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, "trace\n");
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, "result: %zd wrote: %s%s", rc, buf,
98 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 device *dev,
138 struct device_attribute *attr, const char *buf, size_t len)
140 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
144 nvdimm_bus_lock(dev);
145 rc = nd_size_select_store(dev, buf, &nd_pfn->align,
146 nd_pfn_supported_alignments());
147 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
148 buf[len - 1] == '\n' ? "" : "\n");
149 nvdimm_bus_unlock(dev);
152 return rc ? rc : len;
154 static DEVICE_ATTR_RW(align);
156 static ssize_t uuid_show(struct device *dev,
157 struct device_attribute *attr, char *buf)
159 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
162 return sprintf(buf, "%pUb\n", nd_pfn->uuid);
163 return sprintf(buf, "\n");
166 static ssize_t uuid_store(struct device *dev,
167 struct device_attribute *attr, const char *buf, size_t len)
169 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
173 rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
174 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
175 buf[len - 1] == '\n' ? "" : "\n");
178 return rc ? rc : len;
180 static DEVICE_ATTR_RW(uuid);
182 static ssize_t namespace_show(struct device *dev,
183 struct device_attribute *attr, char *buf)
185 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
188 nvdimm_bus_lock(dev);
189 rc = sprintf(buf, "%s\n", nd_pfn->ndns
190 ? dev_name(&nd_pfn->ndns->dev) : "");
191 nvdimm_bus_unlock(dev);
195 static ssize_t namespace_store(struct device *dev,
196 struct device_attribute *attr, const char *buf, size_t len)
198 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
202 nvdimm_bus_lock(dev);
203 rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
204 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
205 buf[len - 1] == '\n' ? "" : "\n");
206 nvdimm_bus_unlock(dev);
211 static DEVICE_ATTR_RW(namespace);
213 static ssize_t resource_show(struct device *dev,
214 struct device_attribute *attr, char *buf)
216 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
221 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
222 u64 offset = __le64_to_cpu(pfn_sb->dataoff);
223 struct nd_namespace_common *ndns = nd_pfn->ndns;
224 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
225 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
227 rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
228 + start_pad + offset);
230 /* no address to convey if the pfn instance is disabled */
237 static DEVICE_ATTR_RO(resource);
239 static ssize_t size_show(struct device *dev,
240 struct device_attribute *attr, char *buf)
242 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
247 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
248 u64 offset = __le64_to_cpu(pfn_sb->dataoff);
249 struct nd_namespace_common *ndns = nd_pfn->ndns;
250 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
251 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
252 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
254 rc = sprintf(buf, "%llu\n", (unsigned long long)
255 resource_size(&nsio->res) - start_pad
256 - end_trunc - offset);
258 /* no size to convey if the pfn instance is disabled */
265 static DEVICE_ATTR_RO(size);
267 static ssize_t supported_alignments_show(struct device *dev,
268 struct device_attribute *attr, char *buf)
270 return nd_size_select_show(0, nd_pfn_supported_alignments(), buf);
272 static DEVICE_ATTR_RO(supported_alignments);
274 static struct attribute *nd_pfn_attributes[] = {
276 &dev_attr_namespace.attr,
278 &dev_attr_align.attr,
279 &dev_attr_resource.attr,
281 &dev_attr_supported_alignments.attr,
285 static umode_t pfn_visible(struct kobject *kobj, struct attribute *a, int n)
287 if (a == &dev_attr_resource.attr)
292 struct attribute_group nd_pfn_attribute_group = {
293 .attrs = nd_pfn_attributes,
294 .is_visible = pfn_visible,
297 static const struct attribute_group *nd_pfn_attribute_groups[] = {
298 &nd_pfn_attribute_group,
299 &nd_device_attribute_group,
300 &nd_numa_attribute_group,
304 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
305 struct nd_namespace_common *ndns)
312 nd_pfn->mode = PFN_MODE_NONE;
313 nd_pfn->align = PFN_DEFAULT_ALIGNMENT;
315 device_initialize(&nd_pfn->dev);
316 if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
317 dev_dbg(&ndns->dev, "failed, already claimed by %s\n",
318 dev_name(ndns->claim));
325 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
327 struct nd_pfn *nd_pfn;
330 nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
334 nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
335 if (nd_pfn->id < 0) {
341 dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
342 dev->groups = nd_pfn_attribute_groups;
343 dev->type = &nd_pfn_device_type;
344 dev->parent = &nd_region->dev;
349 struct device *nd_pfn_create(struct nd_region *nd_region)
351 struct nd_pfn *nd_pfn;
354 if (!is_memory(&nd_region->dev))
357 nd_pfn = nd_pfn_alloc(nd_region);
358 dev = nd_pfn_devinit(nd_pfn, NULL);
360 __nd_device_register(dev);
364 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
366 u64 checksum, offset;
367 enum nd_pfn_mode mode;
368 struct nd_namespace_io *nsio;
369 unsigned long align, start_pad;
370 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
371 struct nd_namespace_common *ndns = nd_pfn->ndns;
372 const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
374 if (!pfn_sb || !ndns)
377 if (!is_memory(nd_pfn->dev.parent))
380 if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
383 if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
386 checksum = le64_to_cpu(pfn_sb->checksum);
387 pfn_sb->checksum = 0;
388 if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
390 pfn_sb->checksum = cpu_to_le64(checksum);
392 if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
395 if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
396 pfn_sb->start_pad = 0;
397 pfn_sb->end_trunc = 0;
400 if (__le16_to_cpu(pfn_sb->version_minor) < 2)
403 switch (le32_to_cpu(pfn_sb->mode)) {
411 align = le32_to_cpu(pfn_sb->align);
412 offset = le64_to_cpu(pfn_sb->dataoff);
413 start_pad = le32_to_cpu(pfn_sb->start_pad);
415 align = 1UL << ilog2(offset);
416 mode = le32_to_cpu(pfn_sb->mode);
420 * When probing a namepace via nd_pfn_probe() the uuid
421 * is NULL (see: nd_pfn_devinit()) we init settings from
424 nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
427 nd_pfn->align = align;
431 * When probing a pfn / dax instance we validate the
432 * live settings against the pfn_sb
434 if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
438 * If the uuid validates, but other settings mismatch
439 * return EINVAL because userspace has managed to change
440 * the configuration without specifying new
443 if (nd_pfn->align != align || nd_pfn->mode != mode) {
444 dev_err(&nd_pfn->dev,
445 "init failed, settings mismatch\n");
446 dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
447 nd_pfn->align, align, nd_pfn->mode,
453 if (align > nvdimm_namespace_capacity(ndns)) {
454 dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
455 align, nvdimm_namespace_capacity(ndns));
460 * These warnings are verbose because they can only trigger in
461 * the case where the physical address alignment of the
462 * namespace has changed since the pfn superblock was
465 nsio = to_nd_namespace_io(&ndns->dev);
466 if (offset >= resource_size(&nsio->res)) {
467 dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
468 dev_name(&ndns->dev));
472 if ((align && !IS_ALIGNED(nsio->res.start + offset + start_pad, align))
473 || !IS_ALIGNED(offset, PAGE_SIZE)) {
474 dev_err(&nd_pfn->dev,
475 "bad offset: %#llx dax disabled align: %#lx\n",
482 EXPORT_SYMBOL(nd_pfn_validate);
484 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
487 struct nd_pfn *nd_pfn;
488 struct device *pfn_dev;
489 struct nd_pfn_sb *pfn_sb;
490 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
495 switch (ndns->claim_class) {
496 case NVDIMM_CCLASS_NONE:
497 case NVDIMM_CCLASS_PFN:
503 nvdimm_bus_lock(&ndns->dev);
504 nd_pfn = nd_pfn_alloc(nd_region);
505 pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
506 nvdimm_bus_unlock(&ndns->dev);
509 pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
510 nd_pfn = to_nd_pfn(pfn_dev);
511 nd_pfn->pfn_sb = pfn_sb;
512 rc = nd_pfn_validate(nd_pfn, PFN_SIG);
513 dev_dbg(dev, "pfn: %s\n", rc == 0 ? dev_name(pfn_dev) : "<none>");
515 nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
518 __nd_device_register(pfn_dev);
522 EXPORT_SYMBOL(nd_pfn_probe);
525 * We hotplug memory at section granularity, pad the reserved area from
526 * the previous section base to the namespace base address.
528 static unsigned long init_altmap_base(resource_size_t base)
530 unsigned long base_pfn = PHYS_PFN(base);
532 return PFN_SECTION_ALIGN_DOWN(base_pfn);
535 static unsigned long init_altmap_reserve(resource_size_t base)
537 unsigned long reserve = PHYS_PFN(SZ_8K);
538 unsigned long base_pfn = PHYS_PFN(base);
540 reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
544 static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
546 struct resource *res = &pgmap->res;
547 struct vmem_altmap *altmap = &pgmap->altmap;
548 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
549 u64 offset = le64_to_cpu(pfn_sb->dataoff);
550 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
551 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
552 struct nd_namespace_common *ndns = nd_pfn->ndns;
553 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
554 resource_size_t base = nsio->res.start + start_pad;
555 struct vmem_altmap __altmap = {
556 .base_pfn = init_altmap_base(base),
557 .reserve = init_altmap_reserve(base),
560 memcpy(res, &nsio->res, sizeof(*res));
561 res->start += start_pad;
562 res->end -= end_trunc;
564 if (nd_pfn->mode == PFN_MODE_RAM) {
567 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
568 pgmap->altmap_valid = false;
569 } else if (nd_pfn->mode == PFN_MODE_PMEM) {
570 nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
571 - offset) / PAGE_SIZE);
572 if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
573 dev_info(&nd_pfn->dev,
574 "number of pfns truncated from %lld to %ld\n",
575 le64_to_cpu(nd_pfn->pfn_sb->npfns),
577 memcpy(altmap, &__altmap, sizeof(*altmap));
578 altmap->free = PHYS_PFN(offset - SZ_8K);
580 pgmap->altmap_valid = true;
587 static u64 phys_pmem_align_down(struct nd_pfn *nd_pfn, u64 phys)
589 return min_t(u64, PHYS_SECTION_ALIGN_DOWN(phys),
590 ALIGN_DOWN(phys, nd_pfn->align));
593 static int nd_pfn_init(struct nd_pfn *nd_pfn)
595 u32 dax_label_reserve = is_nd_dax(&nd_pfn->dev) ? SZ_128K : 0;
596 struct nd_namespace_common *ndns = nd_pfn->ndns;
597 u32 start_pad = 0, end_trunc = 0;
598 resource_size_t start, size;
599 struct nd_namespace_io *nsio;
600 struct nd_region *nd_region;
601 struct nd_pfn_sb *pfn_sb;
608 pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
612 nd_pfn->pfn_sb = pfn_sb;
613 if (is_nd_dax(&nd_pfn->dev))
617 rc = nd_pfn_validate(nd_pfn, sig);
621 /* no info block, do init */;
622 nd_region = to_nd_region(nd_pfn->dev.parent);
624 dev_info(&nd_pfn->dev,
625 "%s is read-only, unable to init metadata\n",
626 dev_name(&nd_region->dev));
630 memset(pfn_sb, 0, sizeof(*pfn_sb));
633 * Check if pmem collides with 'System RAM' when section aligned and
634 * trim it accordingly
636 nsio = to_nd_namespace_io(&ndns->dev);
637 start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
638 size = resource_size(&nsio->res);
639 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
640 IORES_DESC_NONE) == REGION_MIXED) {
641 start = nsio->res.start;
642 start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
645 start = nsio->res.start;
646 size = PHYS_SECTION_ALIGN_UP(start + size) - start;
647 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
648 IORES_DESC_NONE) == REGION_MIXED
649 || !IS_ALIGNED(start + resource_size(&nsio->res),
651 size = resource_size(&nsio->res);
652 end_trunc = start + size - phys_pmem_align_down(nd_pfn,
656 if (start_pad + end_trunc)
657 dev_info(&nd_pfn->dev, "%s alignment 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 int nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
716 if (!nd_pfn->uuid || !nd_pfn->ndns)
719 rc = nd_pfn_init(nd_pfn);
723 /* we need a valid pfn_sb before we can init a dev_pagemap */
724 return __nvdimm_setup_pfn(nd_pfn, pgmap);
726 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);