2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
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.
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.
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>
26 static DEFINE_IDA(dimm_ida);
29 * Retrieve bus and dimm handle and return if this bus supports
30 * get_config_data commands
32 int nvdimm_check_config_data(struct device *dev)
34 struct nvdimm *nvdimm = to_nvdimm(dev);
36 if (!nvdimm->cmd_mask ||
37 !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
38 if (test_bit(NDD_ALIASING, &nvdimm->flags))
47 static int validate_dimm(struct nvdimm_drvdata *ndd)
54 rc = nvdimm_check_config_data(ndd->dev);
56 dev_dbg(ndd->dev, "%pf: %s error: %d\n",
57 __builtin_return_address(0), __func__, rc);
62 * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
63 * @nvdimm: dimm to initialize
65 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
67 struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
68 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
69 struct nvdimm_bus_descriptor *nd_desc;
70 int rc = validate_dimm(ndd);
77 return 0; /* already valid */
79 memset(cmd, 0, sizeof(*cmd));
80 nd_desc = nvdimm_bus->nd_desc;
81 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
82 ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd), &cmd_rc);
88 int nvdimm_get_config_data(struct nvdimm_drvdata *ndd, void *buf,
89 size_t offset, size_t len)
91 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
92 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
93 int rc = validate_dimm(ndd), cmd_rc = 0;
94 struct nd_cmd_get_config_data_hdr *cmd;
95 size_t max_cmd_size, buf_offset;
100 if (offset + len > ndd->nsarea.config_size)
103 max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
104 cmd = kvzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
108 for (buf_offset = 0; len;
109 len -= cmd->in_length, buf_offset += cmd->in_length) {
112 cmd->in_offset = offset + buf_offset;
113 cmd->in_length = min(max_cmd_size, len);
115 cmd_size = sizeof(*cmd) + cmd->in_length;
117 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
118 ND_CMD_GET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
126 /* out_buf should be valid, copy it into our output buffer */
127 memcpy(buf + buf_offset, cmd->out_buf, cmd->in_length);
134 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
135 void *buf, size_t len)
137 size_t max_cmd_size, buf_offset;
138 struct nd_cmd_set_config_hdr *cmd;
139 int rc = validate_dimm(ndd), cmd_rc = 0;
140 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
141 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
146 if (offset + len > ndd->nsarea.config_size)
149 max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
150 cmd = kvzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
154 for (buf_offset = 0; len; len -= cmd->in_length,
155 buf_offset += cmd->in_length) {
158 cmd->in_offset = offset + buf_offset;
159 cmd->in_length = min(max_cmd_size, len);
160 memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
162 /* status is output in the last 4-bytes of the command buffer */
163 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
165 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
166 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
179 void nvdimm_set_aliasing(struct device *dev)
181 struct nvdimm *nvdimm = to_nvdimm(dev);
183 set_bit(NDD_ALIASING, &nvdimm->flags);
186 void nvdimm_set_locked(struct device *dev)
188 struct nvdimm *nvdimm = to_nvdimm(dev);
190 set_bit(NDD_LOCKED, &nvdimm->flags);
193 void nvdimm_clear_locked(struct device *dev)
195 struct nvdimm *nvdimm = to_nvdimm(dev);
197 clear_bit(NDD_LOCKED, &nvdimm->flags);
200 static void nvdimm_release(struct device *dev)
202 struct nvdimm *nvdimm = to_nvdimm(dev);
204 ida_simple_remove(&dimm_ida, nvdimm->id);
208 static struct device_type nvdimm_device_type = {
210 .release = nvdimm_release,
213 bool is_nvdimm(struct device *dev)
215 return dev->type == &nvdimm_device_type;
218 struct nvdimm *to_nvdimm(struct device *dev)
220 struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
222 WARN_ON(!is_nvdimm(dev));
225 EXPORT_SYMBOL_GPL(to_nvdimm);
227 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
229 struct nd_region *nd_region = &ndbr->nd_region;
230 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
232 return nd_mapping->nvdimm;
234 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
236 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
238 /* pmem mapping properties are private to libnvdimm */
239 return ARCH_MEMREMAP_PMEM;
241 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
243 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
245 struct nvdimm *nvdimm = nd_mapping->nvdimm;
247 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
249 return dev_get_drvdata(&nvdimm->dev);
251 EXPORT_SYMBOL(to_ndd);
253 void nvdimm_drvdata_release(struct kref *kref)
255 struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
256 struct device *dev = ndd->dev;
257 struct resource *res, *_r;
259 dev_dbg(dev, "trace\n");
260 nvdimm_bus_lock(dev);
261 for_each_dpa_resource_safe(ndd, res, _r)
262 nvdimm_free_dpa(ndd, res);
263 nvdimm_bus_unlock(dev);
270 void get_ndd(struct nvdimm_drvdata *ndd)
272 kref_get(&ndd->kref);
275 void put_ndd(struct nvdimm_drvdata *ndd)
278 kref_put(&ndd->kref, nvdimm_drvdata_release);
281 const char *nvdimm_name(struct nvdimm *nvdimm)
283 return dev_name(&nvdimm->dev);
285 EXPORT_SYMBOL_GPL(nvdimm_name);
287 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
289 return &nvdimm->dev.kobj;
291 EXPORT_SYMBOL_GPL(nvdimm_kobj);
293 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
295 return nvdimm->cmd_mask;
297 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
299 void *nvdimm_provider_data(struct nvdimm *nvdimm)
302 return nvdimm->provider_data;
305 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
307 static ssize_t commands_show(struct device *dev,
308 struct device_attribute *attr, char *buf)
310 struct nvdimm *nvdimm = to_nvdimm(dev);
313 if (!nvdimm->cmd_mask)
314 return sprintf(buf, "\n");
316 for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
317 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
318 len += sprintf(buf + len, "\n");
321 static DEVICE_ATTR_RO(commands);
323 static ssize_t flags_show(struct device *dev,
324 struct device_attribute *attr, char *buf)
326 struct nvdimm *nvdimm = to_nvdimm(dev);
328 return sprintf(buf, "%s%s\n",
329 test_bit(NDD_ALIASING, &nvdimm->flags) ? "alias " : "",
330 test_bit(NDD_LOCKED, &nvdimm->flags) ? "lock " : "");
332 static DEVICE_ATTR_RO(flags);
334 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
337 struct nvdimm *nvdimm = to_nvdimm(dev);
340 * The state may be in the process of changing, userspace should
341 * quiesce probing if it wants a static answer
343 nvdimm_bus_lock(dev);
344 nvdimm_bus_unlock(dev);
345 return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
346 ? "active" : "idle");
348 static DEVICE_ATTR_RO(state);
350 static ssize_t available_slots_show(struct device *dev,
351 struct device_attribute *attr, char *buf)
353 struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
360 nvdimm_bus_lock(dev);
361 nfree = nd_label_nfree(ndd);
362 if (nfree - 1 > nfree) {
363 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
367 rc = sprintf(buf, "%d\n", nfree);
368 nvdimm_bus_unlock(dev);
371 static DEVICE_ATTR_RO(available_slots);
373 static struct attribute *nvdimm_attributes[] = {
374 &dev_attr_state.attr,
375 &dev_attr_flags.attr,
376 &dev_attr_commands.attr,
377 &dev_attr_available_slots.attr,
381 struct attribute_group nvdimm_attribute_group = {
382 .attrs = nvdimm_attributes,
384 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
386 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
387 const struct attribute_group **groups, unsigned long flags,
388 unsigned long cmd_mask, int num_flush,
389 struct resource *flush_wpq)
391 struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
397 nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
398 if (nvdimm->id < 0) {
402 nvdimm->provider_data = provider_data;
403 nvdimm->flags = flags;
404 nvdimm->cmd_mask = cmd_mask;
405 nvdimm->num_flush = num_flush;
406 nvdimm->flush_wpq = flush_wpq;
407 atomic_set(&nvdimm->busy, 0);
409 dev_set_name(dev, "nmem%d", nvdimm->id);
410 dev->parent = &nvdimm_bus->dev;
411 dev->type = &nvdimm_device_type;
412 dev->devt = MKDEV(nvdimm_major, nvdimm->id);
413 dev->groups = groups;
414 nd_device_register(dev);
418 EXPORT_SYMBOL_GPL(nvdimm_create);
420 int alias_dpa_busy(struct device *dev, void *data)
422 resource_size_t map_end, blk_start, new;
423 struct blk_alloc_info *info = data;
424 struct nd_mapping *nd_mapping;
425 struct nd_region *nd_region;
426 struct nvdimm_drvdata *ndd;
427 struct resource *res;
433 nd_region = to_nd_region(dev);
434 for (i = 0; i < nd_region->ndr_mappings; i++) {
435 nd_mapping = &nd_region->mapping[i];
436 if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
440 if (i >= nd_region->ndr_mappings)
443 ndd = to_ndd(nd_mapping);
444 map_end = nd_mapping->start + nd_mapping->size - 1;
445 blk_start = nd_mapping->start;
448 * In the allocation case ->res is set to free space that we are
449 * looking to validate against PMEM aliasing collision rules
450 * (i.e. BLK is allocated after all aliased PMEM).
453 if (info->res->start >= nd_mapping->start
454 && info->res->start < map_end)
462 * Find the free dpa from the end of the last pmem allocation to
463 * the end of the interleave-set mapping.
465 for_each_dpa_resource(ndd, res) {
466 if (strncmp(res->name, "pmem", 4) != 0)
468 if ((res->start >= blk_start && res->start < map_end)
469 || (res->end >= blk_start
470 && res->end <= map_end)) {
471 new = max(blk_start, min(map_end + 1, res->end + 1));
472 if (new != blk_start) {
479 /* update the free space range with the probed blk_start */
480 if (info->res && blk_start > info->res->start) {
481 info->res->start = max(info->res->start, blk_start);
482 if (info->res->start > info->res->end)
483 info->res->end = info->res->start - 1;
487 info->available -= blk_start - nd_mapping->start;
493 * nd_blk_available_dpa - account the unused dpa of BLK region
494 * @nd_mapping: container of dpa-resource-root + labels
496 * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
497 * we arrange for them to never start at an lower dpa than the last
498 * PMEM allocation in an aliased region.
500 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
502 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
503 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
504 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
505 struct blk_alloc_info info = {
506 .nd_mapping = nd_mapping,
507 .available = nd_mapping->size,
510 struct resource *res;
515 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
517 /* now account for busy blk allocations in unaliased dpa */
518 for_each_dpa_resource(ndd, res) {
519 if (strncmp(res->name, "blk", 3) != 0)
521 info.available -= resource_size(res);
524 return info.available;
528 * nd_pmem_max_contiguous_dpa - For the given dimm+region, return the max
529 * contiguous unallocated dpa range.
530 * @nd_region: constrain available space check to this reference region
531 * @nd_mapping: container of dpa-resource-root + labels
533 resource_size_t nd_pmem_max_contiguous_dpa(struct nd_region *nd_region,
534 struct nd_mapping *nd_mapping)
536 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
537 struct nvdimm_bus *nvdimm_bus;
538 resource_size_t max = 0;
539 struct resource *res;
541 /* if a dimm is disabled the available capacity is zero */
545 nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
546 if (__reserve_free_pmem(&nd_region->dev, nd_mapping->nvdimm))
548 for_each_dpa_resource(ndd, res) {
549 if (strcmp(res->name, "pmem-reserve") != 0)
551 if (resource_size(res) > max)
552 max = resource_size(res);
554 release_free_pmem(nvdimm_bus, nd_mapping);
559 * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
560 * @nd_mapping: container of dpa-resource-root + labels
561 * @nd_region: constrain available space check to this reference region
562 * @overlap: calculate available space assuming this level of overlap
564 * Validate that a PMEM label, if present, aligns with the start of an
565 * interleave set and truncate the available size at the lowest BLK
568 * The expectation is that this routine is called multiple times as it
569 * probes for the largest BLK encroachment for any single member DIMM of
570 * the interleave set. Once that value is determined the PMEM-limit for
571 * the set can be established.
573 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
574 struct nd_mapping *nd_mapping, resource_size_t *overlap)
576 resource_size_t map_start, map_end, busy = 0, available, blk_start;
577 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
578 struct resource *res;
584 map_start = nd_mapping->start;
585 map_end = map_start + nd_mapping->size - 1;
586 blk_start = max(map_start, map_end + 1 - *overlap);
587 for_each_dpa_resource(ndd, res) {
588 if (res->start >= map_start && res->start < map_end) {
589 if (strncmp(res->name, "blk", 3) == 0)
590 blk_start = min(blk_start,
591 max(map_start, res->start));
592 else if (res->end > map_end) {
593 reason = "misaligned to iset";
596 busy += resource_size(res);
597 } else if (res->end >= map_start && res->end <= map_end) {
598 if (strncmp(res->name, "blk", 3) == 0) {
600 * If a BLK allocation overlaps the start of
601 * PMEM the entire interleave set may now only
604 blk_start = map_start;
606 busy += resource_size(res);
607 } else if (map_start > res->start && map_start < res->end) {
608 /* total eclipse of the mapping */
609 busy += nd_mapping->size;
610 blk_start = map_start;
614 *overlap = map_end + 1 - blk_start;
615 available = blk_start - map_start;
616 if (busy < available)
617 return available - busy;
621 nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
625 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
627 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
629 __release_region(&ndd->dpa, res->start, resource_size(res));
632 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
633 struct nd_label_id *label_id, resource_size_t start,
636 char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
637 struct resource *res;
642 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
643 res = __request_region(&ndd->dpa, start, n, name, 0);
650 * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
651 * @nvdimm: container of dpa-resource-root + labels
652 * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
654 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
655 struct nd_label_id *label_id)
657 resource_size_t allocated = 0;
658 struct resource *res;
660 for_each_dpa_resource(ndd, res)
661 if (strcmp(res->name, label_id->id) == 0)
662 allocated += resource_size(res);
667 static int count_dimms(struct device *dev, void *c)
676 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
679 /* Flush any possible dimm registration failures */
682 device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
683 dev_dbg(&nvdimm_bus->dev, "count: %d\n", count);
684 if (count != dimm_count)
688 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
690 void __exit nvdimm_devs_exit(void)
692 ida_destroy(&dimm_ida);