xref: /linux/drivers/nvdimm/pfn_devs.c (revision 0889d07f3e4b171c453b2aaf2b257f9074cdf624)
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 	nd_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 	nd_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 unsigned long *nd_pfn_supported_alignments(unsigned long *alignments)
107 {
108 
109 	alignments[0] = PAGE_SIZE;
110 
111 	if (has_transparent_hugepage()) {
112 		alignments[1] = HPAGE_PMD_SIZE;
113 		if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD))
114 			alignments[2] = HPAGE_PUD_SIZE;
115 	}
116 
117 	return alignments;
118 }
119 
120 /*
121  * Use pmd mapping if supported as default alignment
122  */
123 static unsigned long nd_pfn_default_alignment(void)
124 {
125 
126 	if (has_transparent_hugepage())
127 		return HPAGE_PMD_SIZE;
128 	return PAGE_SIZE;
129 }
130 
131 static ssize_t align_store(struct device *dev,
132 		struct device_attribute *attr, const char *buf, size_t len)
133 {
134 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
135 	unsigned long aligns[MAX_NVDIMM_ALIGN] = { [0] = 0, };
136 	ssize_t rc;
137 
138 	nd_device_lock(dev);
139 	nvdimm_bus_lock(dev);
140 	rc = nd_size_select_store(dev, buf, &nd_pfn->align,
141 			nd_pfn_supported_alignments(aligns));
142 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
143 			buf[len - 1] == '\n' ? "" : "\n");
144 	nvdimm_bus_unlock(dev);
145 	nd_device_unlock(dev);
146 
147 	return rc ? rc : len;
148 }
149 static DEVICE_ATTR_RW(align);
150 
151 static ssize_t uuid_show(struct device *dev,
152 		struct device_attribute *attr, char *buf)
153 {
154 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
155 
156 	if (nd_pfn->uuid)
157 		return sprintf(buf, "%pUb\n", nd_pfn->uuid);
158 	return sprintf(buf, "\n");
159 }
160 
161 static ssize_t uuid_store(struct device *dev,
162 		struct device_attribute *attr, const char *buf, size_t len)
163 {
164 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
165 	ssize_t rc;
166 
167 	nd_device_lock(dev);
168 	rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
169 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
170 			buf[len - 1] == '\n' ? "" : "\n");
171 	nd_device_unlock(dev);
172 
173 	return rc ? rc : len;
174 }
175 static DEVICE_ATTR_RW(uuid);
176 
177 static ssize_t namespace_show(struct device *dev,
178 		struct device_attribute *attr, char *buf)
179 {
180 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
181 	ssize_t rc;
182 
183 	nvdimm_bus_lock(dev);
184 	rc = sprintf(buf, "%s\n", nd_pfn->ndns
185 			? dev_name(&nd_pfn->ndns->dev) : "");
186 	nvdimm_bus_unlock(dev);
187 	return rc;
188 }
189 
190 static ssize_t namespace_store(struct device *dev,
191 		struct device_attribute *attr, const char *buf, size_t len)
192 {
193 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
194 	ssize_t rc;
195 
196 	nd_device_lock(dev);
197 	nvdimm_bus_lock(dev);
198 	rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
199 	dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
200 			buf[len - 1] == '\n' ? "" : "\n");
201 	nvdimm_bus_unlock(dev);
202 	nd_device_unlock(dev);
203 
204 	return rc;
205 }
206 static DEVICE_ATTR_RW(namespace);
207 
208 static ssize_t resource_show(struct device *dev,
209 		struct device_attribute *attr, char *buf)
210 {
211 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
212 	ssize_t rc;
213 
214 	nd_device_lock(dev);
215 	if (dev->driver) {
216 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
217 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
218 		struct nd_namespace_common *ndns = nd_pfn->ndns;
219 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
220 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
221 
222 		rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
223 				+ start_pad + offset);
224 	} else {
225 		/* no address to convey if the pfn instance is disabled */
226 		rc = -ENXIO;
227 	}
228 	nd_device_unlock(dev);
229 
230 	return rc;
231 }
232 static DEVICE_ATTR_RO(resource);
233 
234 static ssize_t size_show(struct device *dev,
235 		struct device_attribute *attr, char *buf)
236 {
237 	struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
238 	ssize_t rc;
239 
240 	nd_device_lock(dev);
241 	if (dev->driver) {
242 		struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
243 		u64 offset = __le64_to_cpu(pfn_sb->dataoff);
244 		struct nd_namespace_common *ndns = nd_pfn->ndns;
245 		u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
246 		u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
247 		struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
248 
249 		rc = sprintf(buf, "%llu\n", (unsigned long long)
250 				resource_size(&nsio->res) - start_pad
251 				- end_trunc - offset);
252 	} else {
253 		/* no size to convey if the pfn instance is disabled */
254 		rc = -ENXIO;
255 	}
256 	nd_device_unlock(dev);
257 
258 	return rc;
259 }
260 static DEVICE_ATTR_RO(size);
261 
262 static ssize_t supported_alignments_show(struct device *dev,
263 		struct device_attribute *attr, char *buf)
264 {
265 	unsigned long aligns[MAX_NVDIMM_ALIGN] = { [0] = 0, };
266 
267 	return nd_size_select_show(0,
268 			nd_pfn_supported_alignments(aligns), buf);
269 }
270 static DEVICE_ATTR_RO(supported_alignments);
271 
272 static struct attribute *nd_pfn_attributes[] = {
273 	&dev_attr_mode.attr,
274 	&dev_attr_namespace.attr,
275 	&dev_attr_uuid.attr,
276 	&dev_attr_align.attr,
277 	&dev_attr_resource.attr,
278 	&dev_attr_size.attr,
279 	&dev_attr_supported_alignments.attr,
280 	NULL,
281 };
282 
283 static umode_t pfn_visible(struct kobject *kobj, struct attribute *a, int n)
284 {
285 	if (a == &dev_attr_resource.attr)
286 		return 0400;
287 	return a->mode;
288 }
289 
290 struct attribute_group nd_pfn_attribute_group = {
291 	.attrs = nd_pfn_attributes,
292 	.is_visible = pfn_visible,
293 };
294 
295 static const struct attribute_group *nd_pfn_attribute_groups[] = {
296 	&nd_pfn_attribute_group,
297 	&nd_device_attribute_group,
298 	&nd_numa_attribute_group,
299 	NULL,
300 };
301 
302 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
303 		struct nd_namespace_common *ndns)
304 {
305 	struct device *dev;
306 
307 	if (!nd_pfn)
308 		return NULL;
309 
310 	nd_pfn->mode = PFN_MODE_NONE;
311 	nd_pfn->align = nd_pfn_default_alignment();
312 	dev = &nd_pfn->dev;
313 	device_initialize(&nd_pfn->dev);
314 	if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
315 		dev_dbg(&ndns->dev, "failed, already claimed by %s\n",
316 				dev_name(ndns->claim));
317 		put_device(dev);
318 		return NULL;
319 	}
320 	return dev;
321 }
322 
323 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
324 {
325 	struct nd_pfn *nd_pfn;
326 	struct device *dev;
327 
328 	nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
329 	if (!nd_pfn)
330 		return NULL;
331 
332 	nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
333 	if (nd_pfn->id < 0) {
334 		kfree(nd_pfn);
335 		return NULL;
336 	}
337 
338 	dev = &nd_pfn->dev;
339 	dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
340 	dev->groups = nd_pfn_attribute_groups;
341 	dev->type = &nd_pfn_device_type;
342 	dev->parent = &nd_region->dev;
343 
344 	return nd_pfn;
345 }
346 
347 struct device *nd_pfn_create(struct nd_region *nd_region)
348 {
349 	struct nd_pfn *nd_pfn;
350 	struct device *dev;
351 
352 	if (!is_memory(&nd_region->dev))
353 		return NULL;
354 
355 	nd_pfn = nd_pfn_alloc(nd_region);
356 	dev = nd_pfn_devinit(nd_pfn, NULL);
357 
358 	__nd_device_register(dev);
359 	return dev;
360 }
361 
362 /*
363  * nd_pfn_clear_memmap_errors() clears any errors in the volatile memmap
364  * space associated with the namespace. If the memmap is set to DRAM, then
365  * this is a no-op. Since the memmap area is freshly initialized during
366  * probe, we have an opportunity to clear any badblocks in this area.
367  */
368 static int nd_pfn_clear_memmap_errors(struct nd_pfn *nd_pfn)
369 {
370 	struct nd_region *nd_region = to_nd_region(nd_pfn->dev.parent);
371 	struct nd_namespace_common *ndns = nd_pfn->ndns;
372 	void *zero_page = page_address(ZERO_PAGE(0));
373 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
374 	int num_bad, meta_num, rc, bb_present;
375 	sector_t first_bad, meta_start;
376 	struct nd_namespace_io *nsio;
377 
378 	if (nd_pfn->mode != PFN_MODE_PMEM)
379 		return 0;
380 
381 	nsio = to_nd_namespace_io(&ndns->dev);
382 	meta_start = (SZ_4K + sizeof(*pfn_sb)) >> 9;
383 	meta_num = (le64_to_cpu(pfn_sb->dataoff) >> 9) - meta_start;
384 
385 	do {
386 		unsigned long zero_len;
387 		u64 nsoff;
388 
389 		bb_present = badblocks_check(&nd_region->bb, meta_start,
390 				meta_num, &first_bad, &num_bad);
391 		if (bb_present) {
392 			dev_dbg(&nd_pfn->dev, "meta: %x badblocks at %llx\n",
393 					num_bad, first_bad);
394 			nsoff = ALIGN_DOWN((nd_region->ndr_start
395 					+ (first_bad << 9)) - nsio->res.start,
396 					PAGE_SIZE);
397 			zero_len = ALIGN(num_bad << 9, PAGE_SIZE);
398 			while (zero_len) {
399 				unsigned long chunk = min(zero_len, PAGE_SIZE);
400 
401 				rc = nvdimm_write_bytes(ndns, nsoff, zero_page,
402 							chunk, 0);
403 				if (rc)
404 					break;
405 
406 				zero_len -= chunk;
407 				nsoff += chunk;
408 			}
409 			if (rc) {
410 				dev_err(&nd_pfn->dev,
411 					"error clearing %x badblocks at %llx\n",
412 					num_bad, first_bad);
413 				return rc;
414 			}
415 		}
416 	} while (bb_present);
417 
418 	return 0;
419 }
420 
421 static bool nd_supported_alignment(unsigned long align)
422 {
423 	int i;
424 	unsigned long supported[MAX_NVDIMM_ALIGN] = { [0] = 0, };
425 
426 	if (align == 0)
427 		return false;
428 
429 	nd_pfn_supported_alignments(supported);
430 	for (i = 0; supported[i]; i++)
431 		if (align == supported[i])
432 			return true;
433 	return false;
434 }
435 
436 /**
437  * nd_pfn_validate - read and validate info-block
438  * @nd_pfn: fsdax namespace runtime state / properties
439  * @sig: 'devdax' or 'fsdax' signature
440  *
441  * Upon return the info-block buffer contents (->pfn_sb) are
442  * indeterminate when validation fails, and a coherent info-block
443  * otherwise.
444  */
445 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
446 {
447 	u64 checksum, offset;
448 	enum nd_pfn_mode mode;
449 	struct nd_namespace_io *nsio;
450 	unsigned long align, start_pad;
451 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
452 	struct nd_namespace_common *ndns = nd_pfn->ndns;
453 	const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
454 
455 	if (!pfn_sb || !ndns)
456 		return -ENODEV;
457 
458 	if (!is_memory(nd_pfn->dev.parent))
459 		return -ENODEV;
460 
461 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
462 		return -ENXIO;
463 
464 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
465 		return -ENODEV;
466 
467 	checksum = le64_to_cpu(pfn_sb->checksum);
468 	pfn_sb->checksum = 0;
469 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
470 		return -ENODEV;
471 	pfn_sb->checksum = cpu_to_le64(checksum);
472 
473 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
474 		return -ENODEV;
475 
476 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
477 		pfn_sb->start_pad = 0;
478 		pfn_sb->end_trunc = 0;
479 	}
480 
481 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
482 		pfn_sb->align = 0;
483 
484 	if (__le16_to_cpu(pfn_sb->version_minor) < 4) {
485 		pfn_sb->page_struct_size = cpu_to_le16(64);
486 		pfn_sb->page_size = cpu_to_le32(PAGE_SIZE);
487 	}
488 
489 	switch (le32_to_cpu(pfn_sb->mode)) {
490 	case PFN_MODE_RAM:
491 	case PFN_MODE_PMEM:
492 		break;
493 	default:
494 		return -ENXIO;
495 	}
496 
497 	align = le32_to_cpu(pfn_sb->align);
498 	offset = le64_to_cpu(pfn_sb->dataoff);
499 	start_pad = le32_to_cpu(pfn_sb->start_pad);
500 	if (align == 0)
501 		align = 1UL << ilog2(offset);
502 	mode = le32_to_cpu(pfn_sb->mode);
503 
504 	if ((le32_to_cpu(pfn_sb->page_size) > PAGE_SIZE) &&
505 			(mode == PFN_MODE_PMEM)) {
506 		dev_err(&nd_pfn->dev,
507 				"init failed, page size mismatch %d\n",
508 				le32_to_cpu(pfn_sb->page_size));
509 		return -EOPNOTSUPP;
510 	}
511 
512 	if ((le16_to_cpu(pfn_sb->page_struct_size) < sizeof(struct page)) &&
513 			(mode == PFN_MODE_PMEM)) {
514 		dev_err(&nd_pfn->dev,
515 				"init failed, struct page size mismatch %d\n",
516 				le16_to_cpu(pfn_sb->page_struct_size));
517 		return -EOPNOTSUPP;
518 	}
519 
520 	/*
521 	 * Check whether the we support the alignment. For Dax if the
522 	 * superblock alignment is not matching, we won't initialize
523 	 * the device.
524 	 */
525 	if (!nd_supported_alignment(align) &&
526 			!memcmp(pfn_sb->signature, DAX_SIG, PFN_SIG_LEN)) {
527 		dev_err(&nd_pfn->dev, "init failed, alignment mismatch: "
528 				"%ld:%ld\n", nd_pfn->align, align);
529 		return -EOPNOTSUPP;
530 	}
531 
532 	if (!nd_pfn->uuid) {
533 		/*
534 		 * When probing a namepace via nd_pfn_probe() the uuid
535 		 * is NULL (see: nd_pfn_devinit()) we init settings from
536 		 * pfn_sb
537 		 */
538 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
539 		if (!nd_pfn->uuid)
540 			return -ENOMEM;
541 		nd_pfn->align = align;
542 		nd_pfn->mode = mode;
543 	} else {
544 		/*
545 		 * When probing a pfn / dax instance we validate the
546 		 * live settings against the pfn_sb
547 		 */
548 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
549 			return -ENODEV;
550 
551 		/*
552 		 * If the uuid validates, but other settings mismatch
553 		 * return EINVAL because userspace has managed to change
554 		 * the configuration without specifying new
555 		 * identification.
556 		 */
557 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
558 			dev_err(&nd_pfn->dev,
559 					"init failed, settings mismatch\n");
560 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
561 					nd_pfn->align, align, nd_pfn->mode,
562 					mode);
563 			return -EINVAL;
564 		}
565 	}
566 
567 	if (align > nvdimm_namespace_capacity(ndns)) {
568 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
569 				align, nvdimm_namespace_capacity(ndns));
570 		return -EINVAL;
571 	}
572 
573 	/*
574 	 * These warnings are verbose because they can only trigger in
575 	 * the case where the physical address alignment of the
576 	 * namespace has changed since the pfn superblock was
577 	 * established.
578 	 */
579 	nsio = to_nd_namespace_io(&ndns->dev);
580 	if (offset >= resource_size(&nsio->res)) {
581 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
582 				dev_name(&ndns->dev));
583 		return -EBUSY;
584 	}
585 
586 	if ((align && !IS_ALIGNED(nsio->res.start + offset + start_pad, align))
587 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
588 		dev_err(&nd_pfn->dev,
589 				"bad offset: %#llx dax disabled align: %#lx\n",
590 				offset, align);
591 		return -ENXIO;
592 	}
593 
594 	return nd_pfn_clear_memmap_errors(nd_pfn);
595 }
596 EXPORT_SYMBOL(nd_pfn_validate);
597 
598 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
599 {
600 	int rc;
601 	struct nd_pfn *nd_pfn;
602 	struct device *pfn_dev;
603 	struct nd_pfn_sb *pfn_sb;
604 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
605 
606 	if (ndns->force_raw)
607 		return -ENODEV;
608 
609 	switch (ndns->claim_class) {
610 	case NVDIMM_CCLASS_NONE:
611 	case NVDIMM_CCLASS_PFN:
612 		break;
613 	default:
614 		return -ENODEV;
615 	}
616 
617 	nvdimm_bus_lock(&ndns->dev);
618 	nd_pfn = nd_pfn_alloc(nd_region);
619 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
620 	nvdimm_bus_unlock(&ndns->dev);
621 	if (!pfn_dev)
622 		return -ENOMEM;
623 	pfn_sb = devm_kmalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
624 	nd_pfn = to_nd_pfn(pfn_dev);
625 	nd_pfn->pfn_sb = pfn_sb;
626 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
627 	dev_dbg(dev, "pfn: %s\n", rc == 0 ? dev_name(pfn_dev) : "<none>");
628 	if (rc < 0) {
629 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
630 		put_device(pfn_dev);
631 	} else
632 		__nd_device_register(pfn_dev);
633 
634 	return rc;
635 }
636 EXPORT_SYMBOL(nd_pfn_probe);
637 
638 static u32 info_block_reserve(void)
639 {
640 	return ALIGN(SZ_8K, PAGE_SIZE);
641 }
642 
643 /*
644  * We hotplug memory at sub-section granularity, pad the reserved area
645  * from the previous section base to the namespace base address.
646  */
647 static unsigned long init_altmap_base(resource_size_t base)
648 {
649 	unsigned long base_pfn = PHYS_PFN(base);
650 
651 	return SUBSECTION_ALIGN_DOWN(base_pfn);
652 }
653 
654 static unsigned long init_altmap_reserve(resource_size_t base)
655 {
656 	unsigned long reserve = info_block_reserve() >> PAGE_SHIFT;
657 	unsigned long base_pfn = PHYS_PFN(base);
658 
659 	reserve += base_pfn - SUBSECTION_ALIGN_DOWN(base_pfn);
660 	return reserve;
661 }
662 
663 static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
664 {
665 	struct resource *res = &pgmap->res;
666 	struct vmem_altmap *altmap = &pgmap->altmap;
667 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
668 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
669 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
670 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
671 	u32 reserve = info_block_reserve();
672 	struct nd_namespace_common *ndns = nd_pfn->ndns;
673 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
674 	resource_size_t base = nsio->res.start + start_pad;
675 	resource_size_t end = nsio->res.end - end_trunc;
676 	struct vmem_altmap __altmap = {
677 		.base_pfn = init_altmap_base(base),
678 		.reserve = init_altmap_reserve(base),
679 		.end_pfn = PHYS_PFN(end),
680 	};
681 
682 	memcpy(res, &nsio->res, sizeof(*res));
683 	res->start += start_pad;
684 	res->end -= end_trunc;
685 
686 	if (nd_pfn->mode == PFN_MODE_RAM) {
687 		if (offset < reserve)
688 			return -EINVAL;
689 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
690 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
691 		nd_pfn->npfns = PHYS_PFN((resource_size(res) - offset));
692 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
693 			dev_info(&nd_pfn->dev,
694 					"number of pfns truncated from %lld to %ld\n",
695 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
696 					nd_pfn->npfns);
697 		memcpy(altmap, &__altmap, sizeof(*altmap));
698 		altmap->free = PHYS_PFN(offset - reserve);
699 		altmap->alloc = 0;
700 		pgmap->flags |= PGMAP_ALTMAP_VALID;
701 	} else
702 		return -ENXIO;
703 
704 	return 0;
705 }
706 
707 static int nd_pfn_init(struct nd_pfn *nd_pfn)
708 {
709 	struct nd_namespace_common *ndns = nd_pfn->ndns;
710 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
711 	resource_size_t start, size;
712 	struct nd_region *nd_region;
713 	unsigned long npfns, align;
714 	u32 end_trunc;
715 	struct nd_pfn_sb *pfn_sb;
716 	phys_addr_t offset;
717 	const char *sig;
718 	u64 checksum;
719 	int rc;
720 
721 	pfn_sb = devm_kmalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
722 	if (!pfn_sb)
723 		return -ENOMEM;
724 
725 	nd_pfn->pfn_sb = pfn_sb;
726 	if (is_nd_dax(&nd_pfn->dev))
727 		sig = DAX_SIG;
728 	else
729 		sig = PFN_SIG;
730 
731 	rc = nd_pfn_validate(nd_pfn, sig);
732 	if (rc != -ENODEV)
733 		return rc;
734 
735 	/* no info block, do init */;
736 	memset(pfn_sb, 0, sizeof(*pfn_sb));
737 
738 	nd_region = to_nd_region(nd_pfn->dev.parent);
739 	if (nd_region->ro) {
740 		dev_info(&nd_pfn->dev,
741 				"%s is read-only, unable to init metadata\n",
742 				dev_name(&nd_region->dev));
743 		return -ENXIO;
744 	}
745 
746 	/*
747 	 * Note, we use 64 here for the standard size of struct page,
748 	 * debugging options may cause it to be larger in which case the
749 	 * implementation will limit the pfns advertised through
750 	 * ->direct_access() to those that are included in the memmap.
751 	 */
752 	start = nsio->res.start;
753 	size = resource_size(&nsio->res);
754 	npfns = PHYS_PFN(size - SZ_8K);
755 	align = max(nd_pfn->align, (1UL << SUBSECTION_SHIFT));
756 	end_trunc = start + size - ALIGN_DOWN(start + size, align);
757 	if (nd_pfn->mode == PFN_MODE_PMEM) {
758 		/*
759 		 * The altmap should be padded out to the block size used
760 		 * when populating the vmemmap. This *should* be equal to
761 		 * PMD_SIZE for most architectures.
762 		 *
763 		 * Also make sure size of struct page is less than 64. We
764 		 * want to make sure we use large enough size here so that
765 		 * we don't have a dynamic reserve space depending on
766 		 * struct page size. But we also want to make sure we notice
767 		 * when we end up adding new elements to struct page.
768 		 */
769 		BUILD_BUG_ON(sizeof(struct page) > MAX_STRUCT_PAGE_SIZE);
770 		offset = ALIGN(start + SZ_8K + MAX_STRUCT_PAGE_SIZE * npfns, align)
771 			- start;
772 	} else if (nd_pfn->mode == PFN_MODE_RAM)
773 		offset = ALIGN(start + SZ_8K, align) - start;
774 	else
775 		return -ENXIO;
776 
777 	if (offset >= size) {
778 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
779 				dev_name(&ndns->dev));
780 		return -ENXIO;
781 	}
782 
783 	npfns = PHYS_PFN(size - offset - end_trunc);
784 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
785 	pfn_sb->dataoff = cpu_to_le64(offset);
786 	pfn_sb->npfns = cpu_to_le64(npfns);
787 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
788 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
789 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
790 	pfn_sb->version_major = cpu_to_le16(1);
791 	pfn_sb->version_minor = cpu_to_le16(4);
792 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
793 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
794 	pfn_sb->page_struct_size = cpu_to_le16(MAX_STRUCT_PAGE_SIZE);
795 	pfn_sb->page_size = cpu_to_le32(PAGE_SIZE);
796 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
797 	pfn_sb->checksum = cpu_to_le64(checksum);
798 
799 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
800 }
801 
802 /*
803  * Determine the effective resource range and vmem_altmap from an nd_pfn
804  * instance.
805  */
806 int nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
807 {
808 	int rc;
809 
810 	if (!nd_pfn->uuid || !nd_pfn->ndns)
811 		return -ENODEV;
812 
813 	rc = nd_pfn_init(nd_pfn);
814 	if (rc)
815 		return rc;
816 
817 	/* we need a valid pfn_sb before we can init a dev_pagemap */
818 	return __nvdimm_setup_pfn(nd_pfn, pgmap);
819 }
820 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
821