xref: /linux/drivers/nvdimm/pfn_devs.c (revision d257f9bf06129613de539ea71ecea60848b662cd)
1 /*
2  * Copyright(c) 2013-2016 Intel Corporation. All rights reserved.
3  *
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.
7  *
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.
12  */
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>
19 #include <linux/fs.h>
20 #include <linux/mm.h>
21 #include "nd-core.h"
22 #include "pfn.h"
23 #include "nd.h"
24 
25 static void nd_pfn_release(struct device *dev)
26 {
27 	struct nd_region *nd_region = to_nd_region(dev->parent);
28 	struct nd_pfn *nd_pfn = to_nd_pfn(dev);
29 
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);
33 	kfree(nd_pfn->uuid);
34 	kfree(nd_pfn);
35 }
36 
37 static struct device_type nd_pfn_device_type = {
38 	.name = "nd_pfn",
39 	.release = nd_pfn_release,
40 };
41 
42 bool is_nd_pfn(struct device *dev)
43 {
44 	return dev ? dev->type == &nd_pfn_device_type : false;
45 }
46 EXPORT_SYMBOL(is_nd_pfn);
47 
48 struct nd_pfn *to_nd_pfn(struct device *dev)
49 {
50 	struct nd_pfn *nd_pfn = container_of(dev, struct nd_pfn, dev);
51 
52 	WARN_ON(!is_nd_pfn(dev));
53 	return nd_pfn;
54 }
55 EXPORT_SYMBOL(to_nd_pfn);
56 
57 static ssize_t mode_show(struct device *dev,
58 		struct device_attribute *attr, char *buf)
59 {
60 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
61 
62 	switch (nd_pfn->mode) {
63 	case PFN_MODE_RAM:
64 		return sprintf(buf, "ram\n");
65 	case PFN_MODE_PMEM:
66 		return sprintf(buf, "pmem\n");
67 	default:
68 		return sprintf(buf, "none\n");
69 	}
70 }
71 
72 static ssize_t mode_store(struct device *dev,
73 		struct device_attribute *attr, const char *buf, size_t len)
74 {
75 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
76 	ssize_t rc = 0;
77 
78 	device_lock(dev);
79 	nvdimm_bus_lock(dev);
80 	if (dev->driver)
81 		rc = -EBUSY;
82 	else {
83 		size_t n = len - 1;
84 
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;
94 		else
95 			rc = -EINVAL;
96 	}
97 	dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
98 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
99 	nvdimm_bus_unlock(dev);
100 	device_unlock(dev);
101 
102 	return rc ? rc : len;
103 }
104 static DEVICE_ATTR_RW(mode);
105 
106 static ssize_t align_show(struct device *dev,
107 		struct device_attribute *attr, char *buf)
108 {
109 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
110 
111 	return sprintf(buf, "%ld\n", nd_pfn->align);
112 }
113 
114 static ssize_t __align_store(struct nd_pfn *nd_pfn, const char *buf)
115 {
116 	unsigned long val;
117 	int rc;
118 
119 	rc = kstrtoul(buf, 0, &val);
120 	if (rc)
121 		return rc;
122 
123 	if (!is_power_of_2(val) || val < PAGE_SIZE || val > SZ_1G)
124 		return -EINVAL;
125 
126 	if (nd_pfn->dev.driver)
127 		return -EBUSY;
128 	else
129 		nd_pfn->align = val;
130 
131 	return 0;
132 }
133 
134 static ssize_t align_store(struct device *dev,
135 		struct device_attribute *attr, const char *buf, size_t len)
136 {
137 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
138 	ssize_t rc;
139 
140 	device_lock(dev);
141 	nvdimm_bus_lock(dev);
142 	rc = __align_store(nd_pfn, buf);
143 	dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
144 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
145 	nvdimm_bus_unlock(dev);
146 	device_unlock(dev);
147 
148 	return rc ? rc : len;
149 }
150 static DEVICE_ATTR_RW(align);
151 
152 static ssize_t uuid_show(struct device *dev,
153 		struct device_attribute *attr, char *buf)
154 {
155 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
156 
157 	if (nd_pfn->uuid)
158 		return sprintf(buf, "%pUb\n", nd_pfn->uuid);
159 	return sprintf(buf, "\n");
160 }
161 
162 static ssize_t uuid_store(struct device *dev,
163 		struct device_attribute *attr, const char *buf, size_t len)
164 {
165 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
166 	ssize_t rc;
167 
168 	device_lock(dev);
169 	rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
170 	dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
171 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
172 	device_unlock(dev);
173 
174 	return rc ? rc : len;
175 }
176 static DEVICE_ATTR_RW(uuid);
177 
178 static ssize_t namespace_show(struct device *dev,
179 		struct device_attribute *attr, char *buf)
180 {
181 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
182 	ssize_t rc;
183 
184 	nvdimm_bus_lock(dev);
185 	rc = sprintf(buf, "%s\n", nd_pfn->ndns
186 			? dev_name(&nd_pfn->ndns->dev) : "");
187 	nvdimm_bus_unlock(dev);
188 	return rc;
189 }
190 
191 static ssize_t namespace_store(struct device *dev,
192 		struct device_attribute *attr, const char *buf, size_t len)
193 {
194 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
195 	ssize_t rc;
196 
197 	device_lock(dev);
198 	nvdimm_bus_lock(dev);
199 	rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
200 	dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
201 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
202 	nvdimm_bus_unlock(dev);
203 	device_unlock(dev);
204 
205 	return rc;
206 }
207 static DEVICE_ATTR_RW(namespace);
208 
209 static ssize_t resource_show(struct device *dev,
210 		struct device_attribute *attr, char *buf)
211 {
212 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
213 	ssize_t rc;
214 
215 	device_lock(dev);
216 	if (dev->driver) {
217 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
218 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
219 		struct nd_namespace_common *ndns = nd_pfn->ndns;
220 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
221 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
222 
223 		rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
224 				+ start_pad + offset);
225 	} else {
226 		/* no address to convey if the pfn instance is disabled */
227 		rc = -ENXIO;
228 	}
229 	device_unlock(dev);
230 
231 	return rc;
232 }
233 static DEVICE_ATTR_RO(resource);
234 
235 static ssize_t size_show(struct device *dev,
236 		struct device_attribute *attr, char *buf)
237 {
238 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
239 	ssize_t rc;
240 
241 	device_lock(dev);
242 	if (dev->driver) {
243 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
244 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
245 		struct nd_namespace_common *ndns = nd_pfn->ndns;
246 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
247 		u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
248 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
249 
250 		rc = sprintf(buf, "%llu\n", (unsigned long long)
251 				resource_size(&nsio->res) - start_pad
252 				- end_trunc - offset);
253 	} else {
254 		/* no size to convey if the pfn instance is disabled */
255 		rc = -ENXIO;
256 	}
257 	device_unlock(dev);
258 
259 	return rc;
260 }
261 static DEVICE_ATTR_RO(size);
262 
263 static struct attribute *nd_pfn_attributes[] = {
264 	&dev_attr_mode.attr,
265 	&dev_attr_namespace.attr,
266 	&dev_attr_uuid.attr,
267 	&dev_attr_align.attr,
268 	&dev_attr_resource.attr,
269 	&dev_attr_size.attr,
270 	NULL,
271 };
272 
273 struct attribute_group nd_pfn_attribute_group = {
274 	.attrs = nd_pfn_attributes,
275 };
276 
277 static const struct attribute_group *nd_pfn_attribute_groups[] = {
278 	&nd_pfn_attribute_group,
279 	&nd_device_attribute_group,
280 	&nd_numa_attribute_group,
281 	NULL,
282 };
283 
284 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
285 		struct nd_namespace_common *ndns)
286 {
287 	struct device *dev = &nd_pfn->dev;
288 
289 	if (!nd_pfn)
290 		return NULL;
291 
292 	nd_pfn->mode = PFN_MODE_NONE;
293 	nd_pfn->align = HPAGE_SIZE;
294 	dev = &nd_pfn->dev;
295 	device_initialize(&nd_pfn->dev);
296 	if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
297 		dev_dbg(&ndns->dev, "%s failed, already claimed by %s\n",
298 				__func__, dev_name(ndns->claim));
299 		put_device(dev);
300 		return NULL;
301 	}
302 	return dev;
303 }
304 
305 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
306 {
307 	struct nd_pfn *nd_pfn;
308 	struct device *dev;
309 
310 	nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
311 	if (!nd_pfn)
312 		return NULL;
313 
314 	nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
315 	if (nd_pfn->id < 0) {
316 		kfree(nd_pfn);
317 		return NULL;
318 	}
319 
320 	dev = &nd_pfn->dev;
321 	dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
322 	dev->groups = nd_pfn_attribute_groups;
323 	dev->type = &nd_pfn_device_type;
324 	dev->parent = &nd_region->dev;
325 
326 	return nd_pfn;
327 }
328 
329 struct device *nd_pfn_create(struct nd_region *nd_region)
330 {
331 	struct nd_pfn *nd_pfn;
332 	struct device *dev;
333 
334 	if (!is_memory(&nd_region->dev))
335 		return NULL;
336 
337 	nd_pfn = nd_pfn_alloc(nd_region);
338 	dev = nd_pfn_devinit(nd_pfn, NULL);
339 
340 	__nd_device_register(dev);
341 	return dev;
342 }
343 
344 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
345 {
346 	u64 checksum, offset;
347 	unsigned long align;
348 	enum nd_pfn_mode mode;
349 	struct nd_namespace_io *nsio;
350 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
351 	struct nd_namespace_common *ndns = nd_pfn->ndns;
352 	const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
353 
354 	if (!pfn_sb || !ndns)
355 		return -ENODEV;
356 
357 	if (!is_memory(nd_pfn->dev.parent))
358 		return -ENODEV;
359 
360 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
361 		return -ENXIO;
362 
363 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
364 		return -ENODEV;
365 
366 	checksum = le64_to_cpu(pfn_sb->checksum);
367 	pfn_sb->checksum = 0;
368 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
369 		return -ENODEV;
370 	pfn_sb->checksum = cpu_to_le64(checksum);
371 
372 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
373 		return -ENODEV;
374 
375 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
376 		pfn_sb->start_pad = 0;
377 		pfn_sb->end_trunc = 0;
378 	}
379 
380 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
381 		pfn_sb->align = 0;
382 
383 	switch (le32_to_cpu(pfn_sb->mode)) {
384 	case PFN_MODE_RAM:
385 	case PFN_MODE_PMEM:
386 		break;
387 	default:
388 		return -ENXIO;
389 	}
390 
391 	align = le32_to_cpu(pfn_sb->align);
392 	offset = le64_to_cpu(pfn_sb->dataoff);
393 	if (align == 0)
394 		align = 1UL << ilog2(offset);
395 	mode = le32_to_cpu(pfn_sb->mode);
396 
397 	if (!nd_pfn->uuid) {
398 		/*
399 		 * When probing a namepace via nd_pfn_probe() the uuid
400 		 * is NULL (see: nd_pfn_devinit()) we init settings from
401 		 * pfn_sb
402 		 */
403 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
404 		if (!nd_pfn->uuid)
405 			return -ENOMEM;
406 		nd_pfn->align = align;
407 		nd_pfn->mode = mode;
408 	} else {
409 		/*
410 		 * When probing a pfn / dax instance we validate the
411 		 * live settings against the pfn_sb
412 		 */
413 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
414 			return -ENODEV;
415 
416 		/*
417 		 * If the uuid validates, but other settings mismatch
418 		 * return EINVAL because userspace has managed to change
419 		 * the configuration without specifying new
420 		 * identification.
421 		 */
422 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
423 			dev_err(&nd_pfn->dev,
424 					"init failed, settings mismatch\n");
425 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
426 					nd_pfn->align, align, nd_pfn->mode,
427 					mode);
428 			return -EINVAL;
429 		}
430 	}
431 
432 	if (align > nvdimm_namespace_capacity(ndns)) {
433 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
434 				align, nvdimm_namespace_capacity(ndns));
435 		return -EINVAL;
436 	}
437 
438 	/*
439 	 * These warnings are verbose because they can only trigger in
440 	 * the case where the physical address alignment of the
441 	 * namespace has changed since the pfn superblock was
442 	 * established.
443 	 */
444 	nsio = to_nd_namespace_io(&ndns->dev);
445 	if (offset >= resource_size(&nsio->res)) {
446 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
447 				dev_name(&ndns->dev));
448 		return -EBUSY;
449 	}
450 
451 	if ((align && !IS_ALIGNED(offset, align))
452 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
453 		dev_err(&nd_pfn->dev,
454 				"bad offset: %#llx dax disabled align: %#lx\n",
455 				offset, align);
456 		return -ENXIO;
457 	}
458 
459 	return 0;
460 }
461 EXPORT_SYMBOL(nd_pfn_validate);
462 
463 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
464 {
465 	int rc;
466 	struct nd_pfn *nd_pfn;
467 	struct device *pfn_dev;
468 	struct nd_pfn_sb *pfn_sb;
469 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
470 
471 	if (ndns->force_raw)
472 		return -ENODEV;
473 
474 	switch (ndns->claim_class) {
475 	case NVDIMM_CCLASS_NONE:
476 	case NVDIMM_CCLASS_PFN:
477 		break;
478 	default:
479 		return -ENODEV;
480 	}
481 
482 	nvdimm_bus_lock(&ndns->dev);
483 	nd_pfn = nd_pfn_alloc(nd_region);
484 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
485 	nvdimm_bus_unlock(&ndns->dev);
486 	if (!pfn_dev)
487 		return -ENOMEM;
488 	pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
489 	nd_pfn = to_nd_pfn(pfn_dev);
490 	nd_pfn->pfn_sb = pfn_sb;
491 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
492 	dev_dbg(dev, "%s: pfn: %s\n", __func__,
493 			rc == 0 ? dev_name(pfn_dev) : "<none>");
494 	if (rc < 0) {
495 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
496 		put_device(pfn_dev);
497 	} else
498 		__nd_device_register(pfn_dev);
499 
500 	return rc;
501 }
502 EXPORT_SYMBOL(nd_pfn_probe);
503 
504 /*
505  * We hotplug memory at section granularity, pad the reserved area from
506  * the previous section base to the namespace base address.
507  */
508 static unsigned long init_altmap_base(resource_size_t base)
509 {
510 	unsigned long base_pfn = PHYS_PFN(base);
511 
512 	return PFN_SECTION_ALIGN_DOWN(base_pfn);
513 }
514 
515 static unsigned long init_altmap_reserve(resource_size_t base)
516 {
517 	unsigned long reserve = PHYS_PFN(SZ_8K);
518 	unsigned long base_pfn = PHYS_PFN(base);
519 
520 	reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
521 	return reserve;
522 }
523 
524 static struct vmem_altmap *__nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
525 		struct resource *res, struct vmem_altmap *altmap)
526 {
527 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
528 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
529 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
530 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
531 	struct nd_namespace_common *ndns = nd_pfn->ndns;
532 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
533 	resource_size_t base = nsio->res.start + start_pad;
534 	struct vmem_altmap __altmap = {
535 		.base_pfn = init_altmap_base(base),
536 		.reserve = init_altmap_reserve(base),
537 	};
538 
539 	memcpy(res, &nsio->res, sizeof(*res));
540 	res->start += start_pad;
541 	res->end -= end_trunc;
542 
543 	if (nd_pfn->mode == PFN_MODE_RAM) {
544 		if (offset < SZ_8K)
545 			return ERR_PTR(-EINVAL);
546 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
547 		altmap = NULL;
548 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
549 		nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
550 					- offset) / PAGE_SIZE);
551 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
552 			dev_info(&nd_pfn->dev,
553 					"number of pfns truncated from %lld to %ld\n",
554 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
555 					nd_pfn->npfns);
556 		memcpy(altmap, &__altmap, sizeof(*altmap));
557 		altmap->free = PHYS_PFN(offset - SZ_8K);
558 		altmap->alloc = 0;
559 	} else
560 		return ERR_PTR(-ENXIO);
561 
562 	return altmap;
563 }
564 
565 static int nd_pfn_init(struct nd_pfn *nd_pfn)
566 {
567 	u32 dax_label_reserve = is_nd_dax(&nd_pfn->dev) ? SZ_128K : 0;
568 	struct nd_namespace_common *ndns = nd_pfn->ndns;
569 	u32 start_pad = 0, end_trunc = 0;
570 	resource_size_t start, size;
571 	struct nd_namespace_io *nsio;
572 	struct nd_region *nd_region;
573 	struct nd_pfn_sb *pfn_sb;
574 	unsigned long npfns;
575 	phys_addr_t offset;
576 	const char *sig;
577 	u64 checksum;
578 	int rc;
579 
580 	pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
581 	if (!pfn_sb)
582 		return -ENOMEM;
583 
584 	nd_pfn->pfn_sb = pfn_sb;
585 	if (is_nd_dax(&nd_pfn->dev))
586 		sig = DAX_SIG;
587 	else
588 		sig = PFN_SIG;
589 	rc = nd_pfn_validate(nd_pfn, sig);
590 	if (rc != -ENODEV)
591 		return rc;
592 
593 	/* no info block, do init */;
594 	nd_region = to_nd_region(nd_pfn->dev.parent);
595 	if (nd_region->ro) {
596 		dev_info(&nd_pfn->dev,
597 				"%s is read-only, unable to init metadata\n",
598 				dev_name(&nd_region->dev));
599 		return -ENXIO;
600 	}
601 
602 	memset(pfn_sb, 0, sizeof(*pfn_sb));
603 
604 	/*
605 	 * Check if pmem collides with 'System RAM' when section aligned and
606 	 * trim it accordingly
607 	 */
608 	nsio = to_nd_namespace_io(&ndns->dev);
609 	start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
610 	size = resource_size(&nsio->res);
611 	if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
612 				IORES_DESC_NONE) == REGION_MIXED) {
613 		start = nsio->res.start;
614 		start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
615 	}
616 
617 	start = nsio->res.start;
618 	size = PHYS_SECTION_ALIGN_UP(start + size) - start;
619 	if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
620 				IORES_DESC_NONE) == REGION_MIXED) {
621 		size = resource_size(&nsio->res);
622 		end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size);
623 	}
624 
625 	if (start_pad + end_trunc)
626 		dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n",
627 				dev_name(&ndns->dev), start_pad + end_trunc);
628 
629 	/*
630 	 * Note, we use 64 here for the standard size of struct page,
631 	 * debugging options may cause it to be larger in which case the
632 	 * implementation will limit the pfns advertised through
633 	 * ->direct_access() to those that are included in the memmap.
634 	 */
635 	start += start_pad;
636 	size = resource_size(&nsio->res);
637 	npfns = PFN_SECTION_ALIGN_UP((size - start_pad - end_trunc - SZ_8K)
638 			/ PAGE_SIZE);
639 	if (nd_pfn->mode == PFN_MODE_PMEM) {
640 		/*
641 		 * vmemmap_populate_hugepages() allocates the memmap array in
642 		 * HPAGE_SIZE chunks.
643 		 */
644 		offset = ALIGN(start + SZ_8K + 64 * npfns + dax_label_reserve,
645 				max(nd_pfn->align, HPAGE_SIZE)) - start;
646 	} else if (nd_pfn->mode == PFN_MODE_RAM)
647 		offset = ALIGN(start + SZ_8K + dax_label_reserve,
648 				nd_pfn->align) - start;
649 	else
650 		return -ENXIO;
651 
652 	if (offset + start_pad + end_trunc >= size) {
653 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
654 				dev_name(&ndns->dev));
655 		return -ENXIO;
656 	}
657 
658 	npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
659 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
660 	pfn_sb->dataoff = cpu_to_le64(offset);
661 	pfn_sb->npfns = cpu_to_le64(npfns);
662 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
663 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
664 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
665 	pfn_sb->version_major = cpu_to_le16(1);
666 	pfn_sb->version_minor = cpu_to_le16(2);
667 	pfn_sb->start_pad = cpu_to_le32(start_pad);
668 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
669 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
670 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
671 	pfn_sb->checksum = cpu_to_le64(checksum);
672 
673 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
674 }
675 
676 /*
677  * Determine the effective resource range and vmem_altmap from an nd_pfn
678  * instance.
679  */
680 struct vmem_altmap *nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
681 		struct resource *res, struct vmem_altmap *altmap)
682 {
683 	int rc;
684 
685 	if (!nd_pfn->uuid || !nd_pfn->ndns)
686 		return ERR_PTR(-ENODEV);
687 
688 	rc = nd_pfn_init(nd_pfn);
689 	if (rc)
690 		return ERR_PTR(rc);
691 
692 	/* we need a valid pfn_sb before we can init a vmem_altmap */
693 	return __nvdimm_setup_pfn(nd_pfn, res, altmap);
694 }
695 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
696