xref: /openbmc/linux/drivers/nvdimm/pfn_devs.c (revision 58f9d806)
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, "trace\n");
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, "result: %zd wrote: %s%s", rc, buf,
98 			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, "result: %zd wrote: %s%s", rc, buf,
148 			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, "result: %zd wrote: %s%s", rc, buf,
175 			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, "result: %zd wrote: %s%s", rc, buf,
205 			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 static umode_t pfn_visible(struct kobject *kobj, struct attribute *a, int n)
286 {
287 	if (a == &dev_attr_resource.attr)
288 		return 0400;
289 	return a->mode;
290 }
291 
292 struct attribute_group nd_pfn_attribute_group = {
293 	.attrs = nd_pfn_attributes,
294 	.is_visible = pfn_visible,
295 };
296 
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,
301 	NULL,
302 };
303 
304 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
305 		struct nd_namespace_common *ndns)
306 {
307 	struct device *dev;
308 
309 	if (!nd_pfn)
310 		return NULL;
311 
312 	nd_pfn->mode = PFN_MODE_NONE;
313 	nd_pfn->align = PFN_DEFAULT_ALIGNMENT;
314 	dev = &nd_pfn->dev;
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));
319 		put_device(dev);
320 		return NULL;
321 	}
322 	return dev;
323 }
324 
325 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
326 {
327 	struct nd_pfn *nd_pfn;
328 	struct device *dev;
329 
330 	nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
331 	if (!nd_pfn)
332 		return NULL;
333 
334 	nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
335 	if (nd_pfn->id < 0) {
336 		kfree(nd_pfn);
337 		return NULL;
338 	}
339 
340 	dev = &nd_pfn->dev;
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;
345 
346 	return nd_pfn;
347 }
348 
349 struct device *nd_pfn_create(struct nd_region *nd_region)
350 {
351 	struct nd_pfn *nd_pfn;
352 	struct device *dev;
353 
354 	if (!is_memory(&nd_region->dev))
355 		return NULL;
356 
357 	nd_pfn = nd_pfn_alloc(nd_region);
358 	dev = nd_pfn_devinit(nd_pfn, NULL);
359 
360 	__nd_device_register(dev);
361 	return dev;
362 }
363 
364 /*
365  * nd_pfn_clear_memmap_errors() clears any errors in the volatile memmap
366  * space associated with the namespace. If the memmap is set to DRAM, then
367  * this is a no-op. Since the memmap area is freshly initialized during
368  * probe, we have an opportunity to clear any badblocks in this area.
369  */
370 static int nd_pfn_clear_memmap_errors(struct nd_pfn *nd_pfn)
371 {
372 	struct nd_region *nd_region = to_nd_region(nd_pfn->dev.parent);
373 	struct nd_namespace_common *ndns = nd_pfn->ndns;
374 	void *zero_page = page_address(ZERO_PAGE(0));
375 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
376 	int num_bad, meta_num, rc, bb_present;
377 	sector_t first_bad, meta_start;
378 	struct nd_namespace_io *nsio;
379 
380 	if (nd_pfn->mode != PFN_MODE_PMEM)
381 		return 0;
382 
383 	nsio = to_nd_namespace_io(&ndns->dev);
384 	meta_start = (SZ_4K + sizeof(*pfn_sb)) >> 9;
385 	meta_num = (le64_to_cpu(pfn_sb->dataoff) >> 9) - meta_start;
386 
387 	do {
388 		unsigned long zero_len;
389 		u64 nsoff;
390 
391 		bb_present = badblocks_check(&nd_region->bb, meta_start,
392 				meta_num, &first_bad, &num_bad);
393 		if (bb_present) {
394 			dev_dbg(&nd_pfn->dev, "meta: %x badblocks at %lx\n",
395 					num_bad, first_bad);
396 			nsoff = ALIGN_DOWN((nd_region->ndr_start
397 					+ (first_bad << 9)) - nsio->res.start,
398 					PAGE_SIZE);
399 			zero_len = ALIGN(num_bad << 9, PAGE_SIZE);
400 			while (zero_len) {
401 				unsigned long chunk = min(zero_len, PAGE_SIZE);
402 
403 				rc = nvdimm_write_bytes(ndns, nsoff, zero_page,
404 							chunk, 0);
405 				if (rc)
406 					break;
407 
408 				zero_len -= chunk;
409 				nsoff += chunk;
410 			}
411 			if (rc) {
412 				dev_err(&nd_pfn->dev,
413 					"error clearing %x badblocks at %lx\n",
414 					num_bad, first_bad);
415 				return rc;
416 			}
417 		}
418 	} while (bb_present);
419 
420 	return 0;
421 }
422 
423 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
424 {
425 	u64 checksum, offset;
426 	enum nd_pfn_mode mode;
427 	struct nd_namespace_io *nsio;
428 	unsigned long align, start_pad;
429 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
430 	struct nd_namespace_common *ndns = nd_pfn->ndns;
431 	const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
432 
433 	if (!pfn_sb || !ndns)
434 		return -ENODEV;
435 
436 	if (!is_memory(nd_pfn->dev.parent))
437 		return -ENODEV;
438 
439 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
440 		return -ENXIO;
441 
442 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
443 		return -ENODEV;
444 
445 	checksum = le64_to_cpu(pfn_sb->checksum);
446 	pfn_sb->checksum = 0;
447 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
448 		return -ENODEV;
449 	pfn_sb->checksum = cpu_to_le64(checksum);
450 
451 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
452 		return -ENODEV;
453 
454 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
455 		pfn_sb->start_pad = 0;
456 		pfn_sb->end_trunc = 0;
457 	}
458 
459 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
460 		pfn_sb->align = 0;
461 
462 	switch (le32_to_cpu(pfn_sb->mode)) {
463 	case PFN_MODE_RAM:
464 	case PFN_MODE_PMEM:
465 		break;
466 	default:
467 		return -ENXIO;
468 	}
469 
470 	align = le32_to_cpu(pfn_sb->align);
471 	offset = le64_to_cpu(pfn_sb->dataoff);
472 	start_pad = le32_to_cpu(pfn_sb->start_pad);
473 	if (align == 0)
474 		align = 1UL << ilog2(offset);
475 	mode = le32_to_cpu(pfn_sb->mode);
476 
477 	if (!nd_pfn->uuid) {
478 		/*
479 		 * When probing a namepace via nd_pfn_probe() the uuid
480 		 * is NULL (see: nd_pfn_devinit()) we init settings from
481 		 * pfn_sb
482 		 */
483 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
484 		if (!nd_pfn->uuid)
485 			return -ENOMEM;
486 		nd_pfn->align = align;
487 		nd_pfn->mode = mode;
488 	} else {
489 		/*
490 		 * When probing a pfn / dax instance we validate the
491 		 * live settings against the pfn_sb
492 		 */
493 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
494 			return -ENODEV;
495 
496 		/*
497 		 * If the uuid validates, but other settings mismatch
498 		 * return EINVAL because userspace has managed to change
499 		 * the configuration without specifying new
500 		 * identification.
501 		 */
502 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
503 			dev_err(&nd_pfn->dev,
504 					"init failed, settings mismatch\n");
505 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
506 					nd_pfn->align, align, nd_pfn->mode,
507 					mode);
508 			return -EINVAL;
509 		}
510 	}
511 
512 	if (align > nvdimm_namespace_capacity(ndns)) {
513 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
514 				align, nvdimm_namespace_capacity(ndns));
515 		return -EINVAL;
516 	}
517 
518 	/*
519 	 * These warnings are verbose because they can only trigger in
520 	 * the case where the physical address alignment of the
521 	 * namespace has changed since the pfn superblock was
522 	 * established.
523 	 */
524 	nsio = to_nd_namespace_io(&ndns->dev);
525 	if (offset >= resource_size(&nsio->res)) {
526 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
527 				dev_name(&ndns->dev));
528 		return -EBUSY;
529 	}
530 
531 	if ((align && !IS_ALIGNED(nsio->res.start + offset + start_pad, align))
532 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
533 		dev_err(&nd_pfn->dev,
534 				"bad offset: %#llx dax disabled align: %#lx\n",
535 				offset, align);
536 		return -ENXIO;
537 	}
538 
539 	return nd_pfn_clear_memmap_errors(nd_pfn);
540 }
541 EXPORT_SYMBOL(nd_pfn_validate);
542 
543 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
544 {
545 	int rc;
546 	struct nd_pfn *nd_pfn;
547 	struct device *pfn_dev;
548 	struct nd_pfn_sb *pfn_sb;
549 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
550 
551 	if (ndns->force_raw)
552 		return -ENODEV;
553 
554 	switch (ndns->claim_class) {
555 	case NVDIMM_CCLASS_NONE:
556 	case NVDIMM_CCLASS_PFN:
557 		break;
558 	default:
559 		return -ENODEV;
560 	}
561 
562 	nvdimm_bus_lock(&ndns->dev);
563 	nd_pfn = nd_pfn_alloc(nd_region);
564 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
565 	nvdimm_bus_unlock(&ndns->dev);
566 	if (!pfn_dev)
567 		return -ENOMEM;
568 	pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
569 	nd_pfn = to_nd_pfn(pfn_dev);
570 	nd_pfn->pfn_sb = pfn_sb;
571 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
572 	dev_dbg(dev, "pfn: %s\n", rc == 0 ? dev_name(pfn_dev) : "<none>");
573 	if (rc < 0) {
574 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
575 		put_device(pfn_dev);
576 	} else
577 		__nd_device_register(pfn_dev);
578 
579 	return rc;
580 }
581 EXPORT_SYMBOL(nd_pfn_probe);
582 
583 /*
584  * We hotplug memory at section granularity, pad the reserved area from
585  * the previous section base to the namespace base address.
586  */
587 static unsigned long init_altmap_base(resource_size_t base)
588 {
589 	unsigned long base_pfn = PHYS_PFN(base);
590 
591 	return PFN_SECTION_ALIGN_DOWN(base_pfn);
592 }
593 
594 static unsigned long init_altmap_reserve(resource_size_t base)
595 {
596 	unsigned long reserve = PHYS_PFN(SZ_8K);
597 	unsigned long base_pfn = PHYS_PFN(base);
598 
599 	reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
600 	return reserve;
601 }
602 
603 static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
604 {
605 	struct resource *res = &pgmap->res;
606 	struct vmem_altmap *altmap = &pgmap->altmap;
607 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
608 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
609 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
610 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
611 	struct nd_namespace_common *ndns = nd_pfn->ndns;
612 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
613 	resource_size_t base = nsio->res.start + start_pad;
614 	struct vmem_altmap __altmap = {
615 		.base_pfn = init_altmap_base(base),
616 		.reserve = init_altmap_reserve(base),
617 	};
618 
619 	memcpy(res, &nsio->res, sizeof(*res));
620 	res->start += start_pad;
621 	res->end -= end_trunc;
622 
623 	if (nd_pfn->mode == PFN_MODE_RAM) {
624 		if (offset < SZ_8K)
625 			return -EINVAL;
626 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
627 		pgmap->altmap_valid = false;
628 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
629 		nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
630 					- offset) / PAGE_SIZE);
631 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
632 			dev_info(&nd_pfn->dev,
633 					"number of pfns truncated from %lld to %ld\n",
634 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
635 					nd_pfn->npfns);
636 		memcpy(altmap, &__altmap, sizeof(*altmap));
637 		altmap->free = PHYS_PFN(offset - SZ_8K);
638 		altmap->alloc = 0;
639 		pgmap->altmap_valid = true;
640 	} else
641 		return -ENXIO;
642 
643 	return 0;
644 }
645 
646 static u64 phys_pmem_align_down(struct nd_pfn *nd_pfn, u64 phys)
647 {
648 	return min_t(u64, PHYS_SECTION_ALIGN_DOWN(phys),
649 			ALIGN_DOWN(phys, nd_pfn->align));
650 }
651 
652 /*
653  * Check if pmem collides with 'System RAM', or other regions when
654  * section aligned.  Trim it accordingly.
655  */
656 static void trim_pfn_device(struct nd_pfn *nd_pfn, u32 *start_pad, u32 *end_trunc)
657 {
658 	struct nd_namespace_common *ndns = nd_pfn->ndns;
659 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
660 	struct nd_region *nd_region = to_nd_region(nd_pfn->dev.parent);
661 	const resource_size_t start = nsio->res.start;
662 	const resource_size_t end = start + resource_size(&nsio->res);
663 	resource_size_t adjust, size;
664 
665 	*start_pad = 0;
666 	*end_trunc = 0;
667 
668 	adjust = start - PHYS_SECTION_ALIGN_DOWN(start);
669 	size = resource_size(&nsio->res) + adjust;
670 	if (region_intersects(start - adjust, size, IORESOURCE_SYSTEM_RAM,
671 				IORES_DESC_NONE) == REGION_MIXED
672 			|| nd_region_conflict(nd_region, start - adjust, size))
673 		*start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
674 
675 	/* Now check that end of the range does not collide. */
676 	adjust = PHYS_SECTION_ALIGN_UP(end) - end;
677 	size = resource_size(&nsio->res) + adjust;
678 	if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
679 				IORES_DESC_NONE) == REGION_MIXED
680 			|| !IS_ALIGNED(end, nd_pfn->align)
681 			|| nd_region_conflict(nd_region, start, size + adjust))
682 		*end_trunc = end - phys_pmem_align_down(nd_pfn, end);
683 }
684 
685 static int nd_pfn_init(struct nd_pfn *nd_pfn)
686 {
687 	u32 dax_label_reserve = is_nd_dax(&nd_pfn->dev) ? SZ_128K : 0;
688 	struct nd_namespace_common *ndns = nd_pfn->ndns;
689 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
690 	resource_size_t start, size;
691 	struct nd_region *nd_region;
692 	u32 start_pad, end_trunc;
693 	struct nd_pfn_sb *pfn_sb;
694 	unsigned long npfns;
695 	phys_addr_t offset;
696 	const char *sig;
697 	u64 checksum;
698 	int rc;
699 
700 	pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
701 	if (!pfn_sb)
702 		return -ENOMEM;
703 
704 	nd_pfn->pfn_sb = pfn_sb;
705 	if (is_nd_dax(&nd_pfn->dev))
706 		sig = DAX_SIG;
707 	else
708 		sig = PFN_SIG;
709 	rc = nd_pfn_validate(nd_pfn, sig);
710 	if (rc != -ENODEV)
711 		return rc;
712 
713 	/* no info block, do init */;
714 	nd_region = to_nd_region(nd_pfn->dev.parent);
715 	if (nd_region->ro) {
716 		dev_info(&nd_pfn->dev,
717 				"%s is read-only, unable to init metadata\n",
718 				dev_name(&nd_region->dev));
719 		return -ENXIO;
720 	}
721 
722 	memset(pfn_sb, 0, sizeof(*pfn_sb));
723 
724 	trim_pfn_device(nd_pfn, &start_pad, &end_trunc);
725 	if (start_pad + end_trunc)
726 		dev_info(&nd_pfn->dev, "%s alignment collision, truncate %d bytes\n",
727 				dev_name(&ndns->dev), start_pad + end_trunc);
728 
729 	/*
730 	 * Note, we use 64 here for the standard size of struct page,
731 	 * debugging options may cause it to be larger in which case the
732 	 * implementation will limit the pfns advertised through
733 	 * ->direct_access() to those that are included in the memmap.
734 	 */
735 	start = nsio->res.start + start_pad;
736 	size = resource_size(&nsio->res);
737 	npfns = PFN_SECTION_ALIGN_UP((size - start_pad - end_trunc - SZ_8K)
738 			/ PAGE_SIZE);
739 	if (nd_pfn->mode == PFN_MODE_PMEM) {
740 		/*
741 		 * The altmap should be padded out to the block size used
742 		 * when populating the vmemmap. This *should* be equal to
743 		 * PMD_SIZE for most architectures.
744 		 */
745 		offset = ALIGN(start + SZ_8K + 64 * npfns + dax_label_reserve,
746 				max(nd_pfn->align, PMD_SIZE)) - start;
747 	} else if (nd_pfn->mode == PFN_MODE_RAM)
748 		offset = ALIGN(start + SZ_8K + dax_label_reserve,
749 				nd_pfn->align) - start;
750 	else
751 		return -ENXIO;
752 
753 	if (offset + start_pad + end_trunc >= size) {
754 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
755 				dev_name(&ndns->dev));
756 		return -ENXIO;
757 	}
758 
759 	npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
760 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
761 	pfn_sb->dataoff = cpu_to_le64(offset);
762 	pfn_sb->npfns = cpu_to_le64(npfns);
763 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
764 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
765 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
766 	pfn_sb->version_major = cpu_to_le16(1);
767 	pfn_sb->version_minor = cpu_to_le16(2);
768 	pfn_sb->start_pad = cpu_to_le32(start_pad);
769 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
770 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
771 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
772 	pfn_sb->checksum = cpu_to_le64(checksum);
773 
774 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
775 }
776 
777 /*
778  * Determine the effective resource range and vmem_altmap from an nd_pfn
779  * instance.
780  */
781 int nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
782 {
783 	int rc;
784 
785 	if (!nd_pfn->uuid || !nd_pfn->ndns)
786 		return -ENODEV;
787 
788 	rc = nd_pfn_init(nd_pfn);
789 	if (rc)
790 		return rc;
791 
792 	/* we need a valid pfn_sb before we can init a dev_pagemap */
793 	return __nvdimm_setup_pfn(nd_pfn, pgmap);
794 }
795 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
796