xref: /openbmc/linux/drivers/nvdimm/pfn_devs.c (revision e553d2a5)
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 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 	nd_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 	nd_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 	nd_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 	nd_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 	nd_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 	nd_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 	nd_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 	nd_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 	nd_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 	nd_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 /**
416  * nd_pfn_validate - read and validate info-block
417  * @nd_pfn: fsdax namespace runtime state / properties
418  * @sig: 'devdax' or 'fsdax' signature
419  *
420  * Upon return the info-block buffer contents (->pfn_sb) are
421  * indeterminate when validation fails, and a coherent info-block
422  * otherwise.
423  */
424 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
425 {
426 	u64 checksum, offset;
427 	enum nd_pfn_mode mode;
428 	struct nd_namespace_io *nsio;
429 	unsigned long align, start_pad;
430 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
431 	struct nd_namespace_common *ndns = nd_pfn->ndns;
432 	const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
433 
434 	if (!pfn_sb || !ndns)
435 		return -ENODEV;
436 
437 	if (!is_memory(nd_pfn->dev.parent))
438 		return -ENODEV;
439 
440 	if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0))
441 		return -ENXIO;
442 
443 	if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
444 		return -ENODEV;
445 
446 	checksum = le64_to_cpu(pfn_sb->checksum);
447 	pfn_sb->checksum = 0;
448 	if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
449 		return -ENODEV;
450 	pfn_sb->checksum = cpu_to_le64(checksum);
451 
452 	if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
453 		return -ENODEV;
454 
455 	if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
456 		pfn_sb->start_pad = 0;
457 		pfn_sb->end_trunc = 0;
458 	}
459 
460 	if (__le16_to_cpu(pfn_sb->version_minor) < 2)
461 		pfn_sb->align = 0;
462 
463 	if (__le16_to_cpu(pfn_sb->version_minor) < 4) {
464 		pfn_sb->page_struct_size = cpu_to_le16(64);
465 		pfn_sb->page_size = cpu_to_le32(PAGE_SIZE);
466 	}
467 
468 	switch (le32_to_cpu(pfn_sb->mode)) {
469 	case PFN_MODE_RAM:
470 	case PFN_MODE_PMEM:
471 		break;
472 	default:
473 		return -ENXIO;
474 	}
475 
476 	align = le32_to_cpu(pfn_sb->align);
477 	offset = le64_to_cpu(pfn_sb->dataoff);
478 	start_pad = le32_to_cpu(pfn_sb->start_pad);
479 	if (align == 0)
480 		align = 1UL << ilog2(offset);
481 	mode = le32_to_cpu(pfn_sb->mode);
482 
483 	if ((le32_to_cpu(pfn_sb->page_size) > PAGE_SIZE) &&
484 			(mode == PFN_MODE_PMEM)) {
485 		dev_err(&nd_pfn->dev,
486 				"init failed, page size mismatch %d\n",
487 				le32_to_cpu(pfn_sb->page_size));
488 		return -EOPNOTSUPP;
489 	}
490 
491 	if ((le16_to_cpu(pfn_sb->page_struct_size) < sizeof(struct page)) &&
492 			(mode == PFN_MODE_PMEM)) {
493 		dev_err(&nd_pfn->dev,
494 				"init failed, struct page size mismatch %d\n",
495 				le16_to_cpu(pfn_sb->page_struct_size));
496 		return -EOPNOTSUPP;
497 	}
498 
499 	if (!nd_pfn->uuid) {
500 		/*
501 		 * When probing a namepace via nd_pfn_probe() the uuid
502 		 * is NULL (see: nd_pfn_devinit()) we init settings from
503 		 * pfn_sb
504 		 */
505 		nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
506 		if (!nd_pfn->uuid)
507 			return -ENOMEM;
508 		nd_pfn->align = align;
509 		nd_pfn->mode = mode;
510 	} else {
511 		/*
512 		 * When probing a pfn / dax instance we validate the
513 		 * live settings against the pfn_sb
514 		 */
515 		if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
516 			return -ENODEV;
517 
518 		/*
519 		 * If the uuid validates, but other settings mismatch
520 		 * return EINVAL because userspace has managed to change
521 		 * the configuration without specifying new
522 		 * identification.
523 		 */
524 		if (nd_pfn->align != align || nd_pfn->mode != mode) {
525 			dev_err(&nd_pfn->dev,
526 					"init failed, settings mismatch\n");
527 			dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
528 					nd_pfn->align, align, nd_pfn->mode,
529 					mode);
530 			return -EINVAL;
531 		}
532 	}
533 
534 	if (align > nvdimm_namespace_capacity(ndns)) {
535 		dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
536 				align, nvdimm_namespace_capacity(ndns));
537 		return -EINVAL;
538 	}
539 
540 	/*
541 	 * These warnings are verbose because they can only trigger in
542 	 * the case where the physical address alignment of the
543 	 * namespace has changed since the pfn superblock was
544 	 * established.
545 	 */
546 	nsio = to_nd_namespace_io(&ndns->dev);
547 	if (offset >= resource_size(&nsio->res)) {
548 		dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
549 				dev_name(&ndns->dev));
550 		return -EBUSY;
551 	}
552 
553 	if ((align && !IS_ALIGNED(nsio->res.start + offset + start_pad, align))
554 			|| !IS_ALIGNED(offset, PAGE_SIZE)) {
555 		dev_err(&nd_pfn->dev,
556 				"bad offset: %#llx dax disabled align: %#lx\n",
557 				offset, align);
558 		return -ENXIO;
559 	}
560 
561 	return nd_pfn_clear_memmap_errors(nd_pfn);
562 }
563 EXPORT_SYMBOL(nd_pfn_validate);
564 
565 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
566 {
567 	int rc;
568 	struct nd_pfn *nd_pfn;
569 	struct device *pfn_dev;
570 	struct nd_pfn_sb *pfn_sb;
571 	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
572 
573 	if (ndns->force_raw)
574 		return -ENODEV;
575 
576 	switch (ndns->claim_class) {
577 	case NVDIMM_CCLASS_NONE:
578 	case NVDIMM_CCLASS_PFN:
579 		break;
580 	default:
581 		return -ENODEV;
582 	}
583 
584 	nvdimm_bus_lock(&ndns->dev);
585 	nd_pfn = nd_pfn_alloc(nd_region);
586 	pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
587 	nvdimm_bus_unlock(&ndns->dev);
588 	if (!pfn_dev)
589 		return -ENOMEM;
590 	pfn_sb = devm_kmalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
591 	nd_pfn = to_nd_pfn(pfn_dev);
592 	nd_pfn->pfn_sb = pfn_sb;
593 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
594 	dev_dbg(dev, "pfn: %s\n", rc == 0 ? dev_name(pfn_dev) : "<none>");
595 	if (rc < 0) {
596 		nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
597 		put_device(pfn_dev);
598 	} else
599 		__nd_device_register(pfn_dev);
600 
601 	return rc;
602 }
603 EXPORT_SYMBOL(nd_pfn_probe);
604 
605 static u32 info_block_reserve(void)
606 {
607 	return ALIGN(SZ_8K, PAGE_SIZE);
608 }
609 
610 /*
611  * We hotplug memory at sub-section granularity, pad the reserved area
612  * from the previous section base to the namespace base address.
613  */
614 static unsigned long init_altmap_base(resource_size_t base)
615 {
616 	unsigned long base_pfn = PHYS_PFN(base);
617 
618 	return SUBSECTION_ALIGN_DOWN(base_pfn);
619 }
620 
621 static unsigned long init_altmap_reserve(resource_size_t base)
622 {
623 	unsigned long reserve = info_block_reserve() >> PAGE_SHIFT;
624 	unsigned long base_pfn = PHYS_PFN(base);
625 
626 	reserve += base_pfn - SUBSECTION_ALIGN_DOWN(base_pfn);
627 	return reserve;
628 }
629 
630 static int __nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
631 {
632 	struct resource *res = &pgmap->res;
633 	struct vmem_altmap *altmap = &pgmap->altmap;
634 	struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
635 	u64 offset = le64_to_cpu(pfn_sb->dataoff);
636 	u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
637 	u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
638 	u32 reserve = info_block_reserve();
639 	struct nd_namespace_common *ndns = nd_pfn->ndns;
640 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
641 	resource_size_t base = nsio->res.start + start_pad;
642 	struct vmem_altmap __altmap = {
643 		.base_pfn = init_altmap_base(base),
644 		.reserve = init_altmap_reserve(base),
645 	};
646 
647 	memcpy(res, &nsio->res, sizeof(*res));
648 	res->start += start_pad;
649 	res->end -= end_trunc;
650 
651 	if (nd_pfn->mode == PFN_MODE_RAM) {
652 		if (offset < reserve)
653 			return -EINVAL;
654 		nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
655 	} else if (nd_pfn->mode == PFN_MODE_PMEM) {
656 		nd_pfn->npfns = PHYS_PFN((resource_size(res) - offset));
657 		if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
658 			dev_info(&nd_pfn->dev,
659 					"number of pfns truncated from %lld to %ld\n",
660 					le64_to_cpu(nd_pfn->pfn_sb->npfns),
661 					nd_pfn->npfns);
662 		memcpy(altmap, &__altmap, sizeof(*altmap));
663 		altmap->free = PHYS_PFN(offset - reserve);
664 		altmap->alloc = 0;
665 		pgmap->flags |= PGMAP_ALTMAP_VALID;
666 	} else
667 		return -ENXIO;
668 
669 	return 0;
670 }
671 
672 static int nd_pfn_init(struct nd_pfn *nd_pfn)
673 {
674 	struct nd_namespace_common *ndns = nd_pfn->ndns;
675 	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
676 	resource_size_t start, size;
677 	struct nd_region *nd_region;
678 	unsigned long npfns, align;
679 	u32 end_trunc;
680 	struct nd_pfn_sb *pfn_sb;
681 	phys_addr_t offset;
682 	const char *sig;
683 	u64 checksum;
684 	int rc;
685 
686 	pfn_sb = devm_kmalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
687 	if (!pfn_sb)
688 		return -ENOMEM;
689 
690 	nd_pfn->pfn_sb = pfn_sb;
691 	if (is_nd_dax(&nd_pfn->dev))
692 		sig = DAX_SIG;
693 	else
694 		sig = PFN_SIG;
695 
696 	rc = nd_pfn_validate(nd_pfn, sig);
697 	if (rc != -ENODEV)
698 		return rc;
699 
700 	/* no info block, do init */;
701 	memset(pfn_sb, 0, sizeof(*pfn_sb));
702 
703 	nd_region = to_nd_region(nd_pfn->dev.parent);
704 	if (nd_region->ro) {
705 		dev_info(&nd_pfn->dev,
706 				"%s is read-only, unable to init metadata\n",
707 				dev_name(&nd_region->dev));
708 		return -ENXIO;
709 	}
710 
711 	/*
712 	 * Note, we use 64 here for the standard size of struct page,
713 	 * debugging options may cause it to be larger in which case the
714 	 * implementation will limit the pfns advertised through
715 	 * ->direct_access() to those that are included in the memmap.
716 	 */
717 	start = nsio->res.start;
718 	size = resource_size(&nsio->res);
719 	npfns = PHYS_PFN(size - SZ_8K);
720 	align = max(nd_pfn->align, (1UL << SUBSECTION_SHIFT));
721 	end_trunc = start + size - ALIGN_DOWN(start + size, align);
722 	if (nd_pfn->mode == PFN_MODE_PMEM) {
723 		/*
724 		 * The altmap should be padded out to the block size used
725 		 * when populating the vmemmap. This *should* be equal to
726 		 * PMD_SIZE for most architectures.
727 		 *
728 		 * Also make sure size of struct page is less than 64. We
729 		 * want to make sure we use large enough size here so that
730 		 * we don't have a dynamic reserve space depending on
731 		 * struct page size. But we also want to make sure we notice
732 		 * when we end up adding new elements to struct page.
733 		 */
734 		BUILD_BUG_ON(sizeof(struct page) > MAX_STRUCT_PAGE_SIZE);
735 		offset = ALIGN(start + SZ_8K + MAX_STRUCT_PAGE_SIZE * npfns, align)
736 			- start;
737 	} else if (nd_pfn->mode == PFN_MODE_RAM)
738 		offset = ALIGN(start + SZ_8K, align) - start;
739 	else
740 		return -ENXIO;
741 
742 	if (offset >= size) {
743 		dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
744 				dev_name(&ndns->dev));
745 		return -ENXIO;
746 	}
747 
748 	npfns = PHYS_PFN(size - offset - end_trunc);
749 	pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
750 	pfn_sb->dataoff = cpu_to_le64(offset);
751 	pfn_sb->npfns = cpu_to_le64(npfns);
752 	memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
753 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
754 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
755 	pfn_sb->version_major = cpu_to_le16(1);
756 	pfn_sb->version_minor = cpu_to_le16(4);
757 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
758 	pfn_sb->align = cpu_to_le32(nd_pfn->align);
759 	pfn_sb->page_struct_size = cpu_to_le16(MAX_STRUCT_PAGE_SIZE);
760 	pfn_sb->page_size = cpu_to_le32(PAGE_SIZE);
761 	checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
762 	pfn_sb->checksum = cpu_to_le64(checksum);
763 
764 	return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb), 0);
765 }
766 
767 /*
768  * Determine the effective resource range and vmem_altmap from an nd_pfn
769  * instance.
770  */
771 int nvdimm_setup_pfn(struct nd_pfn *nd_pfn, struct dev_pagemap *pgmap)
772 {
773 	int rc;
774 
775 	if (!nd_pfn->uuid || !nd_pfn->ndns)
776 		return -ENODEV;
777 
778 	rc = nd_pfn_init(nd_pfn);
779 	if (rc)
780 		return rc;
781 
782 	/* we need a valid pfn_sb before we can init a dev_pagemap */
783 	return __nvdimm_setup_pfn(nd_pfn, pgmap);
784 }
785 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
786