1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4  */
5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6 #include <linux/moduleparam.h>
7 #include <linux/vmalloc.h>
8 #include <linux/device.h>
9 #include <linux/ndctl.h>
10 #include <linux/slab.h>
11 #include <linux/io.h>
12 #include <linux/fs.h>
13 #include <linux/mm.h>
14 #include "nd-core.h"
15 #include "label.h"
16 #include "pmem.h"
17 #include "nd.h"
18 
19 static DEFINE_IDA(dimm_ida);
20 
21 static bool noblk;
22 module_param(noblk, bool, 0444);
23 MODULE_PARM_DESC(noblk, "force disable BLK / local alias support");
24 
25 /*
26  * Retrieve bus and dimm handle and return if this bus supports
27  * get_config_data commands
28  */
nvdimm_check_config_data(struct device * dev)29 int nvdimm_check_config_data(struct device *dev)
30 {
31 	struct nvdimm *nvdimm = to_nvdimm(dev);
32 
33 	if (!nvdimm->cmd_mask ||
34 	    !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
35 		if (test_bit(NDD_ALIASING, &nvdimm->flags))
36 			return -ENXIO;
37 		else
38 			return -ENOTTY;
39 	}
40 
41 	return 0;
42 }
43 
validate_dimm(struct nvdimm_drvdata * ndd)44 static int validate_dimm(struct nvdimm_drvdata *ndd)
45 {
46 	int rc;
47 
48 	if (!ndd)
49 		return -EINVAL;
50 
51 	rc = nvdimm_check_config_data(ndd->dev);
52 	if (rc)
53 		dev_dbg(ndd->dev, "%ps: %s error: %d\n",
54 				__builtin_return_address(0), __func__, rc);
55 	return rc;
56 }
57 
58 /**
59  * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
60  * @nvdimm: dimm to initialize
61  */
nvdimm_init_nsarea(struct nvdimm_drvdata * ndd)62 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
63 {
64 	struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
65 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
66 	struct nvdimm_bus_descriptor *nd_desc;
67 	int rc = validate_dimm(ndd);
68 	int cmd_rc = 0;
69 
70 	if (rc)
71 		return rc;
72 
73 	if (cmd->config_size)
74 		return 0; /* already valid */
75 
76 	memset(cmd, 0, sizeof(*cmd));
77 	nd_desc = nvdimm_bus->nd_desc;
78 	rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
79 			ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd), &cmd_rc);
80 	if (rc < 0)
81 		return rc;
82 	return cmd_rc;
83 }
84 
nvdimm_get_config_data(struct nvdimm_drvdata * ndd,void * buf,size_t offset,size_t len)85 int nvdimm_get_config_data(struct nvdimm_drvdata *ndd, void *buf,
86 			   size_t offset, size_t len)
87 {
88 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
89 	struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
90 	int rc = validate_dimm(ndd), cmd_rc = 0;
91 	struct nd_cmd_get_config_data_hdr *cmd;
92 	size_t max_cmd_size, buf_offset;
93 
94 	if (rc)
95 		return rc;
96 
97 	if (offset + len > ndd->nsarea.config_size)
98 		return -ENXIO;
99 
100 	max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
101 	cmd = kvzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
102 	if (!cmd)
103 		return -ENOMEM;
104 
105 	for (buf_offset = 0; len;
106 	     len -= cmd->in_length, buf_offset += cmd->in_length) {
107 		size_t cmd_size;
108 
109 		cmd->in_offset = offset + buf_offset;
110 		cmd->in_length = min(max_cmd_size, len);
111 
112 		cmd_size = sizeof(*cmd) + cmd->in_length;
113 
114 		rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
115 				ND_CMD_GET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
116 		if (rc < 0)
117 			break;
118 		if (cmd_rc < 0) {
119 			rc = cmd_rc;
120 			break;
121 		}
122 
123 		/* out_buf should be valid, copy it into our output buffer */
124 		memcpy(buf + buf_offset, cmd->out_buf, cmd->in_length);
125 	}
126 	kvfree(cmd);
127 
128 	return rc;
129 }
130 
nvdimm_set_config_data(struct nvdimm_drvdata * ndd,size_t offset,void * buf,size_t len)131 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
132 		void *buf, size_t len)
133 {
134 	size_t max_cmd_size, buf_offset;
135 	struct nd_cmd_set_config_hdr *cmd;
136 	int rc = validate_dimm(ndd), cmd_rc = 0;
137 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
138 	struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
139 
140 	if (rc)
141 		return rc;
142 
143 	if (offset + len > ndd->nsarea.config_size)
144 		return -ENXIO;
145 
146 	max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
147 	cmd = kvzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
148 	if (!cmd)
149 		return -ENOMEM;
150 
151 	for (buf_offset = 0; len; len -= cmd->in_length,
152 			buf_offset += cmd->in_length) {
153 		size_t cmd_size;
154 
155 		cmd->in_offset = offset + buf_offset;
156 		cmd->in_length = min(max_cmd_size, len);
157 		memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
158 
159 		/* status is output in the last 4-bytes of the command buffer */
160 		cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
161 
162 		rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
163 				ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
164 		if (rc < 0)
165 			break;
166 		if (cmd_rc < 0) {
167 			rc = cmd_rc;
168 			break;
169 		}
170 	}
171 	kvfree(cmd);
172 
173 	return rc;
174 }
175 
nvdimm_set_aliasing(struct device * dev)176 void nvdimm_set_aliasing(struct device *dev)
177 {
178 	struct nvdimm *nvdimm = to_nvdimm(dev);
179 
180 	set_bit(NDD_ALIASING, &nvdimm->flags);
181 }
182 
nvdimm_set_locked(struct device * dev)183 void nvdimm_set_locked(struct device *dev)
184 {
185 	struct nvdimm *nvdimm = to_nvdimm(dev);
186 
187 	set_bit(NDD_LOCKED, &nvdimm->flags);
188 }
189 
nvdimm_clear_locked(struct device * dev)190 void nvdimm_clear_locked(struct device *dev)
191 {
192 	struct nvdimm *nvdimm = to_nvdimm(dev);
193 
194 	clear_bit(NDD_LOCKED, &nvdimm->flags);
195 }
196 
nvdimm_release(struct device * dev)197 static void nvdimm_release(struct device *dev)
198 {
199 	struct nvdimm *nvdimm = to_nvdimm(dev);
200 
201 	ida_simple_remove(&dimm_ida, nvdimm->id);
202 	kfree(nvdimm);
203 }
204 
205 static struct device_type nvdimm_device_type = {
206 	.name = "nvdimm",
207 	.release = nvdimm_release,
208 };
209 
is_nvdimm(struct device * dev)210 bool is_nvdimm(struct device *dev)
211 {
212 	return dev->type == &nvdimm_device_type;
213 }
214 
to_nvdimm(struct device * dev)215 struct nvdimm *to_nvdimm(struct device *dev)
216 {
217 	struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
218 
219 	WARN_ON(!is_nvdimm(dev));
220 	return nvdimm;
221 }
222 EXPORT_SYMBOL_GPL(to_nvdimm);
223 
nd_blk_region_to_dimm(struct nd_blk_region * ndbr)224 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
225 {
226 	struct nd_region *nd_region = &ndbr->nd_region;
227 	struct nd_mapping *nd_mapping = &nd_region->mapping[0];
228 
229 	return nd_mapping->nvdimm;
230 }
231 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
232 
nd_blk_memremap_flags(struct nd_blk_region * ndbr)233 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
234 {
235 	/* pmem mapping properties are private to libnvdimm */
236 	return ARCH_MEMREMAP_PMEM;
237 }
238 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
239 
to_ndd(struct nd_mapping * nd_mapping)240 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
241 {
242 	struct nvdimm *nvdimm = nd_mapping->nvdimm;
243 
244 	WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
245 
246 	return dev_get_drvdata(&nvdimm->dev);
247 }
248 EXPORT_SYMBOL(to_ndd);
249 
nvdimm_drvdata_release(struct kref * kref)250 void nvdimm_drvdata_release(struct kref *kref)
251 {
252 	struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
253 	struct device *dev = ndd->dev;
254 	struct resource *res, *_r;
255 
256 	dev_dbg(dev, "trace\n");
257 	nvdimm_bus_lock(dev);
258 	for_each_dpa_resource_safe(ndd, res, _r)
259 		nvdimm_free_dpa(ndd, res);
260 	nvdimm_bus_unlock(dev);
261 
262 	kvfree(ndd->data);
263 	kfree(ndd);
264 	put_device(dev);
265 }
266 
get_ndd(struct nvdimm_drvdata * ndd)267 void get_ndd(struct nvdimm_drvdata *ndd)
268 {
269 	kref_get(&ndd->kref);
270 }
271 
put_ndd(struct nvdimm_drvdata * ndd)272 void put_ndd(struct nvdimm_drvdata *ndd)
273 {
274 	if (ndd)
275 		kref_put(&ndd->kref, nvdimm_drvdata_release);
276 }
277 
nvdimm_name(struct nvdimm * nvdimm)278 const char *nvdimm_name(struct nvdimm *nvdimm)
279 {
280 	return dev_name(&nvdimm->dev);
281 }
282 EXPORT_SYMBOL_GPL(nvdimm_name);
283 
nvdimm_kobj(struct nvdimm * nvdimm)284 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
285 {
286 	return &nvdimm->dev.kobj;
287 }
288 EXPORT_SYMBOL_GPL(nvdimm_kobj);
289 
nvdimm_cmd_mask(struct nvdimm * nvdimm)290 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
291 {
292 	return nvdimm->cmd_mask;
293 }
294 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
295 
nvdimm_provider_data(struct nvdimm * nvdimm)296 void *nvdimm_provider_data(struct nvdimm *nvdimm)
297 {
298 	if (nvdimm)
299 		return nvdimm->provider_data;
300 	return NULL;
301 }
302 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
303 
commands_show(struct device * dev,struct device_attribute * attr,char * buf)304 static ssize_t commands_show(struct device *dev,
305 		struct device_attribute *attr, char *buf)
306 {
307 	struct nvdimm *nvdimm = to_nvdimm(dev);
308 	int cmd, len = 0;
309 
310 	if (!nvdimm->cmd_mask)
311 		return sprintf(buf, "\n");
312 
313 	for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
314 		len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
315 	len += sprintf(buf + len, "\n");
316 	return len;
317 }
318 static DEVICE_ATTR_RO(commands);
319 
flags_show(struct device * dev,struct device_attribute * attr,char * buf)320 static ssize_t flags_show(struct device *dev,
321 		struct device_attribute *attr, char *buf)
322 {
323 	struct nvdimm *nvdimm = to_nvdimm(dev);
324 
325 	return sprintf(buf, "%s%s\n",
326 			test_bit(NDD_ALIASING, &nvdimm->flags) ? "alias " : "",
327 			test_bit(NDD_LOCKED, &nvdimm->flags) ? "lock " : "");
328 }
329 static DEVICE_ATTR_RO(flags);
330 
state_show(struct device * dev,struct device_attribute * attr,char * buf)331 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
332 		char *buf)
333 {
334 	struct nvdimm *nvdimm = to_nvdimm(dev);
335 
336 	/*
337 	 * The state may be in the process of changing, userspace should
338 	 * quiesce probing if it wants a static answer
339 	 */
340 	nvdimm_bus_lock(dev);
341 	nvdimm_bus_unlock(dev);
342 	return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
343 			? "active" : "idle");
344 }
345 static DEVICE_ATTR_RO(state);
346 
available_slots_show(struct device * dev,struct device_attribute * attr,char * buf)347 static ssize_t available_slots_show(struct device *dev,
348 		struct device_attribute *attr, char *buf)
349 {
350 	struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
351 	ssize_t rc;
352 	u32 nfree;
353 
354 	if (!ndd)
355 		return -ENXIO;
356 
357 	nvdimm_bus_lock(dev);
358 	nfree = nd_label_nfree(ndd);
359 	if (nfree - 1 > nfree) {
360 		dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
361 		nfree = 0;
362 	} else
363 		nfree--;
364 	rc = sprintf(buf, "%d\n", nfree);
365 	nvdimm_bus_unlock(dev);
366 	return rc;
367 }
368 static DEVICE_ATTR_RO(available_slots);
369 
security_show(struct device * dev,struct device_attribute * attr,char * buf)370 __weak ssize_t security_show(struct device *dev,
371 		struct device_attribute *attr, char *buf)
372 {
373 	struct nvdimm *nvdimm = to_nvdimm(dev);
374 
375 	if (test_bit(NVDIMM_SECURITY_DISABLED, &nvdimm->sec.flags))
376 		return sprintf(buf, "disabled\n");
377 	if (test_bit(NVDIMM_SECURITY_UNLOCKED, &nvdimm->sec.flags))
378 		return sprintf(buf, "unlocked\n");
379 	if (test_bit(NVDIMM_SECURITY_LOCKED, &nvdimm->sec.flags))
380 		return sprintf(buf, "locked\n");
381 	if (test_bit(NVDIMM_SECURITY_OVERWRITE, &nvdimm->sec.flags))
382 		return sprintf(buf, "overwrite\n");
383 	return -ENOTTY;
384 }
385 
frozen_show(struct device * dev,struct device_attribute * attr,char * buf)386 static ssize_t frozen_show(struct device *dev,
387 		struct device_attribute *attr, char *buf)
388 {
389 	struct nvdimm *nvdimm = to_nvdimm(dev);
390 
391 	return sprintf(buf, "%d\n", test_bit(NVDIMM_SECURITY_FROZEN,
392 				&nvdimm->sec.flags));
393 }
394 static DEVICE_ATTR_RO(frozen);
395 
security_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t len)396 static ssize_t security_store(struct device *dev,
397 		struct device_attribute *attr, const char *buf, size_t len)
398 
399 {
400 	ssize_t rc;
401 
402 	/*
403 	 * Require all userspace triggered security management to be
404 	 * done while probing is idle and the DIMM is not in active use
405 	 * in any region.
406 	 */
407 	nd_device_lock(dev);
408 	nvdimm_bus_lock(dev);
409 	wait_nvdimm_bus_probe_idle(dev);
410 	rc = nvdimm_security_store(dev, buf, len);
411 	nvdimm_bus_unlock(dev);
412 	nd_device_unlock(dev);
413 
414 	return rc;
415 }
416 static DEVICE_ATTR_RW(security);
417 
418 static struct attribute *nvdimm_attributes[] = {
419 	&dev_attr_state.attr,
420 	&dev_attr_flags.attr,
421 	&dev_attr_commands.attr,
422 	&dev_attr_available_slots.attr,
423 	&dev_attr_security.attr,
424 	&dev_attr_frozen.attr,
425 	NULL,
426 };
427 
nvdimm_visible(struct kobject * kobj,struct attribute * a,int n)428 static umode_t nvdimm_visible(struct kobject *kobj, struct attribute *a, int n)
429 {
430 	struct device *dev = container_of(kobj, typeof(*dev), kobj);
431 	struct nvdimm *nvdimm = to_nvdimm(dev);
432 
433 	if (a != &dev_attr_security.attr && a != &dev_attr_frozen.attr)
434 		return a->mode;
435 	if (!nvdimm->sec.flags)
436 		return 0;
437 
438 	if (a == &dev_attr_security.attr) {
439 		/* Are there any state mutation ops (make writable)? */
440 		if (nvdimm->sec.ops->freeze || nvdimm->sec.ops->disable
441 				|| nvdimm->sec.ops->change_key
442 				|| nvdimm->sec.ops->erase
443 				|| nvdimm->sec.ops->overwrite)
444 			return a->mode;
445 		return 0444;
446 	}
447 
448 	if (nvdimm->sec.ops->freeze)
449 		return a->mode;
450 	return 0;
451 }
452 
453 struct attribute_group nvdimm_attribute_group = {
454 	.attrs = nvdimm_attributes,
455 	.is_visible = nvdimm_visible,
456 };
457 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
458 
__nvdimm_create(struct nvdimm_bus * nvdimm_bus,void * provider_data,const struct attribute_group ** groups,unsigned long flags,unsigned long cmd_mask,int num_flush,struct resource * flush_wpq,const char * dimm_id,const struct nvdimm_security_ops * sec_ops)459 struct nvdimm *__nvdimm_create(struct nvdimm_bus *nvdimm_bus,
460 		void *provider_data, const struct attribute_group **groups,
461 		unsigned long flags, unsigned long cmd_mask, int num_flush,
462 		struct resource *flush_wpq, const char *dimm_id,
463 		const struct nvdimm_security_ops *sec_ops)
464 {
465 	struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
466 	struct device *dev;
467 
468 	if (!nvdimm)
469 		return NULL;
470 
471 	nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
472 	if (nvdimm->id < 0) {
473 		kfree(nvdimm);
474 		return NULL;
475 	}
476 
477 	nvdimm->dimm_id = dimm_id;
478 	nvdimm->provider_data = provider_data;
479 	if (noblk)
480 		flags |= 1 << NDD_NOBLK;
481 	nvdimm->flags = flags;
482 	nvdimm->cmd_mask = cmd_mask;
483 	nvdimm->num_flush = num_flush;
484 	nvdimm->flush_wpq = flush_wpq;
485 	atomic_set(&nvdimm->busy, 0);
486 	dev = &nvdimm->dev;
487 	dev_set_name(dev, "nmem%d", nvdimm->id);
488 	dev->parent = &nvdimm_bus->dev;
489 	dev->type = &nvdimm_device_type;
490 	dev->devt = MKDEV(nvdimm_major, nvdimm->id);
491 	dev->groups = groups;
492 	nvdimm->sec.ops = sec_ops;
493 	nvdimm->sec.overwrite_tmo = 0;
494 	INIT_DELAYED_WORK(&nvdimm->dwork, nvdimm_security_overwrite_query);
495 	/*
496 	 * Security state must be initialized before device_add() for
497 	 * attribute visibility.
498 	 */
499 	/* get security state and extended (master) state */
500 	nvdimm->sec.flags = nvdimm_security_flags(nvdimm, NVDIMM_USER);
501 	nvdimm->sec.ext_flags = nvdimm_security_flags(nvdimm, NVDIMM_MASTER);
502 	nd_device_register(dev);
503 
504 	return nvdimm;
505 }
506 EXPORT_SYMBOL_GPL(__nvdimm_create);
507 
shutdown_security_notify(void * data)508 static void shutdown_security_notify(void *data)
509 {
510 	struct nvdimm *nvdimm = data;
511 
512 	sysfs_put(nvdimm->sec.overwrite_state);
513 }
514 
nvdimm_security_setup_events(struct device * dev)515 int nvdimm_security_setup_events(struct device *dev)
516 {
517 	struct nvdimm *nvdimm = to_nvdimm(dev);
518 
519 	if (!nvdimm->sec.flags || !nvdimm->sec.ops
520 			|| !nvdimm->sec.ops->overwrite)
521 		return 0;
522 	nvdimm->sec.overwrite_state = sysfs_get_dirent(dev->kobj.sd, "security");
523 	if (!nvdimm->sec.overwrite_state)
524 		return -ENOMEM;
525 
526 	return devm_add_action_or_reset(dev, shutdown_security_notify, nvdimm);
527 }
528 EXPORT_SYMBOL_GPL(nvdimm_security_setup_events);
529 
nvdimm_in_overwrite(struct nvdimm * nvdimm)530 int nvdimm_in_overwrite(struct nvdimm *nvdimm)
531 {
532 	return test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags);
533 }
534 EXPORT_SYMBOL_GPL(nvdimm_in_overwrite);
535 
nvdimm_security_freeze(struct nvdimm * nvdimm)536 int nvdimm_security_freeze(struct nvdimm *nvdimm)
537 {
538 	int rc;
539 
540 	WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
541 
542 	if (!nvdimm->sec.ops || !nvdimm->sec.ops->freeze)
543 		return -EOPNOTSUPP;
544 
545 	if (!nvdimm->sec.flags)
546 		return -EIO;
547 
548 	if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
549 		dev_warn(&nvdimm->dev, "Overwrite operation in progress.\n");
550 		return -EBUSY;
551 	}
552 
553 	rc = nvdimm->sec.ops->freeze(nvdimm);
554 	nvdimm->sec.flags = nvdimm_security_flags(nvdimm, NVDIMM_USER);
555 
556 	return rc;
557 }
558 
alias_dpa_busy(struct device * dev,void * data)559 int alias_dpa_busy(struct device *dev, void *data)
560 {
561 	resource_size_t map_end, blk_start, new;
562 	struct blk_alloc_info *info = data;
563 	struct nd_mapping *nd_mapping;
564 	struct nd_region *nd_region;
565 	struct nvdimm_drvdata *ndd;
566 	struct resource *res;
567 	int i;
568 
569 	if (!is_memory(dev))
570 		return 0;
571 
572 	nd_region = to_nd_region(dev);
573 	for (i = 0; i < nd_region->ndr_mappings; i++) {
574 		nd_mapping  = &nd_region->mapping[i];
575 		if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
576 			break;
577 	}
578 
579 	if (i >= nd_region->ndr_mappings)
580 		return 0;
581 
582 	ndd = to_ndd(nd_mapping);
583 	map_end = nd_mapping->start + nd_mapping->size - 1;
584 	blk_start = nd_mapping->start;
585 
586 	/*
587 	 * In the allocation case ->res is set to free space that we are
588 	 * looking to validate against PMEM aliasing collision rules
589 	 * (i.e. BLK is allocated after all aliased PMEM).
590 	 */
591 	if (info->res) {
592 		if (info->res->start >= nd_mapping->start
593 				&& info->res->start < map_end)
594 			/* pass */;
595 		else
596 			return 0;
597 	}
598 
599  retry:
600 	/*
601 	 * Find the free dpa from the end of the last pmem allocation to
602 	 * the end of the interleave-set mapping.
603 	 */
604 	for_each_dpa_resource(ndd, res) {
605 		if (strncmp(res->name, "pmem", 4) != 0)
606 			continue;
607 		if ((res->start >= blk_start && res->start < map_end)
608 				|| (res->end >= blk_start
609 					&& res->end <= map_end)) {
610 			new = max(blk_start, min(map_end + 1, res->end + 1));
611 			if (new != blk_start) {
612 				blk_start = new;
613 				goto retry;
614 			}
615 		}
616 	}
617 
618 	/* update the free space range with the probed blk_start */
619 	if (info->res && blk_start > info->res->start) {
620 		info->res->start = max(info->res->start, blk_start);
621 		if (info->res->start > info->res->end)
622 			info->res->end = info->res->start - 1;
623 		return 1;
624 	}
625 
626 	info->available -= blk_start - nd_mapping->start;
627 
628 	return 0;
629 }
630 
631 /**
632  * nd_blk_available_dpa - account the unused dpa of BLK region
633  * @nd_mapping: container of dpa-resource-root + labels
634  *
635  * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
636  * we arrange for them to never start at an lower dpa than the last
637  * PMEM allocation in an aliased region.
638  */
nd_blk_available_dpa(struct nd_region * nd_region)639 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
640 {
641 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
642 	struct nd_mapping *nd_mapping = &nd_region->mapping[0];
643 	struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
644 	struct blk_alloc_info info = {
645 		.nd_mapping = nd_mapping,
646 		.available = nd_mapping->size,
647 		.res = NULL,
648 	};
649 	struct resource *res;
650 
651 	if (!ndd)
652 		return 0;
653 
654 	device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
655 
656 	/* now account for busy blk allocations in unaliased dpa */
657 	for_each_dpa_resource(ndd, res) {
658 		if (strncmp(res->name, "blk", 3) != 0)
659 			continue;
660 		info.available -= resource_size(res);
661 	}
662 
663 	return info.available;
664 }
665 
666 /**
667  * nd_pmem_max_contiguous_dpa - For the given dimm+region, return the max
668  *			   contiguous unallocated dpa range.
669  * @nd_region: constrain available space check to this reference region
670  * @nd_mapping: container of dpa-resource-root + labels
671  */
nd_pmem_max_contiguous_dpa(struct nd_region * nd_region,struct nd_mapping * nd_mapping)672 resource_size_t nd_pmem_max_contiguous_dpa(struct nd_region *nd_region,
673 					   struct nd_mapping *nd_mapping)
674 {
675 	struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
676 	struct nvdimm_bus *nvdimm_bus;
677 	resource_size_t max = 0;
678 	struct resource *res;
679 
680 	/* if a dimm is disabled the available capacity is zero */
681 	if (!ndd)
682 		return 0;
683 
684 	nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
685 	if (__reserve_free_pmem(&nd_region->dev, nd_mapping->nvdimm))
686 		return 0;
687 	for_each_dpa_resource(ndd, res) {
688 		if (strcmp(res->name, "pmem-reserve") != 0)
689 			continue;
690 		if (resource_size(res) > max)
691 			max = resource_size(res);
692 	}
693 	release_free_pmem(nvdimm_bus, nd_mapping);
694 	return max;
695 }
696 
697 /**
698  * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
699  * @nd_mapping: container of dpa-resource-root + labels
700  * @nd_region: constrain available space check to this reference region
701  * @overlap: calculate available space assuming this level of overlap
702  *
703  * Validate that a PMEM label, if present, aligns with the start of an
704  * interleave set and truncate the available size at the lowest BLK
705  * overlap point.
706  *
707  * The expectation is that this routine is called multiple times as it
708  * probes for the largest BLK encroachment for any single member DIMM of
709  * the interleave set.  Once that value is determined the PMEM-limit for
710  * the set can be established.
711  */
nd_pmem_available_dpa(struct nd_region * nd_region,struct nd_mapping * nd_mapping,resource_size_t * overlap)712 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
713 		struct nd_mapping *nd_mapping, resource_size_t *overlap)
714 {
715 	resource_size_t map_start, map_end, busy = 0, available, blk_start;
716 	struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
717 	struct resource *res;
718 	const char *reason;
719 
720 	if (!ndd)
721 		return 0;
722 
723 	map_start = nd_mapping->start;
724 	map_end = map_start + nd_mapping->size - 1;
725 	blk_start = max(map_start, map_end + 1 - *overlap);
726 	for_each_dpa_resource(ndd, res) {
727 		if (res->start >= map_start && res->start < map_end) {
728 			if (strncmp(res->name, "blk", 3) == 0)
729 				blk_start = min(blk_start,
730 						max(map_start, res->start));
731 			else if (res->end > map_end) {
732 				reason = "misaligned to iset";
733 				goto err;
734 			} else
735 				busy += resource_size(res);
736 		} else if (res->end >= map_start && res->end <= map_end) {
737 			if (strncmp(res->name, "blk", 3) == 0) {
738 				/*
739 				 * If a BLK allocation overlaps the start of
740 				 * PMEM the entire interleave set may now only
741 				 * be used for BLK.
742 				 */
743 				blk_start = map_start;
744 			} else
745 				busy += resource_size(res);
746 		} else if (map_start > res->start && map_start < res->end) {
747 			/* total eclipse of the mapping */
748 			busy += nd_mapping->size;
749 			blk_start = map_start;
750 		}
751 	}
752 
753 	*overlap = map_end + 1 - blk_start;
754 	available = blk_start - map_start;
755 	if (busy < available)
756 		return available - busy;
757 	return 0;
758 
759  err:
760 	nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
761 	return 0;
762 }
763 
nvdimm_free_dpa(struct nvdimm_drvdata * ndd,struct resource * res)764 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
765 {
766 	WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
767 	kfree(res->name);
768 	__release_region(&ndd->dpa, res->start, resource_size(res));
769 }
770 
nvdimm_allocate_dpa(struct nvdimm_drvdata * ndd,struct nd_label_id * label_id,resource_size_t start,resource_size_t n)771 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
772 		struct nd_label_id *label_id, resource_size_t start,
773 		resource_size_t n)
774 {
775 	char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
776 	struct resource *res;
777 
778 	if (!name)
779 		return NULL;
780 
781 	WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
782 	res = __request_region(&ndd->dpa, start, n, name, 0);
783 	if (!res)
784 		kfree(name);
785 	return res;
786 }
787 
788 /**
789  * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
790  * @nvdimm: container of dpa-resource-root + labels
791  * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
792  */
nvdimm_allocated_dpa(struct nvdimm_drvdata * ndd,struct nd_label_id * label_id)793 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
794 		struct nd_label_id *label_id)
795 {
796 	resource_size_t allocated = 0;
797 	struct resource *res;
798 
799 	for_each_dpa_resource(ndd, res)
800 		if (strcmp(res->name, label_id->id) == 0)
801 			allocated += resource_size(res);
802 
803 	return allocated;
804 }
805 
count_dimms(struct device * dev,void * c)806 static int count_dimms(struct device *dev, void *c)
807 {
808 	int *count = c;
809 
810 	if (is_nvdimm(dev))
811 		(*count)++;
812 	return 0;
813 }
814 
nvdimm_bus_check_dimm_count(struct nvdimm_bus * nvdimm_bus,int dimm_count)815 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
816 {
817 	int count = 0;
818 	/* Flush any possible dimm registration failures */
819 	nd_synchronize();
820 
821 	device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
822 	dev_dbg(&nvdimm_bus->dev, "count: %d\n", count);
823 	if (count != dimm_count)
824 		return -ENXIO;
825 	return 0;
826 }
827 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
828 
nvdimm_devs_exit(void)829 void __exit nvdimm_devs_exit(void)
830 {
831 	ida_destroy(&dimm_ida);
832 }
833