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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14#include <linux/moduleparam.h>
15#include <linux/vmalloc.h>
16#include <linux/device.h>
17#include <linux/ndctl.h>
18#include <linux/slab.h>
19#include <linux/io.h>
20#include <linux/fs.h>
21#include <linux/mm.h>
22#include "nd-core.h"
23#include "label.h"
24#include "pmem.h"
25#include "nd.h"
26
27static DEFINE_IDA(dimm_ida);
28
29static bool noblk;
30module_param(noblk, bool, 0444);
31MODULE_PARM_DESC(noblk, "force disable BLK / local alias support");
32
33/*
34 * Retrieve bus and dimm handle and return if this bus supports
35 * get_config_data commands
36 */
37int nvdimm_check_config_data(struct device *dev)
38{
39 struct nvdimm *nvdimm = to_nvdimm(dev);
40
41 if (!nvdimm->cmd_mask ||
42 !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
43 if (test_bit(NDD_ALIASING, &nvdimm->flags))
44 return -ENXIO;
45 else
46 return -ENOTTY;
47 }
48
49 return 0;
50}
51
52static int validate_dimm(struct nvdimm_drvdata *ndd)
53{
54 int rc;
55
56 if (!ndd)
57 return -EINVAL;
58
59 rc = nvdimm_check_config_data(ndd->dev);
60 if (rc)
61 dev_dbg(ndd->dev, "%pf: %s error: %d\n",
62 __builtin_return_address(0), __func__, rc);
63 return rc;
64}
65
66/**
67 * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
68 * @nvdimm: dimm to initialize
69 */
70int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
71{
72 struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
73 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
74 struct nvdimm_bus_descriptor *nd_desc;
75 int rc = validate_dimm(ndd);
76 int cmd_rc = 0;
77
78 if (rc)
79 return rc;
80
81 if (cmd->config_size)
82 return 0; /* already valid */
83
84 memset(cmd, 0, sizeof(*cmd));
85 nd_desc = nvdimm_bus->nd_desc;
86 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
87 ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd), &cmd_rc);
88 if (rc < 0)
89 return rc;
90 return cmd_rc;
91}
92
93int nvdimm_get_config_data(struct nvdimm_drvdata *ndd, void *buf,
94 size_t offset, size_t len)
95{
96 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
97 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
98 int rc = validate_dimm(ndd), cmd_rc = 0;
99 struct nd_cmd_get_config_data_hdr *cmd;
100 size_t max_cmd_size, buf_offset;
101
102 if (rc)
103 return rc;
104
105 if (offset + len > ndd->nsarea.config_size)
106 return -ENXIO;
107
108 max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
109 cmd = kvzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
110 if (!cmd)
111 return -ENOMEM;
112
113 for (buf_offset = 0; len;
114 len -= cmd->in_length, buf_offset += cmd->in_length) {
115 size_t cmd_size;
116
117 cmd->in_offset = offset + buf_offset;
118 cmd->in_length = min(max_cmd_size, len);
119
120 cmd_size = sizeof(*cmd) + cmd->in_length;
121
122 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
123 ND_CMD_GET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
124 if (rc < 0)
125 break;
126 if (cmd_rc < 0) {
127 rc = cmd_rc;
128 break;
129 }
130
131 /* out_buf should be valid, copy it into our output buffer */
132 memcpy(buf + buf_offset, cmd->out_buf, cmd->in_length);
133 }
134 kvfree(cmd);
135
136 return rc;
137}
138
139int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
140 void *buf, size_t len)
141{
142 size_t max_cmd_size, buf_offset;
143 struct nd_cmd_set_config_hdr *cmd;
144 int rc = validate_dimm(ndd), cmd_rc = 0;
145 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
146 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
147
148 if (rc)
149 return rc;
150
151 if (offset + len > ndd->nsarea.config_size)
152 return -ENXIO;
153
154 max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
155 cmd = kvzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
156 if (!cmd)
157 return -ENOMEM;
158
159 for (buf_offset = 0; len; len -= cmd->in_length,
160 buf_offset += cmd->in_length) {
161 size_t cmd_size;
162
163 cmd->in_offset = offset + buf_offset;
164 cmd->in_length = min(max_cmd_size, len);
165 memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
166
167 /* status is output in the last 4-bytes of the command buffer */
168 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
169
170 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
171 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
172 if (rc < 0)
173 break;
174 if (cmd_rc < 0) {
175 rc = cmd_rc;
176 break;
177 }
178 }
179 kvfree(cmd);
180
181 return rc;
182}
183
184void nvdimm_set_aliasing(struct device *dev)
185{
186 struct nvdimm *nvdimm = to_nvdimm(dev);
187
188 set_bit(NDD_ALIASING, &nvdimm->flags);
189}
190
191void nvdimm_set_locked(struct device *dev)
192{
193 struct nvdimm *nvdimm = to_nvdimm(dev);
194
195 set_bit(NDD_LOCKED, &nvdimm->flags);
196}
197
198void nvdimm_clear_locked(struct device *dev)
199{
200 struct nvdimm *nvdimm = to_nvdimm(dev);
201
202 clear_bit(NDD_LOCKED, &nvdimm->flags);
203}
204
205static void nvdimm_release(struct device *dev)
206{
207 struct nvdimm *nvdimm = to_nvdimm(dev);
208
209 ida_simple_remove(&dimm_ida, nvdimm->id);
210 kfree(nvdimm);
211}
212
213static struct device_type nvdimm_device_type = {
214 .name = "nvdimm",
215 .release = nvdimm_release,
216};
217
218bool is_nvdimm(struct device *dev)
219{
220 return dev->type == &nvdimm_device_type;
221}
222
223struct nvdimm *to_nvdimm(struct device *dev)
224{
225 struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
226
227 WARN_ON(!is_nvdimm(dev));
228 return nvdimm;
229}
230EXPORT_SYMBOL_GPL(to_nvdimm);
231
232struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
233{
234 struct nd_region *nd_region = &ndbr->nd_region;
235 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
236
237 return nd_mapping->nvdimm;
238}
239EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
240
241unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
242{
243 /* pmem mapping properties are private to libnvdimm */
244 return ARCH_MEMREMAP_PMEM;
245}
246EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
247
248struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
249{
250 struct nvdimm *nvdimm = nd_mapping->nvdimm;
251
252 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
253
254 return dev_get_drvdata(&nvdimm->dev);
255}
256EXPORT_SYMBOL(to_ndd);
257
258void nvdimm_drvdata_release(struct kref *kref)
259{
260 struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
261 struct device *dev = ndd->dev;
262 struct resource *res, *_r;
263
264 dev_dbg(dev, "trace\n");
265 nvdimm_bus_lock(dev);
266 for_each_dpa_resource_safe(ndd, res, _r)
267 nvdimm_free_dpa(ndd, res);
268 nvdimm_bus_unlock(dev);
269
270 kvfree(ndd->data);
271 kfree(ndd);
272 put_device(dev);
273}
274
275void get_ndd(struct nvdimm_drvdata *ndd)
276{
277 kref_get(&ndd->kref);
278}
279
280void put_ndd(struct nvdimm_drvdata *ndd)
281{
282 if (ndd)
283 kref_put(&ndd->kref, nvdimm_drvdata_release);
284}
285
286const char *nvdimm_name(struct nvdimm *nvdimm)
287{
288 return dev_name(&nvdimm->dev);
289}
290EXPORT_SYMBOL_GPL(nvdimm_name);
291
292struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
293{
294 return &nvdimm->dev.kobj;
295}
296EXPORT_SYMBOL_GPL(nvdimm_kobj);
297
298unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
299{
300 return nvdimm->cmd_mask;
301}
302EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
303
304void *nvdimm_provider_data(struct nvdimm *nvdimm)
305{
306 if (nvdimm)
307 return nvdimm->provider_data;
308 return NULL;
309}
310EXPORT_SYMBOL_GPL(nvdimm_provider_data);
311
312static ssize_t commands_show(struct device *dev,
313 struct device_attribute *attr, char *buf)
314{
315 struct nvdimm *nvdimm = to_nvdimm(dev);
316 int cmd, len = 0;
317
318 if (!nvdimm->cmd_mask)
319 return sprintf(buf, "\n");
320
321 for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
322 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
323 len += sprintf(buf + len, "\n");
324 return len;
325}
326static DEVICE_ATTR_RO(commands);
327
328static ssize_t flags_show(struct device *dev,
329 struct device_attribute *attr, char *buf)
330{
331 struct nvdimm *nvdimm = to_nvdimm(dev);
332
333 return sprintf(buf, "%s%s\n",
334 test_bit(NDD_ALIASING, &nvdimm->flags) ? "alias " : "",
335 test_bit(NDD_LOCKED, &nvdimm->flags) ? "lock " : "");
336}
337static DEVICE_ATTR_RO(flags);
338
339static ssize_t state_show(struct device *dev, struct device_attribute *attr,
340 char *buf)
341{
342 struct nvdimm *nvdimm = to_nvdimm(dev);
343
344 /*
345 * The state may be in the process of changing, userspace should
346 * quiesce probing if it wants a static answer
347 */
348 nvdimm_bus_lock(dev);
349 nvdimm_bus_unlock(dev);
350 return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
351 ? "active" : "idle");
352}
353static DEVICE_ATTR_RO(state);
354
355static ssize_t available_slots_show(struct device *dev,
356 struct device_attribute *attr, char *buf)
357{
358 struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
359 ssize_t rc;
360 u32 nfree;
361
362 if (!ndd)
363 return -ENXIO;
364
365 nvdimm_bus_lock(dev);
366 nfree = nd_label_nfree(ndd);
367 if (nfree - 1 > nfree) {
368 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
369 nfree = 0;
370 } else
371 nfree--;
372 rc = sprintf(buf, "%d\n", nfree);
373 nvdimm_bus_unlock(dev);
374 return rc;
375}
376static DEVICE_ATTR_RO(available_slots);
377
378__weak ssize_t security_show(struct device *dev,
379 struct device_attribute *attr, char *buf)
380{
381 struct nvdimm *nvdimm = to_nvdimm(dev);
382
383 switch (nvdimm->sec.state) {
384 case NVDIMM_SECURITY_DISABLED:
385 return sprintf(buf, "disabled\n");
386 case NVDIMM_SECURITY_UNLOCKED:
387 return sprintf(buf, "unlocked\n");
388 case NVDIMM_SECURITY_LOCKED:
389 return sprintf(buf, "locked\n");
390 case NVDIMM_SECURITY_FROZEN:
391 return sprintf(buf, "frozen\n");
392 case NVDIMM_SECURITY_OVERWRITE:
393 return sprintf(buf, "overwrite\n");
394 default:
395 return -ENOTTY;
396 }
397
398 return -ENOTTY;
399}
400
401#define OPS \
402 C( OP_FREEZE, "freeze", 1), \
403 C( OP_DISABLE, "disable", 2), \
404 C( OP_UPDATE, "update", 3), \
405 C( OP_ERASE, "erase", 2), \
406 C( OP_OVERWRITE, "overwrite", 2), \
407 C( OP_MASTER_UPDATE, "master_update", 3), \
408 C( OP_MASTER_ERASE, "master_erase", 2)
409#undef C
410#define C(a, b, c) a
411enum nvdimmsec_op_ids { OPS };
412#undef C
413#define C(a, b, c) { b, c }
414static struct {
415 const char *name;
416 int args;
417} ops[] = { OPS };
418#undef C
419
420#define SEC_CMD_SIZE 32
421#define KEY_ID_SIZE 10
422
423static ssize_t __security_store(struct device *dev, const char *buf, size_t len)
424{
425 struct nvdimm *nvdimm = to_nvdimm(dev);
426 ssize_t rc;
427 char cmd[SEC_CMD_SIZE+1], keystr[KEY_ID_SIZE+1],
428 nkeystr[KEY_ID_SIZE+1];
429 unsigned int key, newkey;
430 int i;
431
432 if (atomic_read(&nvdimm->busy))
433 return -EBUSY;
434
435 rc = sscanf(buf, "%"__stringify(SEC_CMD_SIZE)"s"
436 " %"__stringify(KEY_ID_SIZE)"s"
437 " %"__stringify(KEY_ID_SIZE)"s",
438 cmd, keystr, nkeystr);
439 if (rc < 1)
440 return -EINVAL;
441 for (i = 0; i < ARRAY_SIZE(ops); i++)
442 if (sysfs_streq(cmd, ops[i].name))
443 break;
444 if (i >= ARRAY_SIZE(ops))
445 return -EINVAL;
446 if (ops[i].args > 1)
447 rc = kstrtouint(keystr, 0, &key);
448 if (rc >= 0 && ops[i].args > 2)
449 rc = kstrtouint(nkeystr, 0, &newkey);
450 if (rc < 0)
451 return rc;
452
453 if (i == OP_FREEZE) {
454 dev_dbg(dev, "freeze\n");
455 rc = nvdimm_security_freeze(nvdimm);
456 } else if (i == OP_DISABLE) {
457 dev_dbg(dev, "disable %u\n", key);
458 rc = nvdimm_security_disable(nvdimm, key);
459 } else if (i == OP_UPDATE) {
460 dev_dbg(dev, "update %u %u\n", key, newkey);
461 rc = nvdimm_security_update(nvdimm, key, newkey, NVDIMM_USER);
462 } else if (i == OP_ERASE) {
463 dev_dbg(dev, "erase %u\n", key);
464 rc = nvdimm_security_erase(nvdimm, key, NVDIMM_USER);
465 } else if (i == OP_OVERWRITE) {
466 dev_dbg(dev, "overwrite %u\n", key);
467 rc = nvdimm_security_overwrite(nvdimm, key);
468 } else if (i == OP_MASTER_UPDATE) {
469 dev_dbg(dev, "master_update %u %u\n", key, newkey);
470 rc = nvdimm_security_update(nvdimm, key, newkey,
471 NVDIMM_MASTER);
472 } else if (i == OP_MASTER_ERASE) {
473 dev_dbg(dev, "master_erase %u\n", key);
474 rc = nvdimm_security_erase(nvdimm, key,
475 NVDIMM_MASTER);
476 } else
477 return -EINVAL;
478
479 if (rc == 0)
480 rc = len;
481 return rc;
482}
483
484static ssize_t security_store(struct device *dev,
485 struct device_attribute *attr, const char *buf, size_t len)
486
487{
488 ssize_t rc;
489
490 /*
491 * Require all userspace triggered security management to be
492 * done while probing is idle and the DIMM is not in active use
493 * in any region.
494 */
495 device_lock(dev);
496 nvdimm_bus_lock(dev);
497 wait_nvdimm_bus_probe_idle(dev);
498 rc = __security_store(dev, buf, len);
499 nvdimm_bus_unlock(dev);
500 device_unlock(dev);
501
502 return rc;
503}
504static DEVICE_ATTR_RW(security);
505
506static struct attribute *nvdimm_attributes[] = {
507 &dev_attr_state.attr,
508 &dev_attr_flags.attr,
509 &dev_attr_commands.attr,
510 &dev_attr_available_slots.attr,
511 &dev_attr_security.attr,
512 NULL,
513};
514
515static umode_t nvdimm_visible(struct kobject *kobj, struct attribute *a, int n)
516{
517 struct device *dev = container_of(kobj, typeof(*dev), kobj);
518 struct nvdimm *nvdimm = to_nvdimm(dev);
519
520 if (a != &dev_attr_security.attr)
521 return a->mode;
522 if (nvdimm->sec.state < 0)
523 return 0;
524 /* Are there any state mutation ops? */
525 if (nvdimm->sec.ops->freeze || nvdimm->sec.ops->disable
526 || nvdimm->sec.ops->change_key
527 || nvdimm->sec.ops->erase
528 || nvdimm->sec.ops->overwrite)
529 return a->mode;
530 return 0444;
531}
532
533struct attribute_group nvdimm_attribute_group = {
534 .attrs = nvdimm_attributes,
535 .is_visible = nvdimm_visible,
536};
537EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
538
539struct nvdimm *__nvdimm_create(struct nvdimm_bus *nvdimm_bus,
540 void *provider_data, const struct attribute_group **groups,
541 unsigned long flags, unsigned long cmd_mask, int num_flush,
542 struct resource *flush_wpq, const char *dimm_id,
543 const struct nvdimm_security_ops *sec_ops)
544{
545 struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
546 struct device *dev;
547
548 if (!nvdimm)
549 return NULL;
550
551 nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
552 if (nvdimm->id < 0) {
553 kfree(nvdimm);
554 return NULL;
555 }
556
557 nvdimm->dimm_id = dimm_id;
558 nvdimm->provider_data = provider_data;
559 if (noblk)
560 flags |= 1 << NDD_NOBLK;
561 nvdimm->flags = flags;
562 nvdimm->cmd_mask = cmd_mask;
563 nvdimm->num_flush = num_flush;
564 nvdimm->flush_wpq = flush_wpq;
565 atomic_set(&nvdimm->busy, 0);
566 dev = &nvdimm->dev;
567 dev_set_name(dev, "nmem%d", nvdimm->id);
568 dev->parent = &nvdimm_bus->dev;
569 dev->type = &nvdimm_device_type;
570 dev->devt = MKDEV(nvdimm_major, nvdimm->id);
571 dev->groups = groups;
572 nvdimm->sec.ops = sec_ops;
573 nvdimm->sec.overwrite_tmo = 0;
574 INIT_DELAYED_WORK(&nvdimm->dwork, nvdimm_security_overwrite_query);
575 /*
576 * Security state must be initialized before device_add() for
577 * attribute visibility.
578 */
579 /* get security state and extended (master) state */
580 nvdimm->sec.state = nvdimm_security_state(nvdimm, NVDIMM_USER);
581 nvdimm->sec.ext_state = nvdimm_security_state(nvdimm, NVDIMM_MASTER);
582 nd_device_register(dev);
583
584 return nvdimm;
585}
586EXPORT_SYMBOL_GPL(__nvdimm_create);
587
588static void shutdown_security_notify(void *data)
589{
590 struct nvdimm *nvdimm = data;
591
592 sysfs_put(nvdimm->sec.overwrite_state);
593}
594
595int nvdimm_security_setup_events(struct device *dev)
596{
597 struct nvdimm *nvdimm = to_nvdimm(dev);
598
599 if (nvdimm->sec.state < 0 || !nvdimm->sec.ops
600 || !nvdimm->sec.ops->overwrite)
601 return 0;
602 nvdimm->sec.overwrite_state = sysfs_get_dirent(dev->kobj.sd, "security");
603 if (!nvdimm->sec.overwrite_state)
604 return -ENOMEM;
605
606 return devm_add_action_or_reset(dev, shutdown_security_notify, nvdimm);
607}
608EXPORT_SYMBOL_GPL(nvdimm_security_setup_events);
609
610int nvdimm_in_overwrite(struct nvdimm *nvdimm)
611{
612 return test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags);
613}
614EXPORT_SYMBOL_GPL(nvdimm_in_overwrite);
615
616int nvdimm_security_freeze(struct nvdimm *nvdimm)
617{
618 int rc;
619
620 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
621
622 if (!nvdimm->sec.ops || !nvdimm->sec.ops->freeze)
623 return -EOPNOTSUPP;
624
625 if (nvdimm->sec.state < 0)
626 return -EIO;
627
628 if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
629 dev_warn(&nvdimm->dev, "Overwrite operation in progress.\n");
630 return -EBUSY;
631 }
632
633 rc = nvdimm->sec.ops->freeze(nvdimm);
634 nvdimm->sec.state = nvdimm_security_state(nvdimm, NVDIMM_USER);
635
636 return rc;
637}
638
639int alias_dpa_busy(struct device *dev, void *data)
640{
641 resource_size_t map_end, blk_start, new;
642 struct blk_alloc_info *info = data;
643 struct nd_mapping *nd_mapping;
644 struct nd_region *nd_region;
645 struct nvdimm_drvdata *ndd;
646 struct resource *res;
647 int i;
648
649 if (!is_memory(dev))
650 return 0;
651
652 nd_region = to_nd_region(dev);
653 for (i = 0; i < nd_region->ndr_mappings; i++) {
654 nd_mapping = &nd_region->mapping[i];
655 if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
656 break;
657 }
658
659 if (i >= nd_region->ndr_mappings)
660 return 0;
661
662 ndd = to_ndd(nd_mapping);
663 map_end = nd_mapping->start + nd_mapping->size - 1;
664 blk_start = nd_mapping->start;
665
666 /*
667 * In the allocation case ->res is set to free space that we are
668 * looking to validate against PMEM aliasing collision rules
669 * (i.e. BLK is allocated after all aliased PMEM).
670 */
671 if (info->res) {
672 if (info->res->start >= nd_mapping->start
673 && info->res->start < map_end)
674 /* pass */;
675 else
676 return 0;
677 }
678
679 retry:
680 /*
681 * Find the free dpa from the end of the last pmem allocation to
682 * the end of the interleave-set mapping.
683 */
684 for_each_dpa_resource(ndd, res) {
685 if (strncmp(res->name, "pmem", 4) != 0)
686 continue;
687 if ((res->start >= blk_start && res->start < map_end)
688 || (res->end >= blk_start
689 && res->end <= map_end)) {
690 new = max(blk_start, min(map_end + 1, res->end + 1));
691 if (new != blk_start) {
692 blk_start = new;
693 goto retry;
694 }
695 }
696 }
697
698 /* update the free space range with the probed blk_start */
699 if (info->res && blk_start > info->res->start) {
700 info->res->start = max(info->res->start, blk_start);
701 if (info->res->start > info->res->end)
702 info->res->end = info->res->start - 1;
703 return 1;
704 }
705
706 info->available -= blk_start - nd_mapping->start;
707
708 return 0;
709}
710
711/**
712 * nd_blk_available_dpa - account the unused dpa of BLK region
713 * @nd_mapping: container of dpa-resource-root + labels
714 *
715 * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
716 * we arrange for them to never start at an lower dpa than the last
717 * PMEM allocation in an aliased region.
718 */
719resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
720{
721 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
722 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
723 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
724 struct blk_alloc_info info = {
725 .nd_mapping = nd_mapping,
726 .available = nd_mapping->size,
727 .res = NULL,
728 };
729 struct resource *res;
730
731 if (!ndd)
732 return 0;
733
734 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
735
736 /* now account for busy blk allocations in unaliased dpa */
737 for_each_dpa_resource(ndd, res) {
738 if (strncmp(res->name, "blk", 3) != 0)
739 continue;
740 info.available -= resource_size(res);
741 }
742
743 return info.available;
744}
745
746/**
747 * nd_pmem_max_contiguous_dpa - For the given dimm+region, return the max
748 * contiguous unallocated dpa range.
749 * @nd_region: constrain available space check to this reference region
750 * @nd_mapping: container of dpa-resource-root + labels
751 */
752resource_size_t nd_pmem_max_contiguous_dpa(struct nd_region *nd_region,
753 struct nd_mapping *nd_mapping)
754{
755 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
756 struct nvdimm_bus *nvdimm_bus;
757 resource_size_t max = 0;
758 struct resource *res;
759
760 /* if a dimm is disabled the available capacity is zero */
761 if (!ndd)
762 return 0;
763
764 nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
765 if (__reserve_free_pmem(&nd_region->dev, nd_mapping->nvdimm))
766 return 0;
767 for_each_dpa_resource(ndd, res) {
768 if (strcmp(res->name, "pmem-reserve") != 0)
769 continue;
770 if (resource_size(res) > max)
771 max = resource_size(res);
772 }
773 release_free_pmem(nvdimm_bus, nd_mapping);
774 return max;
775}
776
777/**
778 * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
779 * @nd_mapping: container of dpa-resource-root + labels
780 * @nd_region: constrain available space check to this reference region
781 * @overlap: calculate available space assuming this level of overlap
782 *
783 * Validate that a PMEM label, if present, aligns with the start of an
784 * interleave set and truncate the available size at the lowest BLK
785 * overlap point.
786 *
787 * The expectation is that this routine is called multiple times as it
788 * probes for the largest BLK encroachment for any single member DIMM of
789 * the interleave set. Once that value is determined the PMEM-limit for
790 * the set can be established.
791 */
792resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
793 struct nd_mapping *nd_mapping, resource_size_t *overlap)
794{
795 resource_size_t map_start, map_end, busy = 0, available, blk_start;
796 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
797 struct resource *res;
798 const char *reason;
799
800 if (!ndd)
801 return 0;
802
803 map_start = nd_mapping->start;
804 map_end = map_start + nd_mapping->size - 1;
805 blk_start = max(map_start, map_end + 1 - *overlap);
806 for_each_dpa_resource(ndd, res) {
807 if (res->start >= map_start && res->start < map_end) {
808 if (strncmp(res->name, "blk", 3) == 0)
809 blk_start = min(blk_start,
810 max(map_start, res->start));
811 else if (res->end > map_end) {
812 reason = "misaligned to iset";
813 goto err;
814 } else
815 busy += resource_size(res);
816 } else if (res->end >= map_start && res->end <= map_end) {
817 if (strncmp(res->name, "blk", 3) == 0) {
818 /*
819 * If a BLK allocation overlaps the start of
820 * PMEM the entire interleave set may now only
821 * be used for BLK.
822 */
823 blk_start = map_start;
824 } else
825 busy += resource_size(res);
826 } else if (map_start > res->start && map_start < res->end) {
827 /* total eclipse of the mapping */
828 busy += nd_mapping->size;
829 blk_start = map_start;
830 }
831 }
832
833 *overlap = map_end + 1 - blk_start;
834 available = blk_start - map_start;
835 if (busy < available)
836 return available - busy;
837 return 0;
838
839 err:
840 nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
841 return 0;
842}
843
844void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
845{
846 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
847 kfree(res->name);
848 __release_region(&ndd->dpa, res->start, resource_size(res));
849}
850
851struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
852 struct nd_label_id *label_id, resource_size_t start,
853 resource_size_t n)
854{
855 char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
856 struct resource *res;
857
858 if (!name)
859 return NULL;
860
861 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
862 res = __request_region(&ndd->dpa, start, n, name, 0);
863 if (!res)
864 kfree(name);
865 return res;
866}
867
868/**
869 * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
870 * @nvdimm: container of dpa-resource-root + labels
871 * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
872 */
873resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
874 struct nd_label_id *label_id)
875{
876 resource_size_t allocated = 0;
877 struct resource *res;
878
879 for_each_dpa_resource(ndd, res)
880 if (strcmp(res->name, label_id->id) == 0)
881 allocated += resource_size(res);
882
883 return allocated;
884}
885
886static int count_dimms(struct device *dev, void *c)
887{
888 int *count = c;
889
890 if (is_nvdimm(dev))
891 (*count)++;
892 return 0;
893}
894
895int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
896{
897 int count = 0;
898 /* Flush any possible dimm registration failures */
899 nd_synchronize();
900
901 device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
902 dev_dbg(&nvdimm_bus->dev, "count: %d\n", count);
903 if (count != dimm_count)
904 return -ENXIO;
905 return 0;
906}
907EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
908
909void __exit nvdimm_devs_exit(void)
910{
911 ida_destroy(&dimm_ida);
912}
913