scsi: lpfc: Fix crash when cpu count is 1 and null irq affinity mask
[sfrench/cifs-2.6.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/highmem.h>
7 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/hash.h>
10 #include <linux/sort.h>
11 #include <linux/io.h>
12 #include <linux/nd.h>
13 #include "nd-core.h"
14 #include "nd.h"
15
16 /*
17  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
18  * irrelevant.
19  */
20 #include <linux/io-64-nonatomic-hi-lo.h>
21
22 static DEFINE_IDA(region_ida);
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         ida_simple_remove(&region_ida, 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 static struct device_type nd_blk_device_type = {
144         .name = "nd_blk",
145         .release = nd_region_release,
146 };
147
148 static struct device_type nd_pmem_device_type = {
149         .name = "nd_pmem",
150         .release = nd_region_release,
151 };
152
153 static struct device_type nd_volatile_device_type = {
154         .name = "nd_volatile",
155         .release = nd_region_release,
156 };
157
158 bool is_nd_pmem(struct device *dev)
159 {
160         return dev ? dev->type == &nd_pmem_device_type : false;
161 }
162
163 bool is_nd_blk(struct device *dev)
164 {
165         return dev ? dev->type == &nd_blk_device_type : false;
166 }
167
168 bool is_nd_volatile(struct device *dev)
169 {
170         return dev ? dev->type == &nd_volatile_device_type : false;
171 }
172
173 struct nd_region *to_nd_region(struct device *dev)
174 {
175         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
176
177         WARN_ON(dev->type->release != nd_region_release);
178         return nd_region;
179 }
180 EXPORT_SYMBOL_GPL(to_nd_region);
181
182 struct device *nd_region_dev(struct nd_region *nd_region)
183 {
184         if (!nd_region)
185                 return NULL;
186         return &nd_region->dev;
187 }
188 EXPORT_SYMBOL_GPL(nd_region_dev);
189
190 struct nd_blk_region *to_nd_blk_region(struct device *dev)
191 {
192         struct nd_region *nd_region = to_nd_region(dev);
193
194         WARN_ON(!is_nd_blk(dev));
195         return container_of(nd_region, struct nd_blk_region, nd_region);
196 }
197 EXPORT_SYMBOL_GPL(to_nd_blk_region);
198
199 void *nd_region_provider_data(struct nd_region *nd_region)
200 {
201         return nd_region->provider_data;
202 }
203 EXPORT_SYMBOL_GPL(nd_region_provider_data);
204
205 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
206 {
207         return ndbr->blk_provider_data;
208 }
209 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
210
211 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
212 {
213         ndbr->blk_provider_data = data;
214 }
215 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
216
217 /**
218  * nd_region_to_nstype() - region to an integer namespace type
219  * @nd_region: region-device to interrogate
220  *
221  * This is the 'nstype' attribute of a region as well, an input to the
222  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
223  * namespace devices with namespace drivers.
224  */
225 int nd_region_to_nstype(struct nd_region *nd_region)
226 {
227         if (is_memory(&nd_region->dev)) {
228                 u16 i, alias;
229
230                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
231                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
232                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
233
234                         if (test_bit(NDD_ALIASING, &nvdimm->flags))
235                                 alias++;
236                 }
237                 if (alias)
238                         return ND_DEVICE_NAMESPACE_PMEM;
239                 else
240                         return ND_DEVICE_NAMESPACE_IO;
241         } else if (is_nd_blk(&nd_region->dev)) {
242                 return ND_DEVICE_NAMESPACE_BLK;
243         }
244
245         return 0;
246 }
247 EXPORT_SYMBOL(nd_region_to_nstype);
248
249 static ssize_t size_show(struct device *dev,
250                 struct device_attribute *attr, char *buf)
251 {
252         struct nd_region *nd_region = to_nd_region(dev);
253         unsigned long long size = 0;
254
255         if (is_memory(dev)) {
256                 size = nd_region->ndr_size;
257         } else if (nd_region->ndr_mappings == 1) {
258                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
259
260                 size = nd_mapping->size;
261         }
262
263         return sprintf(buf, "%llu\n", size);
264 }
265 static DEVICE_ATTR_RO(size);
266
267 static ssize_t deep_flush_show(struct device *dev,
268                 struct device_attribute *attr, char *buf)
269 {
270         struct nd_region *nd_region = to_nd_region(dev);
271
272         /*
273          * NOTE: in the nvdimm_has_flush() error case this attribute is
274          * not visible.
275          */
276         return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
277 }
278
279 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
280                 const char *buf, size_t len)
281 {
282         bool flush;
283         int rc = strtobool(buf, &flush);
284         struct nd_region *nd_region = to_nd_region(dev);
285
286         if (rc)
287                 return rc;
288         if (!flush)
289                 return -EINVAL;
290         rc = nvdimm_flush(nd_region, NULL);
291         if (rc)
292                 return rc;
293
294         return len;
295 }
296 static DEVICE_ATTR_RW(deep_flush);
297
298 static ssize_t mappings_show(struct device *dev,
299                 struct device_attribute *attr, char *buf)
300 {
301         struct nd_region *nd_region = to_nd_region(dev);
302
303         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
304 }
305 static DEVICE_ATTR_RO(mappings);
306
307 static ssize_t nstype_show(struct device *dev,
308                 struct device_attribute *attr, char *buf)
309 {
310         struct nd_region *nd_region = to_nd_region(dev);
311
312         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
313 }
314 static DEVICE_ATTR_RO(nstype);
315
316 static ssize_t set_cookie_show(struct device *dev,
317                 struct device_attribute *attr, char *buf)
318 {
319         struct nd_region *nd_region = to_nd_region(dev);
320         struct nd_interleave_set *nd_set = nd_region->nd_set;
321         ssize_t rc = 0;
322
323         if (is_memory(dev) && nd_set)
324                 /* pass, should be precluded by region_visible */;
325         else
326                 return -ENXIO;
327
328         /*
329          * The cookie to show depends on which specification of the
330          * labels we are using. If there are not labels then default to
331          * the v1.1 namespace label cookie definition. To read all this
332          * data we need to wait for probing to settle.
333          */
334         device_lock(dev);
335         nvdimm_bus_lock(dev);
336         wait_nvdimm_bus_probe_idle(dev);
337         if (nd_region->ndr_mappings) {
338                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
339                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
340
341                 if (ndd) {
342                         struct nd_namespace_index *nsindex;
343
344                         nsindex = to_namespace_index(ndd, ndd->ns_current);
345                         rc = sprintf(buf, "%#llx\n",
346                                         nd_region_interleave_set_cookie(nd_region,
347                                                 nsindex));
348                 }
349         }
350         nvdimm_bus_unlock(dev);
351         device_unlock(dev);
352
353         if (rc)
354                 return rc;
355         return sprintf(buf, "%#llx\n", nd_set->cookie1);
356 }
357 static DEVICE_ATTR_RO(set_cookie);
358
359 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
360 {
361         resource_size_t blk_max_overlap = 0, available, overlap;
362         int i;
363
364         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
365
366  retry:
367         available = 0;
368         overlap = blk_max_overlap;
369         for (i = 0; i < nd_region->ndr_mappings; i++) {
370                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
371                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
372
373                 /* if a dimm is disabled the available capacity is zero */
374                 if (!ndd)
375                         return 0;
376
377                 if (is_memory(&nd_region->dev)) {
378                         available += nd_pmem_available_dpa(nd_region,
379                                         nd_mapping, &overlap);
380                         if (overlap > blk_max_overlap) {
381                                 blk_max_overlap = overlap;
382                                 goto retry;
383                         }
384                 } else if (is_nd_blk(&nd_region->dev))
385                         available += nd_blk_available_dpa(nd_region);
386         }
387
388         return available;
389 }
390
391 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
392 {
393         resource_size_t available = 0;
394         int i;
395
396         if (is_memory(&nd_region->dev))
397                 available = PHYS_ADDR_MAX;
398
399         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
400         for (i = 0; i < nd_region->ndr_mappings; i++) {
401                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
402
403                 if (is_memory(&nd_region->dev))
404                         available = min(available,
405                                         nd_pmem_max_contiguous_dpa(nd_region,
406                                                                    nd_mapping));
407                 else if (is_nd_blk(&nd_region->dev))
408                         available += nd_blk_available_dpa(nd_region);
409         }
410         if (is_memory(&nd_region->dev))
411                 return available * nd_region->ndr_mappings;
412         return available;
413 }
414
415 static ssize_t available_size_show(struct device *dev,
416                 struct device_attribute *attr, char *buf)
417 {
418         struct nd_region *nd_region = to_nd_region(dev);
419         unsigned long long available = 0;
420
421         /*
422          * Flush in-flight updates and grab a snapshot of the available
423          * size.  Of course, this value is potentially invalidated the
424          * memory nvdimm_bus_lock() is dropped, but that's userspace's
425          * problem to not race itself.
426          */
427         nvdimm_bus_lock(dev);
428         wait_nvdimm_bus_probe_idle(dev);
429         available = nd_region_available_dpa(nd_region);
430         nvdimm_bus_unlock(dev);
431
432         return sprintf(buf, "%llu\n", available);
433 }
434 static DEVICE_ATTR_RO(available_size);
435
436 static ssize_t max_available_extent_show(struct device *dev,
437                 struct device_attribute *attr, char *buf)
438 {
439         struct nd_region *nd_region = to_nd_region(dev);
440         unsigned long long available = 0;
441
442         nvdimm_bus_lock(dev);
443         wait_nvdimm_bus_probe_idle(dev);
444         available = nd_region_allocatable_dpa(nd_region);
445         nvdimm_bus_unlock(dev);
446
447         return sprintf(buf, "%llu\n", available);
448 }
449 static DEVICE_ATTR_RO(max_available_extent);
450
451 static ssize_t init_namespaces_show(struct device *dev,
452                 struct device_attribute *attr, char *buf)
453 {
454         struct nd_region_data *ndrd = dev_get_drvdata(dev);
455         ssize_t rc;
456
457         nvdimm_bus_lock(dev);
458         if (ndrd)
459                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
460         else
461                 rc = -ENXIO;
462         nvdimm_bus_unlock(dev);
463
464         return rc;
465 }
466 static DEVICE_ATTR_RO(init_namespaces);
467
468 static ssize_t namespace_seed_show(struct device *dev,
469                 struct device_attribute *attr, char *buf)
470 {
471         struct nd_region *nd_region = to_nd_region(dev);
472         ssize_t rc;
473
474         nvdimm_bus_lock(dev);
475         if (nd_region->ns_seed)
476                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
477         else
478                 rc = sprintf(buf, "\n");
479         nvdimm_bus_unlock(dev);
480         return rc;
481 }
482 static DEVICE_ATTR_RO(namespace_seed);
483
484 static ssize_t btt_seed_show(struct device *dev,
485                 struct device_attribute *attr, char *buf)
486 {
487         struct nd_region *nd_region = to_nd_region(dev);
488         ssize_t rc;
489
490         nvdimm_bus_lock(dev);
491         if (nd_region->btt_seed)
492                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
493         else
494                 rc = sprintf(buf, "\n");
495         nvdimm_bus_unlock(dev);
496
497         return rc;
498 }
499 static DEVICE_ATTR_RO(btt_seed);
500
501 static ssize_t pfn_seed_show(struct device *dev,
502                 struct device_attribute *attr, char *buf)
503 {
504         struct nd_region *nd_region = to_nd_region(dev);
505         ssize_t rc;
506
507         nvdimm_bus_lock(dev);
508         if (nd_region->pfn_seed)
509                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
510         else
511                 rc = sprintf(buf, "\n");
512         nvdimm_bus_unlock(dev);
513
514         return rc;
515 }
516 static DEVICE_ATTR_RO(pfn_seed);
517
518 static ssize_t dax_seed_show(struct device *dev,
519                 struct device_attribute *attr, char *buf)
520 {
521         struct nd_region *nd_region = to_nd_region(dev);
522         ssize_t rc;
523
524         nvdimm_bus_lock(dev);
525         if (nd_region->dax_seed)
526                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
527         else
528                 rc = sprintf(buf, "\n");
529         nvdimm_bus_unlock(dev);
530
531         return rc;
532 }
533 static DEVICE_ATTR_RO(dax_seed);
534
535 static ssize_t read_only_show(struct device *dev,
536                 struct device_attribute *attr, char *buf)
537 {
538         struct nd_region *nd_region = to_nd_region(dev);
539
540         return sprintf(buf, "%d\n", nd_region->ro);
541 }
542
543 static ssize_t read_only_store(struct device *dev,
544                 struct device_attribute *attr, const char *buf, size_t len)
545 {
546         bool ro;
547         int rc = strtobool(buf, &ro);
548         struct nd_region *nd_region = to_nd_region(dev);
549
550         if (rc)
551                 return rc;
552
553         nd_region->ro = ro;
554         return len;
555 }
556 static DEVICE_ATTR_RW(read_only);
557
558 static ssize_t region_badblocks_show(struct device *dev,
559                 struct device_attribute *attr, char *buf)
560 {
561         struct nd_region *nd_region = to_nd_region(dev);
562         ssize_t rc;
563
564         device_lock(dev);
565         if (dev->driver)
566                 rc = badblocks_show(&nd_region->bb, buf, 0);
567         else
568                 rc = -ENXIO;
569         device_unlock(dev);
570
571         return rc;
572 }
573 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
574
575 static ssize_t resource_show(struct device *dev,
576                 struct device_attribute *attr, char *buf)
577 {
578         struct nd_region *nd_region = to_nd_region(dev);
579
580         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
581 }
582 static DEVICE_ATTR_RO(resource);
583
584 static ssize_t persistence_domain_show(struct device *dev,
585                 struct device_attribute *attr, char *buf)
586 {
587         struct nd_region *nd_region = to_nd_region(dev);
588
589         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
590                 return sprintf(buf, "cpu_cache\n");
591         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
592                 return sprintf(buf, "memory_controller\n");
593         else
594                 return sprintf(buf, "\n");
595 }
596 static DEVICE_ATTR_RO(persistence_domain);
597
598 static struct attribute *nd_region_attributes[] = {
599         &dev_attr_size.attr,
600         &dev_attr_nstype.attr,
601         &dev_attr_mappings.attr,
602         &dev_attr_btt_seed.attr,
603         &dev_attr_pfn_seed.attr,
604         &dev_attr_dax_seed.attr,
605         &dev_attr_deep_flush.attr,
606         &dev_attr_read_only.attr,
607         &dev_attr_set_cookie.attr,
608         &dev_attr_available_size.attr,
609         &dev_attr_max_available_extent.attr,
610         &dev_attr_namespace_seed.attr,
611         &dev_attr_init_namespaces.attr,
612         &dev_attr_badblocks.attr,
613         &dev_attr_resource.attr,
614         &dev_attr_persistence_domain.attr,
615         NULL,
616 };
617
618 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
619 {
620         struct device *dev = container_of(kobj, typeof(*dev), kobj);
621         struct nd_region *nd_region = to_nd_region(dev);
622         struct nd_interleave_set *nd_set = nd_region->nd_set;
623         int type = nd_region_to_nstype(nd_region);
624
625         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
626                 return 0;
627
628         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
629                 return 0;
630
631         if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
632                 return 0;
633
634         if (a == &dev_attr_resource.attr) {
635                 if (is_nd_pmem(dev))
636                         return 0400;
637                 else
638                         return 0;
639         }
640
641         if (a == &dev_attr_deep_flush.attr) {
642                 int has_flush = nvdimm_has_flush(nd_region);
643
644                 if (has_flush == 1)
645                         return a->mode;
646                 else if (has_flush == 0)
647                         return 0444;
648                 else
649                         return 0;
650         }
651
652         if (a == &dev_attr_persistence_domain.attr) {
653                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
654                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
655                         return 0;
656                 return a->mode;
657         }
658
659         if (a != &dev_attr_set_cookie.attr
660                         && a != &dev_attr_available_size.attr)
661                 return a->mode;
662
663         if ((type == ND_DEVICE_NAMESPACE_PMEM
664                                 || type == ND_DEVICE_NAMESPACE_BLK)
665                         && a == &dev_attr_available_size.attr)
666                 return a->mode;
667         else if (is_memory(dev) && nd_set)
668                 return a->mode;
669
670         return 0;
671 }
672
673 struct attribute_group nd_region_attribute_group = {
674         .attrs = nd_region_attributes,
675         .is_visible = region_visible,
676 };
677 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
678
679 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
680                 struct nd_namespace_index *nsindex)
681 {
682         struct nd_interleave_set *nd_set = nd_region->nd_set;
683
684         if (!nd_set)
685                 return 0;
686
687         if (nsindex && __le16_to_cpu(nsindex->major) == 1
688                         && __le16_to_cpu(nsindex->minor) == 1)
689                 return nd_set->cookie1;
690         return nd_set->cookie2;
691 }
692
693 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
694 {
695         struct nd_interleave_set *nd_set = nd_region->nd_set;
696
697         if (nd_set)
698                 return nd_set->altcookie;
699         return 0;
700 }
701
702 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
703 {
704         struct nd_label_ent *label_ent, *e;
705
706         lockdep_assert_held(&nd_mapping->lock);
707         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
708                 list_del(&label_ent->list);
709                 kfree(label_ent);
710         }
711 }
712
713 /*
714  * Upon successful probe/remove, take/release a reference on the
715  * associated interleave set (if present), and plant new btt + namespace
716  * seeds.  Also, on the removal of a BLK region, notify the provider to
717  * disable the region.
718  */
719 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
720                 struct device *dev, bool probe)
721 {
722         struct nd_region *nd_region;
723
724         if (!probe && is_nd_region(dev)) {
725                 int i;
726
727                 nd_region = to_nd_region(dev);
728                 for (i = 0; i < nd_region->ndr_mappings; i++) {
729                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
730                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
731                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
732
733                         mutex_lock(&nd_mapping->lock);
734                         nd_mapping_free_labels(nd_mapping);
735                         mutex_unlock(&nd_mapping->lock);
736
737                         put_ndd(ndd);
738                         nd_mapping->ndd = NULL;
739                         if (ndd)
740                                 atomic_dec(&nvdimm->busy);
741                 }
742         }
743         if (dev->parent && is_nd_region(dev->parent) && probe) {
744                 nd_region = to_nd_region(dev->parent);
745                 nvdimm_bus_lock(dev);
746                 if (nd_region->ns_seed == dev)
747                         nd_region_create_ns_seed(nd_region);
748                 nvdimm_bus_unlock(dev);
749         }
750         if (is_nd_btt(dev) && probe) {
751                 struct nd_btt *nd_btt = to_nd_btt(dev);
752
753                 nd_region = to_nd_region(dev->parent);
754                 nvdimm_bus_lock(dev);
755                 if (nd_region->btt_seed == dev)
756                         nd_region_create_btt_seed(nd_region);
757                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
758                         nd_region_create_ns_seed(nd_region);
759                 nvdimm_bus_unlock(dev);
760         }
761         if (is_nd_pfn(dev) && probe) {
762                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
763
764                 nd_region = to_nd_region(dev->parent);
765                 nvdimm_bus_lock(dev);
766                 if (nd_region->pfn_seed == dev)
767                         nd_region_create_pfn_seed(nd_region);
768                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
769                         nd_region_create_ns_seed(nd_region);
770                 nvdimm_bus_unlock(dev);
771         }
772         if (is_nd_dax(dev) && probe) {
773                 struct nd_dax *nd_dax = to_nd_dax(dev);
774
775                 nd_region = to_nd_region(dev->parent);
776                 nvdimm_bus_lock(dev);
777                 if (nd_region->dax_seed == dev)
778                         nd_region_create_dax_seed(nd_region);
779                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
780                         nd_region_create_ns_seed(nd_region);
781                 nvdimm_bus_unlock(dev);
782         }
783 }
784
785 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
786 {
787         nd_region_notify_driver_action(nvdimm_bus, dev, true);
788 }
789
790 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
791 {
792         nd_region_notify_driver_action(nvdimm_bus, dev, false);
793 }
794
795 static ssize_t mappingN(struct device *dev, char *buf, int n)
796 {
797         struct nd_region *nd_region = to_nd_region(dev);
798         struct nd_mapping *nd_mapping;
799         struct nvdimm *nvdimm;
800
801         if (n >= nd_region->ndr_mappings)
802                 return -ENXIO;
803         nd_mapping = &nd_region->mapping[n];
804         nvdimm = nd_mapping->nvdimm;
805
806         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
807                         nd_mapping->start, nd_mapping->size,
808                         nd_mapping->position);
809 }
810
811 #define REGION_MAPPING(idx) \
812 static ssize_t mapping##idx##_show(struct device *dev,          \
813                 struct device_attribute *attr, char *buf)       \
814 {                                                               \
815         return mappingN(dev, buf, idx);                         \
816 }                                                               \
817 static DEVICE_ATTR_RO(mapping##idx)
818
819 /*
820  * 32 should be enough for a while, even in the presence of socket
821  * interleave a 32-way interleave set is a degenerate case.
822  */
823 REGION_MAPPING(0);
824 REGION_MAPPING(1);
825 REGION_MAPPING(2);
826 REGION_MAPPING(3);
827 REGION_MAPPING(4);
828 REGION_MAPPING(5);
829 REGION_MAPPING(6);
830 REGION_MAPPING(7);
831 REGION_MAPPING(8);
832 REGION_MAPPING(9);
833 REGION_MAPPING(10);
834 REGION_MAPPING(11);
835 REGION_MAPPING(12);
836 REGION_MAPPING(13);
837 REGION_MAPPING(14);
838 REGION_MAPPING(15);
839 REGION_MAPPING(16);
840 REGION_MAPPING(17);
841 REGION_MAPPING(18);
842 REGION_MAPPING(19);
843 REGION_MAPPING(20);
844 REGION_MAPPING(21);
845 REGION_MAPPING(22);
846 REGION_MAPPING(23);
847 REGION_MAPPING(24);
848 REGION_MAPPING(25);
849 REGION_MAPPING(26);
850 REGION_MAPPING(27);
851 REGION_MAPPING(28);
852 REGION_MAPPING(29);
853 REGION_MAPPING(30);
854 REGION_MAPPING(31);
855
856 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
857 {
858         struct device *dev = container_of(kobj, struct device, kobj);
859         struct nd_region *nd_region = to_nd_region(dev);
860
861         if (n < nd_region->ndr_mappings)
862                 return a->mode;
863         return 0;
864 }
865
866 static struct attribute *mapping_attributes[] = {
867         &dev_attr_mapping0.attr,
868         &dev_attr_mapping1.attr,
869         &dev_attr_mapping2.attr,
870         &dev_attr_mapping3.attr,
871         &dev_attr_mapping4.attr,
872         &dev_attr_mapping5.attr,
873         &dev_attr_mapping6.attr,
874         &dev_attr_mapping7.attr,
875         &dev_attr_mapping8.attr,
876         &dev_attr_mapping9.attr,
877         &dev_attr_mapping10.attr,
878         &dev_attr_mapping11.attr,
879         &dev_attr_mapping12.attr,
880         &dev_attr_mapping13.attr,
881         &dev_attr_mapping14.attr,
882         &dev_attr_mapping15.attr,
883         &dev_attr_mapping16.attr,
884         &dev_attr_mapping17.attr,
885         &dev_attr_mapping18.attr,
886         &dev_attr_mapping19.attr,
887         &dev_attr_mapping20.attr,
888         &dev_attr_mapping21.attr,
889         &dev_attr_mapping22.attr,
890         &dev_attr_mapping23.attr,
891         &dev_attr_mapping24.attr,
892         &dev_attr_mapping25.attr,
893         &dev_attr_mapping26.attr,
894         &dev_attr_mapping27.attr,
895         &dev_attr_mapping28.attr,
896         &dev_attr_mapping29.attr,
897         &dev_attr_mapping30.attr,
898         &dev_attr_mapping31.attr,
899         NULL,
900 };
901
902 struct attribute_group nd_mapping_attribute_group = {
903         .is_visible = mapping_visible,
904         .attrs = mapping_attributes,
905 };
906 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
907
908 int nd_blk_region_init(struct nd_region *nd_region)
909 {
910         struct device *dev = &nd_region->dev;
911         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
912
913         if (!is_nd_blk(dev))
914                 return 0;
915
916         if (nd_region->ndr_mappings < 1) {
917                 dev_dbg(dev, "invalid BLK region\n");
918                 return -ENXIO;
919         }
920
921         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
922 }
923
924 /**
925  * nd_region_acquire_lane - allocate and lock a lane
926  * @nd_region: region id and number of lanes possible
927  *
928  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
929  * We optimize for the common case where there are 256 lanes, one
930  * per-cpu.  For larger systems we need to lock to share lanes.  For now
931  * this implementation assumes the cost of maintaining an allocator for
932  * free lanes is on the order of the lock hold time, so it implements a
933  * static lane = cpu % num_lanes mapping.
934  *
935  * In the case of a BTT instance on top of a BLK namespace a lane may be
936  * acquired recursively.  We lock on the first instance.
937  *
938  * In the case of a BTT instance on top of PMEM, we only acquire a lane
939  * for the BTT metadata updates.
940  */
941 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
942 {
943         unsigned int cpu, lane;
944
945         cpu = get_cpu();
946         if (nd_region->num_lanes < nr_cpu_ids) {
947                 struct nd_percpu_lane *ndl_lock, *ndl_count;
948
949                 lane = cpu % nd_region->num_lanes;
950                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
951                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
952                 if (ndl_count->count++ == 0)
953                         spin_lock(&ndl_lock->lock);
954         } else
955                 lane = cpu;
956
957         return lane;
958 }
959 EXPORT_SYMBOL(nd_region_acquire_lane);
960
961 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
962 {
963         if (nd_region->num_lanes < nr_cpu_ids) {
964                 unsigned int cpu = get_cpu();
965                 struct nd_percpu_lane *ndl_lock, *ndl_count;
966
967                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
968                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
969                 if (--ndl_count->count == 0)
970                         spin_unlock(&ndl_lock->lock);
971                 put_cpu();
972         }
973         put_cpu();
974 }
975 EXPORT_SYMBOL(nd_region_release_lane);
976
977 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
978                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
979                 const char *caller)
980 {
981         struct nd_region *nd_region;
982         struct device *dev;
983         void *region_buf;
984         unsigned int i;
985         int ro = 0;
986
987         for (i = 0; i < ndr_desc->num_mappings; i++) {
988                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
989                 struct nvdimm *nvdimm = mapping->nvdimm;
990
991                 if ((mapping->start | mapping->size) % SZ_4K) {
992                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
993                                         caller, dev_name(&nvdimm->dev), i);
994
995                         return NULL;
996                 }
997
998                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
999                         ro = 1;
1000
1001                 if (test_bit(NDD_NOBLK, &nvdimm->flags)
1002                                 && dev_type == &nd_blk_device_type) {
1003                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
1004                                         caller, dev_name(&nvdimm->dev), i);
1005                         return NULL;
1006                 }
1007         }
1008
1009         if (dev_type == &nd_blk_device_type) {
1010                 struct nd_blk_region_desc *ndbr_desc;
1011                 struct nd_blk_region *ndbr;
1012
1013                 ndbr_desc = to_blk_region_desc(ndr_desc);
1014                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1015                                 * ndr_desc->num_mappings,
1016                                 GFP_KERNEL);
1017                 if (ndbr) {
1018                         nd_region = &ndbr->nd_region;
1019                         ndbr->enable = ndbr_desc->enable;
1020                         ndbr->do_io = ndbr_desc->do_io;
1021                 }
1022                 region_buf = ndbr;
1023         } else {
1024                 nd_region = kzalloc(sizeof(struct nd_region)
1025                                 + sizeof(struct nd_mapping)
1026                                 * ndr_desc->num_mappings,
1027                                 GFP_KERNEL);
1028                 region_buf = nd_region;
1029         }
1030
1031         if (!region_buf)
1032                 return NULL;
1033         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
1034         if (nd_region->id < 0)
1035                 goto err_id;
1036
1037         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1038         if (!nd_region->lane)
1039                 goto err_percpu;
1040
1041         for (i = 0; i < nr_cpu_ids; i++) {
1042                 struct nd_percpu_lane *ndl;
1043
1044                 ndl = per_cpu_ptr(nd_region->lane, i);
1045                 spin_lock_init(&ndl->lock);
1046                 ndl->count = 0;
1047         }
1048
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;
1052
1053                 nd_region->mapping[i].nvdimm = nvdimm;
1054                 nd_region->mapping[i].start = mapping->start;
1055                 nd_region->mapping[i].size = mapping->size;
1056                 nd_region->mapping[i].position = mapping->position;
1057                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1058                 mutex_init(&nd_region->mapping[i].lock);
1059
1060                 get_device(&nvdimm->dev);
1061         }
1062         nd_region->ndr_mappings = ndr_desc->num_mappings;
1063         nd_region->provider_data = ndr_desc->provider_data;
1064         nd_region->nd_set = ndr_desc->nd_set;
1065         nd_region->num_lanes = ndr_desc->num_lanes;
1066         nd_region->flags = ndr_desc->flags;
1067         nd_region->ro = ro;
1068         nd_region->numa_node = ndr_desc->numa_node;
1069         nd_region->target_node = ndr_desc->target_node;
1070         ida_init(&nd_region->ns_ida);
1071         ida_init(&nd_region->btt_ida);
1072         ida_init(&nd_region->pfn_ida);
1073         ida_init(&nd_region->dax_ida);
1074         dev = &nd_region->dev;
1075         dev_set_name(dev, "region%d", nd_region->id);
1076         dev->parent = &nvdimm_bus->dev;
1077         dev->type = dev_type;
1078         dev->groups = ndr_desc->attr_groups;
1079         dev->of_node = ndr_desc->of_node;
1080         nd_region->ndr_size = resource_size(ndr_desc->res);
1081         nd_region->ndr_start = ndr_desc->res->start;
1082         if (ndr_desc->flush)
1083                 nd_region->flush = ndr_desc->flush;
1084         else
1085                 nd_region->flush = NULL;
1086
1087         nd_device_register(dev);
1088
1089         return nd_region;
1090
1091  err_percpu:
1092         ida_simple_remove(&region_ida, nd_region->id);
1093  err_id:
1094         kfree(region_buf);
1095         return NULL;
1096 }
1097
1098 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1099                 struct nd_region_desc *ndr_desc)
1100 {
1101         ndr_desc->num_lanes = ND_MAX_LANES;
1102         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1103                         __func__);
1104 }
1105 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1106
1107 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1108                 struct nd_region_desc *ndr_desc)
1109 {
1110         if (ndr_desc->num_mappings > 1)
1111                 return NULL;
1112         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1113         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1114                         __func__);
1115 }
1116 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1117
1118 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1119                 struct nd_region_desc *ndr_desc)
1120 {
1121         ndr_desc->num_lanes = ND_MAX_LANES;
1122         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1123                         __func__);
1124 }
1125 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1126
1127 int nvdimm_flush(struct nd_region *nd_region, struct bio *bio)
1128 {
1129         int rc = 0;
1130
1131         if (!nd_region->flush)
1132                 rc = generic_nvdimm_flush(nd_region);
1133         else {
1134                 if (nd_region->flush(nd_region, bio))
1135                         rc = -EIO;
1136         }
1137
1138         return rc;
1139 }
1140 /**
1141  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1142  * @nd_region: blk or interleaved pmem region
1143  */
1144 int generic_nvdimm_flush(struct nd_region *nd_region)
1145 {
1146         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1147         int i, idx;
1148
1149         /*
1150          * Try to encourage some diversity in flush hint addresses
1151          * across cpus assuming a limited number of flush hints.
1152          */
1153         idx = this_cpu_read(flush_idx);
1154         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1155
1156         /*
1157          * The first wmb() is needed to 'sfence' all previous writes
1158          * such that they are architecturally visible for the platform
1159          * buffer flush.  Note that we've already arranged for pmem
1160          * writes to avoid the cache via memcpy_flushcache().  The final
1161          * wmb() ensures ordering for the NVDIMM flush write.
1162          */
1163         wmb();
1164         for (i = 0; i < nd_region->ndr_mappings; i++)
1165                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1166                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1167         wmb();
1168
1169         return 0;
1170 }
1171 EXPORT_SYMBOL_GPL(nvdimm_flush);
1172
1173 /**
1174  * nvdimm_has_flush - determine write flushing requirements
1175  * @nd_region: blk or interleaved pmem region
1176  *
1177  * Returns 1 if writes require flushing
1178  * Returns 0 if writes do not require flushing
1179  * Returns -ENXIO if flushing capability can not be determined
1180  */
1181 int nvdimm_has_flush(struct nd_region *nd_region)
1182 {
1183         int i;
1184
1185         /* no nvdimm or pmem api == flushing capability unknown */
1186         if (nd_region->ndr_mappings == 0
1187                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1188                 return -ENXIO;
1189
1190         for (i = 0; i < nd_region->ndr_mappings; i++) {
1191                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1192                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1193
1194                 /* flush hints present / available */
1195                 if (nvdimm->num_flush)
1196                         return 1;
1197         }
1198
1199         /*
1200          * The platform defines dimm devices without hints, assume
1201          * platform persistence mechanism like ADR
1202          */
1203         return 0;
1204 }
1205 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1206
1207 int nvdimm_has_cache(struct nd_region *nd_region)
1208 {
1209         return is_nd_pmem(&nd_region->dev) &&
1210                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1211 }
1212 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1213
1214 bool is_nvdimm_sync(struct nd_region *nd_region)
1215 {
1216         return is_nd_pmem(&nd_region->dev) &&
1217                 !test_bit(ND_REGION_ASYNC, &nd_region->flags);
1218 }
1219 EXPORT_SYMBOL_GPL(is_nvdimm_sync);
1220
1221 struct conflict_context {
1222         struct nd_region *nd_region;
1223         resource_size_t start, size;
1224 };
1225
1226 static int region_conflict(struct device *dev, void *data)
1227 {
1228         struct nd_region *nd_region;
1229         struct conflict_context *ctx = data;
1230         resource_size_t res_end, region_end, region_start;
1231
1232         if (!is_memory(dev))
1233                 return 0;
1234
1235         nd_region = to_nd_region(dev);
1236         if (nd_region == ctx->nd_region)
1237                 return 0;
1238
1239         res_end = ctx->start + ctx->size;
1240         region_start = nd_region->ndr_start;
1241         region_end = region_start + nd_region->ndr_size;
1242         if (ctx->start >= region_start && ctx->start < region_end)
1243                 return -EBUSY;
1244         if (res_end > region_start && res_end <= region_end)
1245                 return -EBUSY;
1246         return 0;
1247 }
1248
1249 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1250                 resource_size_t size)
1251 {
1252         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1253         struct conflict_context ctx = {
1254                 .nd_region = nd_region,
1255                 .start = start,
1256                 .size = size,
1257         };
1258
1259         return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);
1260 }
1261
1262 void __exit nd_region_devs_exit(void)
1263 {
1264         ida_destroy(&region_ida);
1265 }