1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
5 #include <linux/scatterlist.h>
6 #include <linux/memregion.h>
7 #include <linux/highmem.h>
8 #include <linux/sched.h>
9 #include <linux/slab.h>
10 #include <linux/hash.h>
11 #include <linux/sort.h>
18 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
21 #include <linux/io-64-nonatomic-hi-lo.h>
23 static DEFINE_PER_CPU(int, flush_idx);
25 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
26 struct nd_region_data *ndrd)
30 dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
31 nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
32 for (i = 0; i < (1 << ndrd->hints_shift); i++) {
33 struct resource *res = &nvdimm->flush_wpq[i];
34 unsigned long pfn = PHYS_PFN(res->start);
35 void __iomem *flush_page;
37 /* check if flush hints share a page */
38 for (j = 0; j < i; j++) {
39 struct resource *res_j = &nvdimm->flush_wpq[j];
40 unsigned long pfn_j = PHYS_PFN(res_j->start);
47 flush_page = (void __iomem *) ((unsigned long)
48 ndrd_get_flush_wpq(ndrd, dimm, j)
51 flush_page = devm_nvdimm_ioremap(dev,
52 PFN_PHYS(pfn), PAGE_SIZE);
55 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
56 + (res->start & ~PAGE_MASK));
62 int nd_region_activate(struct nd_region *nd_region)
64 int i, j, num_flush = 0;
65 struct nd_region_data *ndrd;
66 struct device *dev = &nd_region->dev;
67 size_t flush_data_size = sizeof(void *);
69 nvdimm_bus_lock(&nd_region->dev);
70 for (i = 0; i < nd_region->ndr_mappings; i++) {
71 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
72 struct nvdimm *nvdimm = nd_mapping->nvdimm;
74 if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
75 nvdimm_bus_unlock(&nd_region->dev);
79 /* at least one null hint slot per-dimm for the "no-hint" case */
80 flush_data_size += sizeof(void *);
81 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
82 if (!nvdimm->num_flush)
84 flush_data_size += nvdimm->num_flush * sizeof(void *);
86 nvdimm_bus_unlock(&nd_region->dev);
88 ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
91 dev_set_drvdata(dev, ndrd);
96 ndrd->hints_shift = ilog2(num_flush);
97 for (i = 0; i < nd_region->ndr_mappings; i++) {
98 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
99 struct nvdimm *nvdimm = nd_mapping->nvdimm;
100 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
107 * Clear out entries that are duplicates. This should prevent the
110 for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
111 /* ignore if NULL already */
112 if (!ndrd_get_flush_wpq(ndrd, i, 0))
115 for (j = i + 1; j < nd_region->ndr_mappings; j++)
116 if (ndrd_get_flush_wpq(ndrd, i, 0) ==
117 ndrd_get_flush_wpq(ndrd, j, 0))
118 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
124 static void nd_region_release(struct device *dev)
126 struct nd_region *nd_region = to_nd_region(dev);
129 for (i = 0; i < nd_region->ndr_mappings; i++) {
130 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
131 struct nvdimm *nvdimm = nd_mapping->nvdimm;
133 put_device(&nvdimm->dev);
135 free_percpu(nd_region->lane);
136 memregion_free(nd_region->id);
138 kfree(to_nd_blk_region(dev));
143 struct nd_region *to_nd_region(struct device *dev)
145 struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
147 WARN_ON(dev->type->release != nd_region_release);
150 EXPORT_SYMBOL_GPL(to_nd_region);
152 struct device *nd_region_dev(struct nd_region *nd_region)
156 return &nd_region->dev;
158 EXPORT_SYMBOL_GPL(nd_region_dev);
160 struct nd_blk_region *to_nd_blk_region(struct device *dev)
162 struct nd_region *nd_region = to_nd_region(dev);
164 WARN_ON(!is_nd_blk(dev));
165 return container_of(nd_region, struct nd_blk_region, nd_region);
167 EXPORT_SYMBOL_GPL(to_nd_blk_region);
169 void *nd_region_provider_data(struct nd_region *nd_region)
171 return nd_region->provider_data;
173 EXPORT_SYMBOL_GPL(nd_region_provider_data);
175 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
177 return ndbr->blk_provider_data;
179 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
181 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
183 ndbr->blk_provider_data = data;
185 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
188 * nd_region_to_nstype() - region to an integer namespace type
189 * @nd_region: region-device to interrogate
191 * This is the 'nstype' attribute of a region as well, an input to the
192 * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
193 * namespace devices with namespace drivers.
195 int nd_region_to_nstype(struct nd_region *nd_region)
197 if (is_memory(&nd_region->dev)) {
200 for (i = 0, label = 0; i < nd_region->ndr_mappings; i++) {
201 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
202 struct nvdimm *nvdimm = nd_mapping->nvdimm;
204 if (test_bit(NDD_LABELING, &nvdimm->flags))
208 return ND_DEVICE_NAMESPACE_PMEM;
210 return ND_DEVICE_NAMESPACE_IO;
211 } else if (is_nd_blk(&nd_region->dev)) {
212 return ND_DEVICE_NAMESPACE_BLK;
217 EXPORT_SYMBOL(nd_region_to_nstype);
219 static unsigned long long region_size(struct nd_region *nd_region)
221 if (is_memory(&nd_region->dev)) {
222 return nd_region->ndr_size;
223 } else if (nd_region->ndr_mappings == 1) {
224 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
226 return nd_mapping->size;
232 static ssize_t size_show(struct device *dev,
233 struct device_attribute *attr, char *buf)
235 struct nd_region *nd_region = to_nd_region(dev);
237 return sprintf(buf, "%llu\n", region_size(nd_region));
239 static DEVICE_ATTR_RO(size);
241 static ssize_t deep_flush_show(struct device *dev,
242 struct device_attribute *attr, char *buf)
244 struct nd_region *nd_region = to_nd_region(dev);
247 * NOTE: in the nvdimm_has_flush() error case this attribute is
250 return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
253 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
254 const char *buf, size_t len)
257 int rc = strtobool(buf, &flush);
258 struct nd_region *nd_region = to_nd_region(dev);
264 rc = nvdimm_flush(nd_region, NULL);
270 static DEVICE_ATTR_RW(deep_flush);
272 static ssize_t mappings_show(struct device *dev,
273 struct device_attribute *attr, char *buf)
275 struct nd_region *nd_region = to_nd_region(dev);
277 return sprintf(buf, "%d\n", nd_region->ndr_mappings);
279 static DEVICE_ATTR_RO(mappings);
281 static ssize_t nstype_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
284 struct nd_region *nd_region = to_nd_region(dev);
286 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
288 static DEVICE_ATTR_RO(nstype);
290 static ssize_t set_cookie_show(struct device *dev,
291 struct device_attribute *attr, char *buf)
293 struct nd_region *nd_region = to_nd_region(dev);
294 struct nd_interleave_set *nd_set = nd_region->nd_set;
297 if (is_memory(dev) && nd_set)
298 /* pass, should be precluded by region_visible */;
303 * The cookie to show depends on which specification of the
304 * labels we are using. If there are not labels then default to
305 * the v1.1 namespace label cookie definition. To read all this
306 * data we need to wait for probing to settle.
309 nvdimm_bus_lock(dev);
310 wait_nvdimm_bus_probe_idle(dev);
311 if (nd_region->ndr_mappings) {
312 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
313 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
316 struct nd_namespace_index *nsindex;
318 nsindex = to_namespace_index(ndd, ndd->ns_current);
319 rc = sprintf(buf, "%#llx\n",
320 nd_region_interleave_set_cookie(nd_region,
324 nvdimm_bus_unlock(dev);
325 nd_device_unlock(dev);
329 return sprintf(buf, "%#llx\n", nd_set->cookie1);
331 static DEVICE_ATTR_RO(set_cookie);
333 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
335 resource_size_t blk_max_overlap = 0, available, overlap;
338 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
342 overlap = blk_max_overlap;
343 for (i = 0; i < nd_region->ndr_mappings; i++) {
344 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
345 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
347 /* if a dimm is disabled the available capacity is zero */
351 if (is_memory(&nd_region->dev)) {
352 available += nd_pmem_available_dpa(nd_region,
353 nd_mapping, &overlap);
354 if (overlap > blk_max_overlap) {
355 blk_max_overlap = overlap;
358 } else if (is_nd_blk(&nd_region->dev))
359 available += nd_blk_available_dpa(nd_region);
365 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
367 resource_size_t available = 0;
370 if (is_memory(&nd_region->dev))
371 available = PHYS_ADDR_MAX;
373 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
374 for (i = 0; i < nd_region->ndr_mappings; i++) {
375 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
377 if (is_memory(&nd_region->dev))
378 available = min(available,
379 nd_pmem_max_contiguous_dpa(nd_region,
381 else if (is_nd_blk(&nd_region->dev))
382 available += nd_blk_available_dpa(nd_region);
384 if (is_memory(&nd_region->dev))
385 return available * nd_region->ndr_mappings;
389 static ssize_t available_size_show(struct device *dev,
390 struct device_attribute *attr, char *buf)
392 struct nd_region *nd_region = to_nd_region(dev);
393 unsigned long long available = 0;
396 * Flush in-flight updates and grab a snapshot of the available
397 * size. Of course, this value is potentially invalidated the
398 * memory nvdimm_bus_lock() is dropped, but that's userspace's
399 * problem to not race itself.
402 nvdimm_bus_lock(dev);
403 wait_nvdimm_bus_probe_idle(dev);
404 available = nd_region_available_dpa(nd_region);
405 nvdimm_bus_unlock(dev);
406 nd_device_unlock(dev);
408 return sprintf(buf, "%llu\n", available);
410 static DEVICE_ATTR_RO(available_size);
412 static ssize_t max_available_extent_show(struct device *dev,
413 struct device_attribute *attr, char *buf)
415 struct nd_region *nd_region = to_nd_region(dev);
416 unsigned long long available = 0;
419 nvdimm_bus_lock(dev);
420 wait_nvdimm_bus_probe_idle(dev);
421 available = nd_region_allocatable_dpa(nd_region);
422 nvdimm_bus_unlock(dev);
423 nd_device_unlock(dev);
425 return sprintf(buf, "%llu\n", available);
427 static DEVICE_ATTR_RO(max_available_extent);
429 static ssize_t init_namespaces_show(struct device *dev,
430 struct device_attribute *attr, char *buf)
432 struct nd_region_data *ndrd = dev_get_drvdata(dev);
435 nvdimm_bus_lock(dev);
437 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
440 nvdimm_bus_unlock(dev);
444 static DEVICE_ATTR_RO(init_namespaces);
446 static ssize_t namespace_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->ns_seed)
454 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
456 rc = sprintf(buf, "\n");
457 nvdimm_bus_unlock(dev);
460 static DEVICE_ATTR_RO(namespace_seed);
462 static ssize_t btt_seed_show(struct device *dev,
463 struct device_attribute *attr, char *buf)
465 struct nd_region *nd_region = to_nd_region(dev);
468 nvdimm_bus_lock(dev);
469 if (nd_region->btt_seed)
470 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
472 rc = sprintf(buf, "\n");
473 nvdimm_bus_unlock(dev);
477 static DEVICE_ATTR_RO(btt_seed);
479 static ssize_t pfn_seed_show(struct device *dev,
480 struct device_attribute *attr, char *buf)
482 struct nd_region *nd_region = to_nd_region(dev);
485 nvdimm_bus_lock(dev);
486 if (nd_region->pfn_seed)
487 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
489 rc = sprintf(buf, "\n");
490 nvdimm_bus_unlock(dev);
494 static DEVICE_ATTR_RO(pfn_seed);
496 static ssize_t dax_seed_show(struct device *dev,
497 struct device_attribute *attr, char *buf)
499 struct nd_region *nd_region = to_nd_region(dev);
502 nvdimm_bus_lock(dev);
503 if (nd_region->dax_seed)
504 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
506 rc = sprintf(buf, "\n");
507 nvdimm_bus_unlock(dev);
511 static DEVICE_ATTR_RO(dax_seed);
513 static ssize_t read_only_show(struct device *dev,
514 struct device_attribute *attr, char *buf)
516 struct nd_region *nd_region = to_nd_region(dev);
518 return sprintf(buf, "%d\n", nd_region->ro);
521 static int revalidate_read_only(struct device *dev, void *data)
523 nd_device_notify(dev, NVDIMM_REVALIDATE_REGION);
527 static ssize_t read_only_store(struct device *dev,
528 struct device_attribute *attr, const char *buf, size_t len)
531 int rc = strtobool(buf, &ro);
532 struct nd_region *nd_region = to_nd_region(dev);
538 device_for_each_child(dev, NULL, revalidate_read_only);
541 static DEVICE_ATTR_RW(read_only);
543 static ssize_t align_show(struct device *dev,
544 struct device_attribute *attr, char *buf)
546 struct nd_region *nd_region = to_nd_region(dev);
548 return sprintf(buf, "%#lx\n", nd_region->align);
551 static ssize_t align_store(struct device *dev,
552 struct device_attribute *attr, const char *buf, size_t len)
554 struct nd_region *nd_region = to_nd_region(dev);
555 unsigned long val, dpa;
559 rc = kstrtoul(buf, 0, &val);
563 if (!nd_region->ndr_mappings)
567 * Ensure space-align is evenly divisible by the region
568 * interleave-width because the kernel typically has no facility
569 * to determine which DIMM(s), dimm-physical-addresses, would
570 * contribute to the tail capacity in system-physical-address
571 * space for the namespace.
573 dpa = div_u64_rem(val, nd_region->ndr_mappings, &remainder);
574 if (!is_power_of_2(dpa) || dpa < PAGE_SIZE
575 || val > region_size(nd_region) || remainder)
579 * Given that space allocation consults this value multiple
580 * times ensure it does not change for the duration of the
583 nvdimm_bus_lock(dev);
584 nd_region->align = val;
585 nvdimm_bus_unlock(dev);
589 static DEVICE_ATTR_RW(align);
591 static ssize_t region_badblocks_show(struct device *dev,
592 struct device_attribute *attr, char *buf)
594 struct nd_region *nd_region = to_nd_region(dev);
599 rc = badblocks_show(&nd_region->bb, buf, 0);
602 nd_device_unlock(dev);
606 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
608 static ssize_t resource_show(struct device *dev,
609 struct device_attribute *attr, char *buf)
611 struct nd_region *nd_region = to_nd_region(dev);
613 return sprintf(buf, "%#llx\n", nd_region->ndr_start);
615 static DEVICE_ATTR_ADMIN_RO(resource);
617 static ssize_t persistence_domain_show(struct device *dev,
618 struct device_attribute *attr, char *buf)
620 struct nd_region *nd_region = to_nd_region(dev);
622 if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
623 return sprintf(buf, "cpu_cache\n");
624 else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
625 return sprintf(buf, "memory_controller\n");
627 return sprintf(buf, "\n");
629 static DEVICE_ATTR_RO(persistence_domain);
631 static struct attribute *nd_region_attributes[] = {
633 &dev_attr_align.attr,
634 &dev_attr_nstype.attr,
635 &dev_attr_mappings.attr,
636 &dev_attr_btt_seed.attr,
637 &dev_attr_pfn_seed.attr,
638 &dev_attr_dax_seed.attr,
639 &dev_attr_deep_flush.attr,
640 &dev_attr_read_only.attr,
641 &dev_attr_set_cookie.attr,
642 &dev_attr_available_size.attr,
643 &dev_attr_max_available_extent.attr,
644 &dev_attr_namespace_seed.attr,
645 &dev_attr_init_namespaces.attr,
646 &dev_attr_badblocks.attr,
647 &dev_attr_resource.attr,
648 &dev_attr_persistence_domain.attr,
652 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
654 struct device *dev = container_of(kobj, typeof(*dev), kobj);
655 struct nd_region *nd_region = to_nd_region(dev);
656 struct nd_interleave_set *nd_set = nd_region->nd_set;
657 int type = nd_region_to_nstype(nd_region);
659 if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
662 if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
665 if (!is_memory(dev) && a == &dev_attr_badblocks.attr)
668 if (a == &dev_attr_resource.attr && !is_memory(dev))
671 if (a == &dev_attr_deep_flush.attr) {
672 int has_flush = nvdimm_has_flush(nd_region);
676 else if (has_flush == 0)
682 if (a == &dev_attr_persistence_domain.attr) {
683 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
684 | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
689 if (a == &dev_attr_align.attr)
692 if (a != &dev_attr_set_cookie.attr
693 && a != &dev_attr_available_size.attr)
696 if ((type == ND_DEVICE_NAMESPACE_PMEM
697 || type == ND_DEVICE_NAMESPACE_BLK)
698 && a == &dev_attr_available_size.attr)
700 else if (is_memory(dev) && nd_set)
706 static ssize_t mappingN(struct device *dev, char *buf, int n)
708 struct nd_region *nd_region = to_nd_region(dev);
709 struct nd_mapping *nd_mapping;
710 struct nvdimm *nvdimm;
712 if (n >= nd_region->ndr_mappings)
714 nd_mapping = &nd_region->mapping[n];
715 nvdimm = nd_mapping->nvdimm;
717 return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
718 nd_mapping->start, nd_mapping->size,
719 nd_mapping->position);
722 #define REGION_MAPPING(idx) \
723 static ssize_t mapping##idx##_show(struct device *dev, \
724 struct device_attribute *attr, char *buf) \
726 return mappingN(dev, buf, idx); \
728 static DEVICE_ATTR_RO(mapping##idx)
731 * 32 should be enough for a while, even in the presence of socket
732 * interleave a 32-way interleave set is a degenerate case.
767 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
769 struct device *dev = container_of(kobj, struct device, kobj);
770 struct nd_region *nd_region = to_nd_region(dev);
772 if (n < nd_region->ndr_mappings)
777 static struct attribute *mapping_attributes[] = {
778 &dev_attr_mapping0.attr,
779 &dev_attr_mapping1.attr,
780 &dev_attr_mapping2.attr,
781 &dev_attr_mapping3.attr,
782 &dev_attr_mapping4.attr,
783 &dev_attr_mapping5.attr,
784 &dev_attr_mapping6.attr,
785 &dev_attr_mapping7.attr,
786 &dev_attr_mapping8.attr,
787 &dev_attr_mapping9.attr,
788 &dev_attr_mapping10.attr,
789 &dev_attr_mapping11.attr,
790 &dev_attr_mapping12.attr,
791 &dev_attr_mapping13.attr,
792 &dev_attr_mapping14.attr,
793 &dev_attr_mapping15.attr,
794 &dev_attr_mapping16.attr,
795 &dev_attr_mapping17.attr,
796 &dev_attr_mapping18.attr,
797 &dev_attr_mapping19.attr,
798 &dev_attr_mapping20.attr,
799 &dev_attr_mapping21.attr,
800 &dev_attr_mapping22.attr,
801 &dev_attr_mapping23.attr,
802 &dev_attr_mapping24.attr,
803 &dev_attr_mapping25.attr,
804 &dev_attr_mapping26.attr,
805 &dev_attr_mapping27.attr,
806 &dev_attr_mapping28.attr,
807 &dev_attr_mapping29.attr,
808 &dev_attr_mapping30.attr,
809 &dev_attr_mapping31.attr,
813 static const struct attribute_group nd_mapping_attribute_group = {
814 .is_visible = mapping_visible,
815 .attrs = mapping_attributes,
818 static const struct attribute_group nd_region_attribute_group = {
819 .attrs = nd_region_attributes,
820 .is_visible = region_visible,
823 static const struct attribute_group *nd_region_attribute_groups[] = {
824 &nd_device_attribute_group,
825 &nd_region_attribute_group,
826 &nd_numa_attribute_group,
827 &nd_mapping_attribute_group,
831 static const struct device_type nd_blk_device_type = {
833 .release = nd_region_release,
834 .groups = nd_region_attribute_groups,
837 static const struct device_type nd_pmem_device_type = {
839 .release = nd_region_release,
840 .groups = nd_region_attribute_groups,
843 static const struct device_type nd_volatile_device_type = {
844 .name = "nd_volatile",
845 .release = nd_region_release,
846 .groups = nd_region_attribute_groups,
849 bool is_nd_pmem(struct device *dev)
851 return dev ? dev->type == &nd_pmem_device_type : false;
854 bool is_nd_blk(struct device *dev)
856 return dev ? dev->type == &nd_blk_device_type : false;
859 bool is_nd_volatile(struct device *dev)
861 return dev ? dev->type == &nd_volatile_device_type : false;
864 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
865 struct nd_namespace_index *nsindex)
867 struct nd_interleave_set *nd_set = nd_region->nd_set;
872 if (nsindex && __le16_to_cpu(nsindex->major) == 1
873 && __le16_to_cpu(nsindex->minor) == 1)
874 return nd_set->cookie1;
875 return nd_set->cookie2;
878 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
880 struct nd_interleave_set *nd_set = nd_region->nd_set;
883 return nd_set->altcookie;
887 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
889 struct nd_label_ent *label_ent, *e;
891 lockdep_assert_held(&nd_mapping->lock);
892 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
893 list_del(&label_ent->list);
899 * When a namespace is activated create new seeds for the next
900 * namespace, or namespace-personality to be configured.
902 void nd_region_advance_seeds(struct nd_region *nd_region, struct device *dev)
904 nvdimm_bus_lock(dev);
905 if (nd_region->ns_seed == dev) {
906 nd_region_create_ns_seed(nd_region);
907 } else if (is_nd_btt(dev)) {
908 struct nd_btt *nd_btt = to_nd_btt(dev);
910 if (nd_region->btt_seed == dev)
911 nd_region_create_btt_seed(nd_region);
912 if (nd_region->ns_seed == &nd_btt->ndns->dev)
913 nd_region_create_ns_seed(nd_region);
914 } else if (is_nd_pfn(dev)) {
915 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
917 if (nd_region->pfn_seed == dev)
918 nd_region_create_pfn_seed(nd_region);
919 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
920 nd_region_create_ns_seed(nd_region);
921 } else if (is_nd_dax(dev)) {
922 struct nd_dax *nd_dax = to_nd_dax(dev);
924 if (nd_region->dax_seed == dev)
925 nd_region_create_dax_seed(nd_region);
926 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
927 nd_region_create_ns_seed(nd_region);
929 nvdimm_bus_unlock(dev);
932 int nd_blk_region_init(struct nd_region *nd_region)
934 struct device *dev = &nd_region->dev;
935 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
940 if (nd_region->ndr_mappings < 1) {
941 dev_dbg(dev, "invalid BLK region\n");
945 return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
949 * nd_region_acquire_lane - allocate and lock a lane
950 * @nd_region: region id and number of lanes possible
952 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
953 * We optimize for the common case where there are 256 lanes, one
954 * per-cpu. For larger systems we need to lock to share lanes. For now
955 * this implementation assumes the cost of maintaining an allocator for
956 * free lanes is on the order of the lock hold time, so it implements a
957 * static lane = cpu % num_lanes mapping.
959 * In the case of a BTT instance on top of a BLK namespace a lane may be
960 * acquired recursively. We lock on the first instance.
962 * In the case of a BTT instance on top of PMEM, we only acquire a lane
963 * for the BTT metadata updates.
965 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
967 unsigned int cpu, lane;
970 if (nd_region->num_lanes < nr_cpu_ids) {
971 struct nd_percpu_lane *ndl_lock, *ndl_count;
973 lane = cpu % nd_region->num_lanes;
974 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
975 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
976 if (ndl_count->count++ == 0)
977 spin_lock(&ndl_lock->lock);
983 EXPORT_SYMBOL(nd_region_acquire_lane);
985 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
987 if (nd_region->num_lanes < nr_cpu_ids) {
988 unsigned int cpu = get_cpu();
989 struct nd_percpu_lane *ndl_lock, *ndl_count;
991 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
992 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
993 if (--ndl_count->count == 0)
994 spin_unlock(&ndl_lock->lock);
999 EXPORT_SYMBOL(nd_region_release_lane);
1002 * PowerPC requires this alignment for memremap_pages(). All other archs
1003 * should be ok with SUBSECTION_SIZE (see memremap_compat_align()).
1005 #define MEMREMAP_COMPAT_ALIGN_MAX SZ_16M
1007 static unsigned long default_align(struct nd_region *nd_region)
1009 unsigned long align;
1013 if (is_nd_blk(&nd_region->dev))
1016 align = MEMREMAP_COMPAT_ALIGN_MAX;
1018 for (i = 0; i < nd_region->ndr_mappings; i++) {
1019 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1020 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1022 if (test_bit(NDD_ALIASING, &nvdimm->flags)) {
1023 align = MEMREMAP_COMPAT_ALIGN_MAX;
1028 mappings = max_t(u16, 1, nd_region->ndr_mappings);
1029 div_u64_rem(align, mappings, &remainder);
1036 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
1037 struct nd_region_desc *ndr_desc,
1038 const struct device_type *dev_type, const char *caller)
1040 struct nd_region *nd_region;
1046 for (i = 0; i < ndr_desc->num_mappings; i++) {
1047 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1048 struct nvdimm *nvdimm = mapping->nvdimm;
1050 if ((mapping->start | mapping->size) % PAGE_SIZE) {
1051 dev_err(&nvdimm_bus->dev,
1052 "%s: %s mapping%d is not %ld aligned\n",
1053 caller, dev_name(&nvdimm->dev), i, PAGE_SIZE);
1057 if (test_bit(NDD_UNARMED, &nvdimm->flags))
1060 if (test_bit(NDD_NOBLK, &nvdimm->flags)
1061 && dev_type == &nd_blk_device_type) {
1062 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
1063 caller, dev_name(&nvdimm->dev), i);
1068 if (dev_type == &nd_blk_device_type) {
1069 struct nd_blk_region_desc *ndbr_desc;
1070 struct nd_blk_region *ndbr;
1072 ndbr_desc = to_blk_region_desc(ndr_desc);
1073 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1074 * ndr_desc->num_mappings,
1077 nd_region = &ndbr->nd_region;
1078 ndbr->enable = ndbr_desc->enable;
1079 ndbr->do_io = ndbr_desc->do_io;
1083 nd_region = kzalloc(struct_size(nd_region, mapping,
1084 ndr_desc->num_mappings),
1086 region_buf = nd_region;
1091 nd_region->id = memregion_alloc(GFP_KERNEL);
1092 if (nd_region->id < 0)
1095 nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1096 if (!nd_region->lane)
1099 for (i = 0; i < nr_cpu_ids; i++) {
1100 struct nd_percpu_lane *ndl;
1102 ndl = per_cpu_ptr(nd_region->lane, i);
1103 spin_lock_init(&ndl->lock);
1107 for (i = 0; i < ndr_desc->num_mappings; i++) {
1108 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1109 struct nvdimm *nvdimm = mapping->nvdimm;
1111 nd_region->mapping[i].nvdimm = nvdimm;
1112 nd_region->mapping[i].start = mapping->start;
1113 nd_region->mapping[i].size = mapping->size;
1114 nd_region->mapping[i].position = mapping->position;
1115 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1116 mutex_init(&nd_region->mapping[i].lock);
1118 get_device(&nvdimm->dev);
1120 nd_region->ndr_mappings = ndr_desc->num_mappings;
1121 nd_region->provider_data = ndr_desc->provider_data;
1122 nd_region->nd_set = ndr_desc->nd_set;
1123 nd_region->num_lanes = ndr_desc->num_lanes;
1124 nd_region->flags = ndr_desc->flags;
1126 nd_region->numa_node = ndr_desc->numa_node;
1127 nd_region->target_node = ndr_desc->target_node;
1128 ida_init(&nd_region->ns_ida);
1129 ida_init(&nd_region->btt_ida);
1130 ida_init(&nd_region->pfn_ida);
1131 ida_init(&nd_region->dax_ida);
1132 dev = &nd_region->dev;
1133 dev_set_name(dev, "region%d", nd_region->id);
1134 dev->parent = &nvdimm_bus->dev;
1135 dev->type = dev_type;
1136 dev->groups = ndr_desc->attr_groups;
1137 dev->of_node = ndr_desc->of_node;
1138 nd_region->ndr_size = resource_size(ndr_desc->res);
1139 nd_region->ndr_start = ndr_desc->res->start;
1140 nd_region->align = default_align(nd_region);
1141 if (ndr_desc->flush)
1142 nd_region->flush = ndr_desc->flush;
1144 nd_region->flush = NULL;
1146 nd_device_register(dev);
1151 memregion_free(nd_region->id);
1157 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1158 struct nd_region_desc *ndr_desc)
1160 ndr_desc->num_lanes = ND_MAX_LANES;
1161 return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1164 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1166 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1167 struct nd_region_desc *ndr_desc)
1169 if (ndr_desc->num_mappings > 1)
1171 ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1172 return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1175 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1177 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1178 struct nd_region_desc *ndr_desc)
1180 ndr_desc->num_lanes = ND_MAX_LANES;
1181 return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1184 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1186 int nvdimm_flush(struct nd_region *nd_region, struct bio *bio)
1190 if (!nd_region->flush)
1191 rc = generic_nvdimm_flush(nd_region);
1193 if (nd_region->flush(nd_region, bio))
1200 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1201 * @nd_region: blk or interleaved pmem region
1203 int generic_nvdimm_flush(struct nd_region *nd_region)
1205 struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1209 * Try to encourage some diversity in flush hint addresses
1210 * across cpus assuming a limited number of flush hints.
1212 idx = this_cpu_read(flush_idx);
1213 idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1216 * The pmem_wmb() is needed to 'sfence' all
1217 * previous writes such that they are architecturally visible for
1218 * the platform buffer flush. Note that we've already arranged for pmem
1219 * writes to avoid the cache via memcpy_flushcache(). The final
1220 * wmb() ensures ordering for the NVDIMM flush write.
1223 for (i = 0; i < nd_region->ndr_mappings; i++)
1224 if (ndrd_get_flush_wpq(ndrd, i, 0))
1225 writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1230 EXPORT_SYMBOL_GPL(nvdimm_flush);
1233 * nvdimm_has_flush - determine write flushing requirements
1234 * @nd_region: blk or interleaved pmem region
1236 * Returns 1 if writes require flushing
1237 * Returns 0 if writes do not require flushing
1238 * Returns -ENXIO if flushing capability can not be determined
1240 int nvdimm_has_flush(struct nd_region *nd_region)
1244 /* no nvdimm or pmem api == flushing capability unknown */
1245 if (nd_region->ndr_mappings == 0
1246 || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1249 /* Test if an explicit flush function is defined */
1250 if (test_bit(ND_REGION_ASYNC, &nd_region->flags) && nd_region->flush)
1253 /* Test if any flush hints for the region are available */
1254 for (i = 0; i < nd_region->ndr_mappings; i++) {
1255 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1256 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1258 /* flush hints present / available */
1259 if (nvdimm->num_flush)
1264 * The platform defines dimm devices without hints nor explicit flush,
1265 * assume platform persistence mechanism like ADR
1269 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1271 int nvdimm_has_cache(struct nd_region *nd_region)
1273 return is_nd_pmem(&nd_region->dev) &&
1274 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1276 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1278 bool is_nvdimm_sync(struct nd_region *nd_region)
1280 if (is_nd_volatile(&nd_region->dev))
1283 return is_nd_pmem(&nd_region->dev) &&
1284 !test_bit(ND_REGION_ASYNC, &nd_region->flags);
1286 EXPORT_SYMBOL_GPL(is_nvdimm_sync);
1288 struct conflict_context {
1289 struct nd_region *nd_region;
1290 resource_size_t start, size;
1293 static int region_conflict(struct device *dev, void *data)
1295 struct nd_region *nd_region;
1296 struct conflict_context *ctx = data;
1297 resource_size_t res_end, region_end, region_start;
1299 if (!is_memory(dev))
1302 nd_region = to_nd_region(dev);
1303 if (nd_region == ctx->nd_region)
1306 res_end = ctx->start + ctx->size;
1307 region_start = nd_region->ndr_start;
1308 region_end = region_start + nd_region->ndr_size;
1309 if (ctx->start >= region_start && ctx->start < region_end)
1311 if (res_end > region_start && res_end <= region_end)
1316 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1317 resource_size_t size)
1319 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1320 struct conflict_context ctx = {
1321 .nd_region = nd_region,
1326 return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);