pmem.c 14.7 KB
Newer Older
1 2 3
/*
 * Persistent Memory Driver
 *
4
 * Copyright (c) 2014-2015, Intel Corporation.
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
 * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>.
 * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 */

#include <asm/cacheflush.h>
#include <linux/blkdev.h>
#include <linux/hdreg.h>
#include <linux/init.h>
#include <linux/platform_device.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
25
#include <linux/badblocks.h>
D
Dan Williams 已提交
26
#include <linux/memremap.h>
27
#include <linux/vmalloc.h>
28
#include <linux/blk-mq.h>
D
Dan Williams 已提交
29
#include <linux/pfn_t.h>
30
#include <linux/slab.h>
31
#include <linux/uio.h>
D
Dan Williams 已提交
32
#include <linux/dax.h>
33
#include <linux/nd.h>
34
#include <linux/backing-dev.h>
35
#include "pmem.h"
36
#include "pfn.h"
37
#include "nd.h"
38
#include "nd-core.h"
39

40 41 42 43 44 45 46 47 48 49 50 51 52
static struct device *to_dev(struct pmem_device *pmem)
{
	/*
	 * nvdimm bus services need a 'dev' parameter, and we record the device
	 * at init in bb.dev.
	 */
	return pmem->bb.dev;
}

static struct nd_region *to_region(struct pmem_device *pmem)
{
	return to_nd_region(to_dev(pmem)->parent);
}
53

54 55
static blk_status_t pmem_clear_poison(struct pmem_device *pmem,
		phys_addr_t offset, unsigned int len)
56
{
57
	struct device *dev = to_dev(pmem);
58 59
	sector_t sector;
	long cleared;
60
	blk_status_t rc = BLK_STS_OK;
61 62 63

	sector = (offset - pmem->data_offset) / 512;

64 65
	cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
	if (cleared < len)
66
		rc = BLK_STS_IOERR;
67
	if (cleared > 0 && cleared / 512) {
68
		cleared /= 512;
69
		dev_dbg(dev, "%#llx clear %ld sector%s\n",
70 71
				(unsigned long long) sector, cleared,
				cleared > 1 ? "s" : "");
72
		badblocks_clear(&pmem->bb, sector, cleared);
73 74
		if (pmem->bb_state)
			sysfs_notify_dirent(pmem->bb_state);
75
	}
76

77
	arch_invalidate_pmem(pmem->virt_addr + offset, len);
78 79

	return rc;
80 81
}

82 83 84
static void write_pmem(void *pmem_addr, struct page *page,
		unsigned int off, unsigned int len)
{
85 86 87 88 89 90 91 92 93 94 95 96 97
	unsigned int chunk;
	void *mem;

	while (len) {
		mem = kmap_atomic(page);
		chunk = min_t(unsigned int, len, PAGE_SIZE);
		memcpy_flushcache(pmem_addr, mem + off, chunk);
		kunmap_atomic(mem);
		len -= chunk;
		off = 0;
		page++;
		pmem_addr += PAGE_SIZE;
	}
98 99
}

100
static blk_status_t read_pmem(struct page *page, unsigned int off,
101 102
		void *pmem_addr, unsigned int len)
{
103
	unsigned int chunk;
104
	unsigned long rem;
105 106 107 108 109
	void *mem;

	while (len) {
		mem = kmap_atomic(page);
		chunk = min_t(unsigned int, len, PAGE_SIZE);
110
		rem = memcpy_mcsafe(mem + off, pmem_addr, chunk);
111
		kunmap_atomic(mem);
112
		if (rem)
113 114 115 116 117 118
			return BLK_STS_IOERR;
		len -= chunk;
		off = 0;
		page++;
		pmem_addr += PAGE_SIZE;
	}
119
	return BLK_STS_OK;
120 121
}

122
static blk_status_t pmem_do_bvec(struct pmem_device *pmem, struct page *page,
123
			unsigned int len, unsigned int off, bool is_write,
124 125
			sector_t sector)
{
126
	blk_status_t rc = BLK_STS_OK;
127
	bool bad_pmem = false;
128
	phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
D
Dan Williams 已提交
129
	void *pmem_addr = pmem->virt_addr + pmem_off;
130

131 132 133
	if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
		bad_pmem = true;

134
	if (!is_write) {
135
		if (unlikely(bad_pmem))
136
			rc = BLK_STS_IOERR;
137
		else {
138
			rc = read_pmem(page, off, pmem_addr, len);
139 140
			flush_dcache_page(page);
		}
141
	} else {
142 143 144 145 146 147 148 149 150 151 152 153 154 155
		/*
		 * Note that we write the data both before and after
		 * clearing poison.  The write before clear poison
		 * handles situations where the latest written data is
		 * preserved and the clear poison operation simply marks
		 * the address range as valid without changing the data.
		 * In this case application software can assume that an
		 * interrupted write will either return the new good
		 * data or an error.
		 *
		 * However, if pmem_clear_poison() leaves the data in an
		 * indeterminate state we need to perform the write
		 * after clear poison.
		 */
156
		flush_dcache_page(page);
157
		write_pmem(pmem_addr, page, off, len);
158
		if (unlikely(bad_pmem)) {
159
			rc = pmem_clear_poison(pmem, pmem_off, len);
160
			write_pmem(pmem_addr, page, off, len);
161
		}
162 163
	}

164
	return rc;
165 166
}

167
static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
168
{
169
	blk_status_t rc = 0;
D
Dan Williams 已提交
170 171
	bool do_acct;
	unsigned long start;
172 173
	struct bio_vec bvec;
	struct bvec_iter iter;
174
	struct pmem_device *pmem = q->queuedata;
175 176
	struct nd_region *nd_region = to_region(pmem);

177
	if (bio->bi_opf & REQ_PREFLUSH)
178
		nvdimm_flush(nd_region);
179

D
Dan Williams 已提交
180
	do_acct = nd_iostat_start(bio, &start);
181 182
	bio_for_each_segment(bvec, bio, iter) {
		rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
183
				bvec.bv_offset, op_is_write(bio_op(bio)),
184 185
				iter.bi_sector);
		if (rc) {
186
			bio->bi_status = rc;
187 188 189
			break;
		}
	}
D
Dan Williams 已提交
190 191
	if (do_acct)
		nd_iostat_end(bio, start);
192

J
Jens Axboe 已提交
193
	if (bio->bi_opf & REQ_FUA)
194
		nvdimm_flush(nd_region);
195

196
	bio_endio(bio);
197
	return BLK_QC_T_NONE;
198 199 200
}

static int pmem_rw_page(struct block_device *bdev, sector_t sector,
201
		       struct page *page, bool is_write)
202
{
203
	struct pmem_device *pmem = bdev->bd_queue->queuedata;
204
	blk_status_t rc;
205

206 207
	rc = pmem_do_bvec(pmem, page, hpage_nr_pages(page) * PAGE_SIZE,
			  0, is_write, sector);
208

209 210 211 212 213 214 215
	/*
	 * The ->rw_page interface is subtle and tricky.  The core
	 * retries on any error, so we can only invoke page_endio() in
	 * the successful completion case.  Otherwise, we'll see crashes
	 * caused by double completion.
	 */
	if (rc == 0)
216
		page_endio(page, is_write, 0);
217

218
	return blk_status_to_errno(rc);
219 220
}

221
/* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
D
Dan Williams 已提交
222 223
__weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
		long nr_pages, void **kaddr, pfn_t *pfn)
224
{
D
Dan Williams 已提交
225
	resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
226

D
Dan Williams 已提交
227 228
	if (unlikely(is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) / 512,
					PFN_PHYS(nr_pages))))
229
		return -EIO;
230
	*kaddr = pmem->virt_addr + offset;
D
Dan Williams 已提交
231
	*pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
232

233 234 235 236 237
	/*
	 * If badblocks are present, limit known good range to the
	 * requested range.
	 */
	if (unlikely(pmem->bb.count))
D
Dan Williams 已提交
238 239
		return nr_pages;
	return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
240 241 242 243 244
}

static const struct block_device_operations pmem_fops = {
	.owner =		THIS_MODULE,
	.rw_page =		pmem_rw_page,
245
	.revalidate_disk =	nvdimm_revalidate_disk,
246 247
};

D
Dan Williams 已提交
248 249 250 251 252 253 254 255
static long pmem_dax_direct_access(struct dax_device *dax_dev,
		pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn)
{
	struct pmem_device *pmem = dax_get_private(dax_dev);

	return __pmem_direct_access(pmem, pgoff, nr_pages, kaddr, pfn);
}

256 257 258 259 260 261
static size_t pmem_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
		void *addr, size_t bytes, struct iov_iter *i)
{
	return copy_from_iter_flushcache(addr, bytes, i);
}

262 263 264
static size_t pmem_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
		void *addr, size_t bytes, struct iov_iter *i)
{
265
	return copy_to_iter_mcsafe(addr, bytes, i);
266 267
}

D
Dan Williams 已提交
268 269
static const struct dax_operations pmem_dax_ops = {
	.direct_access = pmem_dax_direct_access,
270
	.copy_from_iter = pmem_copy_from_iter,
271
	.copy_to_iter = pmem_copy_to_iter,
D
Dan Williams 已提交
272 273
};

274 275 276
static const struct attribute_group *pmem_attribute_groups[] = {
	&dax_attribute_group,
	NULL,
D
Dan Williams 已提交
277 278
};

279 280 281 282 283
static void pmem_release_queue(void *q)
{
	blk_cleanup_queue(q);
}

284 285
static void pmem_freeze_queue(void *q)
{
286
	blk_freeze_queue_start(q);
287 288
}

D
Dan Williams 已提交
289
static void pmem_release_disk(void *__pmem)
290
{
D
Dan Williams 已提交
291 292 293 294 295 296
	struct pmem_device *pmem = __pmem;

	kill_dax(pmem->dax_dev);
	put_dax(pmem->dax_dev);
	del_gendisk(pmem->disk);
	put_disk(pmem->disk);
297 298
}

299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319
static void pmem_release_pgmap_ops(void *__pgmap)
{
	dev_pagemap_put_ops();
}

static void fsdax_pagefree(struct page *page, void *data)
{
	wake_up_var(&page->_refcount);
}

static int setup_pagemap_fsdax(struct device *dev, struct dev_pagemap *pgmap)
{
	dev_pagemap_get_ops();
	if (devm_add_action_or_reset(dev, pmem_release_pgmap_ops, pgmap))
		return -ENOMEM;
	pgmap->type = MEMORY_DEVICE_FS_DAX;
	pgmap->page_free = fsdax_pagefree;

	return 0;
}

320 321
static int pmem_attach_disk(struct device *dev,
		struct nd_namespace_common *ndns)
322
{
323
	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
324
	struct nd_region *nd_region = to_nd_region(dev->parent);
325
	int nid = dev_to_node(dev), fua;
326
	struct resource *res = &nsio->res;
327
	struct resource bb_res;
328
	struct nd_pfn *nd_pfn = NULL;
D
Dan Williams 已提交
329
	struct dax_device *dax_dev;
330
	struct nd_pfn_sb *pfn_sb;
331
	struct pmem_device *pmem;
332
	struct request_queue *q;
333
	struct device *gendev;
334 335
	struct gendisk *disk;
	void *addr;
336 337 338 339 340
	int rc;

	pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
	if (!pmem)
		return -ENOMEM;
341 342 343 344

	/* while nsio_rw_bytes is active, parse a pfn info block if present */
	if (is_nd_pfn(dev)) {
		nd_pfn = to_nd_pfn(dev);
345 346 347
		rc = nvdimm_setup_pfn(nd_pfn, &pmem->pgmap);
		if (rc)
			return rc;
348 349 350 351
	}

	/* we're attaching a block device, disable raw namespace access */
	devm_nsio_disable(dev, nsio);
352

353
	dev_set_drvdata(dev, pmem);
354 355
	pmem->phys_addr = res->start;
	pmem->size = resource_size(res);
356 357
	fua = nvdimm_has_flush(nd_region);
	if (!IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) || fua < 0) {
358
		dev_warn(dev, "unable to guarantee persistence of writes\n");
359 360
		fua = 0;
	}
361

362
	if (!devm_request_mem_region(dev, res->start, resource_size(res),
363
				dev_name(&ndns->dev))) {
364
		dev_warn(dev, "could not reserve region %pR\n", res);
365
		return -EBUSY;
366 367
	}

368
	q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev), NULL);
369
	if (!q)
370
		return -ENOMEM;
371

372 373 374
	if (devm_add_action_or_reset(dev, pmem_release_queue, q))
		return -ENOMEM;

D
Dan Williams 已提交
375
	pmem->pfn_flags = PFN_DEV;
376
	pmem->pgmap.ref = &q->q_usage_counter;
377
	if (is_nd_pfn(dev)) {
378 379
		if (setup_pagemap_fsdax(dev, &pmem->pgmap))
			return -ENOMEM;
380
		addr = devm_memremap_pages(dev, &pmem->pgmap);
381 382
		pfn_sb = nd_pfn->pfn_sb;
		pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
383 384
		pmem->pfn_pad = resource_size(res) -
			resource_size(&pmem->pgmap.res);
385
		pmem->pfn_flags |= PFN_MAP;
386 387
		memcpy(&bb_res, &pmem->pgmap.res, sizeof(bb_res));
		bb_res.start += pmem->data_offset;
388
	} else if (pmem_should_map_pages(dev)) {
389 390
		memcpy(&pmem->pgmap.res, &nsio->res, sizeof(pmem->pgmap.res));
		pmem->pgmap.altmap_valid = false;
391 392
		if (setup_pagemap_fsdax(dev, &pmem->pgmap))
			return -ENOMEM;
393
		addr = devm_memremap_pages(dev, &pmem->pgmap);
D
Dan Williams 已提交
394
		pmem->pfn_flags |= PFN_MAP;
395
		memcpy(&bb_res, &pmem->pgmap.res, sizeof(bb_res));
D
Dan Williams 已提交
396
	} else
397 398
		addr = devm_memremap(dev, pmem->phys_addr,
				pmem->size, ARCH_MEMREMAP_PMEM);
399

400
	/*
401
	 * At release time the queue must be frozen before
402 403
	 * devm_memremap_pages is unwound
	 */
404
	if (devm_add_action_or_reset(dev, pmem_freeze_queue, q))
405
		return -ENOMEM;
406

407 408
	if (IS_ERR(addr))
		return PTR_ERR(addr);
D
Dan Williams 已提交
409
	pmem->virt_addr = addr;
410

411
	blk_queue_write_cache(q, true, fua);
412 413
	blk_queue_make_request(q, pmem_make_request);
	blk_queue_physical_block_size(q, PAGE_SIZE);
414
	blk_queue_logical_block_size(q, pmem_sector_size(ndns));
415
	blk_queue_max_hw_sectors(q, UINT_MAX);
416
	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
417 418
	if (pmem->pfn_flags & PFN_MAP)
		blk_queue_flag_set(QUEUE_FLAG_DAX, q);
419
	q->queuedata = pmem;
420

421
	disk = alloc_disk_node(0, nid);
422 423
	if (!disk)
		return -ENOMEM;
D
Dan Williams 已提交
424
	pmem->disk = disk;
425 426

	disk->fops		= &pmem_fops;
427
	disk->queue		= q;
428
	disk->flags		= GENHD_FL_EXT_DEVT;
429
	disk->queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO;
V
Vishal Verma 已提交
430
	nvdimm_namespace_disk_name(ndns, disk->disk_name);
431 432
	set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
			/ 512);
433 434
	if (devm_init_badblocks(dev, &pmem->bb))
		return -ENOMEM;
435
	nvdimm_badblocks_populate(nd_region, &pmem->bb, &bb_res);
436
	disk->bb = &pmem->bb;
437

D
Dan Williams 已提交
438 439 440 441 442
	dax_dev = alloc_dax(pmem, disk->disk_name, &pmem_dax_ops);
	if (!dax_dev) {
		put_disk(disk);
		return -ENOMEM;
	}
443
	dax_write_cache(dax_dev, nvdimm_has_cache(nd_region));
D
Dan Williams 已提交
444 445
	pmem->dax_dev = dax_dev;

446 447 448
	gendev = disk_to_dev(disk);
	gendev->groups = pmem_attribute_groups;

D
Dan Williams 已提交
449 450
	device_add_disk(dev, disk);
	if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
451 452
		return -ENOMEM;

453
	revalidate_disk(disk);
454

455 456
	pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
					  "badblocks");
457 458
	if (!pmem->bb_state)
		dev_warn(dev, "'badblocks' notification disabled\n");
459

460 461
	return 0;
}
462

463
static int nd_pmem_probe(struct device *dev)
464
{
465
	struct nd_namespace_common *ndns;
466

467 468 469
	ndns = nvdimm_namespace_common_probe(dev);
	if (IS_ERR(ndns))
		return PTR_ERR(ndns);
470

471 472
	if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev)))
		return -ENXIO;
473

474
	if (is_nd_btt(dev))
475 476
		return nvdimm_namespace_attach_btt(ndns);

477
	if (is_nd_pfn(dev))
478
		return pmem_attach_disk(dev, ndns);
479

480
	/* if we find a valid info-block we'll come back as that personality */
481 482
	if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0
			|| nd_dax_probe(dev, ndns) == 0)
483 484
		return -ENXIO;

485 486
	/* ...otherwise we're just a raw pmem device */
	return pmem_attach_disk(dev, ndns);
487 488
}

489
static int nd_pmem_remove(struct device *dev)
490
{
491 492
	struct pmem_device *pmem = dev_get_drvdata(dev);

493
	if (is_nd_btt(dev))
D
Dan Williams 已提交
494
		nvdimm_namespace_detach_btt(to_nd_btt(dev));
495 496 497 498 499 500 501 502
	else {
		/*
		 * Note, this assumes device_lock() context to not race
		 * nd_pmem_notify()
		 */
		sysfs_put(pmem->bb_state);
		pmem->bb_state = NULL;
	}
503 504
	nvdimm_flush(to_nd_region(dev->parent));

505 506 507
	return 0;
}

508 509 510 511 512
static void nd_pmem_shutdown(struct device *dev)
{
	nvdimm_flush(to_nd_region(dev->parent));
}

513 514
static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
{
515
	struct nd_region *nd_region;
D
Dan Williams 已提交
516 517 518 519
	resource_size_t offset = 0, end_trunc = 0;
	struct nd_namespace_common *ndns;
	struct nd_namespace_io *nsio;
	struct resource res;
520
	struct badblocks *bb;
521
	struct kernfs_node *bb_state;
522 523 524 525

	if (event != NVDIMM_REVALIDATE_POISON)
		return;

D
Dan Williams 已提交
526 527 528 529
	if (is_nd_btt(dev)) {
		struct nd_btt *nd_btt = to_nd_btt(dev);

		ndns = nd_btt->ndns;
530 531 532
		nd_region = to_nd_region(ndns->dev.parent);
		nsio = to_nd_namespace_io(&ndns->dev);
		bb = &nsio->bb;
533
		bb_state = NULL;
534 535
	} else {
		struct pmem_device *pmem = dev_get_drvdata(dev);
536

537 538
		nd_region = to_region(pmem);
		bb = &pmem->bb;
539
		bb_state = pmem->bb_state;
540 541 542 543 544 545 546 547 548 549 550 551 552 553 554

		if (is_nd_pfn(dev)) {
			struct nd_pfn *nd_pfn = to_nd_pfn(dev);
			struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;

			ndns = nd_pfn->ndns;
			offset = pmem->data_offset +
					__le32_to_cpu(pfn_sb->start_pad);
			end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
		} else {
			ndns = to_ndns(dev);
		}

		nsio = to_nd_namespace_io(&ndns->dev);
	}
555

D
Dan Williams 已提交
556 557
	res.start = nsio->res.start + offset;
	res.end = nsio->res.end - end_trunc;
558
	nvdimm_badblocks_populate(nd_region, bb, &res);
559 560
	if (bb_state)
		sysfs_notify_dirent(bb_state);
561 562
}

563 564
MODULE_ALIAS("pmem");
MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
565
MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
566 567 568
static struct nd_device_driver nd_pmem_driver = {
	.probe = nd_pmem_probe,
	.remove = nd_pmem_remove,
569
	.notify = nd_pmem_notify,
570
	.shutdown = nd_pmem_shutdown,
571 572
	.drv = {
		.name = "nd_pmem",
573
	},
574
	.type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
575 576
};

577
module_nd_driver(nd_pmem_driver);
578 579 580

MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
MODULE_LICENSE("GPL v2");