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
nd_pfn_release(struct device * dev)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, "%s\n", __func__);
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
is_nd_pfn(struct device * dev)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
to_nd_pfn(struct device * dev)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
mode_show(struct device * dev,struct device_attribute * attr,char * buf)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
mode_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t len)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, "%s: result: %zd wrote: %s%s", __func__,
98 rc, buf, 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
align_show(struct device * dev,struct device_attribute * attr,char * buf)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
__align_store(struct nd_pfn * nd_pfn,const char * buf)114 static ssize_t __align_store(struct nd_pfn *nd_pfn, const char *buf)
115 {
116 unsigned long val;
117 int rc;
118
119 rc = kstrtoul(buf, 0, &val);
120 if (rc)
121 return rc;
122
123 if (!is_power_of_2(val) || val < PAGE_SIZE || val > SZ_1G)
124 return -EINVAL;
125
126 if (nd_pfn->dev.driver)
127 return -EBUSY;
128 else
129 nd_pfn->align = val;
130
131 return 0;
132 }
133
align_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t len)134 static ssize_t align_store(struct device *dev,
135 struct device_attribute *attr, const char *buf, size_t len)
136 {
137 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
138 ssize_t rc;
139
140 device_lock(dev);
141 nvdimm_bus_lock(dev);
142 rc = __align_store(nd_pfn, buf);
143 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
144 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
145 nvdimm_bus_unlock(dev);
146 device_unlock(dev);
147
148 return rc ? rc : len;
149 }
150 static DEVICE_ATTR_RW(align);
151
uuid_show(struct device * dev,struct device_attribute * attr,char * buf)152 static ssize_t uuid_show(struct device *dev,
153 struct device_attribute *attr, char *buf)
154 {
155 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
156
157 if (nd_pfn->uuid)
158 return sprintf(buf, "%pUb\n", nd_pfn->uuid);
159 return sprintf(buf, "\n");
160 }
161
uuid_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t len)162 static ssize_t uuid_store(struct device *dev,
163 struct device_attribute *attr, const char *buf, size_t len)
164 {
165 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
166 ssize_t rc;
167
168 device_lock(dev);
169 rc = nd_uuid_store(dev, &nd_pfn->uuid, buf, len);
170 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
171 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
172 device_unlock(dev);
173
174 return rc ? rc : len;
175 }
176 static DEVICE_ATTR_RW(uuid);
177
namespace_show(struct device * dev,struct device_attribute * attr,char * buf)178 static ssize_t namespace_show(struct device *dev,
179 struct device_attribute *attr, char *buf)
180 {
181 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
182 ssize_t rc;
183
184 nvdimm_bus_lock(dev);
185 rc = sprintf(buf, "%s\n", nd_pfn->ndns
186 ? dev_name(&nd_pfn->ndns->dev) : "");
187 nvdimm_bus_unlock(dev);
188 return rc;
189 }
190
namespace_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t len)191 static ssize_t namespace_store(struct device *dev,
192 struct device_attribute *attr, const char *buf, size_t len)
193 {
194 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
195 ssize_t rc;
196
197 device_lock(dev);
198 nvdimm_bus_lock(dev);
199 rc = nd_namespace_store(dev, &nd_pfn->ndns, buf, len);
200 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
201 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
202 nvdimm_bus_unlock(dev);
203 device_unlock(dev);
204
205 return rc;
206 }
207 static DEVICE_ATTR_RW(namespace);
208
resource_show(struct device * dev,struct device_attribute * attr,char * buf)209 static ssize_t resource_show(struct device *dev,
210 struct device_attribute *attr, char *buf)
211 {
212 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
213 ssize_t rc;
214
215 device_lock(dev);
216 if (dev->driver) {
217 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
218 u64 offset = __le64_to_cpu(pfn_sb->dataoff);
219 struct nd_namespace_common *ndns = nd_pfn->ndns;
220 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
221 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
222
223 rc = sprintf(buf, "%#llx\n", (unsigned long long) nsio->res.start
224 + start_pad + offset);
225 } else {
226 /* no address to convey if the pfn instance is disabled */
227 rc = -ENXIO;
228 }
229 device_unlock(dev);
230
231 return rc;
232 }
233 static DEVICE_ATTR_RO(resource);
234
size_show(struct device * dev,struct device_attribute * attr,char * buf)235 static ssize_t size_show(struct device *dev,
236 struct device_attribute *attr, char *buf)
237 {
238 struct nd_pfn *nd_pfn = to_nd_pfn_safe(dev);
239 ssize_t rc;
240
241 device_lock(dev);
242 if (dev->driver) {
243 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
244 u64 offset = __le64_to_cpu(pfn_sb->dataoff);
245 struct nd_namespace_common *ndns = nd_pfn->ndns;
246 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
247 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
248 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
249
250 rc = sprintf(buf, "%llu\n", (unsigned long long)
251 resource_size(&nsio->res) - start_pad
252 - end_trunc - offset);
253 } else {
254 /* no size to convey if the pfn instance is disabled */
255 rc = -ENXIO;
256 }
257 device_unlock(dev);
258
259 return rc;
260 }
261 static DEVICE_ATTR_RO(size);
262
263 static struct attribute *nd_pfn_attributes[] = {
264 &dev_attr_mode.attr,
265 &dev_attr_namespace.attr,
266 &dev_attr_uuid.attr,
267 &dev_attr_align.attr,
268 &dev_attr_resource.attr,
269 &dev_attr_size.attr,
270 NULL,
271 };
272
pfn_visible(struct kobject * kobj,struct attribute * a,int n)273 static umode_t pfn_visible(struct kobject *kobj, struct attribute *a, int n)
274 {
275 if (a == &dev_attr_resource.attr)
276 return 0400;
277 return a->mode;
278 }
279
280 struct attribute_group nd_pfn_attribute_group = {
281 .attrs = nd_pfn_attributes,
282 .is_visible = pfn_visible,
283 };
284
285 static const struct attribute_group *nd_pfn_attribute_groups[] = {
286 &nd_pfn_attribute_group,
287 &nd_device_attribute_group,
288 &nd_numa_attribute_group,
289 NULL,
290 };
291
nd_pfn_devinit(struct nd_pfn * nd_pfn,struct nd_namespace_common * ndns)292 struct device *nd_pfn_devinit(struct nd_pfn *nd_pfn,
293 struct nd_namespace_common *ndns)
294 {
295 struct device *dev = &nd_pfn->dev;
296
297 if (!nd_pfn)
298 return NULL;
299
300 nd_pfn->mode = PFN_MODE_NONE;
301 nd_pfn->align = HPAGE_SIZE;
302 dev = &nd_pfn->dev;
303 device_initialize(&nd_pfn->dev);
304 if (ndns && !__nd_attach_ndns(&nd_pfn->dev, ndns, &nd_pfn->ndns)) {
305 dev_dbg(&ndns->dev, "%s failed, already claimed by %s\n",
306 __func__, dev_name(ndns->claim));
307 put_device(dev);
308 return NULL;
309 }
310 return dev;
311 }
312
nd_pfn_alloc(struct nd_region * nd_region)313 static struct nd_pfn *nd_pfn_alloc(struct nd_region *nd_region)
314 {
315 struct nd_pfn *nd_pfn;
316 struct device *dev;
317
318 nd_pfn = kzalloc(sizeof(*nd_pfn), GFP_KERNEL);
319 if (!nd_pfn)
320 return NULL;
321
322 nd_pfn->id = ida_simple_get(&nd_region->pfn_ida, 0, 0, GFP_KERNEL);
323 if (nd_pfn->id < 0) {
324 kfree(nd_pfn);
325 return NULL;
326 }
327
328 dev = &nd_pfn->dev;
329 dev_set_name(dev, "pfn%d.%d", nd_region->id, nd_pfn->id);
330 dev->groups = nd_pfn_attribute_groups;
331 dev->type = &nd_pfn_device_type;
332 dev->parent = &nd_region->dev;
333
334 return nd_pfn;
335 }
336
nd_pfn_create(struct nd_region * nd_region)337 struct device *nd_pfn_create(struct nd_region *nd_region)
338 {
339 struct nd_pfn *nd_pfn;
340 struct device *dev;
341
342 if (!is_nd_pmem(&nd_region->dev))
343 return NULL;
344
345 nd_pfn = nd_pfn_alloc(nd_region);
346 dev = nd_pfn_devinit(nd_pfn, NULL);
347
348 __nd_device_register(dev);
349 return dev;
350 }
351
nd_pfn_validate(struct nd_pfn * nd_pfn,const char * sig)352 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
353 {
354 u64 checksum, offset;
355 enum nd_pfn_mode mode;
356 struct nd_namespace_io *nsio;
357 unsigned long align, start_pad;
358 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
359 struct nd_namespace_common *ndns = nd_pfn->ndns;
360 const u8 *parent_uuid = nd_dev_to_uuid(&ndns->dev);
361
362 if (!pfn_sb || !ndns)
363 return -ENODEV;
364
365 if (!is_nd_pmem(nd_pfn->dev.parent))
366 return -ENODEV;
367
368 if (nvdimm_read_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb)))
369 return -ENXIO;
370
371 if (memcmp(pfn_sb->signature, sig, PFN_SIG_LEN) != 0)
372 return -ENODEV;
373
374 checksum = le64_to_cpu(pfn_sb->checksum);
375 pfn_sb->checksum = 0;
376 if (checksum != nd_sb_checksum((struct nd_gen_sb *) pfn_sb))
377 return -ENODEV;
378 pfn_sb->checksum = cpu_to_le64(checksum);
379
380 if (memcmp(pfn_sb->parent_uuid, parent_uuid, 16) != 0)
381 return -ENODEV;
382
383 if (__le16_to_cpu(pfn_sb->version_minor) < 1) {
384 pfn_sb->start_pad = 0;
385 pfn_sb->end_trunc = 0;
386 }
387
388 if (__le16_to_cpu(pfn_sb->version_minor) < 2)
389 pfn_sb->align = 0;
390
391 switch (le32_to_cpu(pfn_sb->mode)) {
392 case PFN_MODE_RAM:
393 case PFN_MODE_PMEM:
394 break;
395 default:
396 return -ENXIO;
397 }
398
399 align = le32_to_cpu(pfn_sb->align);
400 offset = le64_to_cpu(pfn_sb->dataoff);
401 start_pad = le32_to_cpu(pfn_sb->start_pad);
402 if (align == 0)
403 align = 1UL << ilog2(offset);
404 mode = le32_to_cpu(pfn_sb->mode);
405
406 if (!nd_pfn->uuid) {
407 /*
408 * When probing a namepace via nd_pfn_probe() the uuid
409 * is NULL (see: nd_pfn_devinit()) we init settings from
410 * pfn_sb
411 */
412 nd_pfn->uuid = kmemdup(pfn_sb->uuid, 16, GFP_KERNEL);
413 if (!nd_pfn->uuid)
414 return -ENOMEM;
415 nd_pfn->align = align;
416 nd_pfn->mode = mode;
417 } else {
418 /*
419 * When probing a pfn / dax instance we validate the
420 * live settings against the pfn_sb
421 */
422 if (memcmp(nd_pfn->uuid, pfn_sb->uuid, 16) != 0)
423 return -ENODEV;
424
425 /*
426 * If the uuid validates, but other settings mismatch
427 * return EINVAL because userspace has managed to change
428 * the configuration without specifying new
429 * identification.
430 */
431 if (nd_pfn->align != align || nd_pfn->mode != mode) {
432 dev_err(&nd_pfn->dev,
433 "init failed, settings mismatch\n");
434 dev_dbg(&nd_pfn->dev, "align: %lx:%lx mode: %d:%d\n",
435 nd_pfn->align, align, nd_pfn->mode,
436 mode);
437 return -EINVAL;
438 }
439 }
440
441 if (align > nvdimm_namespace_capacity(ndns)) {
442 dev_err(&nd_pfn->dev, "alignment: %lx exceeds capacity %llx\n",
443 align, nvdimm_namespace_capacity(ndns));
444 return -EINVAL;
445 }
446
447 /*
448 * These warnings are verbose because they can only trigger in
449 * the case where the physical address alignment of the
450 * namespace has changed since the pfn superblock was
451 * established.
452 */
453 nsio = to_nd_namespace_io(&ndns->dev);
454 if (offset >= resource_size(&nsio->res)) {
455 dev_err(&nd_pfn->dev, "pfn array size exceeds capacity of %s\n",
456 dev_name(&ndns->dev));
457 return -EBUSY;
458 }
459
460 if ((align && !IS_ALIGNED(nsio->res.start + offset + start_pad, align))
461 || !IS_ALIGNED(offset, PAGE_SIZE)) {
462 dev_err(&nd_pfn->dev,
463 "bad offset: %#llx dax disabled align: %#lx\n",
464 offset, align);
465 return -ENXIO;
466 }
467
468 return 0;
469 }
470 EXPORT_SYMBOL(nd_pfn_validate);
471
nd_pfn_probe(struct device * dev,struct nd_namespace_common * ndns)472 int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
473 {
474 int rc;
475 struct nd_pfn *nd_pfn;
476 struct device *pfn_dev;
477 struct nd_pfn_sb *pfn_sb;
478 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
479
480 if (ndns->force_raw)
481 return -ENODEV;
482
483 nvdimm_bus_lock(&ndns->dev);
484 nd_pfn = nd_pfn_alloc(nd_region);
485 pfn_dev = nd_pfn_devinit(nd_pfn, ndns);
486 nvdimm_bus_unlock(&ndns->dev);
487 if (!pfn_dev)
488 return -ENOMEM;
489 pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
490 nd_pfn = to_nd_pfn(pfn_dev);
491 nd_pfn->pfn_sb = pfn_sb;
492 rc = nd_pfn_validate(nd_pfn, PFN_SIG);
493 dev_dbg(dev, "%s: pfn: %s\n", __func__,
494 rc == 0 ? dev_name(pfn_dev) : "<none>");
495 if (rc < 0) {
496 __nd_detach_ndns(pfn_dev, &nd_pfn->ndns);
497 put_device(pfn_dev);
498 } else
499 __nd_device_register(pfn_dev);
500
501 return rc;
502 }
503 EXPORT_SYMBOL(nd_pfn_probe);
504
505 /*
506 * We hotplug memory at section granularity, pad the reserved area from
507 * the previous section base to the namespace base address.
508 */
init_altmap_base(resource_size_t base)509 static unsigned long init_altmap_base(resource_size_t base)
510 {
511 unsigned long base_pfn = PHYS_PFN(base);
512
513 return PFN_SECTION_ALIGN_DOWN(base_pfn);
514 }
515
init_altmap_reserve(resource_size_t base)516 static unsigned long init_altmap_reserve(resource_size_t base)
517 {
518 unsigned long reserve = PHYS_PFN(SZ_8K);
519 unsigned long base_pfn = PHYS_PFN(base);
520
521 reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
522 return reserve;
523 }
524
__nvdimm_setup_pfn(struct nd_pfn * nd_pfn,struct resource * res,struct vmem_altmap * altmap)525 static struct vmem_altmap *__nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
526 struct resource *res, struct vmem_altmap *altmap)
527 {
528 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
529 u64 offset = le64_to_cpu(pfn_sb->dataoff);
530 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
531 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
532 struct nd_namespace_common *ndns = nd_pfn->ndns;
533 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
534 resource_size_t base = nsio->res.start + start_pad;
535 struct vmem_altmap __altmap = {
536 .base_pfn = init_altmap_base(base),
537 .reserve = init_altmap_reserve(base),
538 };
539
540 memcpy(res, &nsio->res, sizeof(*res));
541 res->start += start_pad;
542 res->end -= end_trunc;
543
544 if (nd_pfn->mode == PFN_MODE_RAM) {
545 if (offset < SZ_8K)
546 return ERR_PTR(-EINVAL);
547 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
548 altmap = NULL;
549 } else if (nd_pfn->mode == PFN_MODE_PMEM) {
550 nd_pfn->npfns = PFN_SECTION_ALIGN_UP((resource_size(res)
551 - offset) / PAGE_SIZE);
552 if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
553 dev_info(&nd_pfn->dev,
554 "number of pfns truncated from %lld to %ld\n",
555 le64_to_cpu(nd_pfn->pfn_sb->npfns),
556 nd_pfn->npfns);
557 memcpy(altmap, &__altmap, sizeof(*altmap));
558 altmap->free = PHYS_PFN(offset - SZ_8K);
559 altmap->alloc = 0;
560 } else
561 return ERR_PTR(-ENXIO);
562
563 return altmap;
564 }
565
phys_pmem_align_down(struct nd_pfn * nd_pfn,u64 phys)566 static u64 phys_pmem_align_down(struct nd_pfn *nd_pfn, u64 phys)
567 {
568 return min_t(u64, PHYS_SECTION_ALIGN_DOWN(phys),
569 ALIGN_DOWN(phys, nd_pfn->align));
570 }
571
nd_pfn_init(struct nd_pfn * nd_pfn)572 static int nd_pfn_init(struct nd_pfn *nd_pfn)
573 {
574 u32 dax_label_reserve = is_nd_dax(&nd_pfn->dev) ? SZ_128K : 0;
575 struct nd_namespace_common *ndns = nd_pfn->ndns;
576 u32 start_pad = 0, end_trunc = 0;
577 resource_size_t start, size;
578 struct nd_namespace_io *nsio;
579 struct nd_region *nd_region;
580 struct nd_pfn_sb *pfn_sb;
581 unsigned long npfns;
582 phys_addr_t offset;
583 const char *sig;
584 u64 checksum;
585 int rc;
586
587 pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
588 if (!pfn_sb)
589 return -ENOMEM;
590
591 nd_pfn->pfn_sb = pfn_sb;
592 if (is_nd_dax(&nd_pfn->dev))
593 sig = DAX_SIG;
594 else
595 sig = PFN_SIG;
596 rc = nd_pfn_validate(nd_pfn, sig);
597 if (rc != -ENODEV)
598 return rc;
599
600 /* no info block, do init */;
601 nd_region = to_nd_region(nd_pfn->dev.parent);
602 if (nd_region->ro) {
603 dev_info(&nd_pfn->dev,
604 "%s is read-only, unable to init metadata\n",
605 dev_name(&nd_region->dev));
606 return -ENXIO;
607 }
608
609 memset(pfn_sb, 0, sizeof(*pfn_sb));
610
611 /*
612 * Check if pmem collides with 'System RAM' when section aligned and
613 * trim it accordingly
614 */
615 nsio = to_nd_namespace_io(&ndns->dev);
616 start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
617 size = resource_size(&nsio->res);
618 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
619 IORES_DESC_NONE) == REGION_MIXED) {
620 start = nsio->res.start;
621 start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
622 }
623
624 start = nsio->res.start;
625 size = PHYS_SECTION_ALIGN_UP(start + size) - start;
626 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
627 IORES_DESC_NONE) == REGION_MIXED
628 || !IS_ALIGNED(start + resource_size(&nsio->res),
629 nd_pfn->align)) {
630 size = resource_size(&nsio->res);
631 end_trunc = start + size - phys_pmem_align_down(nd_pfn,
632 start + size);
633 }
634
635 if (start_pad + end_trunc)
636 dev_info(&nd_pfn->dev, "%s alignment collision, truncate %d bytes\n",
637 dev_name(&ndns->dev), start_pad + end_trunc);
638
639 /*
640 * Note, we use 64 here for the standard size of struct page,
641 * debugging options may cause it to be larger in which case the
642 * implementation will limit the pfns advertised through
643 * ->direct_access() to those that are included in the memmap.
644 */
645 start += start_pad;
646 size = resource_size(&nsio->res);
647 npfns = PFN_SECTION_ALIGN_UP((size - start_pad - end_trunc - SZ_8K)
648 / PAGE_SIZE);
649 if (nd_pfn->mode == PFN_MODE_PMEM) {
650 /*
651 * vmemmap_populate_hugepages() allocates the memmap array in
652 * HPAGE_SIZE chunks.
653 */
654 offset = ALIGN(start + SZ_8K + 64 * npfns + dax_label_reserve,
655 max(nd_pfn->align, HPAGE_SIZE)) - start;
656 } else if (nd_pfn->mode == PFN_MODE_RAM)
657 offset = ALIGN(start + SZ_8K + dax_label_reserve,
658 nd_pfn->align) - start;
659 else
660 return -ENXIO;
661
662 if (offset + start_pad + end_trunc >= size) {
663 dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
664 dev_name(&ndns->dev));
665 return -ENXIO;
666 }
667
668 npfns = (size - offset - start_pad - end_trunc) / SZ_4K;
669 pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
670 pfn_sb->dataoff = cpu_to_le64(offset);
671 pfn_sb->npfns = cpu_to_le64(npfns);
672 memcpy(pfn_sb->signature, sig, PFN_SIG_LEN);
673 memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
674 memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
675 pfn_sb->version_major = cpu_to_le16(1);
676 pfn_sb->version_minor = cpu_to_le16(2);
677 pfn_sb->start_pad = cpu_to_le32(start_pad);
678 pfn_sb->end_trunc = cpu_to_le32(end_trunc);
679 pfn_sb->align = cpu_to_le32(nd_pfn->align);
680 checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
681 pfn_sb->checksum = cpu_to_le64(checksum);
682
683 return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb));
684 }
685
686 /*
687 * Determine the effective resource range and vmem_altmap from an nd_pfn
688 * instance.
689 */
nvdimm_setup_pfn(struct nd_pfn * nd_pfn,struct resource * res,struct vmem_altmap * altmap)690 struct vmem_altmap *nvdimm_setup_pfn(struct nd_pfn *nd_pfn,
691 struct resource *res, struct vmem_altmap *altmap)
692 {
693 int rc;
694
695 if (!nd_pfn->uuid || !nd_pfn->ndns)
696 return ERR_PTR(-ENODEV);
697
698 rc = nd_pfn_init(nd_pfn);
699 if (rc)
700 return ERR_PTR(rc);
701
702 /* we need a valid pfn_sb before we can init a vmem_altmap */
703 return __nvdimm_setup_pfn(nd_pfn, res, altmap);
704 }
705 EXPORT_SYMBOL_GPL(nvdimm_setup_pfn);
706