xref: /linux/drivers/nvdimm/dimm_devs.c (revision a89988a6e00c5a099ee23619cd91dc8dc7ec9328)
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/vmalloc.h>
15 #include <linux/device.h>
16 #include <linux/ndctl.h>
17 #include <linux/slab.h>
18 #include <linux/io.h>
19 #include <linux/fs.h>
20 #include <linux/mm.h>
21 #include "nd-core.h"
22 #include "label.h"
23 #include "nd.h"
24 
25 static DEFINE_IDA(dimm_ida);
26 
27 /*
28  * Retrieve bus and dimm handle and return if this bus supports
29  * get_config_data commands
30  */
31 int nvdimm_check_config_data(struct device *dev)
32 {
33 	struct nvdimm *nvdimm = to_nvdimm(dev);
34 
35 	if (!nvdimm->cmd_mask ||
36 	    !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
37 		if (test_bit(NDD_ALIASING, &nvdimm->flags))
38 			return -ENXIO;
39 		else
40 			return -ENOTTY;
41 	}
42 
43 	return 0;
44 }
45 
46 static int validate_dimm(struct nvdimm_drvdata *ndd)
47 {
48 	int rc;
49 
50 	if (!ndd)
51 		return -EINVAL;
52 
53 	rc = nvdimm_check_config_data(ndd->dev);
54 	if (rc)
55 		dev_dbg(ndd->dev, "%pf: %s error: %d\n",
56 				__builtin_return_address(0), __func__, rc);
57 	return rc;
58 }
59 
60 /**
61  * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
62  * @nvdimm: dimm to initialize
63  */
64 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
65 {
66 	struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
67 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
68 	struct nvdimm_bus_descriptor *nd_desc;
69 	int rc = validate_dimm(ndd);
70 	int cmd_rc = 0;
71 
72 	if (rc)
73 		return rc;
74 
75 	if (cmd->config_size)
76 		return 0; /* already valid */
77 
78 	memset(cmd, 0, sizeof(*cmd));
79 	nd_desc = nvdimm_bus->nd_desc;
80 	rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
81 			ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd), &cmd_rc);
82 	if (rc < 0)
83 		return rc;
84 	return cmd_rc;
85 }
86 
87 int nvdimm_init_config_data(struct nvdimm_drvdata *ndd)
88 {
89 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
90 	struct nd_cmd_get_config_data_hdr *cmd;
91 	struct nvdimm_bus_descriptor *nd_desc;
92 	int rc = validate_dimm(ndd);
93 	u32 max_cmd_size, config_size;
94 	size_t offset;
95 
96 	if (rc)
97 		return rc;
98 
99 	if (ndd->data)
100 		return 0;
101 
102 	if (ndd->nsarea.status || ndd->nsarea.max_xfer == 0
103 			|| ndd->nsarea.config_size < ND_LABEL_MIN_SIZE) {
104 		dev_dbg(ndd->dev, "failed to init config data area: (%d:%d)\n",
105 				ndd->nsarea.max_xfer, ndd->nsarea.config_size);
106 		return -ENXIO;
107 	}
108 
109 	ndd->data = kmalloc(ndd->nsarea.config_size, GFP_KERNEL);
110 	if (!ndd->data)
111 		ndd->data = vmalloc(ndd->nsarea.config_size);
112 
113 	if (!ndd->data)
114 		return -ENOMEM;
115 
116 	max_cmd_size = min_t(u32, PAGE_SIZE, ndd->nsarea.max_xfer);
117 	cmd = kzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
118 	if (!cmd)
119 		return -ENOMEM;
120 
121 	nd_desc = nvdimm_bus->nd_desc;
122 	for (config_size = ndd->nsarea.config_size, offset = 0;
123 			config_size; config_size -= cmd->in_length,
124 			offset += cmd->in_length) {
125 		cmd->in_length = min(config_size, max_cmd_size);
126 		cmd->in_offset = offset;
127 		rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
128 				ND_CMD_GET_CONFIG_DATA, cmd,
129 				cmd->in_length + sizeof(*cmd), NULL);
130 		if (rc || cmd->status) {
131 			rc = -ENXIO;
132 			break;
133 		}
134 		memcpy(ndd->data + offset, cmd->out_buf, cmd->in_length);
135 	}
136 	dev_dbg(ndd->dev, "%s: len: %zu rc: %d\n", __func__, offset, rc);
137 	kfree(cmd);
138 
139 	return rc;
140 }
141 
142 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
143 		void *buf, size_t len)
144 {
145 	int rc = validate_dimm(ndd);
146 	size_t max_cmd_size, buf_offset;
147 	struct nd_cmd_set_config_hdr *cmd;
148 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
149 	struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
150 
151 	if (rc)
152 		return rc;
153 
154 	if (!ndd->data)
155 		return -ENXIO;
156 
157 	if (offset + len > ndd->nsarea.config_size)
158 		return -ENXIO;
159 
160 	max_cmd_size = min_t(u32, PAGE_SIZE, len);
161 	max_cmd_size = min_t(u32, max_cmd_size, ndd->nsarea.max_xfer);
162 	cmd = kzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
163 	if (!cmd)
164 		return -ENOMEM;
165 
166 	for (buf_offset = 0; len; len -= cmd->in_length,
167 			buf_offset += cmd->in_length) {
168 		size_t cmd_size;
169 		u32 *status;
170 
171 		cmd->in_offset = offset + buf_offset;
172 		cmd->in_length = min(max_cmd_size, len);
173 		memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
174 
175 		/* status is output in the last 4-bytes of the command buffer */
176 		cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
177 		status = ((void *) cmd) + cmd_size - sizeof(u32);
178 
179 		rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
180 				ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, NULL);
181 		if (rc || *status) {
182 			rc = rc ? rc : -ENXIO;
183 			break;
184 		}
185 	}
186 	kfree(cmd);
187 
188 	return rc;
189 }
190 
191 void nvdimm_set_aliasing(struct device *dev)
192 {
193 	struct nvdimm *nvdimm = to_nvdimm(dev);
194 
195 	set_bit(NDD_ALIASING, &nvdimm->flags);
196 }
197 
198 void nvdimm_set_locked(struct device *dev)
199 {
200 	struct nvdimm *nvdimm = to_nvdimm(dev);
201 
202 	set_bit(NDD_LOCKED, &nvdimm->flags);
203 }
204 
205 static 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 
213 static struct device_type nvdimm_device_type = {
214 	.name = "nvdimm",
215 	.release = nvdimm_release,
216 };
217 
218 bool is_nvdimm(struct device *dev)
219 {
220 	return dev->type == &nvdimm_device_type;
221 }
222 
223 struct 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 }
230 EXPORT_SYMBOL_GPL(to_nvdimm);
231 
232 struct 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 }
239 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
240 
241 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
242 {
243 	struct nvdimm *nvdimm = nd_mapping->nvdimm;
244 
245 	WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
246 
247 	return dev_get_drvdata(&nvdimm->dev);
248 }
249 EXPORT_SYMBOL(to_ndd);
250 
251 void nvdimm_drvdata_release(struct kref *kref)
252 {
253 	struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
254 	struct device *dev = ndd->dev;
255 	struct resource *res, *_r;
256 
257 	dev_dbg(dev, "%s\n", __func__);
258 
259 	nvdimm_bus_lock(dev);
260 	for_each_dpa_resource_safe(ndd, res, _r)
261 		nvdimm_free_dpa(ndd, res);
262 	nvdimm_bus_unlock(dev);
263 
264 	kvfree(ndd->data);
265 	kfree(ndd);
266 	put_device(dev);
267 }
268 
269 void get_ndd(struct nvdimm_drvdata *ndd)
270 {
271 	kref_get(&ndd->kref);
272 }
273 
274 void put_ndd(struct nvdimm_drvdata *ndd)
275 {
276 	if (ndd)
277 		kref_put(&ndd->kref, nvdimm_drvdata_release);
278 }
279 
280 const char *nvdimm_name(struct nvdimm *nvdimm)
281 {
282 	return dev_name(&nvdimm->dev);
283 }
284 EXPORT_SYMBOL_GPL(nvdimm_name);
285 
286 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
287 {
288 	return &nvdimm->dev.kobj;
289 }
290 EXPORT_SYMBOL_GPL(nvdimm_kobj);
291 
292 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
293 {
294 	return nvdimm->cmd_mask;
295 }
296 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
297 
298 void *nvdimm_provider_data(struct nvdimm *nvdimm)
299 {
300 	if (nvdimm)
301 		return nvdimm->provider_data;
302 	return NULL;
303 }
304 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
305 
306 static ssize_t commands_show(struct device *dev,
307 		struct device_attribute *attr, char *buf)
308 {
309 	struct nvdimm *nvdimm = to_nvdimm(dev);
310 	int cmd, len = 0;
311 
312 	if (!nvdimm->cmd_mask)
313 		return sprintf(buf, "\n");
314 
315 	for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
316 		len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
317 	len += sprintf(buf + len, "\n");
318 	return len;
319 }
320 static DEVICE_ATTR_RO(commands);
321 
322 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
323 		char *buf)
324 {
325 	struct nvdimm *nvdimm = to_nvdimm(dev);
326 
327 	/*
328 	 * The state may be in the process of changing, userspace should
329 	 * quiesce probing if it wants a static answer
330 	 */
331 	nvdimm_bus_lock(dev);
332 	nvdimm_bus_unlock(dev);
333 	return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
334 			? "active" : "idle");
335 }
336 static DEVICE_ATTR_RO(state);
337 
338 static ssize_t available_slots_show(struct device *dev,
339 		struct device_attribute *attr, char *buf)
340 {
341 	struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
342 	ssize_t rc;
343 	u32 nfree;
344 
345 	if (!ndd)
346 		return -ENXIO;
347 
348 	nvdimm_bus_lock(dev);
349 	nfree = nd_label_nfree(ndd);
350 	if (nfree - 1 > nfree) {
351 		dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
352 		nfree = 0;
353 	} else
354 		nfree--;
355 	rc = sprintf(buf, "%d\n", nfree);
356 	nvdimm_bus_unlock(dev);
357 	return rc;
358 }
359 static DEVICE_ATTR_RO(available_slots);
360 
361 static struct attribute *nvdimm_attributes[] = {
362 	&dev_attr_state.attr,
363 	&dev_attr_commands.attr,
364 	&dev_attr_available_slots.attr,
365 	NULL,
366 };
367 
368 struct attribute_group nvdimm_attribute_group = {
369 	.attrs = nvdimm_attributes,
370 };
371 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
372 
373 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
374 		const struct attribute_group **groups, unsigned long flags,
375 		unsigned long cmd_mask, int num_flush,
376 		struct resource *flush_wpq)
377 {
378 	struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
379 	struct device *dev;
380 
381 	if (!nvdimm)
382 		return NULL;
383 
384 	nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
385 	if (nvdimm->id < 0) {
386 		kfree(nvdimm);
387 		return NULL;
388 	}
389 	nvdimm->provider_data = provider_data;
390 	nvdimm->flags = flags;
391 	nvdimm->cmd_mask = cmd_mask;
392 	nvdimm->num_flush = num_flush;
393 	nvdimm->flush_wpq = flush_wpq;
394 	atomic_set(&nvdimm->busy, 0);
395 	dev = &nvdimm->dev;
396 	dev_set_name(dev, "nmem%d", nvdimm->id);
397 	dev->parent = &nvdimm_bus->dev;
398 	dev->type = &nvdimm_device_type;
399 	dev->devt = MKDEV(nvdimm_major, nvdimm->id);
400 	dev->groups = groups;
401 	nd_device_register(dev);
402 
403 	return nvdimm;
404 }
405 EXPORT_SYMBOL_GPL(nvdimm_create);
406 
407 int alias_dpa_busy(struct device *dev, void *data)
408 {
409 	resource_size_t map_end, blk_start, new;
410 	struct blk_alloc_info *info = data;
411 	struct nd_mapping *nd_mapping;
412 	struct nd_region *nd_region;
413 	struct nvdimm_drvdata *ndd;
414 	struct resource *res;
415 	int i;
416 
417 	if (!is_nd_pmem(dev))
418 		return 0;
419 
420 	nd_region = to_nd_region(dev);
421 	for (i = 0; i < nd_region->ndr_mappings; i++) {
422 		nd_mapping  = &nd_region->mapping[i];
423 		if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
424 			break;
425 	}
426 
427 	if (i >= nd_region->ndr_mappings)
428 		return 0;
429 
430 	ndd = to_ndd(nd_mapping);
431 	map_end = nd_mapping->start + nd_mapping->size - 1;
432 	blk_start = nd_mapping->start;
433 
434 	/*
435 	 * In the allocation case ->res is set to free space that we are
436 	 * looking to validate against PMEM aliasing collision rules
437 	 * (i.e. BLK is allocated after all aliased PMEM).
438 	 */
439 	if (info->res) {
440 		if (info->res->start >= nd_mapping->start
441 				&& info->res->start < map_end)
442 			/* pass */;
443 		else
444 			return 0;
445 	}
446 
447  retry:
448 	/*
449 	 * Find the free dpa from the end of the last pmem allocation to
450 	 * the end of the interleave-set mapping.
451 	 */
452 	for_each_dpa_resource(ndd, res) {
453 		if (strncmp(res->name, "pmem", 4) != 0)
454 			continue;
455 		if ((res->start >= blk_start && res->start < map_end)
456 				|| (res->end >= blk_start
457 					&& res->end <= map_end)) {
458 			new = max(blk_start, min(map_end + 1, res->end + 1));
459 			if (new != blk_start) {
460 				blk_start = new;
461 				goto retry;
462 			}
463 		}
464 	}
465 
466 	/* update the free space range with the probed blk_start */
467 	if (info->res && blk_start > info->res->start) {
468 		info->res->start = max(info->res->start, blk_start);
469 		if (info->res->start > info->res->end)
470 			info->res->end = info->res->start - 1;
471 		return 1;
472 	}
473 
474 	info->available -= blk_start - nd_mapping->start;
475 
476 	return 0;
477 }
478 
479 /**
480  * nd_blk_available_dpa - account the unused dpa of BLK region
481  * @nd_mapping: container of dpa-resource-root + labels
482  *
483  * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
484  * we arrange for them to never start at an lower dpa than the last
485  * PMEM allocation in an aliased region.
486  */
487 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
488 {
489 	struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
490 	struct nd_mapping *nd_mapping = &nd_region->mapping[0];
491 	struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
492 	struct blk_alloc_info info = {
493 		.nd_mapping = nd_mapping,
494 		.available = nd_mapping->size,
495 		.res = NULL,
496 	};
497 	struct resource *res;
498 
499 	if (!ndd)
500 		return 0;
501 
502 	device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
503 
504 	/* now account for busy blk allocations in unaliased dpa */
505 	for_each_dpa_resource(ndd, res) {
506 		if (strncmp(res->name, "blk", 3) != 0)
507 			continue;
508 		info.available -= resource_size(res);
509 	}
510 
511 	return info.available;
512 }
513 
514 /**
515  * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
516  * @nd_mapping: container of dpa-resource-root + labels
517  * @nd_region: constrain available space check to this reference region
518  * @overlap: calculate available space assuming this level of overlap
519  *
520  * Validate that a PMEM label, if present, aligns with the start of an
521  * interleave set and truncate the available size at the lowest BLK
522  * overlap point.
523  *
524  * The expectation is that this routine is called multiple times as it
525  * probes for the largest BLK encroachment for any single member DIMM of
526  * the interleave set.  Once that value is determined the PMEM-limit for
527  * the set can be established.
528  */
529 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
530 		struct nd_mapping *nd_mapping, resource_size_t *overlap)
531 {
532 	resource_size_t map_start, map_end, busy = 0, available, blk_start;
533 	struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
534 	struct resource *res;
535 	const char *reason;
536 
537 	if (!ndd)
538 		return 0;
539 
540 	map_start = nd_mapping->start;
541 	map_end = map_start + nd_mapping->size - 1;
542 	blk_start = max(map_start, map_end + 1 - *overlap);
543 	for_each_dpa_resource(ndd, res) {
544 		if (res->start >= map_start && res->start < map_end) {
545 			if (strncmp(res->name, "blk", 3) == 0)
546 				blk_start = min(blk_start,
547 						max(map_start, res->start));
548 			else if (res->end > map_end) {
549 				reason = "misaligned to iset";
550 				goto err;
551 			} else
552 				busy += resource_size(res);
553 		} else if (res->end >= map_start && res->end <= map_end) {
554 			if (strncmp(res->name, "blk", 3) == 0) {
555 				/*
556 				 * If a BLK allocation overlaps the start of
557 				 * PMEM the entire interleave set may now only
558 				 * be used for BLK.
559 				 */
560 				blk_start = map_start;
561 			} else
562 				busy += resource_size(res);
563 		} else if (map_start > res->start && map_start < res->end) {
564 			/* total eclipse of the mapping */
565 			busy += nd_mapping->size;
566 			blk_start = map_start;
567 		}
568 	}
569 
570 	*overlap = map_end + 1 - blk_start;
571 	available = blk_start - map_start;
572 	if (busy < available)
573 		return available - busy;
574 	return 0;
575 
576  err:
577 	nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
578 	return 0;
579 }
580 
581 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
582 {
583 	WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
584 	kfree(res->name);
585 	__release_region(&ndd->dpa, res->start, resource_size(res));
586 }
587 
588 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
589 		struct nd_label_id *label_id, resource_size_t start,
590 		resource_size_t n)
591 {
592 	char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
593 	struct resource *res;
594 
595 	if (!name)
596 		return NULL;
597 
598 	WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
599 	res = __request_region(&ndd->dpa, start, n, name, 0);
600 	if (!res)
601 		kfree(name);
602 	return res;
603 }
604 
605 /**
606  * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
607  * @nvdimm: container of dpa-resource-root + labels
608  * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
609  */
610 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
611 		struct nd_label_id *label_id)
612 {
613 	resource_size_t allocated = 0;
614 	struct resource *res;
615 
616 	for_each_dpa_resource(ndd, res)
617 		if (strcmp(res->name, label_id->id) == 0)
618 			allocated += resource_size(res);
619 
620 	return allocated;
621 }
622 
623 static int count_dimms(struct device *dev, void *c)
624 {
625 	int *count = c;
626 
627 	if (is_nvdimm(dev))
628 		(*count)++;
629 	return 0;
630 }
631 
632 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
633 {
634 	int count = 0;
635 	/* Flush any possible dimm registration failures */
636 	nd_synchronize();
637 
638 	device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
639 	dev_dbg(&nvdimm_bus->dev, "%s: count: %d\n", __func__, count);
640 	if (count != dimm_count)
641 		return -ENXIO;
642 	return 0;
643 }
644 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
645 
646 void __exit nvdimm_devs_exit(void)
647 {
648 	ida_destroy(&dimm_ida);
649 }
650