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 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)) {
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))
233 return ND_DEVICE_NAMESPACE_PMEM;
235 return ND_DEVICE_NAMESPACE_IO;
236 } else if (is_nd_blk(&nd_region->dev)) {
237 return ND_DEVICE_NAMESPACE_BLK;
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
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)
278 int rc = strtobool(buf, &flush);
279 struct nd_region *nd_region = to_nd_region(dev);
285 nvdimm_flush(nd_region);
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;
316 if (is_memory(dev) && nd_set)
317 /* pass, should be precluded by region_visible */;
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.
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);
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,
343 nvdimm_bus_unlock(dev);
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;
357 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
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 */
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;
377 } else if (is_nd_blk(&nd_region->dev))
378 available += nd_blk_available_dpa(nd_region);
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);
411 nvdimm_bus_lock(dev);
413 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
416 nvdimm_bus_unlock(dev);
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);
428 nvdimm_bus_lock(dev);
429 if (nd_region->ns_seed)
430 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
432 rc = sprintf(buf, "\n");
433 nvdimm_bus_unlock(dev);
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);
444 nvdimm_bus_lock(dev);
445 if (nd_region->btt_seed)
446 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
448 rc = sprintf(buf, "\n");
449 nvdimm_bus_unlock(dev);
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);
461 nvdimm_bus_lock(dev);
462 if (nd_region->pfn_seed)
463 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
465 rc = sprintf(buf, "\n");
466 nvdimm_bus_unlock(dev);
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);
478 nvdimm_bus_lock(dev);
479 if (nd_region->dax_seed)
480 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
482 rc = sprintf(buf, "\n");
483 nvdimm_bus_unlock(dev);
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)
501 int rc = strtobool(buf, &ro);
502 struct nd_region *nd_region = to_nd_region(dev);
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 struct attribute *nd_region_attributes[] = {
533 &dev_attr_nstype.attr,
534 &dev_attr_mappings.attr,
535 &dev_attr_btt_seed.attr,
536 &dev_attr_pfn_seed.attr,
537 &dev_attr_dax_seed.attr,
538 &dev_attr_deep_flush.attr,
539 &dev_attr_read_only.attr,
540 &dev_attr_set_cookie.attr,
541 &dev_attr_available_size.attr,
542 &dev_attr_namespace_seed.attr,
543 &dev_attr_init_namespaces.attr,
544 &dev_attr_badblocks.attr,
545 &dev_attr_resource.attr,
549 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
551 struct device *dev = container_of(kobj, typeof(*dev), kobj);
552 struct nd_region *nd_region = to_nd_region(dev);
553 struct nd_interleave_set *nd_set = nd_region->nd_set;
554 int type = nd_region_to_nstype(nd_region);
556 if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
559 if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
562 if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
565 if (!is_nd_pmem(dev) && a == &dev_attr_resource.attr)
568 if (a == &dev_attr_deep_flush.attr) {
569 int has_flush = nvdimm_has_flush(nd_region);
573 else if (has_flush == 0)
579 if (a != &dev_attr_set_cookie.attr
580 && a != &dev_attr_available_size.attr)
583 if ((type == ND_DEVICE_NAMESPACE_PMEM
584 || type == ND_DEVICE_NAMESPACE_BLK)
585 && a == &dev_attr_available_size.attr)
587 else if (is_memory(dev) && nd_set)
593 struct attribute_group nd_region_attribute_group = {
594 .attrs = nd_region_attributes,
595 .is_visible = region_visible,
597 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
599 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
600 struct nd_namespace_index *nsindex)
602 struct nd_interleave_set *nd_set = nd_region->nd_set;
607 if (nsindex && __le16_to_cpu(nsindex->major) == 1
608 && __le16_to_cpu(nsindex->minor) == 1)
609 return nd_set->cookie1;
610 return nd_set->cookie2;
613 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
615 struct nd_interleave_set *nd_set = nd_region->nd_set;
618 return nd_set->altcookie;
622 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
624 struct nd_label_ent *label_ent, *e;
626 lockdep_assert_held(&nd_mapping->lock);
627 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
628 list_del(&label_ent->list);
634 * Upon successful probe/remove, take/release a reference on the
635 * associated interleave set (if present), and plant new btt + namespace
636 * seeds. Also, on the removal of a BLK region, notify the provider to
637 * disable the region.
639 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
640 struct device *dev, bool probe)
642 struct nd_region *nd_region;
644 if (!probe && is_nd_region(dev)) {
647 nd_region = to_nd_region(dev);
648 for (i = 0; i < nd_region->ndr_mappings; i++) {
649 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
650 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
651 struct nvdimm *nvdimm = nd_mapping->nvdimm;
653 mutex_lock(&nd_mapping->lock);
654 nd_mapping_free_labels(nd_mapping);
655 mutex_unlock(&nd_mapping->lock);
658 nd_mapping->ndd = NULL;
660 atomic_dec(&nvdimm->busy);
663 if (dev->parent && is_nd_region(dev->parent) && probe) {
664 nd_region = to_nd_region(dev->parent);
665 nvdimm_bus_lock(dev);
666 if (nd_region->ns_seed == dev)
667 nd_region_create_ns_seed(nd_region);
668 nvdimm_bus_unlock(dev);
670 if (is_nd_btt(dev) && probe) {
671 struct nd_btt *nd_btt = to_nd_btt(dev);
673 nd_region = to_nd_region(dev->parent);
674 nvdimm_bus_lock(dev);
675 if (nd_region->btt_seed == dev)
676 nd_region_create_btt_seed(nd_region);
677 if (nd_region->ns_seed == &nd_btt->ndns->dev)
678 nd_region_create_ns_seed(nd_region);
679 nvdimm_bus_unlock(dev);
681 if (is_nd_pfn(dev) && probe) {
682 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
684 nd_region = to_nd_region(dev->parent);
685 nvdimm_bus_lock(dev);
686 if (nd_region->pfn_seed == dev)
687 nd_region_create_pfn_seed(nd_region);
688 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
689 nd_region_create_ns_seed(nd_region);
690 nvdimm_bus_unlock(dev);
692 if (is_nd_dax(dev) && probe) {
693 struct nd_dax *nd_dax = to_nd_dax(dev);
695 nd_region = to_nd_region(dev->parent);
696 nvdimm_bus_lock(dev);
697 if (nd_region->dax_seed == dev)
698 nd_region_create_dax_seed(nd_region);
699 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
700 nd_region_create_ns_seed(nd_region);
701 nvdimm_bus_unlock(dev);
705 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
707 nd_region_notify_driver_action(nvdimm_bus, dev, true);
710 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
712 nd_region_notify_driver_action(nvdimm_bus, dev, false);
715 static ssize_t mappingN(struct device *dev, char *buf, int n)
717 struct nd_region *nd_region = to_nd_region(dev);
718 struct nd_mapping *nd_mapping;
719 struct nvdimm *nvdimm;
721 if (n >= nd_region->ndr_mappings)
723 nd_mapping = &nd_region->mapping[n];
724 nvdimm = nd_mapping->nvdimm;
726 return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
727 nd_mapping->start, nd_mapping->size);
730 #define REGION_MAPPING(idx) \
731 static ssize_t mapping##idx##_show(struct device *dev, \
732 struct device_attribute *attr, char *buf) \
734 return mappingN(dev, buf, idx); \
736 static DEVICE_ATTR_RO(mapping##idx)
739 * 32 should be enough for a while, even in the presence of socket
740 * interleave a 32-way interleave set is a degenerate case.
775 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
777 struct device *dev = container_of(kobj, struct device, kobj);
778 struct nd_region *nd_region = to_nd_region(dev);
780 if (n < nd_region->ndr_mappings)
785 static struct attribute *mapping_attributes[] = {
786 &dev_attr_mapping0.attr,
787 &dev_attr_mapping1.attr,
788 &dev_attr_mapping2.attr,
789 &dev_attr_mapping3.attr,
790 &dev_attr_mapping4.attr,
791 &dev_attr_mapping5.attr,
792 &dev_attr_mapping6.attr,
793 &dev_attr_mapping7.attr,
794 &dev_attr_mapping8.attr,
795 &dev_attr_mapping9.attr,
796 &dev_attr_mapping10.attr,
797 &dev_attr_mapping11.attr,
798 &dev_attr_mapping12.attr,
799 &dev_attr_mapping13.attr,
800 &dev_attr_mapping14.attr,
801 &dev_attr_mapping15.attr,
802 &dev_attr_mapping16.attr,
803 &dev_attr_mapping17.attr,
804 &dev_attr_mapping18.attr,
805 &dev_attr_mapping19.attr,
806 &dev_attr_mapping20.attr,
807 &dev_attr_mapping21.attr,
808 &dev_attr_mapping22.attr,
809 &dev_attr_mapping23.attr,
810 &dev_attr_mapping24.attr,
811 &dev_attr_mapping25.attr,
812 &dev_attr_mapping26.attr,
813 &dev_attr_mapping27.attr,
814 &dev_attr_mapping28.attr,
815 &dev_attr_mapping29.attr,
816 &dev_attr_mapping30.attr,
817 &dev_attr_mapping31.attr,
821 struct attribute_group nd_mapping_attribute_group = {
822 .is_visible = mapping_visible,
823 .attrs = mapping_attributes,
825 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
827 int nd_blk_region_init(struct nd_region *nd_region)
829 struct device *dev = &nd_region->dev;
830 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
835 if (nd_region->ndr_mappings < 1) {
836 dev_dbg(dev, "invalid BLK region\n");
840 return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
844 * nd_region_acquire_lane - allocate and lock a lane
845 * @nd_region: region id and number of lanes possible
847 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
848 * We optimize for the common case where there are 256 lanes, one
849 * per-cpu. For larger systems we need to lock to share lanes. For now
850 * this implementation assumes the cost of maintaining an allocator for
851 * free lanes is on the order of the lock hold time, so it implements a
852 * static lane = cpu % num_lanes mapping.
854 * In the case of a BTT instance on top of a BLK namespace a lane may be
855 * acquired recursively. We lock on the first instance.
857 * In the case of a BTT instance on top of PMEM, we only acquire a lane
858 * for the BTT metadata updates.
860 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
862 unsigned int cpu, lane;
865 if (nd_region->num_lanes < nr_cpu_ids) {
866 struct nd_percpu_lane *ndl_lock, *ndl_count;
868 lane = cpu % nd_region->num_lanes;
869 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
870 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
871 if (ndl_count->count++ == 0)
872 spin_lock(&ndl_lock->lock);
878 EXPORT_SYMBOL(nd_region_acquire_lane);
880 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
882 if (nd_region->num_lanes < nr_cpu_ids) {
883 unsigned int cpu = get_cpu();
884 struct nd_percpu_lane *ndl_lock, *ndl_count;
886 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
887 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
888 if (--ndl_count->count == 0)
889 spin_unlock(&ndl_lock->lock);
894 EXPORT_SYMBOL(nd_region_release_lane);
896 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
897 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
900 struct nd_region *nd_region;
906 for (i = 0; i < ndr_desc->num_mappings; i++) {
907 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
908 struct nvdimm *nvdimm = mapping->nvdimm;
910 if ((mapping->start | mapping->size) % SZ_4K) {
911 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
912 caller, dev_name(&nvdimm->dev), i);
917 if (test_bit(NDD_UNARMED, &nvdimm->flags))
921 if (dev_type == &nd_blk_device_type) {
922 struct nd_blk_region_desc *ndbr_desc;
923 struct nd_blk_region *ndbr;
925 ndbr_desc = to_blk_region_desc(ndr_desc);
926 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
927 * ndr_desc->num_mappings,
930 nd_region = &ndbr->nd_region;
931 ndbr->enable = ndbr_desc->enable;
932 ndbr->do_io = ndbr_desc->do_io;
936 nd_region = kzalloc(sizeof(struct nd_region)
937 + sizeof(struct nd_mapping)
938 * ndr_desc->num_mappings,
940 region_buf = nd_region;
945 nd_region->id = ida_simple_get(®ion_ida, 0, 0, GFP_KERNEL);
946 if (nd_region->id < 0)
949 nd_region->lane = alloc_percpu(struct nd_percpu_lane);
950 if (!nd_region->lane)
953 for (i = 0; i < nr_cpu_ids; i++) {
954 struct nd_percpu_lane *ndl;
956 ndl = per_cpu_ptr(nd_region->lane, i);
957 spin_lock_init(&ndl->lock);
961 for (i = 0; i < ndr_desc->num_mappings; i++) {
962 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
963 struct nvdimm *nvdimm = mapping->nvdimm;
965 nd_region->mapping[i].nvdimm = nvdimm;
966 nd_region->mapping[i].start = mapping->start;
967 nd_region->mapping[i].size = mapping->size;
968 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
969 mutex_init(&nd_region->mapping[i].lock);
971 get_device(&nvdimm->dev);
973 nd_region->ndr_mappings = ndr_desc->num_mappings;
974 nd_region->provider_data = ndr_desc->provider_data;
975 nd_region->nd_set = ndr_desc->nd_set;
976 nd_region->num_lanes = ndr_desc->num_lanes;
977 nd_region->flags = ndr_desc->flags;
979 nd_region->numa_node = ndr_desc->numa_node;
980 ida_init(&nd_region->ns_ida);
981 ida_init(&nd_region->btt_ida);
982 ida_init(&nd_region->pfn_ida);
983 ida_init(&nd_region->dax_ida);
984 dev = &nd_region->dev;
985 dev_set_name(dev, "region%d", nd_region->id);
986 dev->parent = &nvdimm_bus->dev;
987 dev->type = dev_type;
988 dev->groups = ndr_desc->attr_groups;
989 nd_region->ndr_size = resource_size(ndr_desc->res);
990 nd_region->ndr_start = ndr_desc->res->start;
991 nd_device_register(dev);
996 ida_simple_remove(®ion_ida, nd_region->id);
1002 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1003 struct nd_region_desc *ndr_desc)
1005 ndr_desc->num_lanes = ND_MAX_LANES;
1006 return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1009 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1011 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1012 struct nd_region_desc *ndr_desc)
1014 if (ndr_desc->num_mappings > 1)
1016 ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1017 return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1020 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1022 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1023 struct nd_region_desc *ndr_desc)
1025 ndr_desc->num_lanes = ND_MAX_LANES;
1026 return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1029 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1032 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1033 * @nd_region: blk or interleaved pmem region
1035 void nvdimm_flush(struct nd_region *nd_region)
1037 struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1041 * Try to encourage some diversity in flush hint addresses
1042 * across cpus assuming a limited number of flush hints.
1044 idx = this_cpu_read(flush_idx);
1045 idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1048 * The first wmb() is needed to 'sfence' all previous writes
1049 * such that they are architecturally visible for the platform
1050 * buffer flush. Note that we've already arranged for pmem
1051 * writes to avoid the cache via memcpy_flushcache(). The final
1052 * wmb() ensures ordering for the NVDIMM flush write.
1055 for (i = 0; i < nd_region->ndr_mappings; i++)
1056 if (ndrd_get_flush_wpq(ndrd, i, 0))
1057 writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1060 EXPORT_SYMBOL_GPL(nvdimm_flush);
1063 * nvdimm_has_flush - determine write flushing requirements
1064 * @nd_region: blk or interleaved pmem region
1066 * Returns 1 if writes require flushing
1067 * Returns 0 if writes do not require flushing
1068 * Returns -ENXIO if flushing capability can not be determined
1070 int nvdimm_has_flush(struct nd_region *nd_region)
1074 /* no nvdimm or pmem api == flushing capability unknown */
1075 if (nd_region->ndr_mappings == 0
1076 || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1079 for (i = 0; i < nd_region->ndr_mappings; i++) {
1080 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1081 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1083 /* flush hints present / available */
1084 if (nvdimm->num_flush)
1089 * The platform defines dimm devices without hints, assume
1090 * platform persistence mechanism like ADR
1094 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1096 int nvdimm_has_cache(struct nd_region *nd_region)
1098 return is_nd_pmem(&nd_region->dev);
1100 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1102 void __exit nd_region_devs_exit(void)
1104 ida_destroy(®ion_ida);