xref: /openbmc/linux/drivers/nvdimm/pfn_devs.c (revision a2cab953)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright(c) 2013-2016 Intel Corporation. All rights reserved.
4  */
5 #include <linux/memremap.h>
6 #include <linux/blkdev.h>
7 #include <linux/device.h>
8 #include <linux/sizes.h>
9 #include <linux/slab.h>
10 #include <linux/fs.h>
11 #include <linux/mm.h>
12 #include "nd-core.h"
13 #include "pfn.h"
14 #include "nd.h"
15 
16 static void nd_pfn_release(struct device *dev)
17 {
18 	struct nd_region *nd_region = to_nd_region(dev->parent);
19 	struct nd_pfn *nd_pfn = to_nd_pfn(dev);
20 
21 	dev_dbg(dev, "trace\n");
22 	nd_detach_ndns(&nd_pfn->dev, &nd_pfn->ndns);
23 	ida_simple_remove(&nd_region->pfn_ida, nd_pfn->id);
24 	kfree(nd_pfn->uuid);
25 	kfree(nd_pfn);
26 }
27 
28 struct nd_pfn *to_nd_pfn(struct device *dev)
29 {
30 	struct nd_pfn *nd_pfn = container_of(dev, struct nd_pfn, dev);
31 
32 	WARN_ON(!is_nd_pfn(dev));
33 	return nd_pfn;
34 }
35 EXPORT_SYMBOL(to_nd_pfn);
36 
37 static ssize_t mode_show(struct device *dev,
38 		struct device_attribute *attr, char *buf)
39 {
40 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
41 
42 	switch (nd_pfn->mode) {
43 	case PFN_MODE_RAM:
44 		return sprintf(buf, "ram\n");
45 	case PFN_MODE_PMEM:
46 		return sprintf(buf, "pmem\n");
47 	default:
48 		return sprintf(buf, "none\n");
49 	}
50 }
51 
52 static ssize_t mode_store(struct device *dev,
53 		struct device_attribute *attr, const char *buf, size_t len)
54 {
55 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
56 	ssize_t rc = 0;
57 
58 	device_lock(dev);
59 	nvdimm_bus_lock(dev);
60 	if (dev->driver)
61 		rc = -EBUSY;
62 	else {
63 		size_t n = len - 1;
64 
65 		if (strncmp(buf, "pmem\n", n) == 0
66 				|| strncmp(buf, "pmem", n) == 0) {
67 			nd_pfn->mode = PFN_MODE_PMEM;
68 		} else if (strncmp(buf, "ram\n", n) == 0
69 				|| strncmp(buf, "ram", n) == 0)
70 			nd_pfn->mode = PFN_MODE_RAM;
71 		else if (strncmp(buf, "none\n", n) == 0
72 				|| strncmp(buf, "none", n) == 0)
73 			nd_pfn->mode = PFN_MODE_NONE;
74 		else
75 			rc = -EINVAL;
76 	}
77 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
78 			buf[len - 1] == '\n' ? "" : "\n");
79 	nvdimm_bus_unlock(dev);
80 	device_unlock(dev);
81 
82 	return rc ? rc : len;
83 }
84 static DEVICE_ATTR_RW(mode);
85 
86 static ssize_t align_show(struct device *dev,
87 		struct device_attribute *attr, char *buf)
88 {
89 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
90 
91 	return sprintf(buf, "%ld\n", nd_pfn->align);
92 }
93 
94 static unsigned long *nd_pfn_supported_alignments(unsigned long *alignments)
95 {
96 
97 	alignments[0] = PAGE_SIZE;
98 
99 	if (has_transparent_hugepage()) {
100 		alignments[1] = HPAGE_PMD_SIZE;
101 		if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
102 			alignments[2] = HPAGE_PUD_SIZE;
103 	}
104 
105 	return alignments;
106 }
107 
108 /*
109  * Use pmd mapping if supported as default alignment
110  */
111 static unsigned long nd_pfn_default_alignment(void)
112 {
113 
114 	if (has_transparent_hugepage())
115 		return HPAGE_PMD_SIZE;
116 	return PAGE_SIZE;
117 }
118 
119 static ssize_t align_store(struct device *dev,
120 		struct device_attribute *attr, const char *buf, size_t len)
121 {
122 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
123 	unsigned long aligns[MAX_NVDIMM_ALIGN] = { [0] = 0, };
124 	ssize_t rc;
125 
126 	device_lock(dev);
127 	nvdimm_bus_lock(dev);
128 	rc = nd_size_select_store(dev, buf, &nd_pfn->align,
129 			nd_pfn_supported_alignments(aligns));
130 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
131 			buf[len - 1] == '\n' ? "" : "\n");
132 	nvdimm_bus_unlock(dev);
133 	device_unlock(dev);
134 
135 	return rc ? rc : len;
136 }
137 static DEVICE_ATTR_RW(align);
138 
139 static ssize_t uuid_show(struct device *dev,
140 		struct device_attribute *attr, char *buf)
141 {
142 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
143 
144 	if (nd_pfn->uuid)
145 		return sprintf(buf, "%pUb\n", nd_pfn->uuid);
146 	return sprintf(buf, "\n");
147 }
148 
149 static ssize_t uuid_store(struct device *dev,
150 		struct device_attribute *attr, const char *buf, size_t len)
151 {
152 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
153 	ssize_t rc;
154 
155 	device_lock(dev);
156 	rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
157 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
158 			buf[len - 1] == '\n' ? "" : "\n");
159 	device_unlock(dev);
160 
161 	return rc ? rc : len;
162 }
163 static DEVICE_ATTR_RW(uuid);
164 
165 static ssize_t namespace_show(struct device *dev,
166 		struct device_attribute *attr, char *buf)
167 {
168 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
169 	ssize_t rc;
170 
171 	nvdimm_bus_lock(dev);
172 	rc = sprintf(buf, "%s\n", nd_pfn->ndns
173 			? dev_name(&nd_pfn->ndns->dev) : "");
174 	nvdimm_bus_unlock(dev);
175 	return rc;
176 }
177 
178 static ssize_t namespace_store(struct device *dev,
179 		struct device_attribute *attr, const char *buf, size_t len)
180 {
181 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
182 	ssize_t rc;
183 
184 	device_lock(dev);
185 	nvdimm_bus_lock(dev);
186 	rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
187 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
188 			buf[len - 1] == '\n' ? "" : "\n");
189 	nvdimm_bus_unlock(dev);
190 	device_unlock(dev);
191 
192 	return rc;
193 }
194 static DEVICE_ATTR_RW(namespace);
195 
196 static ssize_t resource_show(struct device *dev,
197 		struct device_attribute *attr, char *buf)
198 {
199 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
200 	ssize_t rc;
201 
202 	device_lock(dev);
203 	if (dev->driver) {
204 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
205 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
206 		struct nd_namespace_common *ndns = nd_pfn->ndns;
207 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
208 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
209 
210 		rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
211 				+ start_pad + offset);
212 	} else {
213 		/* no address to convey if the pfn instance is disabled */
214 		rc = -ENXIO;
215 	}
216 	device_unlock(dev);
217 
218 	return rc;
219 }
220 static DEVICE_ATTR_ADMIN_RO(resource);
221 
222 static ssize_t size_show(struct device *dev,
223 		struct device_attribute *attr, char *buf)
224 {
225 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
226 	ssize_t rc;
227 
228 	device_lock(dev);
229 	if (dev->driver) {
230 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
231 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
232 		struct nd_namespace_common *ndns = nd_pfn->ndns;
233 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
234 		u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
235 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
236 
237 		rc = sprintf(buf, "%llu\n", (unsigned long long)
238 				resource_size(&nsio->res) - start_pad
239 				- end_trunc - offset);
240 	} else {
241 		/* no size to convey if the pfn instance is disabled */
242 		rc = -ENXIO;
243 	}
244 	device_unlock(dev);
245 
246 	return rc;
247 }
248 static DEVICE_ATTR_RO(size);
249 
250 static ssize_t supported_alignments_show(struct device *dev,
251 		struct device_attribute *attr, char *buf)
252 {
253 	unsigned long aligns[MAX_NVDIMM_ALIGN] = { [0] = 0, };
254 
255 	return nd_size_select_show(0,
256 			nd_pfn_supported_alignments(aligns), buf);
257 }
258 static DEVICE_ATTR_RO(supported_alignments);
259 
260 static struct attribute *nd_pfn_attributes[] = {
261 	&dev_attr_mode.attr,
262 	&dev_attr_namespace.attr,
263 	&dev_attr_uuid.attr,
264 	&dev_attr_align.attr,
265 	&dev_attr_resource.attr,
266 	&dev_attr_size.attr,
267 	&dev_attr_supported_alignments.attr,
268 	NULL,
269 };
270 
271 static struct attribute_group nd_pfn_attribute_group = {
272 	.attrs = nd_pfn_attributes,
273 };
274 
275 const struct attribute_group *nd_pfn_attribute_groups[] = {
276 	&nd_pfn_attribute_group,
277 	&nd_device_attribute_group,
278 	&nd_numa_attribute_group,
279 	NULL,
280 };
281 
282 static const struct device_type nd_pfn_device_type = {
283 	.name = "nd_pfn",
284 	.release = nd_pfn_release,
285 	.groups = nd_pfn_attribute_groups,
286 };
287 
288 bool is_nd_pfn(struct device *dev)
289 {
290 	return dev ? dev->type == &nd_pfn_device_type : false;
291 }
292 EXPORT_SYMBOL(is_nd_pfn);
293 
294 static struct lock_class_key nvdimm_pfn_key;
295 
296 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
297 		struct nd_namespace_common *ndns)
298 {
299 	struct device *dev;
300 
301 	if (!nd_pfn)
302 		return NULL;
303 
304 	nd_pfn->mode = PFN_MODE_NONE;
305 	nd_pfn->align = nd_pfn_default_alignment();
306 	dev = &nd_pfn->dev;
307 	device_initialize(&nd_pfn->dev);
308 	lockdep_set_class(&nd_pfn->dev.mutex, &nvdimm_pfn_key);
309 	if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
310 		dev_dbg(&ndns->dev, "failed, already claimed by %s\n",
311 				dev_name(ndns->claim));
312 		put_device(dev);
313 		return NULL;
314 	}
315 	return dev;
316 }
317 
318 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
319 {
320 	struct nd_pfn *nd_pfn;
321 	struct device *dev;
322 
323 	nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
324 	if (!nd_pfn)
325 		return NULL;
326 
327 	nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
328 	if (nd_pfn->id < 0) {
329 		kfree(nd_pfn);
330 		return NULL;
331 	}
332 
333 	dev = &nd_pfn->dev;
334 	dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
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 /*
357  * nd_pfn_clear_memmap_errors() clears any errors in the volatile memmap
358  * space associated with the namespace. If the memmap is set to DRAM, then
359  * this is a no-op. Since the memmap area is freshly initialized during
360  * probe, we have an opportunity to clear any badblocks in this area.
361  */
362 static int nd_pfn_clear_memmap_errors(struct nd_pfn *nd_pfn)
363 {
364 	struct nd_region *nd_region = to_nd_region(nd_pfn->dev.parent);
365 	struct nd_namespace_common *ndns = nd_pfn->ndns;
366 	void *zero_page = page_address(ZERO_PAGE(0));
367 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
368 	int num_bad, meta_num, rc, bb_present;
369 	sector_t first_bad, meta_start;
370 	struct nd_namespace_io *nsio;
371 
372 	if (nd_pfn->mode != PFN_MODE_PMEM)
373 		return 0;
374 
375 	nsio = to_nd_namespace_io(&ndns->dev);
376 	meta_start = (SZ_4K + sizeof(*pfn_sb)) >> 9;
377 	meta_num = (le64_to_cpu(pfn_sb->dataoff) >> 9) - meta_start;
378 
379 	/*
380 	 * re-enable the namespace with correct size so that we can access
381 	 * the device memmap area.
382 	 */
383 	devm_namespace_disable(&nd_pfn->dev, ndns);
384 	rc = devm_namespace_enable(&nd_pfn->dev, ndns, le64_to_cpu(pfn_sb->dataoff));
385 	if (rc)
386 		return rc;
387 
388 	do {
389 		unsigned long zero_len;
390 		u64 nsoff;
391 
392 		bb_present = badblocks_check(&nd_region->bb, meta_start,
393 				meta_num, &first_bad, &num_bad);
394 		if (bb_present) {
395 			dev_dbg(&nd_pfn->dev, "meta: %x badblocks at %llx\n",
396 					num_bad, first_bad);
397 			nsoff = ALIGN_DOWN((nd_region->ndr_start
398 					+ (first_bad << 9)) - nsio->res.start,
399 					PAGE_SIZE);
400 			zero_len = ALIGN(num_bad << 9, PAGE_SIZE);
401 			while (zero_len) {
402 				unsigned long chunk = min(zero_len, PAGE_SIZE);
403 
404 				rc = nvdimm_write_bytes(ndns, nsoff, zero_page,
405 							chunk, 0);
406 				if (rc)
407 					break;
408 
409 				zero_len -= chunk;
410 				nsoff += chunk;
411 			}
412 			if (rc) {
413 				dev_err(&nd_pfn->dev,
414 					"error clearing %x badblocks at %llx\n",
415 					num_bad, first_bad);
416 				return rc;
417 			}
418 		}
419 	} while (bb_present);
420 
421 	return 0;
422 }
423 
424 static bool nd_supported_alignment(unsigned long align)
425 {
426 	int i;
427 	unsigned long supported[MAX_NVDIMM_ALIGN] = { [0] = 0, };
428 
429 	if (align == 0)
430 		return false;
431 
432 	nd_pfn_supported_alignments(supported);
433 	for (i = 0; supported[i]; i++)
434 		if (align == supported[i])
435 			return true;
436 	return false;
437 }
438 
439 /**
440  * nd_pfn_validate - read and validate info-block
441  * @nd_pfn: fsdax namespace runtime state / properties
442  * @sig: 'devdax' or 'fsdax' signature
443  *
444  * Upon return the info-block buffer contents (->pfn_sb) are
445  * indeterminate when validation fails, and a coherent info-block
446  * otherwise.
447  */
448 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
449 {
450 	u64 checksum, offset;
451 	struct resource *res;
452 	enum nd_pfn_mode mode;
453 	struct nd_namespace_io *nsio;
454 	unsigned long align, start_pad;
455 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
456 	struct nd_namespace_common *ndns = nd_pfn->ndns;
457 	const uuid_t *parent_uuid = nd_dev_to_uuid(&ndns->dev);
458 
459 	if (!pfn_sb || !ndns)
460 		return -ENODEV;
461 
462 	if (!is_memory(nd_pfn->dev.parent))
463 		return -ENODEV;
464 
465 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
466 		return -ENXIO;
467 
468 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
469 		return -ENODEV;
470 
471 	checksum = le64_to_cpu(pfn_sb->checksum);
472 	pfn_sb->checksum = 0;
473 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
474 		return -ENODEV;
475 	pfn_sb->checksum = cpu_to_le64(checksum);
476 
477 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
478 		return -ENODEV;
479 
480 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
481 		pfn_sb->start_pad = 0;
482 		pfn_sb->end_trunc = 0;
483 	}
484 
485 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
486 		pfn_sb->align = 0;
487 
488 	if (__le16_to_cpu(pfn_sb->version_minor) < 4) {
489 		pfn_sb->page_struct_size = cpu_to_le16(64);
490 		pfn_sb->page_size = cpu_to_le32(PAGE_SIZE);
491 	}
492 
493 	switch (le32_to_cpu(pfn_sb->mode)) {
494 	case PFN_MODE_RAM:
495 	case PFN_MODE_PMEM:
496 		break;
497 	default:
498 		return -ENXIO;
499 	}
500 
501 	align = le32_to_cpu(pfn_sb->align);
502 	offset = le64_to_cpu(pfn_sb->dataoff);
503 	start_pad = le32_to_cpu(pfn_sb->start_pad);
504 	if (align == 0)
505 		align = 1UL << ilog2(offset);
506 	mode = le32_to_cpu(pfn_sb->mode);
507 
508 	if ((le32_to_cpu(pfn_sb->page_size) > PAGE_SIZE) &&
509 			(mode == PFN_MODE_PMEM)) {
510 		dev_err(&nd_pfn->dev,
511 				"init failed, page size mismatch %d\n",
512 				le32_to_cpu(pfn_sb->page_size));
513 		return -EOPNOTSUPP;
514 	}
515 
516 	if ((le16_to_cpu(pfn_sb->page_struct_size) < sizeof(struct page)) &&
517 			(mode == PFN_MODE_PMEM)) {
518 		dev_err(&nd_pfn->dev,
519 				"init failed, struct page size mismatch %d\n",
520 				le16_to_cpu(pfn_sb->page_struct_size));
521 		return -EOPNOTSUPP;
522 	}
523 
524 	/*
525 	 * Check whether the we support the alignment. For Dax if the
526 	 * superblock alignment is not matching, we won't initialize
527 	 * the device.
528 	 */
529 	if (!nd_supported_alignment(align) &&
530 			!memcmp(pfn_sb->signature, DAX_SIG, PFN_SIG_LEN)) {
531 		dev_err(&nd_pfn->dev, "init failed, alignment mismatch: "
532 				"%ld:%ld\n", nd_pfn->align, align);
533 		return -EOPNOTSUPP;
534 	}
535 
536 	if (!nd_pfn->uuid) {
537 		/*
538 		 * When probing a namepace via nd_pfn_probe() the uuid
539 		 * is NULL (see: nd_pfn_devinit()) we init settings from
540 		 * pfn_sb
541 		 */
542 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
543 		if (!nd_pfn->uuid)
544 			return -ENOMEM;
545 		nd_pfn->align = align;
546 		nd_pfn->mode = mode;
547 	} else {
548 		/*
549 		 * When probing a pfn / dax instance we validate the
550 		 * live settings against the pfn_sb
551 		 */
552 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
553 			return -ENODEV;
554 
555 		/*
556 		 * If the uuid validates, but other settings mismatch
557 		 * return EINVAL because userspace has managed to change
558 		 * the configuration without specifying new
559 		 * identification.
560 		 */
561 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
562 			dev_err(&nd_pfn->dev,
563 					"init failed, settings mismatch\n");
564 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
565 					nd_pfn->align, align, nd_pfn->mode,
566 					mode);
567 			return -EOPNOTSUPP;
568 		}
569 	}
570 
571 	if (align > nvdimm_namespace_capacity(ndns)) {
572 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
573 				align, nvdimm_namespace_capacity(ndns));
574 		return -EOPNOTSUPP;
575 	}
576 
577 	/*
578 	 * These warnings are verbose because they can only trigger in
579 	 * the case where the physical address alignment of the
580 	 * namespace has changed since the pfn superblock was
581 	 * established.
582 	 */
583 	nsio = to_nd_namespace_io(&ndns->dev);
584 	res = &nsio->res;
585 	if (offset >= resource_size(res)) {
586 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
587 				dev_name(&ndns->dev));
588 		return -EOPNOTSUPP;
589 	}
590 
591 	if ((align && !IS_ALIGNED(res->start + offset + start_pad, align))
592 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
593 		dev_err(&nd_pfn->dev,
594 				"bad offset: %#llx dax disabled align: %#lx\n",
595 				offset, align);
596 		return -EOPNOTSUPP;
597 	}
598 
599 	if (!IS_ALIGNED(res->start + le32_to_cpu(pfn_sb->start_pad),
600 				memremap_compat_align())) {
601 		dev_err(&nd_pfn->dev, "resource start misaligned\n");
602 		return -EOPNOTSUPP;
603 	}
604 
605 	if (!IS_ALIGNED(res->end + 1 - le32_to_cpu(pfn_sb->end_trunc),
606 				memremap_compat_align())) {
607 		dev_err(&nd_pfn->dev, "resource end misaligned\n");
608 		return -EOPNOTSUPP;
609 	}
610 
611 	return 0;
612 }
613 EXPORT_SYMBOL(nd_pfn_validate);
614 
615 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
616 {
617 	int rc;
618 	struct nd_pfn *nd_pfn;
619 	struct device *pfn_dev;
620 	struct nd_pfn_sb *pfn_sb;
621 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
622 
623 	if (ndns->force_raw)
624 		return -ENODEV;
625 
626 	switch (ndns->claim_class) {
627 	case NVDIMM_CCLASS_NONE:
628 	case NVDIMM_CCLASS_PFN:
629 		break;
630 	default:
631 		return -ENODEV;
632 	}
633 
634 	nvdimm_bus_lock(&ndns->dev);
635 	nd_pfn = nd_pfn_alloc(nd_region);
636 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
637 	nvdimm_bus_unlock(&ndns->dev);
638 	if (!pfn_dev)
639 		return -ENOMEM;
640 	pfn_sb = devm_kmalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
641 	nd_pfn = to_nd_pfn(pfn_dev);
642 	nd_pfn->pfn_sb = pfn_sb;
643 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
644 	dev_dbg(dev, "pfn: %s\n", rc == 0 ? dev_name(pfn_dev) : "<none>");
645 	if (rc < 0) {
646 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
647 		put_device(pfn_dev);
648 	} else
649 		nd_device_register(pfn_dev);
650 
651 	return rc;
652 }
653 EXPORT_SYMBOL(nd_pfn_probe);
654 
655 /*
656  * We hotplug memory at sub-section granularity, pad the reserved area
657  * from the previous section base to the namespace base address.
658  */
659 static unsigned long init_altmap_base(resource_size_t base)
660 {
661 	unsigned long base_pfn = PHYS_PFN(base);
662 
663 	return SUBSECTION_ALIGN_DOWN(base_pfn);
664 }
665 
666 static unsigned long init_altmap_reserve(resource_size_t base)
667 {
668 	unsigned long reserve = nd_info_block_reserve() >> PAGE_SHIFT;
669 	unsigned long base_pfn = PHYS_PFN(base);
670 
671 	reserve += base_pfn - SUBSECTION_ALIGN_DOWN(base_pfn);
672 	return reserve;
673 }
674 
675 static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
676 {
677 	struct range *range = &pgmap->range;
678 	struct vmem_altmap *altmap = &pgmap->altmap;
679 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
680 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
681 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
682 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
683 	u32 reserve = nd_info_block_reserve();
684 	struct nd_namespace_common *ndns = nd_pfn->ndns;
685 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
686 	resource_size_t base = nsio->res.start + start_pad;
687 	resource_size_t end = nsio->res.end - end_trunc;
688 	struct vmem_altmap __altmap = {
689 		.base_pfn = init_altmap_base(base),
690 		.reserve = init_altmap_reserve(base),
691 		.end_pfn = PHYS_PFN(end),
692 	};
693 
694 	*range = (struct range) {
695 		.start = nsio->res.start + start_pad,
696 		.end = nsio->res.end - end_trunc,
697 	};
698 	pgmap->nr_range = 1;
699 	if (nd_pfn->mode == PFN_MODE_RAM) {
700 		if (offset < reserve)
701 			return -EINVAL;
702 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
703 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
704 		nd_pfn->npfns = PHYS_PFN((range_len(range) - offset));
705 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
706 			dev_info(&nd_pfn->dev,
707 					"number of pfns truncated from %lld to %ld\n",
708 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
709 					nd_pfn->npfns);
710 		memcpy(altmap, &__altmap, sizeof(*altmap));
711 		altmap->free = PHYS_PFN(offset - reserve);
712 		altmap->alloc = 0;
713 		pgmap->flags |= PGMAP_ALTMAP_VALID;
714 	} else
715 		return -ENXIO;
716 
717 	return 0;
718 }
719 
720 static int nd_pfn_init(struct nd_pfn *nd_pfn)
721 {
722 	struct nd_namespace_common *ndns = nd_pfn->ndns;
723 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
724 	resource_size_t start, size;
725 	struct nd_region *nd_region;
726 	unsigned long npfns, align;
727 	u32 end_trunc;
728 	struct nd_pfn_sb *pfn_sb;
729 	phys_addr_t offset;
730 	const char *sig;
731 	u64 checksum;
732 	int rc;
733 
734 	pfn_sb = devm_kmalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
735 	if (!pfn_sb)
736 		return -ENOMEM;
737 
738 	nd_pfn->pfn_sb = pfn_sb;
739 	if (is_nd_dax(&nd_pfn->dev))
740 		sig = DAX_SIG;
741 	else
742 		sig = PFN_SIG;
743 
744 	rc = nd_pfn_validate(nd_pfn, sig);
745 	if (rc == 0)
746 		return nd_pfn_clear_memmap_errors(nd_pfn);
747 	if (rc != -ENODEV)
748 		return rc;
749 
750 	/* no info block, do init */;
751 	memset(pfn_sb, 0, sizeof(*pfn_sb));
752 
753 	nd_region = to_nd_region(nd_pfn->dev.parent);
754 	if (nd_region->ro) {
755 		dev_info(&nd_pfn->dev,
756 				"%s is read-only, unable to init metadata\n",
757 				dev_name(&nd_region->dev));
758 		return -ENXIO;
759 	}
760 
761 	/*
762 	 * Note, we use 64 here for the standard size of struct page,
763 	 * debugging options may cause it to be larger in which case the
764 	 * implementation will limit the pfns advertised through
765 	 * ->direct_access() to those that are included in the memmap.
766 	 */
767 	start = nsio->res.start;
768 	size = resource_size(&nsio->res);
769 	npfns = PHYS_PFN(size - SZ_8K);
770 	align = max(nd_pfn->align, memremap_compat_align());
771 
772 	/*
773 	 * When @start is misaligned fail namespace creation. See
774 	 * the 'struct nd_pfn_sb' commentary on why ->start_pad is not
775 	 * an option.
776 	 */
777 	if (!IS_ALIGNED(start, memremap_compat_align())) {
778 		dev_err(&nd_pfn->dev, "%s: start %pa misaligned to %#lx\n",
779 				dev_name(&ndns->dev), &start,
780 				memremap_compat_align());
781 		return -EINVAL;
782 	}
783 	end_trunc = start + size - ALIGN_DOWN(start + size, align);
784 	if (nd_pfn->mode == PFN_MODE_PMEM) {
785 		/*
786 		 * The altmap should be padded out to the block size used
787 		 * when populating the vmemmap. This *should* be equal to
788 		 * PMD_SIZE for most architectures.
789 		 *
790 		 * Also make sure size of struct page is less than 128. We
791 		 * want to make sure we use large enough size here so that
792 		 * we don't have a dynamic reserve space depending on
793 		 * struct page size. But we also want to make sure we notice
794 		 * when we end up adding new elements to struct page.
795 		 */
796 		BUILD_BUG_ON(sizeof(struct page) > MAX_STRUCT_PAGE_SIZE);
797 		offset = ALIGN(start + SZ_8K + MAX_STRUCT_PAGE_SIZE * npfns, align)
798 			- start;
799 	} else if (nd_pfn->mode == PFN_MODE_RAM)
800 		offset = ALIGN(start + SZ_8K, align) - start;
801 	else
802 		return -ENXIO;
803 
804 	if (offset >= size) {
805 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
806 				dev_name(&ndns->dev));
807 		return -ENXIO;
808 	}
809 
810 	npfns = PHYS_PFN(size - offset - end_trunc);
811 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
812 	pfn_sb->dataoff = cpu_to_le64(offset);
813 	pfn_sb->npfns = cpu_to_le64(npfns);
814 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
815 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
816 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
817 	pfn_sb->version_major = cpu_to_le16(1);
818 	pfn_sb->version_minor = cpu_to_le16(4);
819 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
820 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
821 	pfn_sb->page_struct_size = cpu_to_le16(MAX_STRUCT_PAGE_SIZE);
822 	pfn_sb->page_size = cpu_to_le32(PAGE_SIZE);
823 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
824 	pfn_sb->checksum = cpu_to_le64(checksum);
825 
826 	rc = nd_pfn_clear_memmap_errors(nd_pfn);
827 	if (rc)
828 		return rc;
829 
830 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
831 }
832 
833 /*
834  * Determine the effective resource range and vmem_altmap from an nd_pfn
835  * instance.
836  */
837 int nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
838 {
839 	int rc;
840 
841 	if (!nd_pfn->uuid || !nd_pfn->ndns)
842 		return -ENODEV;
843 
844 	rc = nd_pfn_init(nd_pfn);
845 	if (rc)
846 		return rc;
847 
848 	/* we need a valid pfn_sb before we can init a dev_pagemap */
849 	return __nvdimm_setup_pfn(nd_pfn, pgmap);
850 }
851 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
852