Merge drm/drm-next into drm-misc-next
[linux-2.6-microblaze.git] / drivers / nvdimm / region_devs.c
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"
16
17 /*
18  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
19  * irrelevant.
20  */
21 #include <linux/io-64-nonatomic-hi-lo.h>
22
23 static DEFINE_PER_CPU(int, flush_idx);
24
25 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
26                 struct nd_region_data *ndrd)
27 {
28         int i, j;
29
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;
36
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);
41
42                         if (pfn == pfn_j)
43                                 break;
44                 }
45
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));
57         }
58
59         return 0;
60 }
61
62 int nd_region_activate(struct nd_region *nd_region)
63 {
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 *);
68
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;
73
74                 if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
75                         nvdimm_bus_unlock(&nd_region->dev);
76                         return -EBUSY;
77                 }
78
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 *);
85         }
86         nvdimm_bus_unlock(&nd_region->dev);
87
88         ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
89         if (!ndrd)
90                 return -ENOMEM;
91         dev_set_drvdata(dev, ndrd);
92
93         if (!num_flush)
94                 return 0;
95
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);
101
102                 if (rc)
103                         return rc;
104         }
105
106         /*
107          * Clear out entries that are duplicates. This should prevent the
108          * extra flushings.
109          */
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;
114
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);
119         }
120
121         return 0;
122 }
123
124 static void nd_region_release(struct device *dev)
125 {
126         struct nd_region *nd_region = to_nd_region(dev);
127         u16 i;
128
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;
132
133                 put_device(&nvdimm->dev);
134         }
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);
141 }
142
143 struct nd_region *to_nd_region(struct device *dev)
144 {
145         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
146
147         WARN_ON(dev->type->release != nd_region_release);
148         return nd_region;
149 }
150 EXPORT_SYMBOL_GPL(to_nd_region);
151
152 struct device *nd_region_dev(struct nd_region *nd_region)
153 {
154         if (!nd_region)
155                 return NULL;
156         return &nd_region->dev;
157 }
158 EXPORT_SYMBOL_GPL(nd_region_dev);
159
160 struct nd_blk_region *to_nd_blk_region(struct device *dev)
161 {
162         struct nd_region *nd_region = to_nd_region(dev);
163
164         WARN_ON(!is_nd_blk(dev));
165         return container_of(nd_region, struct nd_blk_region, nd_region);
166 }
167 EXPORT_SYMBOL_GPL(to_nd_blk_region);
168
169 void *nd_region_provider_data(struct nd_region *nd_region)
170 {
171         return nd_region->provider_data;
172 }
173 EXPORT_SYMBOL_GPL(nd_region_provider_data);
174
175 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
176 {
177         return ndbr->blk_provider_data;
178 }
179 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
180
181 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
182 {
183         ndbr->blk_provider_data = data;
184 }
185 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
186
187 /**
188  * nd_region_to_nstype() - region to an integer namespace type
189  * @nd_region: region-device to interrogate
190  *
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.
194  */
195 int nd_region_to_nstype(struct nd_region *nd_region)
196 {
197         if (is_memory(&nd_region->dev)) {
198                 u16 i, label;
199
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;
203
204                         if (test_bit(NDD_LABELING, &nvdimm->flags))
205                                 label++;
206                 }
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;
213         }
214
215         return 0;
216 }
217 EXPORT_SYMBOL(nd_region_to_nstype);
218
219 static unsigned long long region_size(struct nd_region *nd_region)
220 {
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];
225
226                 return nd_mapping->size;
227         }
228
229         return 0;
230 }
231
232 static ssize_t size_show(struct device *dev,
233                 struct device_attribute *attr, char *buf)
234 {
235         struct nd_region *nd_region = to_nd_region(dev);
236
237         return sprintf(buf, "%llu\n", region_size(nd_region));
238 }
239 static DEVICE_ATTR_RO(size);
240
241 static ssize_t deep_flush_show(struct device *dev,
242                 struct device_attribute *attr, char *buf)
243 {
244         struct nd_region *nd_region = to_nd_region(dev);
245
246         /*
247          * NOTE: in the nvdimm_has_flush() error case this attribute is
248          * not visible.
249          */
250         return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
251 }
252
253 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
254                 const char *buf, size_t len)
255 {
256         bool flush;
257         int rc = strtobool(buf, &flush);
258         struct nd_region *nd_region = to_nd_region(dev);
259
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;
267
268         return len;
269 }
270 static DEVICE_ATTR_RW(deep_flush);
271
272 static ssize_t mappings_show(struct device *dev,
273                 struct device_attribute *attr, char *buf)
274 {
275         struct nd_region *nd_region = to_nd_region(dev);
276
277         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
278 }
279 static DEVICE_ATTR_RO(mappings);
280
281 static ssize_t nstype_show(struct device *dev,
282                 struct device_attribute *attr, char *buf)
283 {
284         struct nd_region *nd_region = to_nd_region(dev);
285
286         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
287 }
288 static DEVICE_ATTR_RO(nstype);
289
290 static ssize_t set_cookie_show(struct device *dev,
291                 struct device_attribute *attr, char *buf)
292 {
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;
296
297         if (is_memory(dev) && nd_set)
298                 /* pass, should be precluded by region_visible */;
299         else
300                 return -ENXIO;
301
302         /*
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.
307          */
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);
314
315                 if (ndd) {
316                         struct nd_namespace_index *nsindex;
317
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));
322                 }
323         }
324         nvdimm_bus_unlock(dev);
325         nd_device_unlock(dev);
326
327         if (rc)
328                 return rc;
329         return sprintf(buf, "%#llx\n", nd_set->cookie1);
330 }
331 static DEVICE_ATTR_RO(set_cookie);
332
333 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
334 {
335         resource_size_t blk_max_overlap = 0, available, overlap;
336         int i;
337
338         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
339
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);
346
347                 /* if a dimm is disabled the available capacity is zero */
348                 if (!ndd)
349                         return 0;
350
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;
357                         }
358                 } else if (is_nd_blk(&nd_region->dev))
359                         available += nd_blk_available_dpa(nd_region);
360         }
361
362         return available;
363 }
364
365 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
366 {
367         resource_size_t available = 0;
368         int i;
369
370         if (is_memory(&nd_region->dev))
371                 available = PHYS_ADDR_MAX;
372
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];
376
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);
383         }
384         if (is_memory(&nd_region->dev))
385                 return available * nd_region->ndr_mappings;
386         return available;
387 }
388
389 static ssize_t available_size_show(struct device *dev,
390                 struct device_attribute *attr, char *buf)
391 {
392         struct nd_region *nd_region = to_nd_region(dev);
393         unsigned long long available = 0;
394
395         /*
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.
400          */
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);
407
408         return sprintf(buf, "%llu\n", available);
409 }
410 static DEVICE_ATTR_RO(available_size);
411
412 static ssize_t max_available_extent_show(struct device *dev,
413                 struct device_attribute *attr, char *buf)
414 {
415         struct nd_region *nd_region = to_nd_region(dev);
416         unsigned long long available = 0;
417
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);
424
425         return sprintf(buf, "%llu\n", available);
426 }
427 static DEVICE_ATTR_RO(max_available_extent);
428
429 static ssize_t init_namespaces_show(struct device *dev,
430                 struct device_attribute *attr, char *buf)
431 {
432         struct nd_region_data *ndrd = dev_get_drvdata(dev);
433         ssize_t rc;
434
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);
441
442         return rc;
443 }
444 static DEVICE_ATTR_RO(init_namespaces);
445
446 static ssize_t namespace_seed_show(struct device *dev,
447                 struct device_attribute *attr, char *buf)
448 {
449         struct nd_region *nd_region = to_nd_region(dev);
450         ssize_t rc;
451
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;
459 }
460 static DEVICE_ATTR_RO(namespace_seed);
461
462 static ssize_t btt_seed_show(struct device *dev,
463                 struct device_attribute *attr, char *buf)
464 {
465         struct nd_region *nd_region = to_nd_region(dev);
466         ssize_t rc;
467
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);
474
475         return rc;
476 }
477 static DEVICE_ATTR_RO(btt_seed);
478
479 static ssize_t pfn_seed_show(struct device *dev,
480                 struct device_attribute *attr, char *buf)
481 {
482         struct nd_region *nd_region = to_nd_region(dev);
483         ssize_t rc;
484
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);
491
492         return rc;
493 }
494 static DEVICE_ATTR_RO(pfn_seed);
495
496 static ssize_t dax_seed_show(struct device *dev,
497                 struct device_attribute *attr, char *buf)
498 {
499         struct nd_region *nd_region = to_nd_region(dev);
500         ssize_t rc;
501
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);
508
509         return rc;
510 }
511 static DEVICE_ATTR_RO(dax_seed);
512
513 static ssize_t read_only_show(struct device *dev,
514                 struct device_attribute *attr, char *buf)
515 {
516         struct nd_region *nd_region = to_nd_region(dev);
517
518         return sprintf(buf, "%d\n", nd_region->ro);
519 }
520
521 static int revalidate_read_only(struct device *dev, void *data)
522 {
523         nd_device_notify(dev, NVDIMM_REVALIDATE_REGION);
524         return 0;
525 }
526
527 static ssize_t read_only_store(struct device *dev,
528                 struct device_attribute *attr, const char *buf, size_t len)
529 {
530         bool ro;
531         int rc = strtobool(buf, &ro);
532         struct nd_region *nd_region = to_nd_region(dev);
533
534         if (rc)
535                 return rc;
536
537         nd_region->ro = ro;
538         device_for_each_child(dev, NULL, revalidate_read_only);
539         return len;
540 }
541 static DEVICE_ATTR_RW(read_only);
542
543 static ssize_t align_show(struct device *dev,
544                 struct device_attribute *attr, char *buf)
545 {
546         struct nd_region *nd_region = to_nd_region(dev);
547
548         return sprintf(buf, "%#lx\n", nd_region->align);
549 }
550
551 static ssize_t align_store(struct device *dev,
552                 struct device_attribute *attr, const char *buf, size_t len)
553 {
554         struct nd_region *nd_region = to_nd_region(dev);
555         unsigned long val, dpa;
556         u32 remainder;
557         int rc;
558
559         rc = kstrtoul(buf, 0, &val);
560         if (rc)
561                 return rc;
562
563         if (!nd_region->ndr_mappings)
564                 return -ENXIO;
565
566         /*
567          * Ensure space-align is evenly divisible by the region
568          * interleave-width because the kernel typically has no facility
569          * to determine which DIMM(s), dimm-physical-addresses, would
570          * contribute to the tail capacity in system-physical-address
571          * space for the namespace.
572          */
573         dpa = div_u64_rem(val, nd_region->ndr_mappings, &remainder);
574         if (!is_power_of_2(dpa) || dpa < PAGE_SIZE
575                         || val > region_size(nd_region) || remainder)
576                 return -EINVAL;
577
578         /*
579          * Given that space allocation consults this value multiple
580          * times ensure it does not change for the duration of the
581          * allocation.
582          */
583         nvdimm_bus_lock(dev);
584         nd_region->align = val;
585         nvdimm_bus_unlock(dev);
586
587         return len;
588 }
589 static DEVICE_ATTR_RW(align);
590
591 static ssize_t region_badblocks_show(struct device *dev,
592                 struct device_attribute *attr, char *buf)
593 {
594         struct nd_region *nd_region = to_nd_region(dev);
595         ssize_t rc;
596
597         nd_device_lock(dev);
598         if (dev->driver)
599                 rc = badblocks_show(&nd_region->bb, buf, 0);
600         else
601                 rc = -ENXIO;
602         nd_device_unlock(dev);
603
604         return rc;
605 }
606 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
607
608 static ssize_t resource_show(struct device *dev,
609                 struct device_attribute *attr, char *buf)
610 {
611         struct nd_region *nd_region = to_nd_region(dev);
612
613         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
614 }
615 static DEVICE_ATTR_ADMIN_RO(resource);
616
617 static ssize_t persistence_domain_show(struct device *dev,
618                 struct device_attribute *attr, char *buf)
619 {
620         struct nd_region *nd_region = to_nd_region(dev);
621
622         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
623                 return sprintf(buf, "cpu_cache\n");
624         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
625                 return sprintf(buf, "memory_controller\n");
626         else
627                 return sprintf(buf, "\n");
628 }
629 static DEVICE_ATTR_RO(persistence_domain);
630
631 static struct attribute *nd_region_attributes[] = {
632         &dev_attr_size.attr,
633         &dev_attr_align.attr,
634         &dev_attr_nstype.attr,
635         &dev_attr_mappings.attr,
636         &dev_attr_btt_seed.attr,
637         &dev_attr_pfn_seed.attr,
638         &dev_attr_dax_seed.attr,
639         &dev_attr_deep_flush.attr,
640         &dev_attr_read_only.attr,
641         &dev_attr_set_cookie.attr,
642         &dev_attr_available_size.attr,
643         &dev_attr_max_available_extent.attr,
644         &dev_attr_namespace_seed.attr,
645         &dev_attr_init_namespaces.attr,
646         &dev_attr_badblocks.attr,
647         &dev_attr_resource.attr,
648         &dev_attr_persistence_domain.attr,
649         NULL,
650 };
651
652 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
653 {
654         struct device *dev = container_of(kobj, typeof(*dev), kobj);
655         struct nd_region *nd_region = to_nd_region(dev);
656         struct nd_interleave_set *nd_set = nd_region->nd_set;
657         int type = nd_region_to_nstype(nd_region);
658
659         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
660                 return 0;
661
662         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
663                 return 0;
664
665         if (!is_memory(dev) && a == &dev_attr_badblocks.attr)
666                 return 0;
667
668         if (a == &dev_attr_resource.attr && !is_memory(dev))
669                 return 0;
670
671         if (a == &dev_attr_deep_flush.attr) {
672                 int has_flush = nvdimm_has_flush(nd_region);
673
674                 if (has_flush == 1)
675                         return a->mode;
676                 else if (has_flush == 0)
677                         return 0444;
678                 else
679                         return 0;
680         }
681
682         if (a == &dev_attr_persistence_domain.attr) {
683                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
684                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
685                         return 0;
686                 return a->mode;
687         }
688
689         if (a == &dev_attr_align.attr)
690                 return a->mode;
691
692         if (a != &dev_attr_set_cookie.attr
693                         && a != &dev_attr_available_size.attr)
694                 return a->mode;
695
696         if ((type == ND_DEVICE_NAMESPACE_PMEM
697                                 || type == ND_DEVICE_NAMESPACE_BLK)
698                         && a == &dev_attr_available_size.attr)
699                 return a->mode;
700         else if (is_memory(dev) && nd_set)
701                 return a->mode;
702
703         return 0;
704 }
705
706 static ssize_t mappingN(struct device *dev, char *buf, int n)
707 {
708         struct nd_region *nd_region = to_nd_region(dev);
709         struct nd_mapping *nd_mapping;
710         struct nvdimm *nvdimm;
711
712         if (n >= nd_region->ndr_mappings)
713                 return -ENXIO;
714         nd_mapping = &nd_region->mapping[n];
715         nvdimm = nd_mapping->nvdimm;
716
717         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
718                         nd_mapping->start, nd_mapping->size,
719                         nd_mapping->position);
720 }
721
722 #define REGION_MAPPING(idx) \
723 static ssize_t mapping##idx##_show(struct device *dev,          \
724                 struct device_attribute *attr, char *buf)       \
725 {                                                               \
726         return mappingN(dev, buf, idx);                         \
727 }                                                               \
728 static DEVICE_ATTR_RO(mapping##idx)
729
730 /*
731  * 32 should be enough for a while, even in the presence of socket
732  * interleave a 32-way interleave set is a degenerate case.
733  */
734 REGION_MAPPING(0);
735 REGION_MAPPING(1);
736 REGION_MAPPING(2);
737 REGION_MAPPING(3);
738 REGION_MAPPING(4);
739 REGION_MAPPING(5);
740 REGION_MAPPING(6);
741 REGION_MAPPING(7);
742 REGION_MAPPING(8);
743 REGION_MAPPING(9);
744 REGION_MAPPING(10);
745 REGION_MAPPING(11);
746 REGION_MAPPING(12);
747 REGION_MAPPING(13);
748 REGION_MAPPING(14);
749 REGION_MAPPING(15);
750 REGION_MAPPING(16);
751 REGION_MAPPING(17);
752 REGION_MAPPING(18);
753 REGION_MAPPING(19);
754 REGION_MAPPING(20);
755 REGION_MAPPING(21);
756 REGION_MAPPING(22);
757 REGION_MAPPING(23);
758 REGION_MAPPING(24);
759 REGION_MAPPING(25);
760 REGION_MAPPING(26);
761 REGION_MAPPING(27);
762 REGION_MAPPING(28);
763 REGION_MAPPING(29);
764 REGION_MAPPING(30);
765 REGION_MAPPING(31);
766
767 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
768 {
769         struct device *dev = container_of(kobj, struct device, kobj);
770         struct nd_region *nd_region = to_nd_region(dev);
771
772         if (n < nd_region->ndr_mappings)
773                 return a->mode;
774         return 0;
775 }
776
777 static struct attribute *mapping_attributes[] = {
778         &dev_attr_mapping0.attr,
779         &dev_attr_mapping1.attr,
780         &dev_attr_mapping2.attr,
781         &dev_attr_mapping3.attr,
782         &dev_attr_mapping4.attr,
783         &dev_attr_mapping5.attr,
784         &dev_attr_mapping6.attr,
785         &dev_attr_mapping7.attr,
786         &dev_attr_mapping8.attr,
787         &dev_attr_mapping9.attr,
788         &dev_attr_mapping10.attr,
789         &dev_attr_mapping11.attr,
790         &dev_attr_mapping12.attr,
791         &dev_attr_mapping13.attr,
792         &dev_attr_mapping14.attr,
793         &dev_attr_mapping15.attr,
794         &dev_attr_mapping16.attr,
795         &dev_attr_mapping17.attr,
796         &dev_attr_mapping18.attr,
797         &dev_attr_mapping19.attr,
798         &dev_attr_mapping20.attr,
799         &dev_attr_mapping21.attr,
800         &dev_attr_mapping22.attr,
801         &dev_attr_mapping23.attr,
802         &dev_attr_mapping24.attr,
803         &dev_attr_mapping25.attr,
804         &dev_attr_mapping26.attr,
805         &dev_attr_mapping27.attr,
806         &dev_attr_mapping28.attr,
807         &dev_attr_mapping29.attr,
808         &dev_attr_mapping30.attr,
809         &dev_attr_mapping31.attr,
810         NULL,
811 };
812
813 static const struct attribute_group nd_mapping_attribute_group = {
814         .is_visible = mapping_visible,
815         .attrs = mapping_attributes,
816 };
817
818 static const struct attribute_group nd_region_attribute_group = {
819         .attrs = nd_region_attributes,
820         .is_visible = region_visible,
821 };
822
823 static const struct attribute_group *nd_region_attribute_groups[] = {
824         &nd_device_attribute_group,
825         &nd_region_attribute_group,
826         &nd_numa_attribute_group,
827         &nd_mapping_attribute_group,
828         NULL,
829 };
830
831 static const struct device_type nd_blk_device_type = {
832         .name = "nd_blk",
833         .release = nd_region_release,
834         .groups = nd_region_attribute_groups,
835 };
836
837 static const struct device_type nd_pmem_device_type = {
838         .name = "nd_pmem",
839         .release = nd_region_release,
840         .groups = nd_region_attribute_groups,
841 };
842
843 static const struct device_type nd_volatile_device_type = {
844         .name = "nd_volatile",
845         .release = nd_region_release,
846         .groups = nd_region_attribute_groups,
847 };
848
849 bool is_nd_pmem(struct device *dev)
850 {
851         return dev ? dev->type == &nd_pmem_device_type : false;
852 }
853
854 bool is_nd_blk(struct device *dev)
855 {
856         return dev ? dev->type == &nd_blk_device_type : false;
857 }
858
859 bool is_nd_volatile(struct device *dev)
860 {
861         return dev ? dev->type == &nd_volatile_device_type : false;
862 }
863
864 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
865                 struct nd_namespace_index *nsindex)
866 {
867         struct nd_interleave_set *nd_set = nd_region->nd_set;
868
869         if (!nd_set)
870                 return 0;
871
872         if (nsindex && __le16_to_cpu(nsindex->major) == 1
873                         && __le16_to_cpu(nsindex->minor) == 1)
874                 return nd_set->cookie1;
875         return nd_set->cookie2;
876 }
877
878 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
879 {
880         struct nd_interleave_set *nd_set = nd_region->nd_set;
881
882         if (nd_set)
883                 return nd_set->altcookie;
884         return 0;
885 }
886
887 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
888 {
889         struct nd_label_ent *label_ent, *e;
890
891         lockdep_assert_held(&nd_mapping->lock);
892         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
893                 list_del(&label_ent->list);
894                 kfree(label_ent);
895         }
896 }
897
898 /*
899  * When a namespace is activated create new seeds for the next
900  * namespace, or namespace-personality to be configured.
901  */
902 void nd_region_advance_seeds(struct nd_region *nd_region, struct device *dev)
903 {
904         nvdimm_bus_lock(dev);
905         if (nd_region->ns_seed == dev) {
906                 nd_region_create_ns_seed(nd_region);
907         } else if (is_nd_btt(dev)) {
908                 struct nd_btt *nd_btt = to_nd_btt(dev);
909
910                 if (nd_region->btt_seed == dev)
911                         nd_region_create_btt_seed(nd_region);
912                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
913                         nd_region_create_ns_seed(nd_region);
914         } else if (is_nd_pfn(dev)) {
915                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
916
917                 if (nd_region->pfn_seed == dev)
918                         nd_region_create_pfn_seed(nd_region);
919                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
920                         nd_region_create_ns_seed(nd_region);
921         } else if (is_nd_dax(dev)) {
922                 struct nd_dax *nd_dax = to_nd_dax(dev);
923
924                 if (nd_region->dax_seed == dev)
925                         nd_region_create_dax_seed(nd_region);
926                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
927                         nd_region_create_ns_seed(nd_region);
928         }
929         nvdimm_bus_unlock(dev);
930 }
931
932 int nd_blk_region_init(struct nd_region *nd_region)
933 {
934         struct device *dev = &nd_region->dev;
935         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
936
937         if (!is_nd_blk(dev))
938                 return 0;
939
940         if (nd_region->ndr_mappings < 1) {
941                 dev_dbg(dev, "invalid BLK region\n");
942                 return -ENXIO;
943         }
944
945         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
946 }
947
948 /**
949  * nd_region_acquire_lane - allocate and lock a lane
950  * @nd_region: region id and number of lanes possible
951  *
952  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
953  * We optimize for the common case where there are 256 lanes, one
954  * per-cpu.  For larger systems we need to lock to share lanes.  For now
955  * this implementation assumes the cost of maintaining an allocator for
956  * free lanes is on the order of the lock hold time, so it implements a
957  * static lane = cpu % num_lanes mapping.
958  *
959  * In the case of a BTT instance on top of a BLK namespace a lane may be
960  * acquired recursively.  We lock on the first instance.
961  *
962  * In the case of a BTT instance on top of PMEM, we only acquire a lane
963  * for the BTT metadata updates.
964  */
965 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
966 {
967         unsigned int cpu, lane;
968
969         cpu = get_cpu();
970         if (nd_region->num_lanes < nr_cpu_ids) {
971                 struct nd_percpu_lane *ndl_lock, *ndl_count;
972
973                 lane = cpu % nd_region->num_lanes;
974                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
975                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
976                 if (ndl_count->count++ == 0)
977                         spin_lock(&ndl_lock->lock);
978         } else
979                 lane = cpu;
980
981         return lane;
982 }
983 EXPORT_SYMBOL(nd_region_acquire_lane);
984
985 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
986 {
987         if (nd_region->num_lanes < nr_cpu_ids) {
988                 unsigned int cpu = get_cpu();
989                 struct nd_percpu_lane *ndl_lock, *ndl_count;
990
991                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
992                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
993                 if (--ndl_count->count == 0)
994                         spin_unlock(&ndl_lock->lock);
995                 put_cpu();
996         }
997         put_cpu();
998 }
999 EXPORT_SYMBOL(nd_region_release_lane);
1000
1001 /*
1002  * PowerPC requires this alignment for memremap_pages(). All other archs
1003  * should be ok with SUBSECTION_SIZE (see memremap_compat_align()).
1004  */
1005 #define MEMREMAP_COMPAT_ALIGN_MAX SZ_16M
1006
1007 static unsigned long default_align(struct nd_region *nd_region)
1008 {
1009         unsigned long align;
1010         int i, mappings;
1011         u32 remainder;
1012
1013         if (is_nd_blk(&nd_region->dev))
1014                 align = PAGE_SIZE;
1015         else
1016                 align = MEMREMAP_COMPAT_ALIGN_MAX;
1017
1018         for (i = 0; i < nd_region->ndr_mappings; i++) {
1019                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1020                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1021
1022                 if (test_bit(NDD_ALIASING, &nvdimm->flags)) {
1023                         align = MEMREMAP_COMPAT_ALIGN_MAX;
1024                         break;
1025                 }
1026         }
1027
1028         mappings = max_t(u16, 1, nd_region->ndr_mappings);
1029         div_u64_rem(align, mappings, &remainder);
1030         if (remainder)
1031                 align *= mappings;
1032
1033         return align;
1034 }
1035
1036 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
1037                 struct nd_region_desc *ndr_desc,
1038                 const struct device_type *dev_type, const char *caller)
1039 {
1040         struct nd_region *nd_region;
1041         struct device *dev;
1042         void *region_buf;
1043         unsigned int i;
1044         int ro = 0;
1045
1046         for (i = 0; i < ndr_desc->num_mappings; i++) {
1047                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1048                 struct nvdimm *nvdimm = mapping->nvdimm;
1049
1050                 if ((mapping->start | mapping->size) % PAGE_SIZE) {
1051                         dev_err(&nvdimm_bus->dev,
1052                                 "%s: %s mapping%d is not %ld aligned\n",
1053                                 caller, dev_name(&nvdimm->dev), i, PAGE_SIZE);
1054                         return NULL;
1055                 }
1056
1057                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
1058                         ro = 1;
1059
1060                 if (test_bit(NDD_NOBLK, &nvdimm->flags)
1061                                 && dev_type == &nd_blk_device_type) {
1062                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
1063                                         caller, dev_name(&nvdimm->dev), i);
1064                         return NULL;
1065                 }
1066         }
1067
1068         if (dev_type == &nd_blk_device_type) {
1069                 struct nd_blk_region_desc *ndbr_desc;
1070                 struct nd_blk_region *ndbr;
1071
1072                 ndbr_desc = to_blk_region_desc(ndr_desc);
1073                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1074                                 * ndr_desc->num_mappings,
1075                                 GFP_KERNEL);
1076                 if (ndbr) {
1077                         nd_region = &ndbr->nd_region;
1078                         ndbr->enable = ndbr_desc->enable;
1079                         ndbr->do_io = ndbr_desc->do_io;
1080                 }
1081                 region_buf = ndbr;
1082         } else {
1083                 nd_region = kzalloc(struct_size(nd_region, mapping,
1084                                                 ndr_desc->num_mappings),
1085                                     GFP_KERNEL);
1086                 region_buf = nd_region;
1087         }
1088
1089         if (!region_buf)
1090                 return NULL;
1091         nd_region->id = memregion_alloc(GFP_KERNEL);
1092         if (nd_region->id < 0)
1093                 goto err_id;
1094
1095         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1096         if (!nd_region->lane)
1097                 goto err_percpu;
1098
1099         for (i = 0; i < nr_cpu_ids; i++) {
1100                 struct nd_percpu_lane *ndl;
1101
1102                 ndl = per_cpu_ptr(nd_region->lane, i);
1103                 spin_lock_init(&ndl->lock);
1104                 ndl->count = 0;
1105         }
1106
1107         for (i = 0; i < ndr_desc->num_mappings; i++) {
1108                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1109                 struct nvdimm *nvdimm = mapping->nvdimm;
1110
1111                 nd_region->mapping[i].nvdimm = nvdimm;
1112                 nd_region->mapping[i].start = mapping->start;
1113                 nd_region->mapping[i].size = mapping->size;
1114                 nd_region->mapping[i].position = mapping->position;
1115                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1116                 mutex_init(&nd_region->mapping[i].lock);
1117
1118                 get_device(&nvdimm->dev);
1119         }
1120         nd_region->ndr_mappings = ndr_desc->num_mappings;
1121         nd_region->provider_data = ndr_desc->provider_data;
1122         nd_region->nd_set = ndr_desc->nd_set;
1123         nd_region->num_lanes = ndr_desc->num_lanes;
1124         nd_region->flags = ndr_desc->flags;
1125         nd_region->ro = ro;
1126         nd_region->numa_node = ndr_desc->numa_node;
1127         nd_region->target_node = ndr_desc->target_node;
1128         ida_init(&nd_region->ns_ida);
1129         ida_init(&nd_region->btt_ida);
1130         ida_init(&nd_region->pfn_ida);
1131         ida_init(&nd_region->dax_ida);
1132         dev = &nd_region->dev;
1133         dev_set_name(dev, "region%d", nd_region->id);
1134         dev->parent = &nvdimm_bus->dev;
1135         dev->type = dev_type;
1136         dev->groups = ndr_desc->attr_groups;
1137         dev->of_node = ndr_desc->of_node;
1138         nd_region->ndr_size = resource_size(ndr_desc->res);
1139         nd_region->ndr_start = ndr_desc->res->start;
1140         nd_region->align = default_align(nd_region);
1141         if (ndr_desc->flush)
1142                 nd_region->flush = ndr_desc->flush;
1143         else
1144                 nd_region->flush = NULL;
1145
1146         nd_device_register(dev);
1147
1148         return nd_region;
1149
1150  err_percpu:
1151         memregion_free(nd_region->id);
1152  err_id:
1153         kfree(region_buf);
1154         return NULL;
1155 }
1156
1157 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1158                 struct nd_region_desc *ndr_desc)
1159 {
1160         ndr_desc->num_lanes = ND_MAX_LANES;
1161         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1162                         __func__);
1163 }
1164 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1165
1166 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1167                 struct nd_region_desc *ndr_desc)
1168 {
1169         if (ndr_desc->num_mappings > 1)
1170                 return NULL;
1171         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1172         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1173                         __func__);
1174 }
1175 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1176
1177 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1178                 struct nd_region_desc *ndr_desc)
1179 {
1180         ndr_desc->num_lanes = ND_MAX_LANES;
1181         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1182                         __func__);
1183 }
1184 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1185
1186 int nvdimm_flush(struct nd_region *nd_region, struct bio *bio)
1187 {
1188         int rc = 0;
1189
1190         if (!nd_region->flush)
1191                 rc = generic_nvdimm_flush(nd_region);
1192         else {
1193                 if (nd_region->flush(nd_region, bio))
1194                         rc = -EIO;
1195         }
1196
1197         return rc;
1198 }
1199 /**
1200  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1201  * @nd_region: blk or interleaved pmem region
1202  */
1203 int generic_nvdimm_flush(struct nd_region *nd_region)
1204 {
1205         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1206         int i, idx;
1207
1208         /*
1209          * Try to encourage some diversity in flush hint addresses
1210          * across cpus assuming a limited number of flush hints.
1211          */
1212         idx = this_cpu_read(flush_idx);
1213         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1214
1215         /*
1216          * The pmem_wmb() is needed to 'sfence' all
1217          * previous writes such that they are architecturally visible for
1218          * the platform buffer flush. Note that we've already arranged for pmem
1219          * writes to avoid the cache via memcpy_flushcache().  The final
1220          * wmb() ensures ordering for the NVDIMM flush write.
1221          */
1222         pmem_wmb();
1223         for (i = 0; i < nd_region->ndr_mappings; i++)
1224                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1225                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1226         wmb();
1227
1228         return 0;
1229 }
1230 EXPORT_SYMBOL_GPL(nvdimm_flush);
1231
1232 /**
1233  * nvdimm_has_flush - determine write flushing requirements
1234  * @nd_region: blk or interleaved pmem region
1235  *
1236  * Returns 1 if writes require flushing
1237  * Returns 0 if writes do not require flushing
1238  * Returns -ENXIO if flushing capability can not be determined
1239  */
1240 int nvdimm_has_flush(struct nd_region *nd_region)
1241 {
1242         int i;
1243
1244         /* no nvdimm or pmem api == flushing capability unknown */
1245         if (nd_region->ndr_mappings == 0
1246                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1247                 return -ENXIO;
1248
1249         /* Test if an explicit flush function is defined */
1250         if (test_bit(ND_REGION_ASYNC, &nd_region->flags) && nd_region->flush)
1251                 return 1;
1252
1253         /* Test if any flush hints for the region are available */
1254         for (i = 0; i < nd_region->ndr_mappings; i++) {
1255                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1256                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1257
1258                 /* flush hints present / available */
1259                 if (nvdimm->num_flush)
1260                         return 1;
1261         }
1262
1263         /*
1264          * The platform defines dimm devices without hints nor explicit flush,
1265          * assume platform persistence mechanism like ADR
1266          */
1267         return 0;
1268 }
1269 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1270
1271 int nvdimm_has_cache(struct nd_region *nd_region)
1272 {
1273         return is_nd_pmem(&nd_region->dev) &&
1274                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1275 }
1276 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1277
1278 bool is_nvdimm_sync(struct nd_region *nd_region)
1279 {
1280         if (is_nd_volatile(&nd_region->dev))
1281                 return true;
1282
1283         return is_nd_pmem(&nd_region->dev) &&
1284                 !test_bit(ND_REGION_ASYNC, &nd_region->flags);
1285 }
1286 EXPORT_SYMBOL_GPL(is_nvdimm_sync);
1287
1288 struct conflict_context {
1289         struct nd_region *nd_region;
1290         resource_size_t start, size;
1291 };
1292
1293 static int region_conflict(struct device *dev, void *data)
1294 {
1295         struct nd_region *nd_region;
1296         struct conflict_context *ctx = data;
1297         resource_size_t res_end, region_end, region_start;
1298
1299         if (!is_memory(dev))
1300                 return 0;
1301
1302         nd_region = to_nd_region(dev);
1303         if (nd_region == ctx->nd_region)
1304                 return 0;
1305
1306         res_end = ctx->start + ctx->size;
1307         region_start = nd_region->ndr_start;
1308         region_end = region_start + nd_region->ndr_size;
1309         if (ctx->start >= region_start && ctx->start < region_end)
1310                 return -EBUSY;
1311         if (res_end > region_start && res_end <= region_end)
1312                 return -EBUSY;
1313         return 0;
1314 }
1315
1316 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1317                 resource_size_t size)
1318 {
1319         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1320         struct conflict_context ctx = {
1321                 .nd_region = nd_region,
1322                 .start = start,
1323                 .size = size,
1324         };
1325
1326         return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);
1327 }