ndl;
991 ndl = per_cpu_ptr(nd_region->lane, i);
992 spin_lock_init(&ndl->lock);
996 for (i = 0; i < ndr_desc->num_mappings; i++) {
997 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
998 struct nvdimm *nvdimm = mapping->nvdimm;
1000 nd_region->mapping[i].nvdimm = nvdimm;
1001 nd_region->mapping[i].start = mapping->start;
1002 nd_region->mapping[i].size = mapping->size;
1003 nd_region->mapping[i].position = mapping->position;
1004 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1005 mutex_init(&nd_region->mapping[i].lock);
1007 get_device(&nvdimm->dev);
1009 nd_region->ndr_mappings = ndr_desc->num_mappings;
1010 nd_region->provider_data = ndr_desc->provider_data;
1011 nd_region->nd_set = ndr_desc->nd_set;
1012 nd_region->num_lanes = ndr_desc->num_lanes;
1013 nd_region->flags = ndr_desc->flags;
1015 nd_region->numa_node = ndr_desc->numa_node;
1016 ida_init(&nd_region->ns_ida);
1017 ida_init(&nd_region->btt_ida);
1018 ida_init(&nd_region->pfn_ida);
1019 ida_init(&nd_region->dax_ida);
1020 dev = &nd_region->dev;
1021 dev_set_name(dev, "region%d", nd_region->id);
1022 dev->parent = &nvdimm_bus->dev;
1023 dev->type = dev_type;
1024 dev->groups = ndr_desc->attr_groups;
1025 dev->of_node = ndr_desc->of_node;
1026 nd_region->ndr_size = resource_size(ndr_desc->res);
1027 nd_region->ndr_start = ndr_desc->res->start;
1028 nd_device_register(dev);
1033 ida_simple_remove(®ion_ida, nd_region->id);
1039 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1040 struct nd_region_desc *ndr_desc)
1042 ndr_desc->num_lanes = ND_MAX_LANES;
1043 return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1046 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1048 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1049 struct nd_region_desc *ndr_desc)
1051 if (ndr_desc->num_mappings > 1)
1053 ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1054 return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1057 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1059 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1060 struct nd_region_desc *ndr_desc)
1062 ndr_desc->num_lanes = ND_MAX_LANES;
1063 return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1066 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1069 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1070 * @nd_region: blk or interleaved pmem region
1072 void nvdimm_flush(struct nd_region *nd_region)
1074 struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1078 * Try to encourage some diversity in flush hint addresses
1079 * across cpus assuming a limited number of flush hints.
1081 idx = this_cpu_read(flush_idx);
1082 idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1085 * The first wmb() is needed to 'sfence' all previous writes
1086 * such that they are architecturally visible for the platform
1087 * buffer flush. Note that we've already arranged for pmem
1088 * writes to avoid the cache via memcpy_flushcache(). The final
1089 * wmb() ensures ordering for the NVDIMM flush write.
1092 for (i = 0; i < nd_region->ndr_mappings; i++)
1093 if (ndrd_get_flush_wpq(ndrd, i, 0))
1094 writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1097 EXPORT_SYMBOL_GPL(nvdimm_flush);
1100 * nvdimm_has_flush - determine write flushing requirements
1101 * @nd_region: blk or interleaved pmem region
1103 * Returns 1 if writes require flushing
1104 * Returns 0 if writes do not require flushing
1105 * Returns -ENXIO if flushing capability can not be determined
1107 int nvdimm_has_flush(struct nd_region *nd_region)
1111 /* no nvdimm or pmem api == flushing capability unknown */
1112 if (nd_region->ndr_mappings == 0
1113 || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1116 for (i = 0; i < nd_region->ndr_mappings; i++) {
1117 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1118 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1120 /* flush hints present / available */
1121 if (nvdimm->num_flush)
1126 * The platform defines dimm devices without hints, assume
1127 * platform persistence mechanism like ADR
1131 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1133 int nvdimm_has_cache(struct nd_region *nd_region)
1135 return is_nd_pmem(&nd_region->dev) &&
1136 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1138 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1140 void __exit nd_region_devs_exit(void)
1142 ida_destroy(®ion_ida);