2 * Persistent Memory Driver
4 * Copyright (c) 2014-2015, Intel Corporation.
5 * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>.
6 * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>.
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms and conditions of the GNU General Public License,
10 * version 2, as published by the Free Software Foundation.
12 * This program is distributed in the hope it will be useful, but WITHOUT
13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
18 #include <asm/cacheflush.h>
19 #include <linux/blkdev.h>
20 #include <linux/hdreg.h>
21 #include <linux/init.h>
22 #include <linux/platform_device.h>
23 #include <linux/module.h>
24 #include <linux/moduleparam.h>
25 #include <linux/badblocks.h>
26 #include <linux/memremap.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blk-mq.h>
29 #include <linux/pfn_t.h>
30 #include <linux/slab.h>
31 #include <linux/pmem.h>
32 #include <linux/dax.h>
38 static struct device *to_dev(struct pmem_device *pmem)
41 * nvdimm bus services need a 'dev' parameter, and we record the device
47 static struct nd_region *to_region(struct pmem_device *pmem)
49 return to_nd_region(to_dev(pmem)->parent);
52 static int pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset,
55 struct device *dev = to_dev(pmem);
60 sector = (offset - pmem->data_offset) / 512;
62 cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
65 if (cleared > 0 && cleared / 512) {
67 dev_dbg(dev, "%s: %#llx clear %ld sector%s\n", __func__,
68 (unsigned long long) sector, cleared,
69 cleared > 1 ? "s" : "");
70 badblocks_clear(&pmem->bb, sector, cleared);
72 sysfs_notify_dirent(pmem->bb_state);
75 invalidate_pmem(pmem->virt_addr + offset, len);
80 static void write_pmem(void *pmem_addr, struct page *page,
81 unsigned int off, unsigned int len)
83 void *mem = kmap_atomic(page);
85 memcpy_to_pmem(pmem_addr, mem + off, len);
89 static int read_pmem(struct page *page, unsigned int off,
90 void *pmem_addr, unsigned int len)
93 void *mem = kmap_atomic(page);
95 rc = memcpy_mcsafe(mem + off, pmem_addr, len);
102 static int pmem_do_bvec(struct pmem_device *pmem, struct page *page,
103 unsigned int len, unsigned int off, bool is_write,
107 bool bad_pmem = false;
108 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
109 void *pmem_addr = pmem->virt_addr + pmem_off;
111 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
115 if (unlikely(bad_pmem))
118 rc = read_pmem(page, off, pmem_addr, len);
119 flush_dcache_page(page);
123 * Note that we write the data both before and after
124 * clearing poison. The write before clear poison
125 * handles situations where the latest written data is
126 * preserved and the clear poison operation simply marks
127 * the address range as valid without changing the data.
128 * In this case application software can assume that an
129 * interrupted write will either return the new good
132 * However, if pmem_clear_poison() leaves the data in an
133 * indeterminate state we need to perform the write
134 * after clear poison.
136 flush_dcache_page(page);
137 write_pmem(pmem_addr, page, off, len);
138 if (unlikely(bad_pmem)) {
139 rc = pmem_clear_poison(pmem, pmem_off, len);
140 write_pmem(pmem_addr, page, off, len);
147 /* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */
149 #define REQ_FLUSH REQ_PREFLUSH
152 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
158 struct bvec_iter iter;
159 struct pmem_device *pmem = q->queuedata;
160 struct nd_region *nd_region = to_region(pmem);
162 if (bio->bi_opf & REQ_FLUSH)
163 nvdimm_flush(nd_region);
165 do_acct = nd_iostat_start(bio, &start);
166 bio_for_each_segment(bvec, bio, iter) {
167 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
168 bvec.bv_offset, op_is_write(bio_op(bio)),
176 nd_iostat_end(bio, start);
178 if (bio->bi_opf & REQ_FUA)
179 nvdimm_flush(nd_region);
182 return BLK_QC_T_NONE;
185 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
186 struct page *page, bool is_write)
188 struct pmem_device *pmem = bdev->bd_queue->queuedata;
191 rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, is_write, sector);
194 * The ->rw_page interface is subtle and tricky. The core
195 * retries on any error, so we can only invoke page_endio() in
196 * the successful completion case. Otherwise, we'll see crashes
197 * caused by double completion.
200 page_endio(page, is_write, 0);
205 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
206 __weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
207 long nr_pages, void **kaddr, pfn_t *pfn)
209 resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
211 if (unlikely(is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) / 512,
212 PFN_PHYS(nr_pages))))
214 *kaddr = pmem->virt_addr + offset;
215 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
218 * If badblocks are present, limit known good range to the
221 if (unlikely(pmem->bb.count))
223 return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
226 static const struct block_device_operations pmem_fops = {
227 .owner = THIS_MODULE,
228 .rw_page = pmem_rw_page,
229 .revalidate_disk = nvdimm_revalidate_disk,
232 static long pmem_dax_direct_access(struct dax_device *dax_dev,
233 pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn)
235 struct pmem_device *pmem = dax_get_private(dax_dev);
237 return __pmem_direct_access(pmem, pgoff, nr_pages, kaddr, pfn);
240 static const struct dax_operations pmem_dax_ops = {
241 .direct_access = pmem_dax_direct_access,
244 static void pmem_release_queue(void *q)
246 blk_cleanup_queue(q);
249 static void pmem_freeze_queue(void *q)
251 blk_freeze_queue_start(q);
254 static void pmem_release_disk(void *__pmem)
256 struct pmem_device *pmem = __pmem;
258 kill_dax(pmem->dax_dev);
259 put_dax(pmem->dax_dev);
260 del_gendisk(pmem->disk);
261 put_disk(pmem->disk);
264 static int pmem_attach_disk(struct device *dev,
265 struct nd_namespace_common *ndns)
267 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
268 struct nd_region *nd_region = to_nd_region(dev->parent);
269 struct vmem_altmap __altmap, *altmap = NULL;
270 struct resource *res = &nsio->res;
271 struct nd_pfn *nd_pfn = NULL;
272 struct dax_device *dax_dev;
273 int nid = dev_to_node(dev);
274 struct nd_pfn_sb *pfn_sb;
275 struct pmem_device *pmem;
276 struct resource pfn_res;
277 struct request_queue *q;
278 struct gendisk *disk;
281 /* while nsio_rw_bytes is active, parse a pfn info block if present */
282 if (is_nd_pfn(dev)) {
283 nd_pfn = to_nd_pfn(dev);
284 altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap);
286 return PTR_ERR(altmap);
289 /* we're attaching a block device, disable raw namespace access */
290 devm_nsio_disable(dev, nsio);
292 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
296 dev_set_drvdata(dev, pmem);
297 pmem->phys_addr = res->start;
298 pmem->size = resource_size(res);
299 if (nvdimm_has_flush(nd_region) < 0)
300 dev_warn(dev, "unable to guarantee persistence of writes\n");
302 if (!devm_request_mem_region(dev, res->start, resource_size(res),
303 dev_name(&ndns->dev))) {
304 dev_warn(dev, "could not reserve region %pR\n", res);
308 q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev));
312 if (devm_add_action_or_reset(dev, pmem_release_queue, q))
315 pmem->pfn_flags = PFN_DEV;
316 if (is_nd_pfn(dev)) {
317 addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter,
319 pfn_sb = nd_pfn->pfn_sb;
320 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
321 pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res);
322 pmem->pfn_flags |= PFN_MAP;
323 res = &pfn_res; /* for badblocks populate */
324 res->start += pmem->data_offset;
325 } else if (pmem_should_map_pages(dev)) {
326 addr = devm_memremap_pages(dev, &nsio->res,
327 &q->q_usage_counter, NULL);
328 pmem->pfn_flags |= PFN_MAP;
330 addr = devm_memremap(dev, pmem->phys_addr,
331 pmem->size, ARCH_MEMREMAP_PMEM);
334 * At release time the queue must be frozen before
335 * devm_memremap_pages is unwound
337 if (devm_add_action_or_reset(dev, pmem_freeze_queue, q))
341 return PTR_ERR(addr);
342 pmem->virt_addr = addr;
344 blk_queue_write_cache(q, true, true);
345 blk_queue_make_request(q, pmem_make_request);
346 blk_queue_physical_block_size(q, PAGE_SIZE);
347 blk_queue_logical_block_size(q, pmem_sector_size(ndns));
348 blk_queue_max_hw_sectors(q, UINT_MAX);
349 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
350 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
351 queue_flag_set_unlocked(QUEUE_FLAG_DAX, q);
354 disk = alloc_disk_node(0, nid);
359 disk->fops = &pmem_fops;
361 disk->flags = GENHD_FL_EXT_DEVT;
362 nvdimm_namespace_disk_name(ndns, disk->disk_name);
363 set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
365 if (devm_init_badblocks(dev, &pmem->bb))
367 nvdimm_badblocks_populate(nd_region, &pmem->bb, res);
368 disk->bb = &pmem->bb;
370 dax_dev = alloc_dax(pmem, disk->disk_name, &pmem_dax_ops);
375 pmem->dax_dev = dax_dev;
377 device_add_disk(dev, disk);
378 if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
381 revalidate_disk(disk);
383 pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
386 dev_warn(dev, "'badblocks' notification disabled\n");
391 static int nd_pmem_probe(struct device *dev)
393 struct nd_namespace_common *ndns;
395 ndns = nvdimm_namespace_common_probe(dev);
397 return PTR_ERR(ndns);
399 if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev)))
403 return nvdimm_namespace_attach_btt(ndns);
406 return pmem_attach_disk(dev, ndns);
408 /* if we find a valid info-block we'll come back as that personality */
409 if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0
410 || nd_dax_probe(dev, ndns) == 0)
413 /* ...otherwise we're just a raw pmem device */
414 return pmem_attach_disk(dev, ndns);
417 static int nd_pmem_remove(struct device *dev)
419 struct pmem_device *pmem = dev_get_drvdata(dev);
422 nvdimm_namespace_detach_btt(to_nd_btt(dev));
425 * Note, this assumes device_lock() context to not race
428 sysfs_put(pmem->bb_state);
429 pmem->bb_state = NULL;
431 nvdimm_flush(to_nd_region(dev->parent));
436 static void nd_pmem_shutdown(struct device *dev)
438 nvdimm_flush(to_nd_region(dev->parent));
441 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
443 struct nd_region *nd_region;
444 resource_size_t offset = 0, end_trunc = 0;
445 struct nd_namespace_common *ndns;
446 struct nd_namespace_io *nsio;
448 struct badblocks *bb;
449 struct kernfs_node *bb_state;
451 if (event != NVDIMM_REVALIDATE_POISON)
454 if (is_nd_btt(dev)) {
455 struct nd_btt *nd_btt = to_nd_btt(dev);
458 nd_region = to_nd_region(ndns->dev.parent);
459 nsio = to_nd_namespace_io(&ndns->dev);
463 struct pmem_device *pmem = dev_get_drvdata(dev);
465 nd_region = to_region(pmem);
467 bb_state = pmem->bb_state;
469 if (is_nd_pfn(dev)) {
470 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
471 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
474 offset = pmem->data_offset +
475 __le32_to_cpu(pfn_sb->start_pad);
476 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
481 nsio = to_nd_namespace_io(&ndns->dev);
484 res.start = nsio->res.start + offset;
485 res.end = nsio->res.end - end_trunc;
486 nvdimm_badblocks_populate(nd_region, bb, &res);
488 sysfs_notify_dirent(bb_state);
491 MODULE_ALIAS("pmem");
492 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
493 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
494 static struct nd_device_driver nd_pmem_driver = {
495 .probe = nd_pmem_probe,
496 .remove = nd_pmem_remove,
497 .notify = nd_pmem_notify,
498 .shutdown = nd_pmem_shutdown,
502 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
505 static int __init pmem_init(void)
507 return nd_driver_register(&nd_pmem_driver);
509 module_init(pmem_init);
511 static void pmem_exit(void)
513 driver_unregister(&nd_pmem_driver.drv);
515 module_exit(pmem_exit);
517 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
518 MODULE_LICENSE("GPL v2");