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_init_config_data(struct nvdimm_drvdata *ndd)
90 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
91 int rc = validate_dimm(ndd), cmd_rc = 0;
92 struct nd_cmd_get_config_data_hdr *cmd;
93 struct nvdimm_bus_descriptor *nd_desc;
94 u32 max_cmd_size, config_size;
103 if (ndd->nsarea.status || ndd->nsarea.max_xfer == 0
104 || ndd->nsarea.config_size < ND_LABEL_MIN_SIZE) {
105 dev_dbg(ndd->dev, "failed to init config data area: (%d:%d)\n",
106 ndd->nsarea.max_xfer, ndd->nsarea.config_size);
110 ndd->data = kvmalloc(ndd->nsarea.config_size, GFP_KERNEL);
114 max_cmd_size = min_t(u32, PAGE_SIZE, ndd->nsarea.max_xfer);
115 cmd = kzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
119 nd_desc = nvdimm_bus->nd_desc;
120 for (config_size = ndd->nsarea.config_size, offset = 0;
121 config_size; config_size -= cmd->in_length,
122 offset += cmd->in_length) {
123 cmd->in_length = min(config_size, max_cmd_size);
124 cmd->in_offset = offset;
125 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
126 ND_CMD_GET_CONFIG_DATA, cmd,
127 cmd->in_length + sizeof(*cmd), &cmd_rc);
134 memcpy(ndd->data + offset, cmd->out_buf, cmd->in_length);
136 dev_dbg(ndd->dev, "len: %zu rc: %d\n", offset, rc);
142 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
143 void *buf, size_t len)
145 size_t max_cmd_size, buf_offset;
146 struct nd_cmd_set_config_hdr *cmd;
147 int rc = validate_dimm(ndd), cmd_rc = 0;
148 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
149 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
157 if (offset + len > ndd->nsarea.config_size)
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);
166 for (buf_offset = 0; len; len -= cmd->in_length,
167 buf_offset += cmd->in_length) {
170 cmd->in_offset = offset + buf_offset;
171 cmd->in_length = min(max_cmd_size, len);
172 memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
174 /* status is output in the last 4-bytes of the command buffer */
175 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
177 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
178 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
191 void nvdimm_set_aliasing(struct device *dev)
193 struct nvdimm *nvdimm = to_nvdimm(dev);
195 set_bit(NDD_ALIASING, &nvdimm->flags);
198 void nvdimm_set_locked(struct device *dev)
200 struct nvdimm *nvdimm = to_nvdimm(dev);
202 set_bit(NDD_LOCKED, &nvdimm->flags);
205 void nvdimm_clear_locked(struct device *dev)
207 struct nvdimm *nvdimm = to_nvdimm(dev);
209 clear_bit(NDD_LOCKED, &nvdimm->flags);
212 static void nvdimm_release(struct device *dev)
214 struct nvdimm *nvdimm = to_nvdimm(dev);
216 ida_simple_remove(&dimm_ida, nvdimm->id);
220 static struct device_type nvdimm_device_type = {
222 .release = nvdimm_release,
225 bool is_nvdimm(struct device *dev)
227 return dev->type == &nvdimm_device_type;
230 struct nvdimm *to_nvdimm(struct device *dev)
232 struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
234 WARN_ON(!is_nvdimm(dev));
237 EXPORT_SYMBOL_GPL(to_nvdimm);
239 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
241 struct nd_region *nd_region = &ndbr->nd_region;
242 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
244 return nd_mapping->nvdimm;
246 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
248 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
250 /* pmem mapping properties are private to libnvdimm */
251 return ARCH_MEMREMAP_PMEM;
253 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
255 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
257 struct nvdimm *nvdimm = nd_mapping->nvdimm;
259 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
261 return dev_get_drvdata(&nvdimm->dev);
263 EXPORT_SYMBOL(to_ndd);
265 void nvdimm_drvdata_release(struct kref *kref)
267 struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
268 struct device *dev = ndd->dev;
269 struct resource *res, *_r;
271 dev_dbg(dev, "trace\n");
272 nvdimm_bus_lock(dev);
273 for_each_dpa_resource_safe(ndd, res, _r)
274 nvdimm_free_dpa(ndd, res);
275 nvdimm_bus_unlock(dev);
282 void get_ndd(struct nvdimm_drvdata *ndd)
284 kref_get(&ndd->kref);
287 void put_ndd(struct nvdimm_drvdata *ndd)
290 kref_put(&ndd->kref, nvdimm_drvdata_release);
293 const char *nvdimm_name(struct nvdimm *nvdimm)
295 return dev_name(&nvdimm->dev);
297 EXPORT_SYMBOL_GPL(nvdimm_name);
299 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
301 return &nvdimm->dev.kobj;
303 EXPORT_SYMBOL_GPL(nvdimm_kobj);
305 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
307 return nvdimm->cmd_mask;
309 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
311 void *nvdimm_provider_data(struct nvdimm *nvdimm)
314 return nvdimm->provider_data;
317 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
319 static ssize_t commands_show(struct device *dev,
320 struct device_attribute *attr, char *buf)
322 struct nvdimm *nvdimm = to_nvdimm(dev);
325 if (!nvdimm->cmd_mask)
326 return sprintf(buf, "\n");
328 for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
329 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
330 len += sprintf(buf + len, "\n");
333 static DEVICE_ATTR_RO(commands);
335 static ssize_t flags_show(struct device *dev,
336 struct device_attribute *attr, char *buf)
338 struct nvdimm *nvdimm = to_nvdimm(dev);
340 return sprintf(buf, "%s%s\n",
341 test_bit(NDD_ALIASING, &nvdimm->flags) ? "alias " : "",
342 test_bit(NDD_LOCKED, &nvdimm->flags) ? "lock " : "");
344 static DEVICE_ATTR_RO(flags);
346 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
349 struct nvdimm *nvdimm = to_nvdimm(dev);
352 * The state may be in the process of changing, userspace should
353 * quiesce probing if it wants a static answer
355 nvdimm_bus_lock(dev);
356 nvdimm_bus_unlock(dev);
357 return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
358 ? "active" : "idle");
360 static DEVICE_ATTR_RO(state);
362 static ssize_t __available_slots_show(struct nvdimm_drvdata *ndd, char *buf)
372 nvdimm_bus_lock(dev);
373 nfree = nd_label_nfree(ndd);
374 if (nfree - 1 > nfree) {
375 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
379 rc = sprintf(buf, "%d\n", nfree);
380 nvdimm_bus_unlock(dev);
384 static ssize_t available_slots_show(struct device *dev,
385 struct device_attribute *attr, char *buf)
390 rc = __available_slots_show(dev_get_drvdata(dev), buf);
395 static DEVICE_ATTR_RO(available_slots);
397 static struct attribute *nvdimm_attributes[] = {
398 &dev_attr_state.attr,
399 &dev_attr_flags.attr,
400 &dev_attr_commands.attr,
401 &dev_attr_available_slots.attr,
405 struct attribute_group nvdimm_attribute_group = {
406 .attrs = nvdimm_attributes,
408 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
410 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
411 const struct attribute_group **groups, unsigned long flags,
412 unsigned long cmd_mask, int num_flush,
413 struct resource *flush_wpq)
415 struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
421 nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
422 if (nvdimm->id < 0) {
426 nvdimm->provider_data = provider_data;
427 nvdimm->flags = flags;
428 nvdimm->cmd_mask = cmd_mask;
429 nvdimm->num_flush = num_flush;
430 nvdimm->flush_wpq = flush_wpq;
431 atomic_set(&nvdimm->busy, 0);
433 dev_set_name(dev, "nmem%d", nvdimm->id);
434 dev->parent = &nvdimm_bus->dev;
435 dev->type = &nvdimm_device_type;
436 dev->devt = MKDEV(nvdimm_major, nvdimm->id);
437 dev->groups = groups;
438 nd_device_register(dev);
442 EXPORT_SYMBOL_GPL(nvdimm_create);
444 int alias_dpa_busy(struct device *dev, void *data)
446 resource_size_t map_end, blk_start, new;
447 struct blk_alloc_info *info = data;
448 struct nd_mapping *nd_mapping;
449 struct nd_region *nd_region;
450 struct nvdimm_drvdata *ndd;
451 struct resource *res;
457 nd_region = to_nd_region(dev);
458 for (i = 0; i < nd_region->ndr_mappings; i++) {
459 nd_mapping = &nd_region->mapping[i];
460 if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
464 if (i >= nd_region->ndr_mappings)
467 ndd = to_ndd(nd_mapping);
468 map_end = nd_mapping->start + nd_mapping->size - 1;
469 blk_start = nd_mapping->start;
472 * In the allocation case ->res is set to free space that we are
473 * looking to validate against PMEM aliasing collision rules
474 * (i.e. BLK is allocated after all aliased PMEM).
477 if (info->res->start >= nd_mapping->start
478 && info->res->start < map_end)
486 * Find the free dpa from the end of the last pmem allocation to
487 * the end of the interleave-set mapping.
489 for_each_dpa_resource(ndd, res) {
490 if (strncmp(res->name, "pmem", 4) != 0)
492 if ((res->start >= blk_start && res->start < map_end)
493 || (res->end >= blk_start
494 && res->end <= map_end)) {
495 new = max(blk_start, min(map_end + 1, res->end + 1));
496 if (new != blk_start) {
503 /* update the free space range with the probed blk_start */
504 if (info->res && blk_start > info->res->start) {
505 info->res->start = max(info->res->start, blk_start);
506 if (info->res->start > info->res->end)
507 info->res->end = info->res->start - 1;
511 info->available -= blk_start - nd_mapping->start;
517 * nd_blk_available_dpa - account the unused dpa of BLK region
518 * @nd_mapping: container of dpa-resource-root + labels
520 * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
521 * we arrange for them to never start at an lower dpa than the last
522 * PMEM allocation in an aliased region.
524 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
526 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
527 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
528 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
529 struct blk_alloc_info info = {
530 .nd_mapping = nd_mapping,
531 .available = nd_mapping->size,
534 struct resource *res;
539 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
541 /* now account for busy blk allocations in unaliased dpa */
542 for_each_dpa_resource(ndd, res) {
543 if (strncmp(res->name, "blk", 3) != 0)
545 info.available -= resource_size(res);
548 return info.available;
552 * nd_pmem_max_contiguous_dpa - For the given dimm+region, return the max
553 * contiguous unallocated dpa range.
554 * @nd_region: constrain available space check to this reference region
555 * @nd_mapping: container of dpa-resource-root + labels
557 resource_size_t nd_pmem_max_contiguous_dpa(struct nd_region *nd_region,
558 struct nd_mapping *nd_mapping)
560 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
561 struct nvdimm_bus *nvdimm_bus;
562 resource_size_t max = 0;
563 struct resource *res;
565 /* if a dimm is disabled the available capacity is zero */
569 nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
570 if (__reserve_free_pmem(&nd_region->dev, nd_mapping->nvdimm))
572 for_each_dpa_resource(ndd, res) {
573 if (strcmp(res->name, "pmem-reserve") != 0)
575 if (resource_size(res) > max)
576 max = resource_size(res);
578 release_free_pmem(nvdimm_bus, nd_mapping);
583 * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
584 * @nd_mapping: container of dpa-resource-root + labels
585 * @nd_region: constrain available space check to this reference region
586 * @overlap: calculate available space assuming this level of overlap
588 * Validate that a PMEM label, if present, aligns with the start of an
589 * interleave set and truncate the available size at the lowest BLK
592 * The expectation is that this routine is called multiple times as it
593 * probes for the largest BLK encroachment for any single member DIMM of
594 * the interleave set. Once that value is determined the PMEM-limit for
595 * the set can be established.
597 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
598 struct nd_mapping *nd_mapping, resource_size_t *overlap)
600 resource_size_t map_start, map_end, busy = 0, available, blk_start;
601 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
602 struct resource *res;
608 map_start = nd_mapping->start;
609 map_end = map_start + nd_mapping->size - 1;
610 blk_start = max(map_start, map_end + 1 - *overlap);
611 for_each_dpa_resource(ndd, res) {
612 if (res->start >= map_start && res->start < map_end) {
613 if (strncmp(res->name, "blk", 3) == 0)
614 blk_start = min(blk_start,
615 max(map_start, res->start));
616 else if (res->end > map_end) {
617 reason = "misaligned to iset";
620 busy += resource_size(res);
621 } else if (res->end >= map_start && res->end <= map_end) {
622 if (strncmp(res->name, "blk", 3) == 0) {
624 * If a BLK allocation overlaps the start of
625 * PMEM the entire interleave set may now only
628 blk_start = map_start;
630 busy += resource_size(res);
631 } else if (map_start > res->start && map_start < res->end) {
632 /* total eclipse of the mapping */
633 busy += nd_mapping->size;
634 blk_start = map_start;
638 *overlap = map_end + 1 - blk_start;
639 available = blk_start - map_start;
640 if (busy < available)
641 return available - busy;
645 nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
649 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
651 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
653 __release_region(&ndd->dpa, res->start, resource_size(res));
656 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
657 struct nd_label_id *label_id, resource_size_t start,
660 char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
661 struct resource *res;
666 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
667 res = __request_region(&ndd->dpa, start, n, name, 0);
674 * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
675 * @nvdimm: container of dpa-resource-root + labels
676 * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
678 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
679 struct nd_label_id *label_id)
681 resource_size_t allocated = 0;
682 struct resource *res;
684 for_each_dpa_resource(ndd, res)
685 if (strcmp(res->name, label_id->id) == 0)
686 allocated += resource_size(res);
691 static int count_dimms(struct device *dev, void *c)
700 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
703 /* Flush any possible dimm registration failures */
706 device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
707 dev_dbg(&nvdimm_bus->dev, "count: %d\n", count);
708 if (count != dimm_count)
712 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
714 void __exit nvdimm_devs_exit(void)
716 ida_destroy(&dimm_ida);