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/pmem.h>
19 #include <linux/sort.h>
26 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
29 #include <linux/io-64-nonatomic-hi-lo.h>
31 static DEFINE_IDA(region_ida);
32 static DEFINE_PER_CPU(int, flush_idx);
34 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
35 struct nd_region_data *ndrd)
39 dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
40 nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
41 for (i = 0; i < (1 << ndrd->hints_shift); i++) {
42 struct resource *res = &nvdimm->flush_wpq[i];
43 unsigned long pfn = PHYS_PFN(res->start);
44 void __iomem *flush_page;
46 /* check if flush hints share a page */
47 for (j = 0; j < i; j++) {
48 struct resource *res_j = &nvdimm->flush_wpq[j];
49 unsigned long pfn_j = PHYS_PFN(res_j->start);
56 flush_page = (void __iomem *) ((unsigned long)
57 ndrd_get_flush_wpq(ndrd, dimm, j)
60 flush_page = devm_nvdimm_ioremap(dev,
61 PFN_PHYS(pfn), PAGE_SIZE);
64 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
65 + (res->start & ~PAGE_MASK));
71 int nd_region_activate(struct nd_region *nd_region)
73 int i, j, num_flush = 0;
74 struct nd_region_data *ndrd;
75 struct device *dev = &nd_region->dev;
76 size_t flush_data_size = sizeof(void *);
78 nvdimm_bus_lock(&nd_region->dev);
79 for (i = 0; i < nd_region->ndr_mappings; i++) {
80 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
81 struct nvdimm *nvdimm = nd_mapping->nvdimm;
83 /* at least one null hint slot per-dimm for the "no-hint" case */
84 flush_data_size += sizeof(void *);
85 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
86 if (!nvdimm->num_flush)
88 flush_data_size += nvdimm->num_flush * sizeof(void *);
90 nvdimm_bus_unlock(&nd_region->dev);
92 ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
95 dev_set_drvdata(dev, ndrd);
100 ndrd->hints_shift = ilog2(num_flush);
101 for (i = 0; i < nd_region->ndr_mappings; i++) {
102 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
103 struct nvdimm *nvdimm = nd_mapping->nvdimm;
104 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
111 * Clear out entries that are duplicates. This should prevent the
114 for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
115 /* ignore if NULL already */
116 if (!ndrd_get_flush_wpq(ndrd, i, 0))
119 for (j = i + 1; j < nd_region->ndr_mappings; j++)
120 if (ndrd_get_flush_wpq(ndrd, i, 0) ==
121 ndrd_get_flush_wpq(ndrd, j, 0))
122 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
128 static void nd_region_release(struct device *dev)
130 struct nd_region *nd_region = to_nd_region(dev);
133 for (i = 0; i < nd_region->ndr_mappings; i++) {
134 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
135 struct nvdimm *nvdimm = nd_mapping->nvdimm;
137 put_device(&nvdimm->dev);
139 free_percpu(nd_region->lane);
140 ida_simple_remove(®ion_ida, nd_region->id);
142 kfree(to_nd_blk_region(dev));
147 static struct device_type nd_blk_device_type = {
149 .release = nd_region_release,
152 static struct device_type nd_pmem_device_type = {
154 .release = nd_region_release,
157 static struct device_type nd_volatile_device_type = {
158 .name = "nd_volatile",
159 .release = nd_region_release,
162 bool is_nd_pmem(struct device *dev)
164 return dev ? dev->type == &nd_pmem_device_type : false;
167 bool is_nd_blk(struct device *dev)
169 return dev ? dev->type == &nd_blk_device_type : false;
172 struct nd_region *to_nd_region(struct device *dev)
174 struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
176 WARN_ON(dev->type->release != nd_region_release);
179 EXPORT_SYMBOL_GPL(to_nd_region);
181 struct nd_blk_region *to_nd_blk_region(struct device *dev)
183 struct nd_region *nd_region = to_nd_region(dev);
185 WARN_ON(!is_nd_blk(dev));
186 return container_of(nd_region, struct nd_blk_region, nd_region);
188 EXPORT_SYMBOL_GPL(to_nd_blk_region);
190 void *nd_region_provider_data(struct nd_region *nd_region)
192 return nd_region->provider_data;
194 EXPORT_SYMBOL_GPL(nd_region_provider_data);
196 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
198 return ndbr->blk_provider_data;
200 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
202 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
204 ndbr->blk_provider_data = data;
206 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
209 * nd_region_to_nstype() - region to an integer namespace type
210 * @nd_region: region-device to interrogate
212 * This is the 'nstype' attribute of a region as well, an input to the
213 * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
214 * namespace devices with namespace drivers.
216 int nd_region_to_nstype(struct nd_region *nd_region)
218 if (is_nd_pmem(&nd_region->dev)) {
221 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
222 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
223 struct nvdimm *nvdimm = nd_mapping->nvdimm;
225 if (test_bit(NDD_ALIASING, &nvdimm->flags))
229 return ND_DEVICE_NAMESPACE_PMEM;
231 return ND_DEVICE_NAMESPACE_IO;
232 } else if (is_nd_blk(&nd_region->dev)) {
233 return ND_DEVICE_NAMESPACE_BLK;
238 EXPORT_SYMBOL(nd_region_to_nstype);
240 static ssize_t size_show(struct device *dev,
241 struct device_attribute *attr, char *buf)
243 struct nd_region *nd_region = to_nd_region(dev);
244 unsigned long long size = 0;
246 if (is_nd_pmem(dev)) {
247 size = nd_region->ndr_size;
248 } else if (nd_region->ndr_mappings == 1) {
249 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
251 size = nd_mapping->size;
254 return sprintf(buf, "%llu\n", size);
256 static DEVICE_ATTR_RO(size);
258 static ssize_t deep_flush_show(struct device *dev,
259 struct device_attribute *attr, char *buf)
261 struct nd_region *nd_region = to_nd_region(dev);
264 * NOTE: in the nvdimm_has_flush() error case this attribute is
267 return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
270 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
271 const char *buf, size_t len)
274 int rc = strtobool(buf, &flush);
275 struct nd_region *nd_region = to_nd_region(dev);
281 nvdimm_flush(nd_region);
285 static DEVICE_ATTR_RW(deep_flush);
287 static ssize_t mappings_show(struct device *dev,
288 struct device_attribute *attr, char *buf)
290 struct nd_region *nd_region = to_nd_region(dev);
292 return sprintf(buf, "%d\n", nd_region->ndr_mappings);
294 static DEVICE_ATTR_RO(mappings);
296 static ssize_t nstype_show(struct device *dev,
297 struct device_attribute *attr, char *buf)
299 struct nd_region *nd_region = to_nd_region(dev);
301 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
303 static DEVICE_ATTR_RO(nstype);
305 static ssize_t set_cookie_show(struct device *dev,
306 struct device_attribute *attr, char *buf)
308 struct nd_region *nd_region = to_nd_region(dev);
309 struct nd_interleave_set *nd_set = nd_region->nd_set;
312 if (is_nd_pmem(dev) && nd_set)
313 /* pass, should be precluded by region_visible */;
318 * The cookie to show depends on which specification of the
319 * labels we are using. If there are not labels then default to
320 * the v1.1 namespace label cookie definition. To read all this
321 * data we need to wait for probing to settle.
324 nvdimm_bus_lock(dev);
325 wait_nvdimm_bus_probe_idle(dev);
326 if (nd_region->ndr_mappings) {
327 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
328 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
331 struct nd_namespace_index *nsindex;
333 nsindex = to_namespace_index(ndd, ndd->ns_current);
334 rc = sprintf(buf, "%#llx\n",
335 nd_region_interleave_set_cookie(nd_region,
339 nvdimm_bus_unlock(dev);
344 return sprintf(buf, "%#llx\n", nd_set->cookie1);
346 static DEVICE_ATTR_RO(set_cookie);
348 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
350 resource_size_t blk_max_overlap = 0, available, overlap;
353 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
357 overlap = blk_max_overlap;
358 for (i = 0; i < nd_region->ndr_mappings; i++) {
359 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
360 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
362 /* if a dimm is disabled the available capacity is zero */
366 if (is_nd_pmem(&nd_region->dev)) {
367 available += nd_pmem_available_dpa(nd_region,
368 nd_mapping, &overlap);
369 if (overlap > blk_max_overlap) {
370 blk_max_overlap = overlap;
373 } else if (is_nd_blk(&nd_region->dev))
374 available += nd_blk_available_dpa(nd_region);
380 static ssize_t available_size_show(struct device *dev,
381 struct device_attribute *attr, char *buf)
383 struct nd_region *nd_region = to_nd_region(dev);
384 unsigned long long available = 0;
387 * Flush in-flight updates and grab a snapshot of the available
388 * size. Of course, this value is potentially invalidated the
389 * memory nvdimm_bus_lock() is dropped, but that's userspace's
390 * problem to not race itself.
392 nvdimm_bus_lock(dev);
393 wait_nvdimm_bus_probe_idle(dev);
394 available = nd_region_available_dpa(nd_region);
395 nvdimm_bus_unlock(dev);
397 return sprintf(buf, "%llu\n", available);
399 static DEVICE_ATTR_RO(available_size);
401 static ssize_t init_namespaces_show(struct device *dev,
402 struct device_attribute *attr, char *buf)
404 struct nd_region_data *ndrd = dev_get_drvdata(dev);
407 nvdimm_bus_lock(dev);
409 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
412 nvdimm_bus_unlock(dev);
416 static DEVICE_ATTR_RO(init_namespaces);
418 static ssize_t namespace_seed_show(struct device *dev,
419 struct device_attribute *attr, char *buf)
421 struct nd_region *nd_region = to_nd_region(dev);
424 nvdimm_bus_lock(dev);
425 if (nd_region->ns_seed)
426 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
428 rc = sprintf(buf, "\n");
429 nvdimm_bus_unlock(dev);
432 static DEVICE_ATTR_RO(namespace_seed);
434 static ssize_t btt_seed_show(struct device *dev,
435 struct device_attribute *attr, char *buf)
437 struct nd_region *nd_region = to_nd_region(dev);
440 nvdimm_bus_lock(dev);
441 if (nd_region->btt_seed)
442 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
444 rc = sprintf(buf, "\n");
445 nvdimm_bus_unlock(dev);
449 static DEVICE_ATTR_RO(btt_seed);
451 static ssize_t pfn_seed_show(struct device *dev,
452 struct device_attribute *attr, char *buf)
454 struct nd_region *nd_region = to_nd_region(dev);
457 nvdimm_bus_lock(dev);
458 if (nd_region->pfn_seed)
459 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
461 rc = sprintf(buf, "\n");
462 nvdimm_bus_unlock(dev);
466 static DEVICE_ATTR_RO(pfn_seed);
468 static ssize_t dax_seed_show(struct device *dev,
469 struct device_attribute *attr, char *buf)
471 struct nd_region *nd_region = to_nd_region(dev);
474 nvdimm_bus_lock(dev);
475 if (nd_region->dax_seed)
476 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
478 rc = sprintf(buf, "\n");
479 nvdimm_bus_unlock(dev);
483 static DEVICE_ATTR_RO(dax_seed);
485 static ssize_t read_only_show(struct device *dev,
486 struct device_attribute *attr, char *buf)
488 struct nd_region *nd_region = to_nd_region(dev);
490 return sprintf(buf, "%d\n", nd_region->ro);
493 static ssize_t read_only_store(struct device *dev,
494 struct device_attribute *attr, const char *buf, size_t len)
497 int rc = strtobool(buf, &ro);
498 struct nd_region *nd_region = to_nd_region(dev);
506 static DEVICE_ATTR_RW(read_only);
508 static ssize_t region_badblocks_show(struct device *dev,
509 struct device_attribute *attr, char *buf)
511 struct nd_region *nd_region = to_nd_region(dev);
513 return badblocks_show(&nd_region->bb, buf, 0);
516 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
518 static ssize_t resource_show(struct device *dev,
519 struct device_attribute *attr, char *buf)
521 struct nd_region *nd_region = to_nd_region(dev);
523 return sprintf(buf, "%#llx\n", nd_region->ndr_start);
525 static DEVICE_ATTR_RO(resource);
527 static struct attribute *nd_region_attributes[] = {
529 &dev_attr_nstype.attr,
530 &dev_attr_mappings.attr,
531 &dev_attr_btt_seed.attr,
532 &dev_attr_pfn_seed.attr,
533 &dev_attr_dax_seed.attr,
534 &dev_attr_deep_flush.attr,
535 &dev_attr_read_only.attr,
536 &dev_attr_set_cookie.attr,
537 &dev_attr_available_size.attr,
538 &dev_attr_namespace_seed.attr,
539 &dev_attr_init_namespaces.attr,
540 &dev_attr_badblocks.attr,
541 &dev_attr_resource.attr,
545 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
547 struct device *dev = container_of(kobj, typeof(*dev), kobj);
548 struct nd_region *nd_region = to_nd_region(dev);
549 struct nd_interleave_set *nd_set = nd_region->nd_set;
550 int type = nd_region_to_nstype(nd_region);
552 if (!is_nd_pmem(dev) && a == &dev_attr_pfn_seed.attr)
555 if (!is_nd_pmem(dev) && a == &dev_attr_dax_seed.attr)
558 if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
561 if (!is_nd_pmem(dev) && a == &dev_attr_resource.attr)
564 if (a == &dev_attr_deep_flush.attr) {
565 int has_flush = nvdimm_has_flush(nd_region);
569 else if (has_flush == 0)
575 if (a != &dev_attr_set_cookie.attr
576 && a != &dev_attr_available_size.attr)
579 if ((type == ND_DEVICE_NAMESPACE_PMEM
580 || type == ND_DEVICE_NAMESPACE_BLK)
581 && a == &dev_attr_available_size.attr)
583 else if (is_nd_pmem(dev) && nd_set)
589 struct attribute_group nd_region_attribute_group = {
590 .attrs = nd_region_attributes,
591 .is_visible = region_visible,
593 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
595 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
596 struct nd_namespace_index *nsindex)
598 struct nd_interleave_set *nd_set = nd_region->nd_set;
603 if (nsindex && __le16_to_cpu(nsindex->major) == 1
604 && __le16_to_cpu(nsindex->minor) == 1)
605 return nd_set->cookie1;
606 return nd_set->cookie2;
609 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
611 struct nd_interleave_set *nd_set = nd_region->nd_set;
614 return nd_set->altcookie;
618 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
620 struct nd_label_ent *label_ent, *e;
622 lockdep_assert_held(&nd_mapping->lock);
623 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
624 list_del(&label_ent->list);
630 * Upon successful probe/remove, take/release a reference on the
631 * associated interleave set (if present), and plant new btt + namespace
632 * seeds. Also, on the removal of a BLK region, notify the provider to
633 * disable the region.
635 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
636 struct device *dev, bool probe)
638 struct nd_region *nd_region;
640 if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
643 nd_region = to_nd_region(dev);
644 for (i = 0; i < nd_region->ndr_mappings; i++) {
645 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
646 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
647 struct nvdimm *nvdimm = nd_mapping->nvdimm;
649 mutex_lock(&nd_mapping->lock);
650 nd_mapping_free_labels(nd_mapping);
651 mutex_unlock(&nd_mapping->lock);
654 nd_mapping->ndd = NULL;
656 atomic_dec(&nvdimm->busy);
662 if (dev->parent && (is_nd_blk(dev->parent) || is_nd_pmem(dev->parent))
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 arch_memcpy_to_pmem(). The
1052 * final 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 == flushing capability unknown */
1075 if (nd_region->ndr_mappings == 0)
1078 for (i = 0; i < nd_region->ndr_mappings; i++) {
1079 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1080 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1082 /* flush hints present / available */
1083 if (nvdimm->num_flush)
1088 * The platform defines dimm devices without hints, assume
1089 * platform persistence mechanism like ADR
1093 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1095 void __exit nd_region_devs_exit(void)
1097 ida_destroy(®ion_ida);