2 * Copyright(c) 2013-2016 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 #include <linux/memremap.h>
14 #include <linux/blkdev.h>
15 #include <linux/device.h>
16 #include <linux/genhd.h>
17 #include <linux/sizes.h>
18 #include <linux/slab.h>
25 static void nd_pfn_release(struct device *dev)
27 struct nd_region *nd_region = to_nd_region(dev->parent);
28 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
30 dev_dbg(dev, "%s\n", __func__);
31 nd_detach_ndns(&nd_pfn->dev, &nd_pfn->ndns);
32 ida_simple_remove(&nd_region->pfn_ida, nd_pfn->id);
37 static struct device_type nd_pfn_device_type = {
39 .release = nd_pfn_release,
42 bool is_nd_pfn(struct device *dev)
44 return dev ? dev->type == &nd_pfn_device_type : false;
46 EXPORT_SYMBOL(is_nd_pfn);
48 struct nd_pfn *to_nd_pfn(struct device *dev)
50 struct nd_pfn *nd_pfn = container_of(dev, struct nd_pfn, dev);
52 WARN_ON(!is_nd_pfn(dev));
55 EXPORT_SYMBOL(to_nd_pfn);
57 static ssize_t mode_show(struct device *dev,
58 struct device_attribute *attr, char *buf)
60 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
62 switch (nd_pfn->mode) {
64 return sprintf(buf, "ram\n");
66 return sprintf(buf, "pmem\n");
68 return sprintf(buf, "none\n");
72 static ssize_t mode_store(struct device *dev,
73 struct device_attribute *attr, const char *buf, size_t len)
75 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
85 if (strncmp(buf, "pmem\n", n) == 0
86 || strncmp(buf, "pmem", n) == 0) {
87 nd_pfn->mode = PFN_MODE_PMEM;
88 } else if (strncmp(buf, "ram\n", n) == 0
89 || strncmp(buf, "ram", n) == 0)
90 nd_pfn->mode = PFN_MODE_RAM;
91 else if (strncmp(buf, "none\n", n) == 0
92 || strncmp(buf, "none", n) == 0)
93 nd_pfn->mode = PFN_MODE_NONE;
97 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
98 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
99 nvdimm_bus_unlock(dev);
102 return rc ? rc : len;
104 static DEVICE_ATTR_RW(mode);
106 static ssize_t align_show(struct device *dev,
107 struct device_attribute *attr, char *buf)
109 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
111 return sprintf(buf, "%ld\n", nd_pfn->align);
114 static const unsigned long *nd_pfn_supported_alignments(void)
117 * This needs to be a non-static variable because the *_SIZE
118 * macros aren't always constants.
120 const unsigned long supported_alignments[] = {
122 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
124 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
130 static unsigned long data[ARRAY_SIZE(supported_alignments)];
132 memcpy(data, supported_alignments, sizeof(data));
137 static ssize_t __align_store(struct nd_pfn *nd_pfn, const char *buf)
142 rc = kstrtoul(buf, 0, &val);
146 if (!is_power_of_2(val) || val < PAGE_SIZE || val > SZ_1G)
149 if (nd_pfn->dev.driver)
157 static ssize_t align_store(struct device *dev,
158 struct device_attribute *attr, const char *buf, size_t len)
160 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
164 nvdimm_bus_lock(dev);
165 rc = __align_store(nd_pfn, buf);
166 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
167 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
168 nvdimm_bus_unlock(dev);
171 return rc ? rc : len;
173 static DEVICE_ATTR_RW(align);
175 static ssize_t uuid_show(struct device *dev,
176 struct device_attribute *attr, char *buf)
178 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
181 return sprintf(buf, "%pUb\n", nd_pfn->uuid);
182 return sprintf(buf, "\n");
185 static ssize_t uuid_store(struct device *dev,
186 struct device_attribute *attr, const char *buf, size_t len)
188 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
192 rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
193 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
194 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
197 return rc ? rc : len;
199 static DEVICE_ATTR_RW(uuid);
201 static ssize_t namespace_show(struct device *dev,
202 struct device_attribute *attr, char *buf)
204 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
207 nvdimm_bus_lock(dev);
208 rc = sprintf(buf, "%s\n", nd_pfn->ndns
209 ? dev_name(&nd_pfn->ndns->dev) : "");
210 nvdimm_bus_unlock(dev);
214 static ssize_t namespace_store(struct device *dev,
215 struct device_attribute *attr, const char *buf, size_t len)
217 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
221 nvdimm_bus_lock(dev);
222 rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
223 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
224 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
225 nvdimm_bus_unlock(dev);
230 static DEVICE_ATTR_RW(namespace);
232 static ssize_t resource_show(struct device *dev,
233 struct device_attribute *attr, char *buf)
235 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
240 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
241 u64 offset = __le64_to_cpu(pfn_sb->dataoff);
242 struct nd_namespace_common *ndns = nd_pfn->ndns;
243 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
244 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
246 rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
247 + start_pad + offset);
249 /* no address to convey if the pfn instance is disabled */
256 static DEVICE_ATTR_RO(resource);
258 static ssize_t size_show(struct device *dev,
259 struct device_attribute *attr, char *buf)
261 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
266 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
267 u64 offset = __le64_to_cpu(pfn_sb->dataoff);
268 struct nd_namespace_common *ndns = nd_pfn->ndns;
269 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
270 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
271 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
273 rc = sprintf(buf, "%llu\n", (unsigned long long)
274 resource_size(&nsio->res) - start_pad
275 - end_trunc - offset);
277 /* no size to convey if the pfn instance is disabled */
284 static DEVICE_ATTR_RO(size);
286 static ssize_t supported_alignments_show(struct device *dev,
287 struct device_attribute *attr, char *buf)
289 return nd_size_select_show(0, nd_pfn_supported_alignments(), buf);
291 static DEVICE_ATTR_RO(supported_alignments);
293 static struct attribute *nd_pfn_attributes[] = {
295 &dev_attr_namespace.attr,
297 &dev_attr_align.attr,
298 &dev_attr_resource.attr,
300 &dev_attr_supported_alignments.attr,
304 struct attribute_group nd_pfn_attribute_group = {
305 .attrs = nd_pfn_attributes,
308 static const struct attribute_group *nd_pfn_attribute_groups[] = {
309 &nd_pfn_attribute_group,
310 &nd_device_attribute_group,
311 &nd_numa_attribute_group,
315 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
316 struct nd_namespace_common *ndns)
318 struct device *dev = &nd_pfn->dev;
323 nd_pfn->mode = PFN_MODE_NONE;
324 nd_pfn->align = PFN_DEFAULT_ALIGNMENT;
326 device_initialize(&nd_pfn->dev);
327 if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
328 dev_dbg(&ndns->dev, "%s failed, already claimed by %s\n",
329 __func__, dev_name(ndns->claim));
336 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
338 struct nd_pfn *nd_pfn;
341 nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
345 nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
346 if (nd_pfn->id < 0) {
352 dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
353 dev->groups = nd_pfn_attribute_groups;
354 dev->type = &nd_pfn_device_type;
355 dev->parent = &nd_region->dev;
360 struct device *nd_pfn_create(struct nd_region *nd_region)
362 struct nd_pfn *nd_pfn;
365 if (!is_memory(&nd_region->dev))
368 nd_pfn = nd_pfn_alloc(nd_region);
369 dev = nd_pfn_devinit(nd_pfn, NULL);
371 __nd_device_register(dev);
375 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
377 u64 checksum, offset;
379 enum nd_pfn_mode mode;
380 struct nd_namespace_io *nsio;
381 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
382 struct nd_namespace_common *ndns = nd_pfn->ndns;
383 const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
385 if (!pfn_sb || !ndns)
388 if (!is_memory(nd_pfn->dev.parent))
391 if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
394 if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
397 checksum = le64_to_cpu(pfn_sb->checksum);
398 pfn_sb->checksum = 0;
399 if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
401 pfn_sb->checksum = cpu_to_le64(checksum);
403 if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
406 if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
407 pfn_sb->start_pad = 0;
408 pfn_sb->end_trunc = 0;
411 if (__le16_to_cpu(pfn_sb->version_minor) < 2)
414 switch (le32_to_cpu(pfn_sb->mode)) {
422 align = le32_to_cpu(pfn_sb->align);
423 offset = le64_to_cpu(pfn_sb->dataoff);
425 align = 1UL << ilog2(offset);
426 mode = le32_to_cpu(pfn_sb->mode);
430 * When probing a namepace via nd_pfn_probe() the uuid
431 * is NULL (see: nd_pfn_devinit()) we init settings from
434 nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
437 nd_pfn->align = align;
441 * When probing a pfn / dax instance we validate the
442 * live settings against the pfn_sb
444 if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
448 * If the uuid validates, but other settings mismatch
449 * return EINVAL because userspace has managed to change
450 * the configuration without specifying new
453 if (nd_pfn->align != align || nd_pfn->mode != mode) {
454 dev_err(&nd_pfn->dev,
455 "init failed, settings mismatch\n");
456 dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
457 nd_pfn->align, align, nd_pfn->mode,
463 if (align > nvdimm_namespace_capacity(ndns)) {
464 dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
465 align, nvdimm_namespace_capacity(ndns));
470 * These warnings are verbose because they can only trigger in
471 * the case where the physical address alignment of the
472 * namespace has changed since the pfn superblock was
475 nsio = to_nd_namespace_io(&ndns->dev);
476 if (offset >= resource_size(&nsio->res)) {
477 dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
478 dev_name(&ndns->dev));
482 if ((align && !IS_ALIGNED(offset, align))
483 || !IS_ALIGNED(offset, PAGE_SIZE)) {
484 dev_err(&nd_pfn->dev,
485 "bad offset: %#llx dax disabled align: %#lx\n",
492 EXPORT_SYMBOL(nd_pfn_validate);
494 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
497 struct nd_pfn *nd_pfn;
498 struct device *pfn_dev;
499 struct nd_pfn_sb *pfn_sb;
500 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
505 switch (ndns->claim_class) {
506 case NVDIMM_CCLASS_NONE:
507 case NVDIMM_CCLASS_PFN:
513 nvdimm_bus_lock(&ndns->dev);
514 nd_pfn = nd_pfn_alloc(nd_region);
515 pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
516 nvdimm_bus_unlock(&ndns->dev);
519 pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
520 nd_pfn = to_nd_pfn(pfn_dev);
521 nd_pfn->pfn_sb = pfn_sb;
522 rc = nd_pfn_validate(nd_pfn, PFN_SIG);
523 dev_dbg(dev, "%s: pfn: %s\n", __func__,
524 rc == 0 ? dev_name(pfn_dev) : "<none>");
526 nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
529 __nd_device_register(pfn_dev);
533 EXPORT_SYMBOL(nd_pfn_probe);
536 * We hotplug memory at section granularity, pad the reserved area from
537 * the previous section base to the namespace base address.
539 static unsigned long init_altmap_base(resource_size_t base)
541 unsigned long base_pfn = PHYS_PFN(base);
543 return PFN_SECTION_ALIGN_DOWN(base_pfn);
546 static unsigned long init_altmap_reserve(resource_size_t base)
548 unsigned long reserve = PHYS_PFN(SZ_8K);
549 unsigned long base_pfn = PHYS_PFN(base);
551 reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
555 static struct vmem_altmap *__nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
556 struct resource *res, struct vmem_altmap *altmap)
558 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
559 u64 offset = le64_to_cpu(pfn_sb->dataoff);
560 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
561 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
562 struct nd_namespace_common *ndns = nd_pfn->ndns;
563 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
564 resource_size_t base = nsio->res.start + start_pad;
565 struct vmem_altmap __altmap = {
566 .base_pfn = init_altmap_base(base),
567 .reserve = init_altmap_reserve(base),
570 memcpy(res, &nsio->res, sizeof(*res));
571 res->start += start_pad;
572 res->end -= end_trunc;
574 if (nd_pfn->mode == PFN_MODE_RAM) {
576 return ERR_PTR(-EINVAL);
577 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
579 } else if (nd_pfn->mode == PFN_MODE_PMEM) {
580 nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
581 - offset) / PAGE_SIZE);
582 if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
583 dev_info(&nd_pfn->dev,
584 "number of pfns truncated from %lld to %ld\n",
585 le64_to_cpu(nd_pfn->pfn_sb->npfns),
587 memcpy(altmap, &__altmap, sizeof(*altmap));
588 altmap->free = PHYS_PFN(offset - SZ_8K);
591 return ERR_PTR(-ENXIO);
596 static int nd_pfn_init(struct nd_pfn *nd_pfn)
598 u32 dax_label_reserve = is_nd_dax(&nd_pfn->dev) ? SZ_128K : 0;
599 struct nd_namespace_common *ndns = nd_pfn->ndns;
600 u32 start_pad = 0, end_trunc = 0;
601 resource_size_t start, size;
602 struct nd_namespace_io *nsio;
603 struct nd_region *nd_region;
604 struct nd_pfn_sb *pfn_sb;
611 pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
615 nd_pfn->pfn_sb = pfn_sb;
616 if (is_nd_dax(&nd_pfn->dev))
620 rc = nd_pfn_validate(nd_pfn, sig);
624 /* no info block, do init */;
625 nd_region = to_nd_region(nd_pfn->dev.parent);
627 dev_info(&nd_pfn->dev,
628 "%s is read-only, unable to init metadata\n",
629 dev_name(&nd_region->dev));
633 memset(pfn_sb, 0, sizeof(*pfn_sb));
636 * Check if pmem collides with 'System RAM' when section aligned and
637 * trim it accordingly
639 nsio = to_nd_namespace_io(&ndns->dev);
640 start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
641 size = resource_size(&nsio->res);
642 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
643 IORES_DESC_NONE) == REGION_MIXED) {
644 start = nsio->res.start;
645 start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
648 start = nsio->res.start;
649 size = PHYS_SECTION_ALIGN_UP(start + size) - start;
650 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
651 IORES_DESC_NONE) == REGION_MIXED) {
652 size = resource_size(&nsio->res);
653 end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size);
656 if (start_pad + end_trunc)
657 dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n",
658 dev_name(&ndns->dev), start_pad + end_trunc);
661 * Note, we use 64 here for the standard size of struct page,
662 * debugging options may cause it to be larger in which case the
663 * implementation will limit the pfns advertised through
664 * ->direct_access() to those that are included in the memmap.
667 size = resource_size(&nsio->res);
668 npfns = PFN_SECTION_ALIGN_UP((size - start_pad - end_trunc - SZ_8K)
670 if (nd_pfn->mode == PFN_MODE_PMEM) {
672 * The altmap should be padded out to the block size used
673 * when populating the vmemmap. This *should* be equal to
674 * PMD_SIZE for most architectures.
676 offset = ALIGN(start + SZ_8K + 64 * npfns + dax_label_reserve,
677 max(nd_pfn->align, PMD_SIZE)) - start;
678 } else if (nd_pfn->mode == PFN_MODE_RAM)
679 offset = ALIGN(start + SZ_8K + dax_label_reserve,
680 nd_pfn->align) - start;
684 if (offset + start_pad + end_trunc >= size) {
685 dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
686 dev_name(&ndns->dev));
690 npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
691 pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
692 pfn_sb->dataoff = cpu_to_le64(offset);
693 pfn_sb->npfns = cpu_to_le64(npfns);
694 memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
695 memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
696 memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
697 pfn_sb->version_major = cpu_to_le16(1);
698 pfn_sb->version_minor = cpu_to_le16(2);
699 pfn_sb->start_pad = cpu_to_le32(start_pad);
700 pfn_sb->end_trunc = cpu_to_le32(end_trunc);
701 pfn_sb->align = cpu_to_le32(nd_pfn->align);
702 checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
703 pfn_sb->checksum = cpu_to_le64(checksum);
705 return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
709 * Determine the effective resource range and vmem_altmap from an nd_pfn
712 struct vmem_altmap *nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
713 struct resource *res, struct vmem_altmap *altmap)
717 if (!nd_pfn->uuid || !nd_pfn->ndns)
718 return ERR_PTR(-ENODEV);
720 rc = nd_pfn_init(nd_pfn);
724 /* we need a valid pfn_sb before we can init a vmem_altmap */
725 return __nvdimm_setup_pfn(nd_pfn, res, altmap);
727 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);