Commit | Line | Data |
---|---|---|
9e853f23 RZ |
1 | /* |
2 | * Persistent Memory Driver | |
3 | * | |
9f53f9fa | 4 | * Copyright (c) 2014-2015, Intel Corporation. |
9e853f23 RZ |
5 | * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>. |
6 | * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>. | |
7 | * | |
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. | |
11 | * | |
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 | |
15 | * more details. | |
16 | */ | |
17 | ||
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> | |
b95f5f43 | 25 | #include <linux/badblocks.h> |
9476df7d | 26 | #include <linux/memremap.h> |
32ab0a3f | 27 | #include <linux/vmalloc.h> |
34c0fd54 | 28 | #include <linux/pfn_t.h> |
9e853f23 | 29 | #include <linux/slab.h> |
61031952 | 30 | #include <linux/pmem.h> |
9f53f9fa | 31 | #include <linux/nd.h> |
f295e53b | 32 | #include "pmem.h" |
32ab0a3f | 33 | #include "pfn.h" |
9f53f9fa | 34 | #include "nd.h" |
9e853f23 | 35 | |
f284a4f2 DW |
36 | static struct device *to_dev(struct pmem_device *pmem) |
37 | { | |
38 | /* | |
39 | * nvdimm bus services need a 'dev' parameter, and we record the device | |
40 | * at init in bb.dev. | |
41 | */ | |
42 | return pmem->bb.dev; | |
43 | } | |
44 | ||
45 | static struct nd_region *to_region(struct pmem_device *pmem) | |
46 | { | |
47 | return to_nd_region(to_dev(pmem)->parent); | |
48 | } | |
49 | ||
59e64739 DW |
50 | static void pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset, |
51 | unsigned int len) | |
52 | { | |
f284a4f2 | 53 | struct device *dev = to_dev(pmem); |
59e64739 DW |
54 | sector_t sector; |
55 | long cleared; | |
56 | ||
57 | sector = (offset - pmem->data_offset) / 512; | |
58 | cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len); | |
59 | ||
60 | if (cleared > 0 && cleared / 512) { | |
61 | dev_dbg(dev, "%s: %llx clear %ld sector%s\n", | |
62 | __func__, (unsigned long long) sector, | |
63 | cleared / 512, cleared / 512 > 1 ? "s" : ""); | |
64 | badblocks_clear(&pmem->bb, sector, cleared / 512); | |
65 | } | |
66 | invalidate_pmem(pmem->virt_addr + offset, len); | |
67 | } | |
68 | ||
e10624f8 | 69 | static int pmem_do_bvec(struct pmem_device *pmem, struct page *page, |
9e853f23 RZ |
70 | unsigned int len, unsigned int off, int rw, |
71 | sector_t sector) | |
72 | { | |
b5ebc8ec | 73 | int rc = 0; |
59e64739 | 74 | bool bad_pmem = false; |
9e853f23 | 75 | void *mem = kmap_atomic(page); |
32ab0a3f | 76 | phys_addr_t pmem_off = sector * 512 + pmem->data_offset; |
61031952 | 77 | void __pmem *pmem_addr = pmem->virt_addr + pmem_off; |
9e853f23 | 78 | |
59e64739 DW |
79 | if (unlikely(is_bad_pmem(&pmem->bb, sector, len))) |
80 | bad_pmem = true; | |
81 | ||
9e853f23 | 82 | if (rw == READ) { |
59e64739 | 83 | if (unlikely(bad_pmem)) |
b5ebc8ec DW |
84 | rc = -EIO; |
85 | else { | |
fc0c2028 | 86 | rc = memcpy_from_pmem(mem + off, pmem_addr, len); |
b5ebc8ec DW |
87 | flush_dcache_page(page); |
88 | } | |
9e853f23 | 89 | } else { |
0a370d26 DW |
90 | /* |
91 | * Note that we write the data both before and after | |
92 | * clearing poison. The write before clear poison | |
93 | * handles situations where the latest written data is | |
94 | * preserved and the clear poison operation simply marks | |
95 | * the address range as valid without changing the data. | |
96 | * In this case application software can assume that an | |
97 | * interrupted write will either return the new good | |
98 | * data or an error. | |
99 | * | |
100 | * However, if pmem_clear_poison() leaves the data in an | |
101 | * indeterminate state we need to perform the write | |
102 | * after clear poison. | |
103 | */ | |
9e853f23 | 104 | flush_dcache_page(page); |
61031952 | 105 | memcpy_to_pmem(pmem_addr, mem + off, len); |
59e64739 DW |
106 | if (unlikely(bad_pmem)) { |
107 | pmem_clear_poison(pmem, pmem_off, len); | |
108 | memcpy_to_pmem(pmem_addr, mem + off, len); | |
109 | } | |
9e853f23 RZ |
110 | } |
111 | ||
112 | kunmap_atomic(mem); | |
b5ebc8ec | 113 | return rc; |
9e853f23 RZ |
114 | } |
115 | ||
dece1635 | 116 | static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio) |
9e853f23 | 117 | { |
e10624f8 | 118 | int rc = 0; |
f0dc089c DW |
119 | bool do_acct; |
120 | unsigned long start; | |
9e853f23 | 121 | struct bio_vec bvec; |
9e853f23 | 122 | struct bvec_iter iter; |
bd842b8c | 123 | struct pmem_device *pmem = q->queuedata; |
9e853f23 | 124 | |
f0dc089c | 125 | do_acct = nd_iostat_start(bio, &start); |
e10624f8 DW |
126 | bio_for_each_segment(bvec, bio, iter) { |
127 | rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, | |
128 | bvec.bv_offset, bio_data_dir(bio), | |
129 | iter.bi_sector); | |
130 | if (rc) { | |
131 | bio->bi_error = rc; | |
132 | break; | |
133 | } | |
134 | } | |
f0dc089c DW |
135 | if (do_acct) |
136 | nd_iostat_end(bio, start); | |
61031952 RZ |
137 | |
138 | if (bio_data_dir(bio)) | |
f284a4f2 | 139 | nvdimm_flush(to_region(pmem)); |
61031952 | 140 | |
4246a0b6 | 141 | bio_endio(bio); |
dece1635 | 142 | return BLK_QC_T_NONE; |
9e853f23 RZ |
143 | } |
144 | ||
145 | static int pmem_rw_page(struct block_device *bdev, sector_t sector, | |
146 | struct page *page, int rw) | |
147 | { | |
bd842b8c | 148 | struct pmem_device *pmem = bdev->bd_queue->queuedata; |
e10624f8 | 149 | int rc; |
9e853f23 | 150 | |
09cbfeaf | 151 | rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, rw, sector); |
ba8fe0f8 | 152 | if (rw & WRITE) |
f284a4f2 | 153 | nvdimm_flush(to_region(pmem)); |
9e853f23 | 154 | |
e10624f8 DW |
155 | /* |
156 | * The ->rw_page interface is subtle and tricky. The core | |
157 | * retries on any error, so we can only invoke page_endio() in | |
158 | * the successful completion case. Otherwise, we'll see crashes | |
159 | * caused by double completion. | |
160 | */ | |
161 | if (rc == 0) | |
162 | page_endio(page, rw & WRITE, 0); | |
163 | ||
164 | return rc; | |
9e853f23 RZ |
165 | } |
166 | ||
f295e53b DW |
167 | /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */ |
168 | __weak long pmem_direct_access(struct block_device *bdev, sector_t sector, | |
0a70bd43 | 169 | void __pmem **kaddr, pfn_t *pfn, long size) |
9e853f23 | 170 | { |
bd842b8c | 171 | struct pmem_device *pmem = bdev->bd_queue->queuedata; |
32ab0a3f | 172 | resource_size_t offset = sector * 512 + pmem->data_offset; |
589e75d1 | 173 | |
0a70bd43 DW |
174 | if (unlikely(is_bad_pmem(&pmem->bb, sector, size))) |
175 | return -EIO; | |
e2e05394 | 176 | *kaddr = pmem->virt_addr + offset; |
34c0fd54 | 177 | *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags); |
9e853f23 | 178 | |
0a70bd43 DW |
179 | /* |
180 | * If badblocks are present, limit known good range to the | |
181 | * requested range. | |
182 | */ | |
183 | if (unlikely(pmem->bb.count)) | |
184 | return size; | |
cfe30b87 | 185 | return pmem->size - pmem->pfn_pad - offset; |
9e853f23 RZ |
186 | } |
187 | ||
188 | static const struct block_device_operations pmem_fops = { | |
189 | .owner = THIS_MODULE, | |
190 | .rw_page = pmem_rw_page, | |
191 | .direct_access = pmem_direct_access, | |
58138820 | 192 | .revalidate_disk = nvdimm_revalidate_disk, |
9e853f23 RZ |
193 | }; |
194 | ||
030b99e3 DW |
195 | static void pmem_release_queue(void *q) |
196 | { | |
197 | blk_cleanup_queue(q); | |
198 | } | |
199 | ||
f02716db | 200 | static void pmem_release_disk(void *disk) |
030b99e3 DW |
201 | { |
202 | del_gendisk(disk); | |
203 | put_disk(disk); | |
204 | } | |
205 | ||
200c79da DW |
206 | static int pmem_attach_disk(struct device *dev, |
207 | struct nd_namespace_common *ndns) | |
9e853f23 | 208 | { |
200c79da | 209 | struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); |
f284a4f2 | 210 | struct nd_region *nd_region = to_nd_region(dev->parent); |
200c79da DW |
211 | struct vmem_altmap __altmap, *altmap = NULL; |
212 | struct resource *res = &nsio->res; | |
213 | struct nd_pfn *nd_pfn = NULL; | |
214 | int nid = dev_to_node(dev); | |
215 | struct nd_pfn_sb *pfn_sb; | |
9e853f23 | 216 | struct pmem_device *pmem; |
200c79da | 217 | struct resource pfn_res; |
468ded03 | 218 | struct request_queue *q; |
200c79da DW |
219 | struct gendisk *disk; |
220 | void *addr; | |
221 | ||
222 | /* while nsio_rw_bytes is active, parse a pfn info block if present */ | |
223 | if (is_nd_pfn(dev)) { | |
224 | nd_pfn = to_nd_pfn(dev); | |
225 | altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap); | |
226 | if (IS_ERR(altmap)) | |
227 | return PTR_ERR(altmap); | |
228 | } | |
229 | ||
230 | /* we're attaching a block device, disable raw namespace access */ | |
231 | devm_nsio_disable(dev, nsio); | |
9e853f23 | 232 | |
708ab62b | 233 | pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL); |
9e853f23 | 234 | if (!pmem) |
200c79da | 235 | return -ENOMEM; |
9e853f23 | 236 | |
200c79da | 237 | dev_set_drvdata(dev, pmem); |
9e853f23 RZ |
238 | pmem->phys_addr = res->start; |
239 | pmem->size = resource_size(res); | |
f284a4f2 | 240 | if (nvdimm_has_flush(nd_region) < 0) |
61031952 | 241 | dev_warn(dev, "unable to guarantee persistence of writes\n"); |
9e853f23 | 242 | |
947df02d DW |
243 | if (!devm_request_mem_region(dev, res->start, resource_size(res), |
244 | dev_name(dev))) { | |
245 | dev_warn(dev, "could not reserve region %pR\n", res); | |
200c79da | 246 | return -EBUSY; |
9e853f23 RZ |
247 | } |
248 | ||
468ded03 DW |
249 | q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev)); |
250 | if (!q) | |
200c79da | 251 | return -ENOMEM; |
468ded03 | 252 | |
34c0fd54 | 253 | pmem->pfn_flags = PFN_DEV; |
200c79da DW |
254 | if (is_nd_pfn(dev)) { |
255 | addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter, | |
256 | altmap); | |
257 | pfn_sb = nd_pfn->pfn_sb; | |
258 | pmem->data_offset = le64_to_cpu(pfn_sb->dataoff); | |
259 | pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res); | |
260 | pmem->pfn_flags |= PFN_MAP; | |
261 | res = &pfn_res; /* for badblocks populate */ | |
262 | res->start += pmem->data_offset; | |
263 | } else if (pmem_should_map_pages(dev)) { | |
264 | addr = devm_memremap_pages(dev, &nsio->res, | |
5c2c2587 | 265 | &q->q_usage_counter, NULL); |
34c0fd54 DW |
266 | pmem->pfn_flags |= PFN_MAP; |
267 | } else | |
200c79da DW |
268 | addr = devm_memremap(dev, pmem->phys_addr, |
269 | pmem->size, ARCH_MEMREMAP_PMEM); | |
b36f4761 | 270 | |
030b99e3 DW |
271 | /* |
272 | * At release time the queue must be dead before | |
273 | * devm_memremap_pages is unwound | |
274 | */ | |
f02716db | 275 | if (devm_add_action_or_reset(dev, pmem_release_queue, q)) |
200c79da | 276 | return -ENOMEM; |
8c2f7e86 | 277 | |
200c79da DW |
278 | if (IS_ERR(addr)) |
279 | return PTR_ERR(addr); | |
280 | pmem->virt_addr = (void __pmem *) addr; | |
9e853f23 | 281 | |
5a92289f DW |
282 | blk_queue_make_request(q, pmem_make_request); |
283 | blk_queue_physical_block_size(q, PAGE_SIZE); | |
284 | blk_queue_max_hw_sectors(q, UINT_MAX); | |
285 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); | |
286 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q); | |
287 | q->queuedata = pmem; | |
9e853f23 | 288 | |
538ea4aa | 289 | disk = alloc_disk_node(0, nid); |
030b99e3 DW |
290 | if (!disk) |
291 | return -ENOMEM; | |
9e853f23 | 292 | |
9e853f23 | 293 | disk->fops = &pmem_fops; |
5a92289f | 294 | disk->queue = q; |
9e853f23 | 295 | disk->flags = GENHD_FL_EXT_DEVT; |
5212e11f | 296 | nvdimm_namespace_disk_name(ndns, disk->disk_name); |
32ab0a3f | 297 | disk->driverfs_dev = dev; |
cfe30b87 DW |
298 | set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset) |
299 | / 512); | |
b95f5f43 DW |
300 | if (devm_init_badblocks(dev, &pmem->bb)) |
301 | return -ENOMEM; | |
f284a4f2 | 302 | nvdimm_badblocks_populate(nd_region, &pmem->bb, res); |
57f7f317 | 303 | disk->bb = &pmem->bb; |
9e853f23 | 304 | add_disk(disk); |
f02716db DW |
305 | |
306 | if (devm_add_action_or_reset(dev, pmem_release_disk, disk)) | |
307 | return -ENOMEM; | |
308 | ||
58138820 | 309 | revalidate_disk(disk); |
9e853f23 | 310 | |
8c2f7e86 DW |
311 | return 0; |
312 | } | |
9e853f23 | 313 | |
9f53f9fa | 314 | static int nd_pmem_probe(struct device *dev) |
9e853f23 | 315 | { |
8c2f7e86 | 316 | struct nd_namespace_common *ndns; |
9e853f23 | 317 | |
8c2f7e86 DW |
318 | ndns = nvdimm_namespace_common_probe(dev); |
319 | if (IS_ERR(ndns)) | |
320 | return PTR_ERR(ndns); | |
bf9bccc1 | 321 | |
200c79da DW |
322 | if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev))) |
323 | return -ENXIO; | |
708ab62b | 324 | |
200c79da | 325 | if (is_nd_btt(dev)) |
708ab62b CH |
326 | return nvdimm_namespace_attach_btt(ndns); |
327 | ||
32ab0a3f | 328 | if (is_nd_pfn(dev)) |
200c79da | 329 | return pmem_attach_disk(dev, ndns); |
32ab0a3f | 330 | |
200c79da | 331 | /* if we find a valid info-block we'll come back as that personality */ |
c5ed9268 DW |
332 | if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0 |
333 | || nd_dax_probe(dev, ndns) == 0) | |
32ab0a3f | 334 | return -ENXIO; |
32ab0a3f | 335 | |
200c79da DW |
336 | /* ...otherwise we're just a raw pmem device */ |
337 | return pmem_attach_disk(dev, ndns); | |
9e853f23 RZ |
338 | } |
339 | ||
9f53f9fa | 340 | static int nd_pmem_remove(struct device *dev) |
9e853f23 | 341 | { |
8c2f7e86 | 342 | if (is_nd_btt(dev)) |
298f2bc5 | 343 | nvdimm_namespace_detach_btt(to_nd_btt(dev)); |
9e853f23 RZ |
344 | return 0; |
345 | } | |
346 | ||
71999466 DW |
347 | static void nd_pmem_notify(struct device *dev, enum nvdimm_event event) |
348 | { | |
298f2bc5 | 349 | struct pmem_device *pmem = dev_get_drvdata(dev); |
f284a4f2 | 350 | struct nd_region *nd_region = to_region(pmem); |
298f2bc5 DW |
351 | resource_size_t offset = 0, end_trunc = 0; |
352 | struct nd_namespace_common *ndns; | |
353 | struct nd_namespace_io *nsio; | |
354 | struct resource res; | |
71999466 DW |
355 | |
356 | if (event != NVDIMM_REVALIDATE_POISON) | |
357 | return; | |
358 | ||
298f2bc5 DW |
359 | if (is_nd_btt(dev)) { |
360 | struct nd_btt *nd_btt = to_nd_btt(dev); | |
361 | ||
362 | ndns = nd_btt->ndns; | |
363 | } else if (is_nd_pfn(dev)) { | |
a3901802 DW |
364 | struct nd_pfn *nd_pfn = to_nd_pfn(dev); |
365 | struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb; | |
366 | ||
298f2bc5 DW |
367 | ndns = nd_pfn->ndns; |
368 | offset = pmem->data_offset + __le32_to_cpu(pfn_sb->start_pad); | |
369 | end_trunc = __le32_to_cpu(pfn_sb->end_trunc); | |
370 | } else | |
371 | ndns = to_ndns(dev); | |
a3901802 | 372 | |
298f2bc5 DW |
373 | nsio = to_nd_namespace_io(&ndns->dev); |
374 | res.start = nsio->res.start + offset; | |
375 | res.end = nsio->res.end - end_trunc; | |
a3901802 | 376 | nvdimm_badblocks_populate(nd_region, &pmem->bb, &res); |
71999466 DW |
377 | } |
378 | ||
9f53f9fa DW |
379 | MODULE_ALIAS("pmem"); |
380 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); | |
bf9bccc1 | 381 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); |
9f53f9fa DW |
382 | static struct nd_device_driver nd_pmem_driver = { |
383 | .probe = nd_pmem_probe, | |
384 | .remove = nd_pmem_remove, | |
71999466 | 385 | .notify = nd_pmem_notify, |
9f53f9fa DW |
386 | .drv = { |
387 | .name = "nd_pmem", | |
9e853f23 | 388 | }, |
bf9bccc1 | 389 | .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, |
9e853f23 RZ |
390 | }; |
391 | ||
392 | static int __init pmem_init(void) | |
393 | { | |
55155291 | 394 | return nd_driver_register(&nd_pmem_driver); |
9e853f23 RZ |
395 | } |
396 | module_init(pmem_init); | |
397 | ||
398 | static void pmem_exit(void) | |
399 | { | |
9f53f9fa | 400 | driver_unregister(&nd_pmem_driver.drv); |
9e853f23 RZ |
401 | } |
402 | module_exit(pmem_exit); | |
403 | ||
404 | MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); | |
405 | MODULE_LICENSE("GPL v2"); |