1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
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>
19 static DEFINE_IDA(dimm_ida);
22 module_param(noblk, bool, 0444);
23 MODULE_PARM_DESC(noblk, "force disable BLK / local alias support");
26 * Retrieve bus and dimm handle and return if this bus supports
27 * get_config_data commands
29 int nvdimm_check_config_data(struct device *dev)
31 struct nvdimm *nvdimm = to_nvdimm(dev);
33 if (!nvdimm->cmd_mask ||
34 !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
35 if (test_bit(NDD_ALIASING, &nvdimm->flags))
44 static int validate_dimm(struct nvdimm_drvdata *ndd)
51 rc = nvdimm_check_config_data(ndd->dev);
53 dev_dbg(ndd->dev, "%ps: %s error: %d\n",
54 __builtin_return_address(0), __func__, rc);
59 * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
60 * @nvdimm: dimm to initialize
62 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
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);
74 return 0; /* already valid */
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);
85 int nvdimm_get_config_data(struct nvdimm_drvdata *ndd, void *buf,
86 size_t offset, size_t len)
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;
97 if (offset + len > ndd->nsarea.config_size)
100 max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
101 cmd = kvzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
105 for (buf_offset = 0; len;
106 len -= cmd->in_length, buf_offset += cmd->in_length) {
109 cmd->in_offset = offset + buf_offset;
110 cmd->in_length = min(max_cmd_size, len);
112 cmd_size = sizeof(*cmd) + cmd->in_length;
114 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
115 ND_CMD_GET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
123 /* out_buf should be valid, copy it into our output buffer */
124 memcpy(buf + buf_offset, cmd->out_buf, cmd->in_length);
131 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
132 void *buf, size_t len)
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;
143 if (offset + len > ndd->nsarea.config_size)
146 max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
147 cmd = kvzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
151 for (buf_offset = 0; len; len -= cmd->in_length,
152 buf_offset += cmd->in_length) {
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);
159 /* status is output in the last 4-bytes of the command buffer */
160 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
162 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
163 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
176 void nvdimm_set_aliasing(struct device *dev)
178 struct nvdimm *nvdimm = to_nvdimm(dev);
180 set_bit(NDD_ALIASING, &nvdimm->flags);
183 void nvdimm_set_locked(struct device *dev)
185 struct nvdimm *nvdimm = to_nvdimm(dev);
187 set_bit(NDD_LOCKED, &nvdimm->flags);
190 void nvdimm_clear_locked(struct device *dev)
192 struct nvdimm *nvdimm = to_nvdimm(dev);
194 clear_bit(NDD_LOCKED, &nvdimm->flags);
197 static void nvdimm_release(struct device *dev)
199 struct nvdimm *nvdimm = to_nvdimm(dev);
201 ida_simple_remove(&dimm_ida, nvdimm->id);
205 static struct device_type nvdimm_device_type = {
207 .release = nvdimm_release,
210 bool is_nvdimm(struct device *dev)
212 return dev->type == &nvdimm_device_type;
215 struct nvdimm *to_nvdimm(struct device *dev)
217 struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
219 WARN_ON(!is_nvdimm(dev));
222 EXPORT_SYMBOL_GPL(to_nvdimm);
224 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
226 struct nd_region *nd_region = &ndbr->nd_region;
227 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
229 return nd_mapping->nvdimm;
231 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
233 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
235 /* pmem mapping properties are private to libnvdimm */
236 return ARCH_MEMREMAP_PMEM;
238 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
240 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
242 struct nvdimm *nvdimm = nd_mapping->nvdimm;
244 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
246 return dev_get_drvdata(&nvdimm->dev);
248 EXPORT_SYMBOL(to_ndd);
250 void nvdimm_drvdata_release(struct kref *kref)
252 struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
253 struct device *dev = ndd->dev;
254 struct resource *res, *_r;
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);
267 void get_ndd(struct nvdimm_drvdata *ndd)
269 kref_get(&ndd->kref);
272 void put_ndd(struct nvdimm_drvdata *ndd)
275 kref_put(&ndd->kref, nvdimm_drvdata_release);
278 const char *nvdimm_name(struct nvdimm *nvdimm)
280 return dev_name(&nvdimm->dev);
282 EXPORT_SYMBOL_GPL(nvdimm_name);
284 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
286 return &nvdimm->dev.kobj;
288 EXPORT_SYMBOL_GPL(nvdimm_kobj);
290 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
292 return nvdimm->cmd_mask;
294 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
296 void *nvdimm_provider_data(struct nvdimm *nvdimm)
299 return nvdimm->provider_data;
302 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
304 static ssize_t commands_show(struct device *dev,
305 struct device_attribute *attr, char *buf)
307 struct nvdimm *nvdimm = to_nvdimm(dev);
310 if (!nvdimm->cmd_mask)
311 return sprintf(buf, "\n");
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");
318 static DEVICE_ATTR_RO(commands);
320 static ssize_t flags_show(struct device *dev,
321 struct device_attribute *attr, char *buf)
323 struct nvdimm *nvdimm = to_nvdimm(dev);
325 return sprintf(buf, "%s%s\n",
326 test_bit(NDD_ALIASING, &nvdimm->flags) ? "alias " : "",
327 test_bit(NDD_LOCKED, &nvdimm->flags) ? "lock " : "");
329 static DEVICE_ATTR_RO(flags);
331 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
334 struct nvdimm *nvdimm = to_nvdimm(dev);
337 * The state may be in the process of changing, userspace should
338 * quiesce probing if it wants a static answer
340 nvdimm_bus_lock(dev);
341 nvdimm_bus_unlock(dev);
342 return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
343 ? "active" : "idle");
345 static DEVICE_ATTR_RO(state);
347 static ssize_t available_slots_show(struct device *dev,
348 struct device_attribute *attr, char *buf)
350 struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
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");
364 rc = sprintf(buf, "%d\n", nfree);
365 nvdimm_bus_unlock(dev);
368 static DEVICE_ATTR_RO(available_slots);
370 __weak ssize_t security_show(struct device *dev,
371 struct device_attribute *attr, char *buf)
373 struct nvdimm *nvdimm = to_nvdimm(dev);
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");
386 static ssize_t frozen_show(struct device *dev,
387 struct device_attribute *attr, char *buf)
389 struct nvdimm *nvdimm = to_nvdimm(dev);
391 return sprintf(buf, "%d\n", test_bit(NVDIMM_SECURITY_FROZEN,
392 &nvdimm->sec.flags));
394 static DEVICE_ATTR_RO(frozen);
396 static ssize_t security_store(struct device *dev,
397 struct device_attribute *attr, const char *buf, size_t len)
403 * Require all userspace triggered security management to be
404 * done while probing is idle and the DIMM is not in active use
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);
416 static DEVICE_ATTR_RW(security);
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,
428 static umode_t nvdimm_visible(struct kobject *kobj, struct attribute *a, int n)
430 struct device *dev = container_of(kobj, typeof(*dev), kobj);
431 struct nvdimm *nvdimm = to_nvdimm(dev);
433 if (a != &dev_attr_security.attr && a != &dev_attr_frozen.attr)
435 if (!nvdimm->sec.flags)
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)
448 if (nvdimm->sec.ops->freeze)
453 struct attribute_group nvdimm_attribute_group = {
454 .attrs = nvdimm_attributes,
455 .is_visible = nvdimm_visible,
457 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
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)
465 struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
471 nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
472 if (nvdimm->id < 0) {
477 nvdimm->dimm_id = dimm_id;
478 nvdimm->provider_data = provider_data;
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);
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);
496 * Security state must be initialized before device_add() for
497 * attribute visibility.
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);
506 EXPORT_SYMBOL_GPL(__nvdimm_create);
508 static void shutdown_security_notify(void *data)
510 struct nvdimm *nvdimm = data;
512 sysfs_put(nvdimm->sec.overwrite_state);
515 int nvdimm_security_setup_events(struct device *dev)
517 struct nvdimm *nvdimm = to_nvdimm(dev);
519 if (!nvdimm->sec.flags || !nvdimm->sec.ops
520 || !nvdimm->sec.ops->overwrite)
522 nvdimm->sec.overwrite_state = sysfs_get_dirent(dev->kobj.sd, "security");
523 if (!nvdimm->sec.overwrite_state)
526 return devm_add_action_or_reset(dev, shutdown_security_notify, nvdimm);
528 EXPORT_SYMBOL_GPL(nvdimm_security_setup_events);
530 int nvdimm_in_overwrite(struct nvdimm *nvdimm)
532 return test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags);
534 EXPORT_SYMBOL_GPL(nvdimm_in_overwrite);
536 int nvdimm_security_freeze(struct nvdimm *nvdimm)
540 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
542 if (!nvdimm->sec.ops || !nvdimm->sec.ops->freeze)
545 if (!nvdimm->sec.flags)
548 if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
549 dev_warn(&nvdimm->dev, "Overwrite operation in progress.\n");
553 rc = nvdimm->sec.ops->freeze(nvdimm);
554 nvdimm->sec.flags = nvdimm_security_flags(nvdimm, NVDIMM_USER);
559 int alias_dpa_busy(struct device *dev, void *data)
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;
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)
579 if (i >= nd_region->ndr_mappings)
582 ndd = to_ndd(nd_mapping);
583 map_end = nd_mapping->start + nd_mapping->size - 1;
584 blk_start = nd_mapping->start;
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).
592 if (info->res->start >= nd_mapping->start
593 && info->res->start < map_end)
601 * Find the free dpa from the end of the last pmem allocation to
602 * the end of the interleave-set mapping.
604 for_each_dpa_resource(ndd, res) {
605 if (strncmp(res->name, "pmem", 4) != 0)
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) {
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;
626 info->available -= blk_start - nd_mapping->start;
632 * nd_blk_available_dpa - account the unused dpa of BLK region
633 * @nd_mapping: container of dpa-resource-root + labels
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.
639 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
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,
649 struct resource *res;
654 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
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)
660 info.available -= resource_size(res);
663 return info.available;
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
672 resource_size_t nd_pmem_max_contiguous_dpa(struct nd_region *nd_region,
673 struct nd_mapping *nd_mapping)
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;
680 /* if a dimm is disabled the available capacity is zero */
684 nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
685 if (__reserve_free_pmem(&nd_region->dev, nd_mapping->nvdimm))
687 for_each_dpa_resource(ndd, res) {
688 if (strcmp(res->name, "pmem-reserve") != 0)
690 if (resource_size(res) > max)
691 max = resource_size(res);
693 release_free_pmem(nvdimm_bus, nd_mapping);
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
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
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.
712 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
713 struct nd_mapping *nd_mapping, resource_size_t *overlap)
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;
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";
735 busy += resource_size(res);
736 } else if (res->end >= map_start && res->end <= map_end) {
737 if (strncmp(res->name, "blk", 3) == 0) {
739 * If a BLK allocation overlaps the start of
740 * PMEM the entire interleave set may now only
743 blk_start = map_start;
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;
753 *overlap = map_end + 1 - blk_start;
754 available = blk_start - map_start;
755 if (busy < available)
756 return available - busy;
760 nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
764 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
766 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
768 __release_region(&ndd->dpa, res->start, resource_size(res));
771 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
772 struct nd_label_id *label_id, resource_size_t start,
775 char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
776 struct resource *res;
781 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
782 res = __request_region(&ndd->dpa, start, n, name, 0);
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>
793 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
794 struct nd_label_id *label_id)
796 resource_size_t allocated = 0;
797 struct resource *res;
799 for_each_dpa_resource(ndd, res)
800 if (strcmp(res->name, label_id->id) == 0)
801 allocated += resource_size(res);
806 static int count_dimms(struct device *dev, void *c)
815 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
818 /* Flush any possible dimm registration failures */
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)
827 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
829 void __exit nvdimm_devs_exit(void)
831 ida_destroy(&dimm_ida);