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 if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
83 nvdimm_bus_unlock(&nd_region->dev);
87 /* at least one null hint slot per-dimm for the "no-hint" case */
88 flush_data_size += sizeof(void *);
89 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
90 if (!nvdimm->num_flush)
92 flush_data_size += nvdimm->num_flush * sizeof(void *);
94 nvdimm_bus_unlock(&nd_region->dev);
96 ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
99 dev_set_drvdata(dev, ndrd);
104 ndrd->hints_shift = ilog2(num_flush);
105 for (i = 0; i < nd_region->ndr_mappings; i++) {
106 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
107 struct nvdimm *nvdimm = nd_mapping->nvdimm;
108 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
115 * Clear out entries that are duplicates. This should prevent the
118 for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
119 /* ignore if NULL already */
120 if (!ndrd_get_flush_wpq(ndrd, i, 0))
123 for (j = i + 1; j < nd_region->ndr_mappings; j++)
124 if (ndrd_get_flush_wpq(ndrd, i, 0) ==
125 ndrd_get_flush_wpq(ndrd, j, 0))
126 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
132 static void nd_region_release(struct device *dev)
134 struct nd_region *nd_region = to_nd_region(dev);
137 for (i = 0; i < nd_region->ndr_mappings; i++) {
138 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
139 struct nvdimm *nvdimm = nd_mapping->nvdimm;
141 put_device(&nvdimm->dev);
143 free_percpu(nd_region->lane);
144 ida_simple_remove(®ion_ida, nd_region->id);
146 kfree(to_nd_blk_region(dev));
151 static struct device_type nd_blk_device_type = {
153 .release = nd_region_release,
156 static struct device_type nd_pmem_device_type = {
158 .release = nd_region_release,
161 static struct device_type nd_volatile_device_type = {
162 .name = "nd_volatile",
163 .release = nd_region_release,
166 bool is_nd_pmem(struct device *dev)
168 return dev ? dev->type == &nd_pmem_device_type : false;
171 bool is_nd_blk(struct device *dev)
173 return dev ? dev->type == &nd_blk_device_type : false;
176 bool is_nd_volatile(struct device *dev)
178 return dev ? dev->type == &nd_volatile_device_type : false;
181 struct nd_region *to_nd_region(struct device *dev)
183 struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
185 WARN_ON(dev->type->release != nd_region_release);
188 EXPORT_SYMBOL_GPL(to_nd_region);
190 struct device *nd_region_dev(struct nd_region *nd_region)
194 return &nd_region->dev;
196 EXPORT_SYMBOL_GPL(nd_region_dev);
198 struct nd_blk_region *to_nd_blk_region(struct device *dev)
200 struct nd_region *nd_region = to_nd_region(dev);
202 WARN_ON(!is_nd_blk(dev));
203 return container_of(nd_region, struct nd_blk_region, nd_region);
205 EXPORT_SYMBOL_GPL(to_nd_blk_region);
207 void *nd_region_provider_data(struct nd_region *nd_region)
209 return nd_region->provider_data;
211 EXPORT_SYMBOL_GPL(nd_region_provider_data);
213 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
215 return ndbr->blk_provider_data;
217 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
219 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
221 ndbr->blk_provider_data = data;
223 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
226 * nd_region_to_nstype() - region to an integer namespace type
227 * @nd_region: region-device to interrogate
229 * This is the 'nstype' attribute of a region as well, an input to the
230 * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
231 * namespace devices with namespace drivers.
233 int nd_region_to_nstype(struct nd_region *nd_region)
235 if (is_memory(&nd_region->dev)) {
238 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
239 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
240 struct nvdimm *nvdimm = nd_mapping->nvdimm;
242 if (test_bit(NDD_ALIASING, &nvdimm->flags))
246 return ND_DEVICE_NAMESPACE_PMEM;
248 return ND_DEVICE_NAMESPACE_IO;
249 } else if (is_nd_blk(&nd_region->dev)) {
250 return ND_DEVICE_NAMESPACE_BLK;
255 EXPORT_SYMBOL(nd_region_to_nstype);
257 static ssize_t size_show(struct device *dev,
258 struct device_attribute *attr, char *buf)
260 struct nd_region *nd_region = to_nd_region(dev);
261 unsigned long long size = 0;
263 if (is_memory(dev)) {
264 size = nd_region->ndr_size;
265 } else if (nd_region->ndr_mappings == 1) {
266 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
268 size = nd_mapping->size;
271 return sprintf(buf, "%llu\n", size);
273 static DEVICE_ATTR_RO(size);
275 static ssize_t deep_flush_show(struct device *dev,
276 struct device_attribute *attr, char *buf)
278 struct nd_region *nd_region = to_nd_region(dev);
281 * NOTE: in the nvdimm_has_flush() error case this attribute is
284 return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
287 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
288 const char *buf, size_t len)
291 int rc = strtobool(buf, &flush);
292 struct nd_region *nd_region = to_nd_region(dev);
298 nvdimm_flush(nd_region);
302 static DEVICE_ATTR_RW(deep_flush);
304 static ssize_t mappings_show(struct device *dev,
305 struct device_attribute *attr, char *buf)
307 struct nd_region *nd_region = to_nd_region(dev);
309 return sprintf(buf, "%d\n", nd_region->ndr_mappings);
311 static DEVICE_ATTR_RO(mappings);
313 static ssize_t nstype_show(struct device *dev,
314 struct device_attribute *attr, char *buf)
316 struct nd_region *nd_region = to_nd_region(dev);
318 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
320 static DEVICE_ATTR_RO(nstype);
322 static ssize_t set_cookie_show(struct device *dev,
323 struct device_attribute *attr, char *buf)
325 struct nd_region *nd_region = to_nd_region(dev);
326 struct nd_interleave_set *nd_set = nd_region->nd_set;
329 if (is_memory(dev) && nd_set)
330 /* pass, should be precluded by region_visible */;
335 * The cookie to show depends on which specification of the
336 * labels we are using. If there are not labels then default to
337 * the v1.1 namespace label cookie definition. To read all this
338 * data we need to wait for probing to settle.
341 nvdimm_bus_lock(dev);
342 wait_nvdimm_bus_probe_idle(dev);
343 if (nd_region->ndr_mappings) {
344 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
345 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
348 struct nd_namespace_index *nsindex;
350 nsindex = to_namespace_index(ndd, ndd->ns_current);
351 rc = sprintf(buf, "%#llx\n",
352 nd_region_interleave_set_cookie(nd_region,
356 nvdimm_bus_unlock(dev);
361 return sprintf(buf, "%#llx\n", nd_set->cookie1);
363 static DEVICE_ATTR_RO(set_cookie);
365 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
367 resource_size_t blk_max_overlap = 0, available, overlap;
370 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
374 overlap = blk_max_overlap;
375 for (i = 0; i < nd_region->ndr_mappings; i++) {
376 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
377 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
379 /* if a dimm is disabled the available capacity is zero */
383 if (is_memory(&nd_region->dev)) {
384 available += nd_pmem_available_dpa(nd_region,
385 nd_mapping, &overlap);
386 if (overlap > blk_max_overlap) {
387 blk_max_overlap = overlap;
390 } else if (is_nd_blk(&nd_region->dev))
391 available += nd_blk_available_dpa(nd_region);
397 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
399 resource_size_t available = 0;
402 if (is_memory(&nd_region->dev))
403 available = PHYS_ADDR_MAX;
405 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
406 for (i = 0; i < nd_region->ndr_mappings; i++) {
407 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
409 if (is_memory(&nd_region->dev))
410 available = min(available,
411 nd_pmem_max_contiguous_dpa(nd_region,
413 else if (is_nd_blk(&nd_region->dev))
414 available += nd_blk_available_dpa(nd_region);
416 if (is_memory(&nd_region->dev))
417 return available * nd_region->ndr_mappings;
421 static ssize_t available_size_show(struct device *dev,
422 struct device_attribute *attr, char *buf)
424 struct nd_region *nd_region = to_nd_region(dev);
425 unsigned long long available = 0;
428 * Flush in-flight updates and grab a snapshot of the available
429 * size. Of course, this value is potentially invalidated the
430 * memory nvdimm_bus_lock() is dropped, but that's userspace's
431 * problem to not race itself.
433 nvdimm_bus_lock(dev);
434 wait_nvdimm_bus_probe_idle(dev);
435 available = nd_region_available_dpa(nd_region);
436 nvdimm_bus_unlock(dev);
438 return sprintf(buf, "%llu\n", available);
440 static DEVICE_ATTR_RO(available_size);
442 static ssize_t max_available_extent_show(struct device *dev,
443 struct device_attribute *attr, char *buf)
445 struct nd_region *nd_region = to_nd_region(dev);
446 unsigned long long available = 0;
448 nvdimm_bus_lock(dev);
449 wait_nvdimm_bus_probe_idle(dev);
450 available = nd_region_allocatable_dpa(nd_region);
451 nvdimm_bus_unlock(dev);
453 return sprintf(buf, "%llu\n", available);
455 static DEVICE_ATTR_RO(max_available_extent);
457 static ssize_t init_namespaces_show(struct device *dev,
458 struct device_attribute *attr, char *buf)
460 struct nd_region_data *ndrd = dev_get_drvdata(dev);
463 nvdimm_bus_lock(dev);
465 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
468 nvdimm_bus_unlock(dev);
472 static DEVICE_ATTR_RO(init_namespaces);
474 static ssize_t namespace_seed_show(struct device *dev,
475 struct device_attribute *attr, char *buf)
477 struct nd_region *nd_region = to_nd_region(dev);
480 nvdimm_bus_lock(dev);
481 if (nd_region->ns_seed)
482 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
484 rc = sprintf(buf, "\n");
485 nvdimm_bus_unlock(dev);
488 static DEVICE_ATTR_RO(namespace_seed);
490 static ssize_t btt_seed_show(struct device *dev,
491 struct device_attribute *attr, char *buf)
493 struct nd_region *nd_region = to_nd_region(dev);
496 nvdimm_bus_lock(dev);
497 if (nd_region->btt_seed)
498 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
500 rc = sprintf(buf, "\n");
501 nvdimm_bus_unlock(dev);
505 static DEVICE_ATTR_RO(btt_seed);
507 static ssize_t pfn_seed_show(struct device *dev,
508 struct device_attribute *attr, char *buf)
510 struct nd_region *nd_region = to_nd_region(dev);
513 nvdimm_bus_lock(dev);
514 if (nd_region->pfn_seed)
515 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
517 rc = sprintf(buf, "\n");
518 nvdimm_bus_unlock(dev);
522 static DEVICE_ATTR_RO(pfn_seed);
524 static ssize_t dax_seed_show(struct device *dev,
525 struct device_attribute *attr, char *buf)
527 struct nd_region *nd_region = to_nd_region(dev);
530 nvdimm_bus_lock(dev);
531 if (nd_region->dax_seed)
532 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
534 rc = sprintf(buf, "\n");
535 nvdimm_bus_unlock(dev);
539 static DEVICE_ATTR_RO(dax_seed);
541 static ssize_t read_only_show(struct device *dev,
542 struct device_attribute *attr, char *buf)
544 struct nd_region *nd_region = to_nd_region(dev);
546 return sprintf(buf, "%d\n", nd_region->ro);
549 static ssize_t read_only_store(struct device *dev,
550 struct device_attribute *attr, const char *buf, size_t len)
553 int rc = strtobool(buf, &ro);
554 struct nd_region *nd_region = to_nd_region(dev);
562 static DEVICE_ATTR_RW(read_only);
564 static ssize_t region_badblocks_show(struct device *dev,
565 struct device_attribute *attr, char *buf)
567 struct nd_region *nd_region = to_nd_region(dev);
572 rc = badblocks_show(&nd_region->bb, buf, 0);
579 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
581 static ssize_t resource_show(struct device *dev,
582 struct device_attribute *attr, char *buf)
584 struct nd_region *nd_region = to_nd_region(dev);
586 return sprintf(buf, "%#llx\n", nd_region->ndr_start);
588 static DEVICE_ATTR_RO(resource);
590 static ssize_t persistence_domain_show(struct device *dev,
591 struct device_attribute *attr, char *buf)
593 struct nd_region *nd_region = to_nd_region(dev);
595 if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
596 return sprintf(buf, "cpu_cache\n");
597 else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
598 return sprintf(buf, "memory_controller\n");
600 return sprintf(buf, "\n");
602 static DEVICE_ATTR_RO(persistence_domain);
604 static struct attribute *nd_region_attributes[] = {
606 &dev_attr_nstype.attr,
607 &dev_attr_mappings.attr,
608 &dev_attr_btt_seed.attr,
609 &dev_attr_pfn_seed.attr,
610 &dev_attr_dax_seed.attr,
611 &dev_attr_deep_flush.attr,
612 &dev_attr_read_only.attr,
613 &dev_attr_set_cookie.attr,
614 &dev_attr_available_size.attr,
615 &dev_attr_max_available_extent.attr,
616 &dev_attr_namespace_seed.attr,
617 &dev_attr_init_namespaces.attr,
618 &dev_attr_badblocks.attr,
619 &dev_attr_resource.attr,
620 &dev_attr_persistence_domain.attr,
624 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
626 struct device *dev = container_of(kobj, typeof(*dev), kobj);
627 struct nd_region *nd_region = to_nd_region(dev);
628 struct nd_interleave_set *nd_set = nd_region->nd_set;
629 int type = nd_region_to_nstype(nd_region);
631 if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
634 if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
637 if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
640 if (a == &dev_attr_resource.attr) {
647 if (a == &dev_attr_deep_flush.attr) {
648 int has_flush = nvdimm_has_flush(nd_region);
652 else if (has_flush == 0)
658 if (a == &dev_attr_persistence_domain.attr) {
659 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
660 | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
665 if (a != &dev_attr_set_cookie.attr
666 && a != &dev_attr_available_size.attr)
669 if ((type == ND_DEVICE_NAMESPACE_PMEM
670 || type == ND_DEVICE_NAMESPACE_BLK)
671 && a == &dev_attr_available_size.attr)
673 else if (is_memory(dev) && nd_set)
679 struct attribute_group nd_region_attribute_group = {
680 .attrs = nd_region_attributes,
681 .is_visible = region_visible,
683 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
685 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
686 struct nd_namespace_index *nsindex)
688 struct nd_interleave_set *nd_set = nd_region->nd_set;
693 if (nsindex && __le16_to_cpu(nsindex->major) == 1
694 && __le16_to_cpu(nsindex->minor) == 1)
695 return nd_set->cookie1;
696 return nd_set->cookie2;
699 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
701 struct nd_interleave_set *nd_set = nd_region->nd_set;
704 return nd_set->altcookie;
708 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
710 struct nd_label_ent *label_ent, *e;
712 lockdep_assert_held(&nd_mapping->lock);
713 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
714 list_del(&label_ent->list);
720 * Upon successful probe/remove, take/release a reference on the
721 * associated interleave set (if present), and plant new btt + namespace
722 * seeds. Also, on the removal of a BLK region, notify the provider to
723 * disable the region.
725 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
726 struct device *dev, bool probe)
728 struct nd_region *nd_region;
730 if (!probe && is_nd_region(dev)) {
733 nd_region = to_nd_region(dev);
734 for (i = 0; i < nd_region->ndr_mappings; i++) {
735 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
736 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
737 struct nvdimm *nvdimm = nd_mapping->nvdimm;
739 mutex_lock(&nd_mapping->lock);
740 nd_mapping_free_labels(nd_mapping);
741 mutex_unlock(&nd_mapping->lock);
744 nd_mapping->ndd = NULL;
746 atomic_dec(&nvdimm->busy);
749 if (dev->parent && is_nd_region(dev->parent) && probe) {
750 nd_region = to_nd_region(dev->parent);
751 nvdimm_bus_lock(dev);
752 if (nd_region->ns_seed == dev)
753 nd_region_create_ns_seed(nd_region);
754 nvdimm_bus_unlock(dev);
756 if (is_nd_btt(dev) && probe) {
757 struct nd_btt *nd_btt = to_nd_btt(dev);
759 nd_region = to_nd_region(dev->parent);
760 nvdimm_bus_lock(dev);
761 if (nd_region->btt_seed == dev)
762 nd_region_create_btt_seed(nd_region);
763 if (nd_region->ns_seed == &nd_btt->ndns->dev)
764 nd_region_create_ns_seed(nd_region);
765 nvdimm_bus_unlock(dev);
767 if (is_nd_pfn(dev) && probe) {
768 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
770 nd_region = to_nd_region(dev->parent);
771 nvdimm_bus_lock(dev);
772 if (nd_region->pfn_seed == dev)
773 nd_region_create_pfn_seed(nd_region);
774 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
775 nd_region_create_ns_seed(nd_region);
776 nvdimm_bus_unlock(dev);
778 if (is_nd_dax(dev) && probe) {
779 struct nd_dax *nd_dax = to_nd_dax(dev);
781 nd_region = to_nd_region(dev->parent);
782 nvdimm_bus_lock(dev);
783 if (nd_region->dax_seed == dev)
784 nd_region_create_dax_seed(nd_region);
785 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
786 nd_region_create_ns_seed(nd_region);
787 nvdimm_bus_unlock(dev);
791 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
793 nd_region_notify_driver_action(nvdimm_bus, dev, true);
796 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
798 nd_region_notify_driver_action(nvdimm_bus, dev, false);
801 static ssize_t mappingN(struct device *dev, char *buf, int n)
803 struct nd_region *nd_region = to_nd_region(dev);
804 struct nd_mapping *nd_mapping;
805 struct nvdimm *nvdimm;
807 if (n >= nd_region->ndr_mappings)
809 nd_mapping = &nd_region->mapping[n];
810 nvdimm = nd_mapping->nvdimm;
812 return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
813 nd_mapping->start, nd_mapping->size,
814 nd_mapping->position);
817 #define REGION_MAPPING(idx) \
818 static ssize_t mapping##idx##_show(struct device *dev, \
819 struct device_attribute *attr, char *buf) \
821 return mappingN(dev, buf, idx); \
823 static DEVICE_ATTR_RO(mapping##idx)
826 * 32 should be enough for a while, even in the presence of socket
827 * interleave a 32-way interleave set is a degenerate case.
862 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
864 struct device *dev = container_of(kobj, struct device, kobj);
865 struct nd_region *nd_region = to_nd_region(dev);
867 if (n < nd_region->ndr_mappings)
872 static struct attribute *mapping_attributes[] = {
873 &dev_attr_mapping0.attr,
874 &dev_attr_mapping1.attr,
875 &dev_attr_mapping2.attr,
876 &dev_attr_mapping3.attr,
877 &dev_attr_mapping4.attr,
878 &dev_attr_mapping5.attr,
879 &dev_attr_mapping6.attr,
880 &dev_attr_mapping7.attr,
881 &dev_attr_mapping8.attr,
882 &dev_attr_mapping9.attr,
883 &dev_attr_mapping10.attr,
884 &dev_attr_mapping11.attr,
885 &dev_attr_mapping12.attr,
886 &dev_attr_mapping13.attr,
887 &dev_attr_mapping14.attr,
888 &dev_attr_mapping15.attr,
889 &dev_attr_mapping16.attr,
890 &dev_attr_mapping17.attr,
891 &dev_attr_mapping18.attr,
892 &dev_attr_mapping19.attr,
893 &dev_attr_mapping20.attr,
894 &dev_attr_mapping21.attr,
895 &dev_attr_mapping22.attr,
896 &dev_attr_mapping23.attr,
897 &dev_attr_mapping24.attr,
898 &dev_attr_mapping25.attr,
899 &dev_attr_mapping26.attr,
900 &dev_attr_mapping27.attr,
901 &dev_attr_mapping28.attr,
902 &dev_attr_mapping29.attr,
903 &dev_attr_mapping30.attr,
904 &dev_attr_mapping31.attr,
908 struct attribute_group nd_mapping_attribute_group = {
909 .is_visible = mapping_visible,
910 .attrs = mapping_attributes,
912 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
914 int nd_blk_region_init(struct nd_region *nd_region)
916 struct device *dev = &nd_region->dev;
917 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
922 if (nd_region->ndr_mappings < 1) {
923 dev_dbg(dev, "invalid BLK region\n");
927 return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
931 * nd_region_acquire_lane - allocate and lock a lane
932 * @nd_region: region id and number of lanes possible
934 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
935 * We optimize for the common case where there are 256 lanes, one
936 * per-cpu. For larger systems we need to lock to share lanes. For now
937 * this implementation assumes the cost of maintaining an allocator for
938 * free lanes is on the order of the lock hold time, so it implements a
939 * static lane = cpu % num_lanes mapping.
941 * In the case of a BTT instance on top of a BLK namespace a lane may be
942 * acquired recursively. We lock on the first instance.
944 * In the case of a BTT instance on top of PMEM, we only acquire a lane
945 * for the BTT metadata updates.
947 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
949 unsigned int cpu, lane;
952 if (nd_region->num_lanes < nr_cpu_ids) {
953 struct nd_percpu_lane *ndl_lock, *ndl_count;
955 lane = cpu % nd_region->num_lanes;
956 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
957 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
958 if (ndl_count->count++ == 0)
959 spin_lock(&ndl_lock->lock);
965 EXPORT_SYMBOL(nd_region_acquire_lane);
967 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
969 if (nd_region->num_lanes < nr_cpu_ids) {
970 unsigned int cpu = get_cpu();
971 struct nd_percpu_lane *ndl_lock, *ndl_count;
973 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
974 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
975 if (--ndl_count->count == 0)
976 spin_unlock(&ndl_lock->lock);
981 EXPORT_SYMBOL(nd_region_release_lane);
983 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
984 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
987 struct nd_region *nd_region;
993 for (i = 0; i < ndr_desc->num_mappings; i++) {
994 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
995 struct nvdimm *nvdimm = mapping->nvdimm;
997 if ((mapping->start | mapping->size) % SZ_4K) {
998 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
999 caller, dev_name(&nvdimm->dev), i);
1004 if (test_bit(NDD_UNARMED, &nvdimm->flags))
1007 if (test_bit(NDD_NOBLK, &nvdimm->flags)
1008 && dev_type == &nd_blk_device_type) {
1009 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
1010 caller, dev_name(&nvdimm->dev), i);
1015 if (dev_type == &nd_blk_device_type) {
1016 struct nd_blk_region_desc *ndbr_desc;
1017 struct nd_blk_region *ndbr;
1019 ndbr_desc = to_blk_region_desc(ndr_desc);
1020 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1021 * ndr_desc->num_mappings,
1024 nd_region = &ndbr->nd_region;
1025 ndbr->enable = ndbr_desc->enable;
1026 ndbr->do_io = ndbr_desc->do_io;
1030 nd_region = kzalloc(sizeof(struct nd_region)
1031 + sizeof(struct nd_mapping)
1032 * ndr_desc->num_mappings,
1034 region_buf = nd_region;
1039 nd_region->id = ida_simple_get(®ion_ida, 0, 0, GFP_KERNEL);
1040 if (nd_region->id < 0)
1043 nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1044 if (!nd_region->lane)
1047 for (i = 0; i < nr_cpu_ids; i++) {
1048 struct nd_percpu_lane *ndl;
1050 ndl = per_cpu_ptr(nd_region->lane, i);
1051 spin_lock_init(&ndl->lock);
1055 for (i = 0; i < ndr_desc->num_mappings; i++) {
1056 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1057 struct nvdimm *nvdimm = mapping->nvdimm;
1059 nd_region->mapping[i].nvdimm = nvdimm;
1060 nd_region->mapping[i].start = mapping->start;
1061 nd_region->mapping[i].size = mapping->size;
1062 nd_region->mapping[i].position = mapping->position;
1063 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1064 mutex_init(&nd_region->mapping[i].lock);
1066 get_device(&nvdimm->dev);
1068 nd_region->ndr_mappings = ndr_desc->num_mappings;
1069 nd_region->provider_data = ndr_desc->provider_data;
1070 nd_region->nd_set = ndr_desc->nd_set;
1071 nd_region->num_lanes = ndr_desc->num_lanes;
1072 nd_region->flags = ndr_desc->flags;
1074 nd_region->numa_node = ndr_desc->numa_node;
1075 nd_region->target_node = ndr_desc->target_node;
1076 ida_init(&nd_region->ns_ida);
1077 ida_init(&nd_region->btt_ida);
1078 ida_init(&nd_region->pfn_ida);
1079 ida_init(&nd_region->dax_ida);
1080 dev = &nd_region->dev;
1081 dev_set_name(dev, "region%d", nd_region->id);
1082 dev->parent = &nvdimm_bus->dev;
1083 dev->type = dev_type;
1084 dev->groups = ndr_desc->attr_groups;
1085 dev->of_node = ndr_desc->of_node;
1086 nd_region->ndr_size = resource_size(ndr_desc->res);
1087 nd_region->ndr_start = ndr_desc->res->start;
1088 nd_device_register(dev);
1093 ida_simple_remove(®ion_ida, nd_region->id);
1099 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1100 struct nd_region_desc *ndr_desc)
1102 ndr_desc->num_lanes = ND_MAX_LANES;
1103 return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1106 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1108 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1109 struct nd_region_desc *ndr_desc)
1111 if (ndr_desc->num_mappings > 1)
1113 ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1114 return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1117 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1119 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1120 struct nd_region_desc *ndr_desc)
1122 ndr_desc->num_lanes = ND_MAX_LANES;
1123 return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1126 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1129 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1130 * @nd_region: blk or interleaved pmem region
1132 void nvdimm_flush(struct nd_region *nd_region)
1134 struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1138 * Try to encourage some diversity in flush hint addresses
1139 * across cpus assuming a limited number of flush hints.
1141 idx = this_cpu_read(flush_idx);
1142 idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1145 * The first wmb() is needed to 'sfence' all previous writes
1146 * such that they are architecturally visible for the platform
1147 * buffer flush. Note that we've already arranged for pmem
1148 * writes to avoid the cache via memcpy_flushcache(). The final
1149 * wmb() ensures ordering for the NVDIMM flush write.
1152 for (i = 0; i < nd_region->ndr_mappings; i++)
1153 if (ndrd_get_flush_wpq(ndrd, i, 0))
1154 writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1157 EXPORT_SYMBOL_GPL(nvdimm_flush);
1160 * nvdimm_has_flush - determine write flushing requirements
1161 * @nd_region: blk or interleaved pmem region
1163 * Returns 1 if writes require flushing
1164 * Returns 0 if writes do not require flushing
1165 * Returns -ENXIO if flushing capability can not be determined
1167 int nvdimm_has_flush(struct nd_region *nd_region)
1171 /* no nvdimm or pmem api == flushing capability unknown */
1172 if (nd_region->ndr_mappings == 0
1173 || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1176 for (i = 0; i < nd_region->ndr_mappings; i++) {
1177 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1178 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1180 /* flush hints present / available */
1181 if (nvdimm->num_flush)
1186 * The platform defines dimm devices without hints, assume
1187 * platform persistence mechanism like ADR
1191 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1193 int nvdimm_has_cache(struct nd_region *nd_region)
1195 return is_nd_pmem(&nd_region->dev) &&
1196 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1198 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1200 struct conflict_context {
1201 struct nd_region *nd_region;
1202 resource_size_t start, size;
1205 static int region_conflict(struct device *dev, void *data)
1207 struct nd_region *nd_region;
1208 struct conflict_context *ctx = data;
1209 resource_size_t res_end, region_end, region_start;
1211 if (!is_memory(dev))
1214 nd_region = to_nd_region(dev);
1215 if (nd_region == ctx->nd_region)
1218 res_end = ctx->start + ctx->size;
1219 region_start = nd_region->ndr_start;
1220 region_end = region_start + nd_region->ndr_size;
1221 if (ctx->start >= region_start && ctx->start < region_end)
1223 if (res_end > region_start && res_end <= region_end)
1228 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1229 resource_size_t size)
1231 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1232 struct conflict_context ctx = {
1233 .nd_region = nd_region,
1238 return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);
1241 void __exit nd_region_devs_exit(void)
1243 ida_destroy(®ion_ida);