source: src/linux/universal/linux-4.9/drivers/nvdimm/region_devs.c @ 31662

Last change on this file since 31662 was 31662, checked in by brainslayer, 4 months ago

use new squashfs in all kernels

File size: 25.5 KB
Line 
1/*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11 * General Public License for more details.
12 */
13#include <linux/scatterlist.h>
14#include <linux/highmem.h>
15#include <linux/sched.h>
16#include <linux/slab.h>
17#include <linux/hash.h>
18#include <linux/pmem.h>
19#include <linux/sort.h>
20#include <linux/io.h>
21#include <linux/nd.h>
22#include "nd-core.h"
23#include "nd.h"
24
25/*
26 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
27 * irrelevant.
28 */
29#include <linux/io-64-nonatomic-hi-lo.h>
30
31static DEFINE_IDA(region_ida);
32static DEFINE_PER_CPU(int, flush_idx);
33
34static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
35                struct nd_region_data *ndrd)
36{
37        int i, j;
38
39        dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
40                        nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
41        for (i = 0; i < (1 << ndrd->hints_shift); i++) {
42                struct resource *res = &nvdimm->flush_wpq[i];
43                unsigned long pfn = PHYS_PFN(res->start);
44                void __iomem *flush_page;
45
46                /* check if flush hints share a page */
47                for (j = 0; j < i; j++) {
48                        struct resource *res_j = &nvdimm->flush_wpq[j];
49                        unsigned long pfn_j = PHYS_PFN(res_j->start);
50
51                        if (pfn == pfn_j)
52                                break;
53                }
54
55                if (j < i)
56                        flush_page = (void __iomem *) ((unsigned long)
57                                        ndrd_get_flush_wpq(ndrd, dimm, j)
58                                        & PAGE_MASK);
59                else
60                        flush_page = devm_nvdimm_ioremap(dev,
61                                        PFN_PHYS(pfn), PAGE_SIZE);
62                if (!flush_page)
63                        return -ENXIO;
64                ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
65                                + (res->start & ~PAGE_MASK));
66        }
67
68        return 0;
69}
70
71int nd_region_activate(struct nd_region *nd_region)
72{
73        int i, j, num_flush = 0;
74        struct nd_region_data *ndrd;
75        struct device *dev = &nd_region->dev;
76        size_t flush_data_size = sizeof(void *);
77
78        nvdimm_bus_lock(&nd_region->dev);
79        for (i = 0; i < nd_region->ndr_mappings; i++) {
80                struct nd_mapping *nd_mapping = &nd_region->mapping[i];
81                struct nvdimm *nvdimm = nd_mapping->nvdimm;
82
83                /* at least one null hint slot per-dimm for the "no-hint" case */
84                flush_data_size += sizeof(void *);
85                num_flush = min_not_zero(num_flush, nvdimm->num_flush);
86                if (!nvdimm->num_flush)
87                        continue;
88                flush_data_size += nvdimm->num_flush * sizeof(void *);
89        }
90        nvdimm_bus_unlock(&nd_region->dev);
91
92        ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
93        if (!ndrd)
94                return -ENOMEM;
95        dev_set_drvdata(dev, ndrd);
96
97        if (!num_flush)
98                return 0;
99
100        ndrd->hints_shift = ilog2(num_flush);
101        for (i = 0; i < nd_region->ndr_mappings; i++) {
102                struct nd_mapping *nd_mapping = &nd_region->mapping[i];
103                struct nvdimm *nvdimm = nd_mapping->nvdimm;
104                int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
105
106                if (rc)
107                        return rc;
108        }
109
110        /*
111         * Clear out entries that are duplicates. This should prevent the
112         * extra flushings.
113         */
114        for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
115                /* ignore if NULL already */
116                if (!ndrd_get_flush_wpq(ndrd, i, 0))
117                        continue;
118
119                for (j = i + 1; j < nd_region->ndr_mappings; j++)
120                        if (ndrd_get_flush_wpq(ndrd, i, 0) ==
121                            ndrd_get_flush_wpq(ndrd, j, 0))
122                                ndrd_set_flush_wpq(ndrd, j, 0, NULL);
123        }
124
125        return 0;
126}
127
128static void nd_region_release(struct device *dev)
129{
130        struct nd_region *nd_region = to_nd_region(dev);
131        u16 i;
132
133        for (i = 0; i < nd_region->ndr_mappings; i++) {
134                struct nd_mapping *nd_mapping = &nd_region->mapping[i];
135                struct nvdimm *nvdimm = nd_mapping->nvdimm;
136
137                put_device(&nvdimm->dev);
138        }
139        free_percpu(nd_region->lane);
140        ida_simple_remove(&region_ida, nd_region->id);
141        if (is_nd_blk(dev))
142                kfree(to_nd_blk_region(dev));
143        else
144                kfree(nd_region);
145}
146
147static struct device_type nd_blk_device_type = {
148        .name = "nd_blk",
149        .release = nd_region_release,
150};
151
152static struct device_type nd_pmem_device_type = {
153        .name = "nd_pmem",
154        .release = nd_region_release,
155};
156
157static struct device_type nd_volatile_device_type = {
158        .name = "nd_volatile",
159        .release = nd_region_release,
160};
161
162bool is_nd_pmem(struct device *dev)
163{
164        return dev ? dev->type == &nd_pmem_device_type : false;
165}
166
167bool is_nd_blk(struct device *dev)
168{
169        return dev ? dev->type == &nd_blk_device_type : false;
170}
171
172struct nd_region *to_nd_region(struct device *dev)
173{
174        struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
175
176        WARN_ON(dev->type->release != nd_region_release);
177        return nd_region;
178}
179EXPORT_SYMBOL_GPL(to_nd_region);
180
181struct nd_blk_region *to_nd_blk_region(struct device *dev)
182{
183        struct nd_region *nd_region = to_nd_region(dev);
184
185        WARN_ON(!is_nd_blk(dev));
186        return container_of(nd_region, struct nd_blk_region, nd_region);
187}
188EXPORT_SYMBOL_GPL(to_nd_blk_region);
189
190void *nd_region_provider_data(struct nd_region *nd_region)
191{
192        return nd_region->provider_data;
193}
194EXPORT_SYMBOL_GPL(nd_region_provider_data);
195
196void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
197{
198        return ndbr->blk_provider_data;
199}
200EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
201
202void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
203{
204        ndbr->blk_provider_data = data;
205}
206EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
207
208/**
209 * nd_region_to_nstype() - region to an integer namespace type
210 * @nd_region: region-device to interrogate
211 *
212 * This is the 'nstype' attribute of a region as well, an input to the
213 * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
214 * namespace devices with namespace drivers.
215 */
216int nd_region_to_nstype(struct nd_region *nd_region)
217{
218        if (is_nd_pmem(&nd_region->dev)) {
219                u16 i, alias;
220
221                for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
222                        struct nd_mapping *nd_mapping = &nd_region->mapping[i];
223                        struct nvdimm *nvdimm = nd_mapping->nvdimm;
224
225                        if (nvdimm->flags & NDD_ALIASING)
226                                alias++;
227                }
228                if (alias)
229                        return ND_DEVICE_NAMESPACE_PMEM;
230                else
231                        return ND_DEVICE_NAMESPACE_IO;
232        } else if (is_nd_blk(&nd_region->dev)) {
233                return ND_DEVICE_NAMESPACE_BLK;
234        }
235
236        return 0;
237}
238EXPORT_SYMBOL(nd_region_to_nstype);
239
240static ssize_t size_show(struct device *dev,
241                struct device_attribute *attr, char *buf)
242{
243        struct nd_region *nd_region = to_nd_region(dev);
244        unsigned long long size = 0;
245
246        if (is_nd_pmem(dev)) {
247                size = nd_region->ndr_size;
248        } else if (nd_region->ndr_mappings == 1) {
249                struct nd_mapping *nd_mapping = &nd_region->mapping[0];
250
251                size = nd_mapping->size;
252        }
253
254        return sprintf(buf, "%llu\n", size);
255}
256static DEVICE_ATTR_RO(size);
257
258static ssize_t mappings_show(struct device *dev,
259                struct device_attribute *attr, char *buf)
260{
261        struct nd_region *nd_region = to_nd_region(dev);
262
263        return sprintf(buf, "%d\n", nd_region->ndr_mappings);
264}
265static DEVICE_ATTR_RO(mappings);
266
267static ssize_t nstype_show(struct device *dev,
268                struct device_attribute *attr, char *buf)
269{
270        struct nd_region *nd_region = to_nd_region(dev);
271
272        return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
273}
274static DEVICE_ATTR_RO(nstype);
275
276static ssize_t set_cookie_show(struct device *dev,
277                struct device_attribute *attr, char *buf)
278{
279        struct nd_region *nd_region = to_nd_region(dev);
280        struct nd_interleave_set *nd_set = nd_region->nd_set;
281
282        if (is_nd_pmem(dev) && nd_set)
283                /* pass, should be precluded by region_visible */;
284        else
285                return -ENXIO;
286
287        return sprintf(buf, "%#llx\n", nd_set->cookie);
288}
289static DEVICE_ATTR_RO(set_cookie);
290
291resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
292{
293        resource_size_t blk_max_overlap = 0, available, overlap;
294        int i;
295
296        WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
297
298 retry:
299        available = 0;
300        overlap = blk_max_overlap;
301        for (i = 0; i < nd_region->ndr_mappings; i++) {
302                struct nd_mapping *nd_mapping = &nd_region->mapping[i];
303                struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
304
305                /* if a dimm is disabled the available capacity is zero */
306                if (!ndd)
307                        return 0;
308
309                if (is_nd_pmem(&nd_region->dev)) {
310                        available += nd_pmem_available_dpa(nd_region,
311                                        nd_mapping, &overlap);
312                        if (overlap > blk_max_overlap) {
313                                blk_max_overlap = overlap;
314                                goto retry;
315                        }
316                } else if (is_nd_blk(&nd_region->dev))
317                        available += nd_blk_available_dpa(nd_region);
318        }
319
320        return available;
321}
322
323static ssize_t available_size_show(struct device *dev,
324                struct device_attribute *attr, char *buf)
325{
326        struct nd_region *nd_region = to_nd_region(dev);
327        unsigned long long available = 0;
328
329        /*
330         * Flush in-flight updates and grab a snapshot of the available
331         * size.  Of course, this value is potentially invalidated the
332         * memory nvdimm_bus_lock() is dropped, but that's userspace's
333         * problem to not race itself.
334         */
335        nvdimm_bus_lock(dev);
336        wait_nvdimm_bus_probe_idle(dev);
337        available = nd_region_available_dpa(nd_region);
338        nvdimm_bus_unlock(dev);
339
340        return sprintf(buf, "%llu\n", available);
341}
342static DEVICE_ATTR_RO(available_size);
343
344static ssize_t init_namespaces_show(struct device *dev,
345                struct device_attribute *attr, char *buf)
346{
347        struct nd_region_data *ndrd = dev_get_drvdata(dev);
348        ssize_t rc;
349
350        nvdimm_bus_lock(dev);
351        if (ndrd)
352                rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
353        else
354                rc = -ENXIO;
355        nvdimm_bus_unlock(dev);
356
357        return rc;
358}
359static DEVICE_ATTR_RO(init_namespaces);
360
361static ssize_t namespace_seed_show(struct device *dev,
362                struct device_attribute *attr, char *buf)
363{
364        struct nd_region *nd_region = to_nd_region(dev);
365        ssize_t rc;
366
367        nvdimm_bus_lock(dev);
368        if (nd_region->ns_seed)
369                rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
370        else
371                rc = sprintf(buf, "\n");
372        nvdimm_bus_unlock(dev);
373        return rc;
374}
375static DEVICE_ATTR_RO(namespace_seed);
376
377static ssize_t btt_seed_show(struct device *dev,
378                struct device_attribute *attr, char *buf)
379{
380        struct nd_region *nd_region = to_nd_region(dev);
381        ssize_t rc;
382
383        nvdimm_bus_lock(dev);
384        if (nd_region->btt_seed)
385                rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
386        else
387                rc = sprintf(buf, "\n");
388        nvdimm_bus_unlock(dev);
389
390        return rc;
391}
392static DEVICE_ATTR_RO(btt_seed);
393
394static ssize_t pfn_seed_show(struct device *dev,
395                struct device_attribute *attr, char *buf)
396{
397        struct nd_region *nd_region = to_nd_region(dev);
398        ssize_t rc;
399
400        nvdimm_bus_lock(dev);
401        if (nd_region->pfn_seed)
402                rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
403        else
404                rc = sprintf(buf, "\n");
405        nvdimm_bus_unlock(dev);
406
407        return rc;
408}
409static DEVICE_ATTR_RO(pfn_seed);
410
411static ssize_t dax_seed_show(struct device *dev,
412                struct device_attribute *attr, char *buf)
413{
414        struct nd_region *nd_region = to_nd_region(dev);
415        ssize_t rc;
416
417        nvdimm_bus_lock(dev);
418        if (nd_region->dax_seed)
419                rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
420        else
421                rc = sprintf(buf, "\n");
422        nvdimm_bus_unlock(dev);
423
424        return rc;
425}
426static DEVICE_ATTR_RO(dax_seed);
427
428static ssize_t read_only_show(struct device *dev,
429                struct device_attribute *attr, char *buf)
430{
431        struct nd_region *nd_region = to_nd_region(dev);
432
433        return sprintf(buf, "%d\n", nd_region->ro);
434}
435
436static ssize_t read_only_store(struct device *dev,
437                struct device_attribute *attr, const char *buf, size_t len)
438{
439        bool ro;
440        int rc = strtobool(buf, &ro);
441        struct nd_region *nd_region = to_nd_region(dev);
442
443        if (rc)
444                return rc;
445
446        nd_region->ro = ro;
447        return len;
448}
449static DEVICE_ATTR_RW(read_only);
450
451static struct attribute *nd_region_attributes[] = {
452        &dev_attr_size.attr,
453        &dev_attr_nstype.attr,
454        &dev_attr_mappings.attr,
455        &dev_attr_btt_seed.attr,
456        &dev_attr_pfn_seed.attr,
457        &dev_attr_dax_seed.attr,
458        &dev_attr_read_only.attr,
459        &dev_attr_set_cookie.attr,
460        &dev_attr_available_size.attr,
461        &dev_attr_namespace_seed.attr,
462        &dev_attr_init_namespaces.attr,
463        NULL,
464};
465
466static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
467{
468        struct device *dev = container_of(kobj, typeof(*dev), kobj);
469        struct nd_region *nd_region = to_nd_region(dev);
470        struct nd_interleave_set *nd_set = nd_region->nd_set;
471        int type = nd_region_to_nstype(nd_region);
472
473        if (!is_nd_pmem(dev) && a == &dev_attr_pfn_seed.attr)
474                return 0;
475
476        if (!is_nd_pmem(dev) && a == &dev_attr_dax_seed.attr)
477                return 0;
478
479        if (a != &dev_attr_set_cookie.attr
480                        && a != &dev_attr_available_size.attr)
481                return a->mode;
482
483        if ((type == ND_DEVICE_NAMESPACE_PMEM
484                                || type == ND_DEVICE_NAMESPACE_BLK)
485                        && a == &dev_attr_available_size.attr)
486                return a->mode;
487        else if (is_nd_pmem(dev) && nd_set)
488                return a->mode;
489
490        return 0;
491}
492
493struct attribute_group nd_region_attribute_group = {
494        .attrs = nd_region_attributes,
495        .is_visible = region_visible,
496};
497EXPORT_SYMBOL_GPL(nd_region_attribute_group);
498
499u64 nd_region_interleave_set_cookie(struct nd_region *nd_region)
500{
501        struct nd_interleave_set *nd_set = nd_region->nd_set;
502
503        if (nd_set)
504                return nd_set->cookie;
505        return 0;
506}
507
508u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
509{
510        struct nd_interleave_set *nd_set = nd_region->nd_set;
511
512        if (nd_set)
513                return nd_set->altcookie;
514        return 0;
515}
516
517void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
518{
519        struct nd_label_ent *label_ent, *e;
520
521        WARN_ON(!mutex_is_locked(&nd_mapping->lock));
522        list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
523                list_del(&label_ent->list);
524                kfree(label_ent);
525        }
526}
527
528/*
529 * Upon successful probe/remove, take/release a reference on the
530 * associated interleave set (if present), and plant new btt + namespace
531 * seeds.  Also, on the removal of a BLK region, notify the provider to
532 * disable the region.
533 */
534static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
535                struct device *dev, bool probe)
536{
537        struct nd_region *nd_region;
538
539        if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
540                int i;
541
542                nd_region = to_nd_region(dev);
543                for (i = 0; i < nd_region->ndr_mappings; i++) {
544                        struct nd_mapping *nd_mapping = &nd_region->mapping[i];
545                        struct nvdimm_drvdata *ndd = nd_mapping->ndd;
546                        struct nvdimm *nvdimm = nd_mapping->nvdimm;
547
548                        mutex_lock(&nd_mapping->lock);
549                        nd_mapping_free_labels(nd_mapping);
550                        mutex_unlock(&nd_mapping->lock);
551
552                        put_ndd(ndd);
553                        nd_mapping->ndd = NULL;
554                        if (ndd)
555                                atomic_dec(&nvdimm->busy);
556                }
557
558                if (is_nd_pmem(dev))
559                        return;
560        }
561        if (dev->parent && (is_nd_blk(dev->parent) || is_nd_pmem(dev->parent))
562                        && probe) {
563                nd_region = to_nd_region(dev->parent);
564                nvdimm_bus_lock(dev);
565                if (nd_region->ns_seed == dev)
566                        nd_region_create_ns_seed(nd_region);
567                nvdimm_bus_unlock(dev);
568        }
569        if (is_nd_btt(dev) && probe) {
570                struct nd_btt *nd_btt = to_nd_btt(dev);
571
572                nd_region = to_nd_region(dev->parent);
573                nvdimm_bus_lock(dev);
574                if (nd_region->btt_seed == dev)
575                        nd_region_create_btt_seed(nd_region);
576                if (nd_region->ns_seed == &nd_btt->ndns->dev)
577                        nd_region_create_ns_seed(nd_region);
578                nvdimm_bus_unlock(dev);
579        }
580        if (is_nd_pfn(dev) && probe) {
581                struct nd_pfn *nd_pfn = to_nd_pfn(dev);
582
583                nd_region = to_nd_region(dev->parent);
584                nvdimm_bus_lock(dev);
585                if (nd_region->pfn_seed == dev)
586                        nd_region_create_pfn_seed(nd_region);
587                if (nd_region->ns_seed == &nd_pfn->ndns->dev)
588                        nd_region_create_ns_seed(nd_region);
589                nvdimm_bus_unlock(dev);
590        }
591        if (is_nd_dax(dev) && probe) {
592                struct nd_dax *nd_dax = to_nd_dax(dev);
593
594                nd_region = to_nd_region(dev->parent);
595                nvdimm_bus_lock(dev);
596                if (nd_region->dax_seed == dev)
597                        nd_region_create_dax_seed(nd_region);
598                if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
599                        nd_region_create_ns_seed(nd_region);
600                nvdimm_bus_unlock(dev);
601        }
602}
603
604void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
605{
606        nd_region_notify_driver_action(nvdimm_bus, dev, true);
607}
608
609void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
610{
611        nd_region_notify_driver_action(nvdimm_bus, dev, false);
612}
613
614static ssize_t mappingN(struct device *dev, char *buf, int n)
615{
616        struct nd_region *nd_region = to_nd_region(dev);
617        struct nd_mapping *nd_mapping;
618        struct nvdimm *nvdimm;
619
620        if (n >= nd_region->ndr_mappings)
621                return -ENXIO;
622        nd_mapping = &nd_region->mapping[n];
623        nvdimm = nd_mapping->nvdimm;
624
625        return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
626                        nd_mapping->start, nd_mapping->size);
627}
628
629#define REGION_MAPPING(idx) \
630static ssize_t mapping##idx##_show(struct device *dev,          \
631                struct device_attribute *attr, char *buf)       \
632{                                                               \
633        return mappingN(dev, buf, idx);                         \
634}                                                               \
635static DEVICE_ATTR_RO(mapping##idx)
636
637/*
638 * 32 should be enough for a while, even in the presence of socket
639 * interleave a 32-way interleave set is a degenerate case.
640 */
641REGION_MAPPING(0);
642REGION_MAPPING(1);
643REGION_MAPPING(2);
644REGION_MAPPING(3);
645REGION_MAPPING(4);
646REGION_MAPPING(5);
647REGION_MAPPING(6);
648REGION_MAPPING(7);
649REGION_MAPPING(8);
650REGION_MAPPING(9);
651REGION_MAPPING(10);
652REGION_MAPPING(11);
653REGION_MAPPING(12);
654REGION_MAPPING(13);
655REGION_MAPPING(14);
656REGION_MAPPING(15);
657REGION_MAPPING(16);
658REGION_MAPPING(17);
659REGION_MAPPING(18);
660REGION_MAPPING(19);
661REGION_MAPPING(20);
662REGION_MAPPING(21);
663REGION_MAPPING(22);
664REGION_MAPPING(23);
665REGION_MAPPING(24);
666REGION_MAPPING(25);
667REGION_MAPPING(26);
668REGION_MAPPING(27);
669REGION_MAPPING(28);
670REGION_MAPPING(29);
671REGION_MAPPING(30);
672REGION_MAPPING(31);
673
674static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
675{
676        struct device *dev = container_of(kobj, struct device, kobj);
677        struct nd_region *nd_region = to_nd_region(dev);
678
679        if (n < nd_region->ndr_mappings)
680                return a->mode;
681        return 0;
682}
683
684static struct attribute *mapping_attributes[] = {
685        &dev_attr_mapping0.attr,
686        &dev_attr_mapping1.attr,
687        &dev_attr_mapping2.attr,
688        &dev_attr_mapping3.attr,
689        &dev_attr_mapping4.attr,
690        &dev_attr_mapping5.attr,
691        &dev_attr_mapping6.attr,
692        &dev_attr_mapping7.attr,
693        &dev_attr_mapping8.attr,
694        &dev_attr_mapping9.attr,
695        &dev_attr_mapping10.attr,
696        &dev_attr_mapping11.attr,
697        &dev_attr_mapping12.attr,
698        &dev_attr_mapping13.attr,
699        &dev_attr_mapping14.attr,
700        &dev_attr_mapping15.attr,
701        &dev_attr_mapping16.attr,
702        &dev_attr_mapping17.attr,
703        &dev_attr_mapping18.attr,
704        &dev_attr_mapping19.attr,
705        &dev_attr_mapping20.attr,
706        &dev_attr_mapping21.attr,
707        &dev_attr_mapping22.attr,
708        &dev_attr_mapping23.attr,
709        &dev_attr_mapping24.attr,
710        &dev_attr_mapping25.attr,
711        &dev_attr_mapping26.attr,
712        &dev_attr_mapping27.attr,
713        &dev_attr_mapping28.attr,
714        &dev_attr_mapping29.attr,
715        &dev_attr_mapping30.attr,
716        &dev_attr_mapping31.attr,
717        NULL,
718};
719
720struct attribute_group nd_mapping_attribute_group = {
721        .is_visible = mapping_visible,
722        .attrs = mapping_attributes,
723};
724EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
725
726int nd_blk_region_init(struct nd_region *nd_region)
727{
728        struct device *dev = &nd_region->dev;
729        struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
730
731        if (!is_nd_blk(dev))
732                return 0;
733
734        if (nd_region->ndr_mappings < 1) {
735                dev_err(dev, "invalid BLK region\n");
736                return -ENXIO;
737        }
738
739        return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
740}
741
742/**
743 * nd_region_acquire_lane - allocate and lock a lane
744 * @nd_region: region id and number of lanes possible
745 *
746 * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
747 * We optimize for the common case where there are 256 lanes, one
748 * per-cpu.  For larger systems we need to lock to share lanes.  For now
749 * this implementation assumes the cost of maintaining an allocator for
750 * free lanes is on the order of the lock hold time, so it implements a
751 * static lane = cpu % num_lanes mapping.
752 *
753 * In the case of a BTT instance on top of a BLK namespace a lane may be
754 * acquired recursively.  We lock on the first instance.
755 *
756 * In the case of a BTT instance on top of PMEM, we only acquire a lane
757 * for the BTT metadata updates.
758 */
759unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
760{
761        unsigned int cpu, lane;
762
763        cpu = get_cpu();
764        if (nd_region->num_lanes < nr_cpu_ids) {
765                struct nd_percpu_lane *ndl_lock, *ndl_count;
766
767                lane = cpu % nd_region->num_lanes;
768                ndl_count = per_cpu_ptr(nd_region->lane, cpu);
769                ndl_lock = per_cpu_ptr(nd_region->lane, lane);
770                if (ndl_count->count++ == 0)
771                        spin_lock(&ndl_lock->lock);
772        } else
773                lane = cpu;
774
775        return lane;
776}
777EXPORT_SYMBOL(nd_region_acquire_lane);
778
779void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
780{
781        if (nd_region->num_lanes < nr_cpu_ids) {
782                unsigned int cpu = get_cpu();
783                struct nd_percpu_lane *ndl_lock, *ndl_count;
784
785                ndl_count = per_cpu_ptr(nd_region->lane, cpu);
786                ndl_lock = per_cpu_ptr(nd_region->lane, lane);
787                if (--ndl_count->count == 0)
788                        spin_unlock(&ndl_lock->lock);
789                put_cpu();
790        }
791        put_cpu();
792}
793EXPORT_SYMBOL(nd_region_release_lane);
794
795static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
796                struct nd_region_desc *ndr_desc, struct device_type *dev_type,
797                const char *caller)
798{
799        struct nd_region *nd_region;
800        struct device *dev;
801        void *region_buf;
802        unsigned int i;
803        int ro = 0;
804
805        for (i = 0; i < ndr_desc->num_mappings; i++) {
806                struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
807                struct nvdimm *nvdimm = mapping->nvdimm;
808
809                if ((mapping->start | mapping->size) % SZ_4K) {
810                        dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
811                                        caller, dev_name(&nvdimm->dev), i);
812
813                        return NULL;
814                }
815
816                if (nvdimm->flags & NDD_UNARMED)
817                        ro = 1;
818        }
819
820        if (dev_type == &nd_blk_device_type) {
821                struct nd_blk_region_desc *ndbr_desc;
822                struct nd_blk_region *ndbr;
823
824                ndbr_desc = to_blk_region_desc(ndr_desc);
825                ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
826                                * ndr_desc->num_mappings,
827                                GFP_KERNEL);
828                if (ndbr) {
829                        nd_region = &ndbr->nd_region;
830                        ndbr->enable = ndbr_desc->enable;
831                        ndbr->do_io = ndbr_desc->do_io;
832                }
833                region_buf = ndbr;
834        } else {
835                nd_region = kzalloc(sizeof(struct nd_region)
836                                + sizeof(struct nd_mapping)
837                                * ndr_desc->num_mappings,
838                                GFP_KERNEL);
839                region_buf = nd_region;
840        }
841
842        if (!region_buf)
843                return NULL;
844        nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
845        if (nd_region->id < 0)
846                goto err_id;
847
848        nd_region->lane = alloc_percpu(struct nd_percpu_lane);
849        if (!nd_region->lane)
850                goto err_percpu;
851
852        for (i = 0; i < nr_cpu_ids; i++) {
853                struct nd_percpu_lane *ndl;
854
855                ndl = per_cpu_ptr(nd_region->lane, i);
856                spin_lock_init(&ndl->lock);
857                ndl->count = 0;
858        }
859
860        for (i = 0; i < ndr_desc->num_mappings; i++) {
861                struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
862                struct nvdimm *nvdimm = mapping->nvdimm;
863
864                nd_region->mapping[i].nvdimm = nvdimm;
865                nd_region->mapping[i].start = mapping->start;
866                nd_region->mapping[i].size = mapping->size;
867                INIT_LIST_HEAD(&nd_region->mapping[i].labels);
868                mutex_init(&nd_region->mapping[i].lock);
869
870                get_device(&nvdimm->dev);
871        }
872        nd_region->ndr_mappings = ndr_desc->num_mappings;
873        nd_region->provider_data = ndr_desc->provider_data;
874        nd_region->nd_set = ndr_desc->nd_set;
875        nd_region->num_lanes = ndr_desc->num_lanes;
876        nd_region->flags = ndr_desc->flags;
877        nd_region->ro = ro;
878        nd_region->numa_node = ndr_desc->numa_node;
879        ida_init(&nd_region->ns_ida);
880        ida_init(&nd_region->btt_ida);
881        ida_init(&nd_region->pfn_ida);
882        ida_init(&nd_region->dax_ida);
883        dev = &nd_region->dev;
884        dev_set_name(dev, "region%d", nd_region->id);
885        dev->parent = &nvdimm_bus->dev;
886        dev->type = dev_type;
887        dev->groups = ndr_desc->attr_groups;
888        nd_region->ndr_size = resource_size(ndr_desc->res);
889        nd_region->ndr_start = ndr_desc->res->start;
890        nd_device_register(dev);
891
892        return nd_region;
893
894 err_percpu:
895        ida_simple_remove(&region_ida, nd_region->id);
896 err_id:
897        kfree(region_buf);
898        return NULL;
899}
900
901struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
902                struct nd_region_desc *ndr_desc)
903{
904        ndr_desc->num_lanes = ND_MAX_LANES;
905        return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
906                        __func__);
907}
908EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
909
910struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
911                struct nd_region_desc *ndr_desc)
912{
913        if (ndr_desc->num_mappings > 1)
914                return NULL;
915        ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
916        return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
917                        __func__);
918}
919EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
920
921struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
922                struct nd_region_desc *ndr_desc)
923{
924        ndr_desc->num_lanes = ND_MAX_LANES;
925        return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
926                        __func__);
927}
928EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
929
930/**
931 * nvdimm_flush - flush any posted write queues between the cpu and pmem media
932 * @nd_region: blk or interleaved pmem region
933 */
934void nvdimm_flush(struct nd_region *nd_region)
935{
936        struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
937        int i, idx;
938
939        /*
940         * Try to encourage some diversity in flush hint addresses
941         * across cpus assuming a limited number of flush hints.
942         */
943        idx = this_cpu_read(flush_idx);
944        idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
945
946        /*
947         * The first wmb() is needed to 'sfence' all previous writes
948         * such that they are architecturally visible for the platform
949         * buffer flush.  Note that we've already arranged for pmem
950         * writes to avoid the cache via arch_memcpy_to_pmem().  The
951         * final wmb() ensures ordering for the NVDIMM flush write.
952         */
953        wmb();
954        for (i = 0; i < nd_region->ndr_mappings; i++)
955                if (ndrd_get_flush_wpq(ndrd, i, 0))
956                        writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
957        wmb();
958}
959EXPORT_SYMBOL_GPL(nvdimm_flush);
960
961/**
962 * nvdimm_has_flush - determine write flushing requirements
963 * @nd_region: blk or interleaved pmem region
964 *
965 * Returns 1 if writes require flushing
966 * Returns 0 if writes do not require flushing
967 * Returns -ENXIO if flushing capability can not be determined
968 */
969int nvdimm_has_flush(struct nd_region *nd_region)
970{
971        struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
972        int i;
973
974        /* no nvdimm == flushing capability unknown */
975        if (nd_region->ndr_mappings == 0)
976                return -ENXIO;
977
978        for (i = 0; i < nd_region->ndr_mappings; i++)
979                /* flush hints present, flushing required */
980                if (ndrd_get_flush_wpq(ndrd, i, 0))
981                        return 1;
982
983        /*
984         * The platform defines dimm devices without hints, assume
985         * platform persistence mechanism like ADR
986         */
987        return 0;
988}
989EXPORT_SYMBOL_GPL(nvdimm_has_flush);
990
991void __exit nd_region_devs_exit(void)
992{
993        ida_destroy(&region_ida);
994}
Note: See TracBrowser for help on using the repository browser.