spi: spi-fsl-qspi: Fix return value check of devm_ioremap() in probe
[linux/fpc-iii.git] / drivers / nvdimm / region_devs.c
blobccbb5b43b8b2c2368b8a5465de4d0fbba5e31e05
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 */
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>
12 #include <linux/io.h>
13 #include <linux/nd.h>
14 #include "nd-core.h"
15 #include "nd.h"
18 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
19 * irrelevant.
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)
28 int i, j;
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);
42 if (pfn == pfn_j)
43 break;
46 if (j < i)
47 flush_page = (void __iomem *) ((unsigned long)
48 ndrd_get_flush_wpq(ndrd, dimm, j)
49 & PAGE_MASK);
50 else
51 flush_page = devm_nvdimm_ioremap(dev,
52 PFN_PHYS(pfn), PAGE_SIZE);
53 if (!flush_page)
54 return -ENXIO;
55 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
56 + (res->start & ~PAGE_MASK));
59 return 0;
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);
76 return -EBUSY;
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)
83 continue;
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);
89 if (!ndrd)
90 return -ENOMEM;
91 dev_set_drvdata(dev, ndrd);
93 if (!num_flush)
94 return 0;
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);
102 if (rc)
103 return rc;
107 * Clear out entries that are duplicates. This should prevent the
108 * extra flushings.
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))
113 continue;
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);
121 return 0;
124 static void nd_region_release(struct device *dev)
126 struct nd_region *nd_region = to_nd_region(dev);
127 u16 i;
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);
137 if (is_nd_blk(dev))
138 kfree(to_nd_blk_region(dev));
139 else
140 kfree(nd_region);
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);
148 return nd_region;
150 EXPORT_SYMBOL_GPL(to_nd_region);
152 struct device *nd_region_dev(struct nd_region *nd_region)
154 if (!nd_region)
155 return NULL;
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)) {
198 u16 i, label;
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))
205 label++;
207 if (label)
208 return ND_DEVICE_NAMESPACE_PMEM;
209 else
210 return ND_DEVICE_NAMESPACE_IO;
211 } else if (is_nd_blk(&nd_region->dev)) {
212 return ND_DEVICE_NAMESPACE_BLK;
215 return 0;
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;
229 return 0;
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
248 * not visible.
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)
256 bool flush;
257 int rc = strtobool(buf, &flush);
258 struct nd_region *nd_region = to_nd_region(dev);
260 if (rc)
261 return rc;
262 if (!flush)
263 return -EINVAL;
264 rc = nvdimm_flush(nd_region, NULL);
265 if (rc)
266 return rc;
268 return len;
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;
295 ssize_t rc = 0;
297 if (is_memory(dev) && nd_set)
298 /* pass, should be precluded by region_visible */;
299 else
300 return -ENXIO;
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.
308 nd_device_lock(dev);
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);
315 if (ndd) {
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,
321 nsindex));
324 nvdimm_bus_unlock(dev);
325 nd_device_unlock(dev);
327 if (rc)
328 return rc;
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;
336 int i;
338 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
340 retry:
341 available = 0;
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 */
348 if (!ndd)
349 return 0;
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;
356 goto retry;
358 } else if (is_nd_blk(&nd_region->dev))
359 available += nd_blk_available_dpa(nd_region);
362 return available;
365 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
367 resource_size_t available = 0;
368 int i;
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,
380 nd_mapping));
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;
386 return available;
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.
401 nd_device_lock(dev);
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;
418 nd_device_lock(dev);
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);
433 ssize_t rc;
435 nvdimm_bus_lock(dev);
436 if (ndrd)
437 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
438 else
439 rc = -ENXIO;
440 nvdimm_bus_unlock(dev);
442 return rc;
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);
450 ssize_t rc;
452 nvdimm_bus_lock(dev);
453 if (nd_region->ns_seed)
454 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
455 else
456 rc = sprintf(buf, "\n");
457 nvdimm_bus_unlock(dev);
458 return rc;
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);
466 ssize_t rc;
468 nvdimm_bus_lock(dev);
469 if (nd_region->btt_seed)
470 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
471 else
472 rc = sprintf(buf, "\n");
473 nvdimm_bus_unlock(dev);
475 return rc;
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);
483 ssize_t rc;
485 nvdimm_bus_lock(dev);
486 if (nd_region->pfn_seed)
487 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
488 else
489 rc = sprintf(buf, "\n");
490 nvdimm_bus_unlock(dev);
492 return rc;
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);
500 ssize_t rc;
502 nvdimm_bus_lock(dev);
503 if (nd_region->dax_seed)
504 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
505 else
506 rc = sprintf(buf, "\n");
507 nvdimm_bus_unlock(dev);
509 return rc;
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 ssize_t read_only_store(struct device *dev,
522 struct device_attribute *attr, const char *buf, size_t len)
524 bool ro;
525 int rc = strtobool(buf, &ro);
526 struct nd_region *nd_region = to_nd_region(dev);
528 if (rc)
529 return rc;
531 nd_region->ro = ro;
532 return len;
534 static DEVICE_ATTR_RW(read_only);
536 static ssize_t align_show(struct device *dev,
537 struct device_attribute *attr, char *buf)
539 struct nd_region *nd_region = to_nd_region(dev);
541 return sprintf(buf, "%#lx\n", nd_region->align);
544 static ssize_t align_store(struct device *dev,
545 struct device_attribute *attr, const char *buf, size_t len)
547 struct nd_region *nd_region = to_nd_region(dev);
548 unsigned long val, dpa;
549 u32 remainder;
550 int rc;
552 rc = kstrtoul(buf, 0, &val);
553 if (rc)
554 return rc;
556 if (!nd_region->ndr_mappings)
557 return -ENXIO;
560 * Ensure space-align is evenly divisible by the region
561 * interleave-width because the kernel typically has no facility
562 * to determine which DIMM(s), dimm-physical-addresses, would
563 * contribute to the tail capacity in system-physical-address
564 * space for the namespace.
566 dpa = div_u64_rem(val, nd_region->ndr_mappings, &remainder);
567 if (!is_power_of_2(dpa) || dpa < PAGE_SIZE
568 || val > region_size(nd_region) || remainder)
569 return -EINVAL;
572 * Given that space allocation consults this value multiple
573 * times ensure it does not change for the duration of the
574 * allocation.
576 nvdimm_bus_lock(dev);
577 nd_region->align = val;
578 nvdimm_bus_unlock(dev);
580 return len;
582 static DEVICE_ATTR_RW(align);
584 static ssize_t region_badblocks_show(struct device *dev,
585 struct device_attribute *attr, char *buf)
587 struct nd_region *nd_region = to_nd_region(dev);
588 ssize_t rc;
590 nd_device_lock(dev);
591 if (dev->driver)
592 rc = badblocks_show(&nd_region->bb, buf, 0);
593 else
594 rc = -ENXIO;
595 nd_device_unlock(dev);
597 return rc;
599 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
601 static ssize_t resource_show(struct device *dev,
602 struct device_attribute *attr, char *buf)
604 struct nd_region *nd_region = to_nd_region(dev);
606 return sprintf(buf, "%#llx\n", nd_region->ndr_start);
608 static DEVICE_ATTR(resource, 0400, resource_show, NULL);
610 static ssize_t persistence_domain_show(struct device *dev,
611 struct device_attribute *attr, char *buf)
613 struct nd_region *nd_region = to_nd_region(dev);
615 if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
616 return sprintf(buf, "cpu_cache\n");
617 else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
618 return sprintf(buf, "memory_controller\n");
619 else
620 return sprintf(buf, "\n");
622 static DEVICE_ATTR_RO(persistence_domain);
624 static struct attribute *nd_region_attributes[] = {
625 &dev_attr_size.attr,
626 &dev_attr_align.attr,
627 &dev_attr_nstype.attr,
628 &dev_attr_mappings.attr,
629 &dev_attr_btt_seed.attr,
630 &dev_attr_pfn_seed.attr,
631 &dev_attr_dax_seed.attr,
632 &dev_attr_deep_flush.attr,
633 &dev_attr_read_only.attr,
634 &dev_attr_set_cookie.attr,
635 &dev_attr_available_size.attr,
636 &dev_attr_max_available_extent.attr,
637 &dev_attr_namespace_seed.attr,
638 &dev_attr_init_namespaces.attr,
639 &dev_attr_badblocks.attr,
640 &dev_attr_resource.attr,
641 &dev_attr_persistence_domain.attr,
642 NULL,
645 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
647 struct device *dev = container_of(kobj, typeof(*dev), kobj);
648 struct nd_region *nd_region = to_nd_region(dev);
649 struct nd_interleave_set *nd_set = nd_region->nd_set;
650 int type = nd_region_to_nstype(nd_region);
652 if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
653 return 0;
655 if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
656 return 0;
658 if (!is_memory(dev) && a == &dev_attr_badblocks.attr)
659 return 0;
661 if (a == &dev_attr_resource.attr && !is_memory(dev))
662 return 0;
664 if (a == &dev_attr_deep_flush.attr) {
665 int has_flush = nvdimm_has_flush(nd_region);
667 if (has_flush == 1)
668 return a->mode;
669 else if (has_flush == 0)
670 return 0444;
671 else
672 return 0;
675 if (a == &dev_attr_persistence_domain.attr) {
676 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
677 | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
678 return 0;
679 return a->mode;
682 if (a == &dev_attr_align.attr) {
683 int i;
685 for (i = 0; i < nd_region->ndr_mappings; i++) {
686 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
687 struct nvdimm *nvdimm = nd_mapping->nvdimm;
689 if (test_bit(NDD_LABELING, &nvdimm->flags))
690 return a->mode;
692 return 0;
695 if (a != &dev_attr_set_cookie.attr
696 && a != &dev_attr_available_size.attr)
697 return a->mode;
699 if ((type == ND_DEVICE_NAMESPACE_PMEM
700 || type == ND_DEVICE_NAMESPACE_BLK)
701 && a == &dev_attr_available_size.attr)
702 return a->mode;
703 else if (is_memory(dev) && nd_set)
704 return a->mode;
706 return 0;
709 static ssize_t mappingN(struct device *dev, char *buf, int n)
711 struct nd_region *nd_region = to_nd_region(dev);
712 struct nd_mapping *nd_mapping;
713 struct nvdimm *nvdimm;
715 if (n >= nd_region->ndr_mappings)
716 return -ENXIO;
717 nd_mapping = &nd_region->mapping[n];
718 nvdimm = nd_mapping->nvdimm;
720 return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
721 nd_mapping->start, nd_mapping->size,
722 nd_mapping->position);
725 #define REGION_MAPPING(idx) \
726 static ssize_t mapping##idx##_show(struct device *dev, \
727 struct device_attribute *attr, char *buf) \
729 return mappingN(dev, buf, idx); \
731 static DEVICE_ATTR_RO(mapping##idx)
734 * 32 should be enough for a while, even in the presence of socket
735 * interleave a 32-way interleave set is a degenerate case.
737 REGION_MAPPING(0);
738 REGION_MAPPING(1);
739 REGION_MAPPING(2);
740 REGION_MAPPING(3);
741 REGION_MAPPING(4);
742 REGION_MAPPING(5);
743 REGION_MAPPING(6);
744 REGION_MAPPING(7);
745 REGION_MAPPING(8);
746 REGION_MAPPING(9);
747 REGION_MAPPING(10);
748 REGION_MAPPING(11);
749 REGION_MAPPING(12);
750 REGION_MAPPING(13);
751 REGION_MAPPING(14);
752 REGION_MAPPING(15);
753 REGION_MAPPING(16);
754 REGION_MAPPING(17);
755 REGION_MAPPING(18);
756 REGION_MAPPING(19);
757 REGION_MAPPING(20);
758 REGION_MAPPING(21);
759 REGION_MAPPING(22);
760 REGION_MAPPING(23);
761 REGION_MAPPING(24);
762 REGION_MAPPING(25);
763 REGION_MAPPING(26);
764 REGION_MAPPING(27);
765 REGION_MAPPING(28);
766 REGION_MAPPING(29);
767 REGION_MAPPING(30);
768 REGION_MAPPING(31);
770 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
772 struct device *dev = container_of(kobj, struct device, kobj);
773 struct nd_region *nd_region = to_nd_region(dev);
775 if (n < nd_region->ndr_mappings)
776 return a->mode;
777 return 0;
780 static struct attribute *mapping_attributes[] = {
781 &dev_attr_mapping0.attr,
782 &dev_attr_mapping1.attr,
783 &dev_attr_mapping2.attr,
784 &dev_attr_mapping3.attr,
785 &dev_attr_mapping4.attr,
786 &dev_attr_mapping5.attr,
787 &dev_attr_mapping6.attr,
788 &dev_attr_mapping7.attr,
789 &dev_attr_mapping8.attr,
790 &dev_attr_mapping9.attr,
791 &dev_attr_mapping10.attr,
792 &dev_attr_mapping11.attr,
793 &dev_attr_mapping12.attr,
794 &dev_attr_mapping13.attr,
795 &dev_attr_mapping14.attr,
796 &dev_attr_mapping15.attr,
797 &dev_attr_mapping16.attr,
798 &dev_attr_mapping17.attr,
799 &dev_attr_mapping18.attr,
800 &dev_attr_mapping19.attr,
801 &dev_attr_mapping20.attr,
802 &dev_attr_mapping21.attr,
803 &dev_attr_mapping22.attr,
804 &dev_attr_mapping23.attr,
805 &dev_attr_mapping24.attr,
806 &dev_attr_mapping25.attr,
807 &dev_attr_mapping26.attr,
808 &dev_attr_mapping27.attr,
809 &dev_attr_mapping28.attr,
810 &dev_attr_mapping29.attr,
811 &dev_attr_mapping30.attr,
812 &dev_attr_mapping31.attr,
813 NULL,
816 static const struct attribute_group nd_mapping_attribute_group = {
817 .is_visible = mapping_visible,
818 .attrs = mapping_attributes,
821 static const struct attribute_group nd_region_attribute_group = {
822 .attrs = nd_region_attributes,
823 .is_visible = region_visible,
826 static const struct attribute_group *nd_region_attribute_groups[] = {
827 &nd_device_attribute_group,
828 &nd_region_attribute_group,
829 &nd_numa_attribute_group,
830 &nd_mapping_attribute_group,
831 NULL,
834 static const struct device_type nd_blk_device_type = {
835 .name = "nd_blk",
836 .release = nd_region_release,
837 .groups = nd_region_attribute_groups,
840 static const struct device_type nd_pmem_device_type = {
841 .name = "nd_pmem",
842 .release = nd_region_release,
843 .groups = nd_region_attribute_groups,
846 static const struct device_type nd_volatile_device_type = {
847 .name = "nd_volatile",
848 .release = nd_region_release,
849 .groups = nd_region_attribute_groups,
852 bool is_nd_pmem(struct device *dev)
854 return dev ? dev->type == &nd_pmem_device_type : false;
857 bool is_nd_blk(struct device *dev)
859 return dev ? dev->type == &nd_blk_device_type : false;
862 bool is_nd_volatile(struct device *dev)
864 return dev ? dev->type == &nd_volatile_device_type : false;
867 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
868 struct nd_namespace_index *nsindex)
870 struct nd_interleave_set *nd_set = nd_region->nd_set;
872 if (!nd_set)
873 return 0;
875 if (nsindex && __le16_to_cpu(nsindex->major) == 1
876 && __le16_to_cpu(nsindex->minor) == 1)
877 return nd_set->cookie1;
878 return nd_set->cookie2;
881 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
883 struct nd_interleave_set *nd_set = nd_region->nd_set;
885 if (nd_set)
886 return nd_set->altcookie;
887 return 0;
890 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
892 struct nd_label_ent *label_ent, *e;
894 lockdep_assert_held(&nd_mapping->lock);
895 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
896 list_del(&label_ent->list);
897 kfree(label_ent);
902 * When a namespace is activated create new seeds for the next
903 * namespace, or namespace-personality to be configured.
905 void nd_region_advance_seeds(struct nd_region *nd_region, struct device *dev)
907 nvdimm_bus_lock(dev);
908 if (nd_region->ns_seed == dev) {
909 nd_region_create_ns_seed(nd_region);
910 } else if (is_nd_btt(dev)) {
911 struct nd_btt *nd_btt = to_nd_btt(dev);
913 if (nd_region->btt_seed == dev)
914 nd_region_create_btt_seed(nd_region);
915 if (nd_region->ns_seed == &nd_btt->ndns->dev)
916 nd_region_create_ns_seed(nd_region);
917 } else if (is_nd_pfn(dev)) {
918 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
920 if (nd_region->pfn_seed == dev)
921 nd_region_create_pfn_seed(nd_region);
922 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
923 nd_region_create_ns_seed(nd_region);
924 } else if (is_nd_dax(dev)) {
925 struct nd_dax *nd_dax = to_nd_dax(dev);
927 if (nd_region->dax_seed == dev)
928 nd_region_create_dax_seed(nd_region);
929 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
930 nd_region_create_ns_seed(nd_region);
932 nvdimm_bus_unlock(dev);
935 int nd_blk_region_init(struct nd_region *nd_region)
937 struct device *dev = &nd_region->dev;
938 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
940 if (!is_nd_blk(dev))
941 return 0;
943 if (nd_region->ndr_mappings < 1) {
944 dev_dbg(dev, "invalid BLK region\n");
945 return -ENXIO;
948 return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
952 * nd_region_acquire_lane - allocate and lock a lane
953 * @nd_region: region id and number of lanes possible
955 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
956 * We optimize for the common case where there are 256 lanes, one
957 * per-cpu. For larger systems we need to lock to share lanes. For now
958 * this implementation assumes the cost of maintaining an allocator for
959 * free lanes is on the order of the lock hold time, so it implements a
960 * static lane = cpu % num_lanes mapping.
962 * In the case of a BTT instance on top of a BLK namespace a lane may be
963 * acquired recursively. We lock on the first instance.
965 * In the case of a BTT instance on top of PMEM, we only acquire a lane
966 * for the BTT metadata updates.
968 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
970 unsigned int cpu, lane;
972 cpu = get_cpu();
973 if (nd_region->num_lanes < nr_cpu_ids) {
974 struct nd_percpu_lane *ndl_lock, *ndl_count;
976 lane = cpu % nd_region->num_lanes;
977 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
978 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
979 if (ndl_count->count++ == 0)
980 spin_lock(&ndl_lock->lock);
981 } else
982 lane = cpu;
984 return lane;
986 EXPORT_SYMBOL(nd_region_acquire_lane);
988 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
990 if (nd_region->num_lanes < nr_cpu_ids) {
991 unsigned int cpu = get_cpu();
992 struct nd_percpu_lane *ndl_lock, *ndl_count;
994 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
995 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
996 if (--ndl_count->count == 0)
997 spin_unlock(&ndl_lock->lock);
998 put_cpu();
1000 put_cpu();
1002 EXPORT_SYMBOL(nd_region_release_lane);
1005 * PowerPC requires this alignment for memremap_pages(). All other archs
1006 * should be ok with SUBSECTION_SIZE (see memremap_compat_align()).
1008 #define MEMREMAP_COMPAT_ALIGN_MAX SZ_16M
1010 static unsigned long default_align(struct nd_region *nd_region)
1012 unsigned long align;
1013 int i, mappings;
1014 u32 remainder;
1016 if (is_nd_blk(&nd_region->dev))
1017 align = PAGE_SIZE;
1018 else
1019 align = MEMREMAP_COMPAT_ALIGN_MAX;
1021 for (i = 0; i < nd_region->ndr_mappings; i++) {
1022 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1023 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1025 if (test_bit(NDD_ALIASING, &nvdimm->flags)) {
1026 align = MEMREMAP_COMPAT_ALIGN_MAX;
1027 break;
1031 mappings = max_t(u16, 1, nd_region->ndr_mappings);
1032 div_u64_rem(align, mappings, &remainder);
1033 if (remainder)
1034 align *= mappings;
1036 return align;
1039 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
1040 struct nd_region_desc *ndr_desc,
1041 const struct device_type *dev_type, const char *caller)
1043 struct nd_region *nd_region;
1044 struct device *dev;
1045 void *region_buf;
1046 unsigned int i;
1047 int ro = 0;
1049 for (i = 0; i < ndr_desc->num_mappings; i++) {
1050 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1051 struct nvdimm *nvdimm = mapping->nvdimm;
1053 if ((mapping->start | mapping->size) % PAGE_SIZE) {
1054 dev_err(&nvdimm_bus->dev,
1055 "%s: %s mapping%d is not %ld aligned\n",
1056 caller, dev_name(&nvdimm->dev), i, PAGE_SIZE);
1057 return NULL;
1060 if (test_bit(NDD_UNARMED, &nvdimm->flags))
1061 ro = 1;
1063 if (test_bit(NDD_NOBLK, &nvdimm->flags)
1064 && dev_type == &nd_blk_device_type) {
1065 dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
1066 caller, dev_name(&nvdimm->dev), i);
1067 return NULL;
1071 if (dev_type == &nd_blk_device_type) {
1072 struct nd_blk_region_desc *ndbr_desc;
1073 struct nd_blk_region *ndbr;
1075 ndbr_desc = to_blk_region_desc(ndr_desc);
1076 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1077 * ndr_desc->num_mappings,
1078 GFP_KERNEL);
1079 if (ndbr) {
1080 nd_region = &ndbr->nd_region;
1081 ndbr->enable = ndbr_desc->enable;
1082 ndbr->do_io = ndbr_desc->do_io;
1084 region_buf = ndbr;
1085 } else {
1086 nd_region = kzalloc(struct_size(nd_region, mapping,
1087 ndr_desc->num_mappings),
1088 GFP_KERNEL);
1089 region_buf = nd_region;
1092 if (!region_buf)
1093 return NULL;
1094 nd_region->id = memregion_alloc(GFP_KERNEL);
1095 if (nd_region->id < 0)
1096 goto err_id;
1098 nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1099 if (!nd_region->lane)
1100 goto err_percpu;
1102 for (i = 0; i < nr_cpu_ids; i++) {
1103 struct nd_percpu_lane *ndl;
1105 ndl = per_cpu_ptr(nd_region->lane, i);
1106 spin_lock_init(&ndl->lock);
1107 ndl->count = 0;
1110 for (i = 0; i < ndr_desc->num_mappings; i++) {
1111 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1112 struct nvdimm *nvdimm = mapping->nvdimm;
1114 nd_region->mapping[i].nvdimm = nvdimm;
1115 nd_region->mapping[i].start = mapping->start;
1116 nd_region->mapping[i].size = mapping->size;
1117 nd_region->mapping[i].position = mapping->position;
1118 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1119 mutex_init(&nd_region->mapping[i].lock);
1121 get_device(&nvdimm->dev);
1123 nd_region->ndr_mappings = ndr_desc->num_mappings;
1124 nd_region->provider_data = ndr_desc->provider_data;
1125 nd_region->nd_set = ndr_desc->nd_set;
1126 nd_region->num_lanes = ndr_desc->num_lanes;
1127 nd_region->flags = ndr_desc->flags;
1128 nd_region->ro = ro;
1129 nd_region->numa_node = ndr_desc->numa_node;
1130 nd_region->target_node = ndr_desc->target_node;
1131 ida_init(&nd_region->ns_ida);
1132 ida_init(&nd_region->btt_ida);
1133 ida_init(&nd_region->pfn_ida);
1134 ida_init(&nd_region->dax_ida);
1135 dev = &nd_region->dev;
1136 dev_set_name(dev, "region%d", nd_region->id);
1137 dev->parent = &nvdimm_bus->dev;
1138 dev->type = dev_type;
1139 dev->groups = ndr_desc->attr_groups;
1140 dev->of_node = ndr_desc->of_node;
1141 nd_region->ndr_size = resource_size(ndr_desc->res);
1142 nd_region->ndr_start = ndr_desc->res->start;
1143 nd_region->align = default_align(nd_region);
1144 if (ndr_desc->flush)
1145 nd_region->flush = ndr_desc->flush;
1146 else
1147 nd_region->flush = NULL;
1149 nd_device_register(dev);
1151 return nd_region;
1153 err_percpu:
1154 memregion_free(nd_region->id);
1155 err_id:
1156 kfree(region_buf);
1157 return NULL;
1160 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1161 struct nd_region_desc *ndr_desc)
1163 ndr_desc->num_lanes = ND_MAX_LANES;
1164 return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1165 __func__);
1167 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1169 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1170 struct nd_region_desc *ndr_desc)
1172 if (ndr_desc->num_mappings > 1)
1173 return NULL;
1174 ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1175 return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1176 __func__);
1178 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1180 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1181 struct nd_region_desc *ndr_desc)
1183 ndr_desc->num_lanes = ND_MAX_LANES;
1184 return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1185 __func__);
1187 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1189 int nvdimm_flush(struct nd_region *nd_region, struct bio *bio)
1191 int rc = 0;
1193 if (!nd_region->flush)
1194 rc = generic_nvdimm_flush(nd_region);
1195 else {
1196 if (nd_region->flush(nd_region, bio))
1197 rc = -EIO;
1200 return rc;
1203 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1204 * @nd_region: blk or interleaved pmem region
1206 int generic_nvdimm_flush(struct nd_region *nd_region)
1208 struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1209 int i, idx;
1212 * Try to encourage some diversity in flush hint addresses
1213 * across cpus assuming a limited number of flush hints.
1215 idx = this_cpu_read(flush_idx);
1216 idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1219 * The first wmb() is needed to 'sfence' all previous writes
1220 * such that they are architecturally visible for the platform
1221 * buffer flush. Note that we've already arranged for pmem
1222 * writes to avoid the cache via memcpy_flushcache(). The final
1223 * wmb() ensures ordering for the NVDIMM flush write.
1225 wmb();
1226 for (i = 0; i < nd_region->ndr_mappings; i++)
1227 if (ndrd_get_flush_wpq(ndrd, i, 0))
1228 writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1229 wmb();
1231 return 0;
1233 EXPORT_SYMBOL_GPL(nvdimm_flush);
1236 * nvdimm_has_flush - determine write flushing requirements
1237 * @nd_region: blk or interleaved pmem region
1239 * Returns 1 if writes require flushing
1240 * Returns 0 if writes do not require flushing
1241 * Returns -ENXIO if flushing capability can not be determined
1243 int nvdimm_has_flush(struct nd_region *nd_region)
1245 int i;
1247 /* no nvdimm or pmem api == flushing capability unknown */
1248 if (nd_region->ndr_mappings == 0
1249 || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1250 return -ENXIO;
1252 for (i = 0; i < nd_region->ndr_mappings; i++) {
1253 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1254 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1256 /* flush hints present / available */
1257 if (nvdimm->num_flush)
1258 return 1;
1262 * The platform defines dimm devices without hints, assume
1263 * platform persistence mechanism like ADR
1265 return 0;
1267 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1269 int nvdimm_has_cache(struct nd_region *nd_region)
1271 return is_nd_pmem(&nd_region->dev) &&
1272 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1274 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1276 bool is_nvdimm_sync(struct nd_region *nd_region)
1278 if (is_nd_volatile(&nd_region->dev))
1279 return true;
1281 return is_nd_pmem(&nd_region->dev) &&
1282 !test_bit(ND_REGION_ASYNC, &nd_region->flags);
1284 EXPORT_SYMBOL_GPL(is_nvdimm_sync);
1286 struct conflict_context {
1287 struct nd_region *nd_region;
1288 resource_size_t start, size;
1291 static int region_conflict(struct device *dev, void *data)
1293 struct nd_region *nd_region;
1294 struct conflict_context *ctx = data;
1295 resource_size_t res_end, region_end, region_start;
1297 if (!is_memory(dev))
1298 return 0;
1300 nd_region = to_nd_region(dev);
1301 if (nd_region == ctx->nd_region)
1302 return 0;
1304 res_end = ctx->start + ctx->size;
1305 region_start = nd_region->ndr_start;
1306 region_end = region_start + nd_region->ndr_size;
1307 if (ctx->start >= region_start && ctx->start < region_end)
1308 return -EBUSY;
1309 if (res_end > region_start && res_end <= region_end)
1310 return -EBUSY;
1311 return 0;
1314 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1315 resource_size_t size)
1317 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1318 struct conflict_context ctx = {
1319 .nd_region = nd_region,
1320 .start = start,
1321 .size = size,
1324 return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);