xref: /openbmc/linux/drivers/nvdimm/pfn_devs.c (revision 0d07cf5e)
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/genhd.h>
9 #include <linux/sizes.h>
10 #include <linux/slab.h>
11 #include <linux/fs.h>
12 #include <linux/mm.h>
13 #include "nd-core.h"
14 #include "pfn.h"
15 #include "nd.h"
16 
17 static void nd_pfn_release(struct device *dev)
18 {
19 	struct nd_region *nd_region = to_nd_region(dev->parent);
20 	struct nd_pfn *nd_pfn = to_nd_pfn(dev);
21 
22 	dev_dbg(dev, "trace\n");
23 	nd_detach_ndns(&nd_pfn->dev, &nd_pfn->ndns);
24 	ida_simple_remove(&nd_region->pfn_ida, nd_pfn->id);
25 	kfree(nd_pfn->uuid);
26 	kfree(nd_pfn);
27 }
28 
29 static struct device_type nd_pfn_device_type = {
30 	.name = "nd_pfn",
31 	.release = nd_pfn_release,
32 };
33 
34 bool is_nd_pfn(struct device *dev)
35 {
36 	return dev ? dev->type == &nd_pfn_device_type : false;
37 }
38 EXPORT_SYMBOL(is_nd_pfn);
39 
40 struct nd_pfn *to_nd_pfn(struct device *dev)
41 {
42 	struct nd_pfn *nd_pfn = container_of(dev, struct nd_pfn, dev);
43 
44 	WARN_ON(!is_nd_pfn(dev));
45 	return nd_pfn;
46 }
47 EXPORT_SYMBOL(to_nd_pfn);
48 
49 static ssize_t mode_show(struct device *dev,
50 		struct device_attribute *attr, char *buf)
51 {
52 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
53 
54 	switch (nd_pfn->mode) {
55 	case PFN_MODE_RAM:
56 		return sprintf(buf, "ram\n");
57 	case PFN_MODE_PMEM:
58 		return sprintf(buf, "pmem\n");
59 	default:
60 		return sprintf(buf, "none\n");
61 	}
62 }
63 
64 static ssize_t mode_store(struct device *dev,
65 		struct device_attribute *attr, const char *buf, size_t len)
66 {
67 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
68 	ssize_t rc = 0;
69 
70 	device_lock(dev);
71 	nvdimm_bus_lock(dev);
72 	if (dev->driver)
73 		rc = -EBUSY;
74 	else {
75 		size_t n = len - 1;
76 
77 		if (strncmp(buf, "pmem\n", n) == 0
78 				|| strncmp(buf, "pmem", n) == 0) {
79 			nd_pfn->mode = PFN_MODE_PMEM;
80 		} else if (strncmp(buf, "ram\n", n) == 0
81 				|| strncmp(buf, "ram", n) == 0)
82 			nd_pfn->mode = PFN_MODE_RAM;
83 		else if (strncmp(buf, "none\n", n) == 0
84 				|| strncmp(buf, "none", n) == 0)
85 			nd_pfn->mode = PFN_MODE_NONE;
86 		else
87 			rc = -EINVAL;
88 	}
89 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
90 			buf[len - 1] == '\n' ? "" : "\n");
91 	nvdimm_bus_unlock(dev);
92 	device_unlock(dev);
93 
94 	return rc ? rc : len;
95 }
96 static DEVICE_ATTR_RW(mode);
97 
98 static ssize_t align_show(struct device *dev,
99 		struct device_attribute *attr, char *buf)
100 {
101 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
102 
103 	return sprintf(buf, "%ld\n", nd_pfn->align);
104 }
105 
106 static const unsigned long *nd_pfn_supported_alignments(void)
107 {
108 	/*
109 	 * This needs to be a non-static variable because the *_SIZE
110 	 * macros aren't always constants.
111 	 */
112 	const unsigned long supported_alignments[] = {
113 		PAGE_SIZE,
114 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
115 		HPAGE_PMD_SIZE,
116 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD
117 		HPAGE_PUD_SIZE,
118 #endif
119 #endif
120 		0,
121 	};
122 	static unsigned long data[ARRAY_SIZE(supported_alignments)];
123 
124 	memcpy(data, supported_alignments, sizeof(data));
125 
126 	return data;
127 }
128 
129 static ssize_t align_store(struct device *dev,
130 		struct device_attribute *attr, const char *buf, size_t len)
131 {
132 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
133 	ssize_t rc;
134 
135 	device_lock(dev);
136 	nvdimm_bus_lock(dev);
137 	rc = nd_size_select_store(dev, buf, &nd_pfn->align,
138 			nd_pfn_supported_alignments());
139 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
140 			buf[len - 1] == '\n' ? "" : "\n");
141 	nvdimm_bus_unlock(dev);
142 	device_unlock(dev);
143 
144 	return rc ? rc : len;
145 }
146 static DEVICE_ATTR_RW(align);
147 
148 static ssize_t uuid_show(struct device *dev,
149 		struct device_attribute *attr, char *buf)
150 {
151 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
152 
153 	if (nd_pfn->uuid)
154 		return sprintf(buf, "%pUb\n", nd_pfn->uuid);
155 	return sprintf(buf, "\n");
156 }
157 
158 static ssize_t uuid_store(struct device *dev,
159 		struct device_attribute *attr, const char *buf, size_t len)
160 {
161 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
162 	ssize_t rc;
163 
164 	device_lock(dev);
165 	rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
166 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
167 			buf[len - 1] == '\n' ? "" : "\n");
168 	device_unlock(dev);
169 
170 	return rc ? rc : len;
171 }
172 static DEVICE_ATTR_RW(uuid);
173 
174 static ssize_t namespace_show(struct device *dev,
175 		struct device_attribute *attr, char *buf)
176 {
177 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
178 	ssize_t rc;
179 
180 	nvdimm_bus_lock(dev);
181 	rc = sprintf(buf, "%s\n", nd_pfn->ndns
182 			? dev_name(&nd_pfn->ndns->dev) : "");
183 	nvdimm_bus_unlock(dev);
184 	return rc;
185 }
186 
187 static ssize_t namespace_store(struct device *dev,
188 		struct device_attribute *attr, const char *buf, size_t len)
189 {
190 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
191 	ssize_t rc;
192 
193 	device_lock(dev);
194 	nvdimm_bus_lock(dev);
195 	rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
196 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
197 			buf[len - 1] == '\n' ? "" : "\n");
198 	nvdimm_bus_unlock(dev);
199 	device_unlock(dev);
200 
201 	return rc;
202 }
203 static DEVICE_ATTR_RW(namespace);
204 
205 static ssize_t resource_show(struct device *dev,
206 		struct device_attribute *attr, char *buf)
207 {
208 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
209 	ssize_t rc;
210 
211 	device_lock(dev);
212 	if (dev->driver) {
213 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
214 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
215 		struct nd_namespace_common *ndns = nd_pfn->ndns;
216 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
217 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
218 
219 		rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
220 				+ start_pad + offset);
221 	} else {
222 		/* no address to convey if the pfn instance is disabled */
223 		rc = -ENXIO;
224 	}
225 	device_unlock(dev);
226 
227 	return rc;
228 }
229 static DEVICE_ATTR_RO(resource);
230 
231 static ssize_t size_show(struct device *dev,
232 		struct device_attribute *attr, char *buf)
233 {
234 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
235 	ssize_t rc;
236 
237 	device_lock(dev);
238 	if (dev->driver) {
239 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
240 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
241 		struct nd_namespace_common *ndns = nd_pfn->ndns;
242 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
243 		u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
244 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
245 
246 		rc = sprintf(buf, "%llu\n", (unsigned long long)
247 				resource_size(&nsio->res) - start_pad
248 				- end_trunc - offset);
249 	} else {
250 		/* no size to convey if the pfn instance is disabled */
251 		rc = -ENXIO;
252 	}
253 	device_unlock(dev);
254 
255 	return rc;
256 }
257 static DEVICE_ATTR_RO(size);
258 
259 static ssize_t supported_alignments_show(struct device *dev,
260 		struct device_attribute *attr, char *buf)
261 {
262 	return nd_size_select_show(0, nd_pfn_supported_alignments(), buf);
263 }
264 static DEVICE_ATTR_RO(supported_alignments);
265 
266 static struct attribute *nd_pfn_attributes[] = {
267 	&dev_attr_mode.attr,
268 	&dev_attr_namespace.attr,
269 	&dev_attr_uuid.attr,
270 	&dev_attr_align.attr,
271 	&dev_attr_resource.attr,
272 	&dev_attr_size.attr,
273 	&dev_attr_supported_alignments.attr,
274 	NULL,
275 };
276 
277 static umode_t pfn_visible(struct kobject *kobj, struct attribute *a, int n)
278 {
279 	if (a == &dev_attr_resource.attr)
280 		return 0400;
281 	return a->mode;
282 }
283 
284 struct attribute_group nd_pfn_attribute_group = {
285 	.attrs = nd_pfn_attributes,
286 	.is_visible = pfn_visible,
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;
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, "failed, already claimed by %s\n",
310 				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 /*
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 	do {
380 		unsigned long zero_len;
381 		u64 nsoff;
382 
383 		bb_present = badblocks_check(&nd_region->bb, meta_start,
384 				meta_num, &first_bad, &num_bad);
385 		if (bb_present) {
386 			dev_dbg(&nd_pfn->dev, "meta: %x badblocks at %llx\n",
387 					num_bad, first_bad);
388 			nsoff = ALIGN_DOWN((nd_region->ndr_start
389 					+ (first_bad << 9)) - nsio->res.start,
390 					PAGE_SIZE);
391 			zero_len = ALIGN(num_bad << 9, PAGE_SIZE);
392 			while (zero_len) {
393 				unsigned long chunk = min(zero_len, PAGE_SIZE);
394 
395 				rc = nvdimm_write_bytes(ndns, nsoff, zero_page,
396 							chunk, 0);
397 				if (rc)
398 					break;
399 
400 				zero_len -= chunk;
401 				nsoff += chunk;
402 			}
403 			if (rc) {
404 				dev_err(&nd_pfn->dev,
405 					"error clearing %x badblocks at %llx\n",
406 					num_bad, first_bad);
407 				return rc;
408 			}
409 		}
410 	} while (bb_present);
411 
412 	return 0;
413 }
414 
415 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
416 {
417 	u64 checksum, offset;
418 	enum nd_pfn_mode mode;
419 	struct nd_namespace_io *nsio;
420 	unsigned long align, start_pad;
421 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
422 	struct nd_namespace_common *ndns = nd_pfn->ndns;
423 	const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
424 
425 	if (!pfn_sb || !ndns)
426 		return -ENODEV;
427 
428 	if (!is_memory(nd_pfn->dev.parent))
429 		return -ENODEV;
430 
431 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
432 		return -ENXIO;
433 
434 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
435 		return -ENODEV;
436 
437 	checksum = le64_to_cpu(pfn_sb->checksum);
438 	pfn_sb->checksum = 0;
439 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
440 		return -ENODEV;
441 	pfn_sb->checksum = cpu_to_le64(checksum);
442 
443 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
444 		return -ENODEV;
445 
446 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
447 		pfn_sb->start_pad = 0;
448 		pfn_sb->end_trunc = 0;
449 	}
450 
451 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
452 		pfn_sb->align = 0;
453 
454 	switch (le32_to_cpu(pfn_sb->mode)) {
455 	case PFN_MODE_RAM:
456 	case PFN_MODE_PMEM:
457 		break;
458 	default:
459 		return -ENXIO;
460 	}
461 
462 	align = le32_to_cpu(pfn_sb->align);
463 	offset = le64_to_cpu(pfn_sb->dataoff);
464 	start_pad = le32_to_cpu(pfn_sb->start_pad);
465 	if (align == 0)
466 		align = 1UL << ilog2(offset);
467 	mode = le32_to_cpu(pfn_sb->mode);
468 
469 	if (!nd_pfn->uuid) {
470 		/*
471 		 * When probing a namepace via nd_pfn_probe() the uuid
472 		 * is NULL (see: nd_pfn_devinit()) we init settings from
473 		 * pfn_sb
474 		 */
475 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
476 		if (!nd_pfn->uuid)
477 			return -ENOMEM;
478 		nd_pfn->align = align;
479 		nd_pfn->mode = mode;
480 	} else {
481 		/*
482 		 * When probing a pfn / dax instance we validate the
483 		 * live settings against the pfn_sb
484 		 */
485 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
486 			return -ENODEV;
487 
488 		/*
489 		 * If the uuid validates, but other settings mismatch
490 		 * return EINVAL because userspace has managed to change
491 		 * the configuration without specifying new
492 		 * identification.
493 		 */
494 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
495 			dev_err(&nd_pfn->dev,
496 					"init failed, settings mismatch\n");
497 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
498 					nd_pfn->align, align, nd_pfn->mode,
499 					mode);
500 			return -EINVAL;
501 		}
502 	}
503 
504 	if (align > nvdimm_namespace_capacity(ndns)) {
505 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
506 				align, nvdimm_namespace_capacity(ndns));
507 		return -EINVAL;
508 	}
509 
510 	/*
511 	 * These warnings are verbose because they can only trigger in
512 	 * the case where the physical address alignment of the
513 	 * namespace has changed since the pfn superblock was
514 	 * established.
515 	 */
516 	nsio = to_nd_namespace_io(&ndns->dev);
517 	if (offset >= resource_size(&nsio->res)) {
518 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
519 				dev_name(&ndns->dev));
520 		return -EBUSY;
521 	}
522 
523 	if ((align && !IS_ALIGNED(nsio->res.start + offset + start_pad, align))
524 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
525 		dev_err(&nd_pfn->dev,
526 				"bad offset: %#llx dax disabled align: %#lx\n",
527 				offset, align);
528 		return -ENXIO;
529 	}
530 
531 	return nd_pfn_clear_memmap_errors(nd_pfn);
532 }
533 EXPORT_SYMBOL(nd_pfn_validate);
534 
535 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
536 {
537 	int rc;
538 	struct nd_pfn *nd_pfn;
539 	struct device *pfn_dev;
540 	struct nd_pfn_sb *pfn_sb;
541 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
542 
543 	if (ndns->force_raw)
544 		return -ENODEV;
545 
546 	switch (ndns->claim_class) {
547 	case NVDIMM_CCLASS_NONE:
548 	case NVDIMM_CCLASS_PFN:
549 		break;
550 	default:
551 		return -ENODEV;
552 	}
553 
554 	nvdimm_bus_lock(&ndns->dev);
555 	nd_pfn = nd_pfn_alloc(nd_region);
556 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
557 	nvdimm_bus_unlock(&ndns->dev);
558 	if (!pfn_dev)
559 		return -ENOMEM;
560 	pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
561 	nd_pfn = to_nd_pfn(pfn_dev);
562 	nd_pfn->pfn_sb = pfn_sb;
563 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
564 	dev_dbg(dev, "pfn: %s\n", rc == 0 ? dev_name(pfn_dev) : "<none>");
565 	if (rc < 0) {
566 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
567 		put_device(pfn_dev);
568 	} else
569 		__nd_device_register(pfn_dev);
570 
571 	return rc;
572 }
573 EXPORT_SYMBOL(nd_pfn_probe);
574 
575 static u32 info_block_reserve(void)
576 {
577 	return ALIGN(SZ_8K, PAGE_SIZE);
578 }
579 
580 /*
581  * We hotplug memory at section granularity, pad the reserved area from
582  * the previous section base to the namespace base address.
583  */
584 static unsigned long init_altmap_base(resource_size_t base)
585 {
586 	unsigned long base_pfn = PHYS_PFN(base);
587 
588 	return PFN_SECTION_ALIGN_DOWN(base_pfn);
589 }
590 
591 static unsigned long init_altmap_reserve(resource_size_t base)
592 {
593 	unsigned long reserve = info_block_reserve() >> PAGE_SHIFT;
594 	unsigned long base_pfn = PHYS_PFN(base);
595 
596 	reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
597 	return reserve;
598 }
599 
600 static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
601 {
602 	struct resource *res = &pgmap->res;
603 	struct vmem_altmap *altmap = &pgmap->altmap;
604 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
605 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
606 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
607 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
608 	u32 reserve = info_block_reserve();
609 	struct nd_namespace_common *ndns = nd_pfn->ndns;
610 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
611 	resource_size_t base = nsio->res.start + start_pad;
612 	struct vmem_altmap __altmap = {
613 		.base_pfn = init_altmap_base(base),
614 		.reserve = init_altmap_reserve(base),
615 	};
616 
617 	memcpy(res, &nsio->res, sizeof(*res));
618 	res->start += start_pad;
619 	res->end -= end_trunc;
620 
621 	if (nd_pfn->mode == PFN_MODE_RAM) {
622 		if (offset < reserve)
623 			return -EINVAL;
624 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
625 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
626 		nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
627 					- offset) / PAGE_SIZE);
628 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
629 			dev_info(&nd_pfn->dev,
630 					"number of pfns truncated from %lld to %ld\n",
631 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
632 					nd_pfn->npfns);
633 		memcpy(altmap, &__altmap, sizeof(*altmap));
634 		altmap->free = PHYS_PFN(offset - reserve);
635 		altmap->alloc = 0;
636 		pgmap->flags |= PGMAP_ALTMAP_VALID;
637 	} else
638 		return -ENXIO;
639 
640 	return 0;
641 }
642 
643 static u64 phys_pmem_align_down(struct nd_pfn *nd_pfn, u64 phys)
644 {
645 	return min_t(u64, PHYS_SECTION_ALIGN_DOWN(phys),
646 			ALIGN_DOWN(phys, nd_pfn->align));
647 }
648 
649 /*
650  * Check if pmem collides with 'System RAM', or other regions when
651  * section aligned.  Trim it accordingly.
652  */
653 static void trim_pfn_device(struct nd_pfn *nd_pfn, u32 *start_pad, u32 *end_trunc)
654 {
655 	struct nd_namespace_common *ndns = nd_pfn->ndns;
656 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
657 	struct nd_region *nd_region = to_nd_region(nd_pfn->dev.parent);
658 	const resource_size_t start = nsio->res.start;
659 	const resource_size_t end = start + resource_size(&nsio->res);
660 	resource_size_t adjust, size;
661 
662 	*start_pad = 0;
663 	*end_trunc = 0;
664 
665 	adjust = start - PHYS_SECTION_ALIGN_DOWN(start);
666 	size = resource_size(&nsio->res) + adjust;
667 	if (region_intersects(start - adjust, size, IORESOURCE_SYSTEM_RAM,
668 				IORES_DESC_NONE) == REGION_MIXED
669 			|| nd_region_conflict(nd_region, start - adjust, size))
670 		*start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
671 
672 	/* Now check that end of the range does not collide. */
673 	adjust = PHYS_SECTION_ALIGN_UP(end) - end;
674 	size = resource_size(&nsio->res) + adjust;
675 	if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
676 				IORES_DESC_NONE) == REGION_MIXED
677 			|| !IS_ALIGNED(end, nd_pfn->align)
678 			|| nd_region_conflict(nd_region, start, size))
679 		*end_trunc = end - phys_pmem_align_down(nd_pfn, end);
680 }
681 
682 static int nd_pfn_init(struct nd_pfn *nd_pfn)
683 {
684 	struct nd_namespace_common *ndns = nd_pfn->ndns;
685 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
686 	u32 start_pad, end_trunc, reserve = info_block_reserve();
687 	resource_size_t start, size;
688 	struct nd_region *nd_region;
689 	struct nd_pfn_sb *pfn_sb;
690 	unsigned long npfns;
691 	phys_addr_t offset;
692 	const char *sig;
693 	u64 checksum;
694 	int rc;
695 
696 	pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
697 	if (!pfn_sb)
698 		return -ENOMEM;
699 
700 	nd_pfn->pfn_sb = pfn_sb;
701 	if (is_nd_dax(&nd_pfn->dev))
702 		sig = DAX_SIG;
703 	else
704 		sig = PFN_SIG;
705 	rc = nd_pfn_validate(nd_pfn, sig);
706 	if (rc != -ENODEV)
707 		return rc;
708 
709 	/* no info block, do init */;
710 	nd_region = to_nd_region(nd_pfn->dev.parent);
711 	if (nd_region->ro) {
712 		dev_info(&nd_pfn->dev,
713 				"%s is read-only, unable to init metadata\n",
714 				dev_name(&nd_region->dev));
715 		return -ENXIO;
716 	}
717 
718 	memset(pfn_sb, 0, sizeof(*pfn_sb));
719 
720 	trim_pfn_device(nd_pfn, &start_pad, &end_trunc);
721 	if (start_pad + end_trunc)
722 		dev_info(&nd_pfn->dev, "%s alignment collision, truncate %d bytes\n",
723 				dev_name(&ndns->dev), start_pad + end_trunc);
724 
725 	/*
726 	 * Note, we use 64 here for the standard size of struct page,
727 	 * debugging options may cause it to be larger in which case the
728 	 * implementation will limit the pfns advertised through
729 	 * ->direct_access() to those that are included in the memmap.
730 	 */
731 	start = nsio->res.start + start_pad;
732 	size = resource_size(&nsio->res);
733 	npfns = PFN_SECTION_ALIGN_UP((size - start_pad - end_trunc - reserve)
734 			/ PAGE_SIZE);
735 	if (nd_pfn->mode == PFN_MODE_PMEM) {
736 		/*
737 		 * The altmap should be padded out to the block size used
738 		 * when populating the vmemmap. This *should* be equal to
739 		 * PMD_SIZE for most architectures.
740 		 */
741 		offset = ALIGN(start + reserve + 64 * npfns,
742 				max(nd_pfn->align, PMD_SIZE)) - start;
743 	} else if (nd_pfn->mode == PFN_MODE_RAM)
744 		offset = ALIGN(start + reserve, nd_pfn->align) - start;
745 	else
746 		return -ENXIO;
747 
748 	if (offset + start_pad + end_trunc >= size) {
749 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
750 				dev_name(&ndns->dev));
751 		return -ENXIO;
752 	}
753 
754 	npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
755 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
756 	pfn_sb->dataoff = cpu_to_le64(offset);
757 	pfn_sb->npfns = cpu_to_le64(npfns);
758 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
759 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
760 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
761 	pfn_sb->version_major = cpu_to_le16(1);
762 	pfn_sb->version_minor = cpu_to_le16(2);
763 	pfn_sb->start_pad = cpu_to_le32(start_pad);
764 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
765 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
766 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
767 	pfn_sb->checksum = cpu_to_le64(checksum);
768 
769 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
770 }
771 
772 /*
773  * Determine the effective resource range and vmem_altmap from an nd_pfn
774  * instance.
775  */
776 int nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
777 {
778 	int rc;
779 
780 	if (!nd_pfn->uuid || !nd_pfn->ndns)
781 		return -ENODEV;
782 
783 	rc = nd_pfn_init(nd_pfn);
784 	if (rc)
785 		return rc;
786 
787 	/* we need a valid pfn_sb before we can init a dev_pagemap */
788 	return __nvdimm_setup_pfn(nd_pfn, pgmap);
789 }
790 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
791