xref: /openbmc/linux/drivers/nvdimm/pfn_devs.c (revision 48c926cd)
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 const unsigned long *nd_pfn_supported_alignments(void)
115 {
116 	/*
117 	 * This needs to be a non-static variable because the *_SIZE
118 	 * macros aren't always constants.
119 	 */
120 	const unsigned long supported_alignments[] = {
121 		PAGE_SIZE,
122 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
123 		HPAGE_PMD_SIZE,
124 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
125 		HPAGE_PUD_SIZE,
126 #endif
127 #endif
128 		0,
129 	};
130 	static unsigned long data[ARRAY_SIZE(supported_alignments)];
131 
132 	memcpy(data, supported_alignments, sizeof(data));
133 
134 	return data;
135 }
136 
137 static ssize_t align_store(struct device *dev,
138 		struct device_attribute *attr, const char *buf, size_t len)
139 {
140 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
141 	ssize_t rc;
142 
143 	device_lock(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, "%s: result: %zd wrote: %s%s", __func__,
148 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
149 	nvdimm_bus_unlock(dev);
150 	device_unlock(dev);
151 
152 	return rc ? rc : len;
153 }
154 static DEVICE_ATTR_RW(align);
155 
156 static ssize_t uuid_show(struct device *dev,
157 		struct device_attribute *attr, char *buf)
158 {
159 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
160 
161 	if (nd_pfn->uuid)
162 		return sprintf(buf, "%pUb\n", nd_pfn->uuid);
163 	return sprintf(buf, "\n");
164 }
165 
166 static ssize_t uuid_store(struct device *dev,
167 		struct device_attribute *attr, const char *buf, size_t len)
168 {
169 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
170 	ssize_t rc;
171 
172 	device_lock(dev);
173 	rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
174 	dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
175 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
176 	device_unlock(dev);
177 
178 	return rc ? rc : len;
179 }
180 static DEVICE_ATTR_RW(uuid);
181 
182 static ssize_t namespace_show(struct device *dev,
183 		struct device_attribute *attr, char *buf)
184 {
185 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
186 	ssize_t rc;
187 
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);
192 	return rc;
193 }
194 
195 static ssize_t namespace_store(struct device *dev,
196 		struct device_attribute *attr, const char *buf, size_t len)
197 {
198 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
199 	ssize_t rc;
200 
201 	device_lock(dev);
202 	nvdimm_bus_lock(dev);
203 	rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
204 	dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
205 			rc, buf, buf[len - 1] == '\n' ? "" : "\n");
206 	nvdimm_bus_unlock(dev);
207 	device_unlock(dev);
208 
209 	return rc;
210 }
211 static DEVICE_ATTR_RW(namespace);
212 
213 static ssize_t resource_show(struct device *dev,
214 		struct device_attribute *attr, char *buf)
215 {
216 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
217 	ssize_t rc;
218 
219 	device_lock(dev);
220 	if (dev->driver) {
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);
226 
227 		rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
228 				+ start_pad + offset);
229 	} else {
230 		/* no address to convey if the pfn instance is disabled */
231 		rc = -ENXIO;
232 	}
233 	device_unlock(dev);
234 
235 	return rc;
236 }
237 static DEVICE_ATTR_RO(resource);
238 
239 static ssize_t size_show(struct device *dev,
240 		struct device_attribute *attr, char *buf)
241 {
242 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
243 	ssize_t rc;
244 
245 	device_lock(dev);
246 	if (dev->driver) {
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);
253 
254 		rc = sprintf(buf, "%llu\n", (unsigned long long)
255 				resource_size(&nsio->res) - start_pad
256 				- end_trunc - offset);
257 	} else {
258 		/* no size to convey if the pfn instance is disabled */
259 		rc = -ENXIO;
260 	}
261 	device_unlock(dev);
262 
263 	return rc;
264 }
265 static DEVICE_ATTR_RO(size);
266 
267 static ssize_t supported_alignments_show(struct device *dev,
268 		struct device_attribute *attr, char *buf)
269 {
270 	return nd_size_select_show(0, nd_pfn_supported_alignments(), buf);
271 }
272 static DEVICE_ATTR_RO(supported_alignments);
273 
274 static struct attribute *nd_pfn_attributes[] = {
275 	&dev_attr_mode.attr,
276 	&dev_attr_namespace.attr,
277 	&dev_attr_uuid.attr,
278 	&dev_attr_align.attr,
279 	&dev_attr_resource.attr,
280 	&dev_attr_size.attr,
281 	&dev_attr_supported_alignments.attr,
282 	NULL,
283 };
284 
285 struct attribute_group nd_pfn_attribute_group = {
286 	.attrs = nd_pfn_attributes,
287 };
288 
289 static const struct attribute_group *nd_pfn_attribute_groups[] = {
290 	&nd_pfn_attribute_group,
291 	&nd_device_attribute_group,
292 	&nd_numa_attribute_group,
293 	NULL,
294 };
295 
296 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
297 		struct nd_namespace_common *ndns)
298 {
299 	struct device *dev = &nd_pfn->dev;
300 
301 	if (!nd_pfn)
302 		return NULL;
303 
304 	nd_pfn->mode = PFN_MODE_NONE;
305 	nd_pfn->align = PFN_DEFAULT_ALIGNMENT;
306 	dev = &nd_pfn->dev;
307 	device_initialize(&nd_pfn->dev);
308 	if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
309 		dev_dbg(&ndns->dev, "%s failed, already claimed by %s\n",
310 				__func__, dev_name(ndns->claim));
311 		put_device(dev);
312 		return NULL;
313 	}
314 	return dev;
315 }
316 
317 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
318 {
319 	struct nd_pfn *nd_pfn;
320 	struct device *dev;
321 
322 	nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
323 	if (!nd_pfn)
324 		return NULL;
325 
326 	nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
327 	if (nd_pfn->id < 0) {
328 		kfree(nd_pfn);
329 		return NULL;
330 	}
331 
332 	dev = &nd_pfn->dev;
333 	dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
334 	dev->groups = nd_pfn_attribute_groups;
335 	dev->type = &nd_pfn_device_type;
336 	dev->parent = &nd_region->dev;
337 
338 	return nd_pfn;
339 }
340 
341 struct device *nd_pfn_create(struct nd_region *nd_region)
342 {
343 	struct nd_pfn *nd_pfn;
344 	struct device *dev;
345 
346 	if (!is_memory(&nd_region->dev))
347 		return NULL;
348 
349 	nd_pfn = nd_pfn_alloc(nd_region);
350 	dev = nd_pfn_devinit(nd_pfn, NULL);
351 
352 	__nd_device_register(dev);
353 	return dev;
354 }
355 
356 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
357 {
358 	u64 checksum, offset;
359 	unsigned long align;
360 	enum nd_pfn_mode mode;
361 	struct nd_namespace_io *nsio;
362 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
363 	struct nd_namespace_common *ndns = nd_pfn->ndns;
364 	const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
365 
366 	if (!pfn_sb || !ndns)
367 		return -ENODEV;
368 
369 	if (!is_memory(nd_pfn->dev.parent))
370 		return -ENODEV;
371 
372 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
373 		return -ENXIO;
374 
375 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
376 		return -ENODEV;
377 
378 	checksum = le64_to_cpu(pfn_sb->checksum);
379 	pfn_sb->checksum = 0;
380 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
381 		return -ENODEV;
382 	pfn_sb->checksum = cpu_to_le64(checksum);
383 
384 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
385 		return -ENODEV;
386 
387 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
388 		pfn_sb->start_pad = 0;
389 		pfn_sb->end_trunc = 0;
390 	}
391 
392 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
393 		pfn_sb->align = 0;
394 
395 	switch (le32_to_cpu(pfn_sb->mode)) {
396 	case PFN_MODE_RAM:
397 	case PFN_MODE_PMEM:
398 		break;
399 	default:
400 		return -ENXIO;
401 	}
402 
403 	align = le32_to_cpu(pfn_sb->align);
404 	offset = le64_to_cpu(pfn_sb->dataoff);
405 	if (align == 0)
406 		align = 1UL << ilog2(offset);
407 	mode = le32_to_cpu(pfn_sb->mode);
408 
409 	if (!nd_pfn->uuid) {
410 		/*
411 		 * When probing a namepace via nd_pfn_probe() the uuid
412 		 * is NULL (see: nd_pfn_devinit()) we init settings from
413 		 * pfn_sb
414 		 */
415 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
416 		if (!nd_pfn->uuid)
417 			return -ENOMEM;
418 		nd_pfn->align = align;
419 		nd_pfn->mode = mode;
420 	} else {
421 		/*
422 		 * When probing a pfn / dax instance we validate the
423 		 * live settings against the pfn_sb
424 		 */
425 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
426 			return -ENODEV;
427 
428 		/*
429 		 * If the uuid validates, but other settings mismatch
430 		 * return EINVAL because userspace has managed to change
431 		 * the configuration without specifying new
432 		 * identification.
433 		 */
434 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
435 			dev_err(&nd_pfn->dev,
436 					"init failed, settings mismatch\n");
437 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
438 					nd_pfn->align, align, nd_pfn->mode,
439 					mode);
440 			return -EINVAL;
441 		}
442 	}
443 
444 	if (align > nvdimm_namespace_capacity(ndns)) {
445 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
446 				align, nvdimm_namespace_capacity(ndns));
447 		return -EINVAL;
448 	}
449 
450 	/*
451 	 * These warnings are verbose because they can only trigger in
452 	 * the case where the physical address alignment of the
453 	 * namespace has changed since the pfn superblock was
454 	 * established.
455 	 */
456 	nsio = to_nd_namespace_io(&ndns->dev);
457 	if (offset >= resource_size(&nsio->res)) {
458 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
459 				dev_name(&ndns->dev));
460 		return -EBUSY;
461 	}
462 
463 	if ((align && !IS_ALIGNED(offset, align))
464 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
465 		dev_err(&nd_pfn->dev,
466 				"bad offset: %#llx dax disabled align: %#lx\n",
467 				offset, align);
468 		return -ENXIO;
469 	}
470 
471 	return 0;
472 }
473 EXPORT_SYMBOL(nd_pfn_validate);
474 
475 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
476 {
477 	int rc;
478 	struct nd_pfn *nd_pfn;
479 	struct device *pfn_dev;
480 	struct nd_pfn_sb *pfn_sb;
481 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
482 
483 	if (ndns->force_raw)
484 		return -ENODEV;
485 
486 	switch (ndns->claim_class) {
487 	case NVDIMM_CCLASS_NONE:
488 	case NVDIMM_CCLASS_PFN:
489 		break;
490 	default:
491 		return -ENODEV;
492 	}
493 
494 	nvdimm_bus_lock(&ndns->dev);
495 	nd_pfn = nd_pfn_alloc(nd_region);
496 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
497 	nvdimm_bus_unlock(&ndns->dev);
498 	if (!pfn_dev)
499 		return -ENOMEM;
500 	pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
501 	nd_pfn = to_nd_pfn(pfn_dev);
502 	nd_pfn->pfn_sb = pfn_sb;
503 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
504 	dev_dbg(dev, "%s: pfn: %s\n", __func__,
505 			rc == 0 ? dev_name(pfn_dev) : "<none>");
506 	if (rc < 0) {
507 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
508 		put_device(pfn_dev);
509 	} else
510 		__nd_device_register(pfn_dev);
511 
512 	return rc;
513 }
514 EXPORT_SYMBOL(nd_pfn_probe);
515 
516 /*
517  * We hotplug memory at section granularity, pad the reserved area from
518  * the previous section base to the namespace base address.
519  */
520 static unsigned long init_altmap_base(resource_size_t base)
521 {
522 	unsigned long base_pfn = PHYS_PFN(base);
523 
524 	return PFN_SECTION_ALIGN_DOWN(base_pfn);
525 }
526 
527 static unsigned long init_altmap_reserve(resource_size_t base)
528 {
529 	unsigned long reserve = PHYS_PFN(SZ_8K);
530 	unsigned long base_pfn = PHYS_PFN(base);
531 
532 	reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
533 	return reserve;
534 }
535 
536 static struct vmem_altmap *__nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
537 		struct resource *res, struct vmem_altmap *altmap)
538 {
539 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
540 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
541 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
542 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
543 	struct nd_namespace_common *ndns = nd_pfn->ndns;
544 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
545 	resource_size_t base = nsio->res.start + start_pad;
546 	struct vmem_altmap __altmap = {
547 		.base_pfn = init_altmap_base(base),
548 		.reserve = init_altmap_reserve(base),
549 	};
550 
551 	memcpy(res, &nsio->res, sizeof(*res));
552 	res->start += start_pad;
553 	res->end -= end_trunc;
554 
555 	if (nd_pfn->mode == PFN_MODE_RAM) {
556 		if (offset < SZ_8K)
557 			return ERR_PTR(-EINVAL);
558 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
559 		altmap = NULL;
560 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
561 		nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
562 					- offset) / PAGE_SIZE);
563 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
564 			dev_info(&nd_pfn->dev,
565 					"number of pfns truncated from %lld to %ld\n",
566 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
567 					nd_pfn->npfns);
568 		memcpy(altmap, &__altmap, sizeof(*altmap));
569 		altmap->free = PHYS_PFN(offset - SZ_8K);
570 		altmap->alloc = 0;
571 	} else
572 		return ERR_PTR(-ENXIO);
573 
574 	return altmap;
575 }
576 
577 static int nd_pfn_init(struct nd_pfn *nd_pfn)
578 {
579 	u32 dax_label_reserve = is_nd_dax(&nd_pfn->dev) ? SZ_128K : 0;
580 	struct nd_namespace_common *ndns = nd_pfn->ndns;
581 	u32 start_pad = 0, end_trunc = 0;
582 	resource_size_t start, size;
583 	struct nd_namespace_io *nsio;
584 	struct nd_region *nd_region;
585 	struct nd_pfn_sb *pfn_sb;
586 	unsigned long npfns;
587 	phys_addr_t offset;
588 	const char *sig;
589 	u64 checksum;
590 	int rc;
591 
592 	pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
593 	if (!pfn_sb)
594 		return -ENOMEM;
595 
596 	nd_pfn->pfn_sb = pfn_sb;
597 	if (is_nd_dax(&nd_pfn->dev))
598 		sig = DAX_SIG;
599 	else
600 		sig = PFN_SIG;
601 	rc = nd_pfn_validate(nd_pfn, sig);
602 	if (rc != -ENODEV)
603 		return rc;
604 
605 	/* no info block, do init */;
606 	nd_region = to_nd_region(nd_pfn->dev.parent);
607 	if (nd_region->ro) {
608 		dev_info(&nd_pfn->dev,
609 				"%s is read-only, unable to init metadata\n",
610 				dev_name(&nd_region->dev));
611 		return -ENXIO;
612 	}
613 
614 	memset(pfn_sb, 0, sizeof(*pfn_sb));
615 
616 	/*
617 	 * Check if pmem collides with 'System RAM' when section aligned and
618 	 * trim it accordingly
619 	 */
620 	nsio = to_nd_namespace_io(&ndns->dev);
621 	start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
622 	size = resource_size(&nsio->res);
623 	if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
624 				IORES_DESC_NONE) == REGION_MIXED) {
625 		start = nsio->res.start;
626 		start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
627 	}
628 
629 	start = nsio->res.start;
630 	size = PHYS_SECTION_ALIGN_UP(start + size) - start;
631 	if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
632 				IORES_DESC_NONE) == REGION_MIXED) {
633 		size = resource_size(&nsio->res);
634 		end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size);
635 	}
636 
637 	if (start_pad + end_trunc)
638 		dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n",
639 				dev_name(&ndns->dev), start_pad + end_trunc);
640 
641 	/*
642 	 * Note, we use 64 here for the standard size of struct page,
643 	 * debugging options may cause it to be larger in which case the
644 	 * implementation will limit the pfns advertised through
645 	 * ->direct_access() to those that are included in the memmap.
646 	 */
647 	start += start_pad;
648 	size = resource_size(&nsio->res);
649 	npfns = PFN_SECTION_ALIGN_UP((size - start_pad - end_trunc - SZ_8K)
650 			/ PAGE_SIZE);
651 	if (nd_pfn->mode == PFN_MODE_PMEM) {
652 		/*
653 		 * The altmap should be padded out to the block size used
654 		 * when populating the vmemmap. This *should* be equal to
655 		 * PMD_SIZE for most architectures.
656 		 */
657 		offset = ALIGN(start + SZ_8K + 64 * npfns + dax_label_reserve,
658 				max(nd_pfn->align, PMD_SIZE)) - start;
659 	} else if (nd_pfn->mode == PFN_MODE_RAM)
660 		offset = ALIGN(start + SZ_8K + dax_label_reserve,
661 				nd_pfn->align) - start;
662 	else
663 		return -ENXIO;
664 
665 	if (offset + start_pad + end_trunc >= size) {
666 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
667 				dev_name(&ndns->dev));
668 		return -ENXIO;
669 	}
670 
671 	npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
672 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
673 	pfn_sb->dataoff = cpu_to_le64(offset);
674 	pfn_sb->npfns = cpu_to_le64(npfns);
675 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
676 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
677 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
678 	pfn_sb->version_major = cpu_to_le16(1);
679 	pfn_sb->version_minor = cpu_to_le16(2);
680 	pfn_sb->start_pad = cpu_to_le32(start_pad);
681 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
682 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
683 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
684 	pfn_sb->checksum = cpu_to_le64(checksum);
685 
686 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
687 }
688 
689 /*
690  * Determine the effective resource range and vmem_altmap from an nd_pfn
691  * instance.
692  */
693 struct vmem_altmap *nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
694 		struct resource *res, struct vmem_altmap *altmap)
695 {
696 	int rc;
697 
698 	if (!nd_pfn->uuid || !nd_pfn->ndns)
699 		return ERR_PTR(-ENODEV);
700 
701 	rc = nd_pfn_init(nd_pfn);
702 	if (rc)
703 		return ERR_PTR(rc);
704 
705 	/* we need a valid pfn_sb before we can init a vmem_altmap */
706 	return __nvdimm_setup_pfn(nd_pfn, res, altmap);
707 }
708 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
709