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> | |
32ab0a3f | 24 | #include <linux/memory_hotplug.h> |
9e853f23 | 25 | #include <linux/moduleparam.h> |
32ab0a3f | 26 | #include <linux/vmalloc.h> |
9e853f23 | 27 | #include <linux/slab.h> |
61031952 | 28 | #include <linux/pmem.h> |
9f53f9fa | 29 | #include <linux/nd.h> |
32ab0a3f | 30 | #include "pfn.h" |
9f53f9fa | 31 | #include "nd.h" |
9e853f23 RZ |
32 | |
33 | struct pmem_device { | |
34 | struct request_queue *pmem_queue; | |
35 | struct gendisk *pmem_disk; | |
32ab0a3f | 36 | struct nd_namespace_common *ndns; |
9e853f23 RZ |
37 | |
38 | /* One contiguous memory region per device */ | |
39 | phys_addr_t phys_addr; | |
32ab0a3f DW |
40 | /* when non-zero this device is hosting a 'pfn' instance */ |
41 | phys_addr_t data_offset; | |
61031952 | 42 | void __pmem *virt_addr; |
9e853f23 RZ |
43 | size_t size; |
44 | }; | |
45 | ||
46 | static int pmem_major; | |
9e853f23 RZ |
47 | |
48 | static void pmem_do_bvec(struct pmem_device *pmem, struct page *page, | |
49 | unsigned int len, unsigned int off, int rw, | |
50 | sector_t sector) | |
51 | { | |
52 | void *mem = kmap_atomic(page); | |
32ab0a3f | 53 | phys_addr_t pmem_off = sector * 512 + pmem->data_offset; |
61031952 | 54 | void __pmem *pmem_addr = pmem->virt_addr + pmem_off; |
9e853f23 RZ |
55 | |
56 | if (rw == READ) { | |
61031952 | 57 | memcpy_from_pmem(mem + off, pmem_addr, len); |
9e853f23 RZ |
58 | flush_dcache_page(page); |
59 | } else { | |
60 | flush_dcache_page(page); | |
61031952 | 61 | memcpy_to_pmem(pmem_addr, mem + off, len); |
9e853f23 RZ |
62 | } |
63 | ||
64 | kunmap_atomic(mem); | |
65 | } | |
66 | ||
67 | static void pmem_make_request(struct request_queue *q, struct bio *bio) | |
68 | { | |
f0dc089c DW |
69 | bool do_acct; |
70 | unsigned long start; | |
9e853f23 | 71 | struct bio_vec bvec; |
9e853f23 | 72 | struct bvec_iter iter; |
edc870e5 DW |
73 | struct block_device *bdev = bio->bi_bdev; |
74 | struct pmem_device *pmem = bdev->bd_disk->private_data; | |
9e853f23 | 75 | |
f0dc089c | 76 | do_acct = nd_iostat_start(bio, &start); |
edc870e5 | 77 | bio_for_each_segment(bvec, bio, iter) |
9e853f23 | 78 | pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, bvec.bv_offset, |
edc870e5 | 79 | bio_data_dir(bio), iter.bi_sector); |
f0dc089c DW |
80 | if (do_acct) |
81 | nd_iostat_end(bio, start); | |
61031952 RZ |
82 | |
83 | if (bio_data_dir(bio)) | |
84 | wmb_pmem(); | |
85 | ||
4246a0b6 | 86 | bio_endio(bio); |
9e853f23 RZ |
87 | } |
88 | ||
89 | static int pmem_rw_page(struct block_device *bdev, sector_t sector, | |
90 | struct page *page, int rw) | |
91 | { | |
92 | struct pmem_device *pmem = bdev->bd_disk->private_data; | |
93 | ||
94 | pmem_do_bvec(pmem, page, PAGE_CACHE_SIZE, 0, rw, sector); | |
ba8fe0f8 RZ |
95 | if (rw & WRITE) |
96 | wmb_pmem(); | |
9e853f23 RZ |
97 | page_endio(page, rw & WRITE, 0); |
98 | ||
99 | return 0; | |
100 | } | |
101 | ||
102 | static long pmem_direct_access(struct block_device *bdev, sector_t sector, | |
cb389b9c | 103 | void __pmem **kaddr, unsigned long *pfn) |
9e853f23 RZ |
104 | { |
105 | struct pmem_device *pmem = bdev->bd_disk->private_data; | |
32ab0a3f DW |
106 | resource_size_t offset = sector * 512 + pmem->data_offset; |
107 | resource_size_t size; | |
108 | ||
109 | if (pmem->data_offset) { | |
110 | /* | |
111 | * Limit the direct_access() size to what is covered by | |
112 | * the memmap | |
113 | */ | |
114 | size = (pmem->size - offset) & ~ND_PFN_MASK; | |
115 | } else | |
116 | size = pmem->size - offset; | |
9e853f23 | 117 | |
61031952 | 118 | /* FIXME convert DAX to comprehend that this mapping has a lifetime */ |
e2e05394 | 119 | *kaddr = pmem->virt_addr + offset; |
9e853f23 RZ |
120 | *pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT; |
121 | ||
32ab0a3f | 122 | return size; |
9e853f23 RZ |
123 | } |
124 | ||
125 | static const struct block_device_operations pmem_fops = { | |
126 | .owner = THIS_MODULE, | |
127 | .rw_page = pmem_rw_page, | |
128 | .direct_access = pmem_direct_access, | |
58138820 | 129 | .revalidate_disk = nvdimm_revalidate_disk, |
9e853f23 RZ |
130 | }; |
131 | ||
9f53f9fa DW |
132 | static struct pmem_device *pmem_alloc(struct device *dev, |
133 | struct resource *res, int id) | |
9e853f23 RZ |
134 | { |
135 | struct pmem_device *pmem; | |
9e853f23 | 136 | |
708ab62b | 137 | pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL); |
9e853f23 | 138 | if (!pmem) |
8c2f7e86 | 139 | return ERR_PTR(-ENOMEM); |
9e853f23 RZ |
140 | |
141 | pmem->phys_addr = res->start; | |
142 | pmem->size = resource_size(res); | |
96601adb | 143 | if (!arch_has_wmb_pmem()) |
61031952 | 144 | dev_warn(dev, "unable to guarantee persistence of writes\n"); |
9e853f23 | 145 | |
708ab62b CH |
146 | if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size, |
147 | dev_name(dev))) { | |
9f53f9fa DW |
148 | dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n", |
149 | &pmem->phys_addr, pmem->size); | |
8c2f7e86 | 150 | return ERR_PTR(-EBUSY); |
9e853f23 RZ |
151 | } |
152 | ||
004f1afb DW |
153 | if (pmem_should_map_pages(dev)) { |
154 | void *addr = devm_memremap_pages(dev, res); | |
155 | ||
156 | if (IS_ERR(addr)) | |
157 | return addr; | |
158 | pmem->virt_addr = (void __pmem *) addr; | |
159 | } else { | |
160 | pmem->virt_addr = memremap_pmem(dev, pmem->phys_addr, | |
161 | pmem->size); | |
162 | if (!pmem->virt_addr) | |
163 | return ERR_PTR(-ENXIO); | |
8c2f7e86 DW |
164 | } |
165 | ||
166 | return pmem; | |
167 | } | |
168 | ||
169 | static void pmem_detach_disk(struct pmem_device *pmem) | |
170 | { | |
32ab0a3f DW |
171 | if (!pmem->pmem_disk) |
172 | return; | |
173 | ||
8c2f7e86 DW |
174 | del_gendisk(pmem->pmem_disk); |
175 | put_disk(pmem->pmem_disk); | |
176 | blk_cleanup_queue(pmem->pmem_queue); | |
177 | } | |
178 | ||
32ab0a3f DW |
179 | static int pmem_attach_disk(struct device *dev, |
180 | struct nd_namespace_common *ndns, struct pmem_device *pmem) | |
8c2f7e86 | 181 | { |
8c2f7e86 | 182 | struct gendisk *disk; |
9e853f23 RZ |
183 | |
184 | pmem->pmem_queue = blk_alloc_queue(GFP_KERNEL); | |
185 | if (!pmem->pmem_queue) | |
8c2f7e86 | 186 | return -ENOMEM; |
9e853f23 RZ |
187 | |
188 | blk_queue_make_request(pmem->pmem_queue, pmem_make_request); | |
6b47496a | 189 | blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE); |
43d3fa3a | 190 | blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX); |
9e853f23 | 191 | blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY); |
0f51c4fa | 192 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue); |
9e853f23 | 193 | |
9f53f9fa | 194 | disk = alloc_disk(0); |
8c2f7e86 DW |
195 | if (!disk) { |
196 | blk_cleanup_queue(pmem->pmem_queue); | |
197 | return -ENOMEM; | |
198 | } | |
9e853f23 | 199 | |
9e853f23 | 200 | disk->major = pmem_major; |
9f53f9fa | 201 | disk->first_minor = 0; |
9e853f23 RZ |
202 | disk->fops = &pmem_fops; |
203 | disk->private_data = pmem; | |
204 | disk->queue = pmem->pmem_queue; | |
205 | disk->flags = GENHD_FL_EXT_DEVT; | |
5212e11f | 206 | nvdimm_namespace_disk_name(ndns, disk->disk_name); |
32ab0a3f DW |
207 | disk->driverfs_dev = dev; |
208 | set_capacity(disk, (pmem->size - pmem->data_offset) / 512); | |
9e853f23 RZ |
209 | pmem->pmem_disk = disk; |
210 | ||
211 | add_disk(disk); | |
58138820 | 212 | revalidate_disk(disk); |
9e853f23 | 213 | |
8c2f7e86 DW |
214 | return 0; |
215 | } | |
9e853f23 | 216 | |
8c2f7e86 DW |
217 | static int pmem_rw_bytes(struct nd_namespace_common *ndns, |
218 | resource_size_t offset, void *buf, size_t size, int rw) | |
219 | { | |
220 | struct pmem_device *pmem = dev_get_drvdata(ndns->claim); | |
221 | ||
222 | if (unlikely(offset + size > pmem->size)) { | |
223 | dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n"); | |
224 | return -EFAULT; | |
225 | } | |
226 | ||
227 | if (rw == READ) | |
61031952 RZ |
228 | memcpy_from_pmem(buf, pmem->virt_addr + offset, size); |
229 | else { | |
230 | memcpy_to_pmem(pmem->virt_addr + offset, buf, size); | |
231 | wmb_pmem(); | |
232 | } | |
8c2f7e86 DW |
233 | |
234 | return 0; | |
235 | } | |
236 | ||
32ab0a3f DW |
237 | static int nd_pfn_init(struct nd_pfn *nd_pfn) |
238 | { | |
239 | struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL); | |
240 | struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev); | |
241 | struct nd_namespace_common *ndns = nd_pfn->ndns; | |
242 | struct nd_region *nd_region; | |
243 | unsigned long npfns; | |
244 | phys_addr_t offset; | |
245 | u64 checksum; | |
246 | int rc; | |
247 | ||
248 | if (!pfn_sb) | |
249 | return -ENOMEM; | |
250 | ||
251 | nd_pfn->pfn_sb = pfn_sb; | |
252 | rc = nd_pfn_validate(nd_pfn); | |
253 | if (rc == 0 || rc == -EBUSY) | |
254 | return rc; | |
255 | ||
256 | /* section alignment for simple hotplug */ | |
257 | if (nvdimm_namespace_capacity(ndns) < ND_PFN_ALIGN | |
258 | || pmem->phys_addr & ND_PFN_MASK) | |
259 | return -ENODEV; | |
260 | ||
261 | nd_region = to_nd_region(nd_pfn->dev.parent); | |
262 | if (nd_region->ro) { | |
263 | dev_info(&nd_pfn->dev, | |
264 | "%s is read-only, unable to init metadata\n", | |
265 | dev_name(&nd_region->dev)); | |
266 | goto err; | |
267 | } | |
268 | ||
269 | memset(pfn_sb, 0, sizeof(*pfn_sb)); | |
270 | npfns = (pmem->size - SZ_8K) / SZ_4K; | |
271 | /* | |
272 | * Note, we use 64 here for the standard size of struct page, | |
273 | * debugging options may cause it to be larger in which case the | |
274 | * implementation will limit the pfns advertised through | |
275 | * ->direct_access() to those that are included in the memmap. | |
276 | */ | |
277 | if (nd_pfn->mode == PFN_MODE_PMEM) | |
278 | offset = ALIGN(SZ_8K + 64 * npfns, PMD_SIZE); | |
279 | else if (nd_pfn->mode == PFN_MODE_RAM) | |
280 | offset = SZ_8K; | |
281 | else | |
282 | goto err; | |
283 | ||
284 | npfns = (pmem->size - offset) / SZ_4K; | |
285 | pfn_sb->mode = cpu_to_le32(nd_pfn->mode); | |
286 | pfn_sb->dataoff = cpu_to_le64(offset); | |
287 | pfn_sb->npfns = cpu_to_le64(npfns); | |
288 | memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN); | |
289 | memcpy(pfn_sb->uuid, nd_pfn->uuid, 16); | |
290 | pfn_sb->version_major = cpu_to_le16(1); | |
291 | checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb); | |
292 | pfn_sb->checksum = cpu_to_le64(checksum); | |
293 | ||
294 | rc = nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb)); | |
295 | if (rc) | |
296 | goto err; | |
297 | ||
298 | return 0; | |
299 | err: | |
300 | nd_pfn->pfn_sb = NULL; | |
301 | kfree(pfn_sb); | |
302 | return -ENXIO; | |
303 | } | |
304 | ||
305 | static int nvdimm_namespace_detach_pfn(struct nd_namespace_common *ndns) | |
306 | { | |
307 | struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim); | |
308 | struct pmem_device *pmem; | |
309 | ||
310 | /* free pmem disk */ | |
311 | pmem = dev_get_drvdata(&nd_pfn->dev); | |
312 | pmem_detach_disk(pmem); | |
313 | ||
314 | /* release nd_pfn resources */ | |
315 | kfree(nd_pfn->pfn_sb); | |
316 | nd_pfn->pfn_sb = NULL; | |
317 | ||
318 | return 0; | |
319 | } | |
320 | ||
321 | static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns) | |
9e853f23 | 322 | { |
32ab0a3f DW |
323 | struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); |
324 | struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim); | |
325 | struct device *dev = &nd_pfn->dev; | |
326 | struct vmem_altmap *altmap; | |
327 | struct nd_region *nd_region; | |
328 | struct nd_pfn_sb *pfn_sb; | |
329 | struct pmem_device *pmem; | |
330 | phys_addr_t offset; | |
331 | int rc; | |
332 | ||
333 | if (!nd_pfn->uuid || !nd_pfn->ndns) | |
334 | return -ENODEV; | |
335 | ||
336 | nd_region = to_nd_region(dev->parent); | |
337 | rc = nd_pfn_init(nd_pfn); | |
338 | if (rc) | |
339 | return rc; | |
340 | ||
341 | if (PAGE_SIZE != SZ_4K) { | |
342 | dev_err(dev, "only supported on systems with 4K PAGE_SIZE\n"); | |
343 | return -ENXIO; | |
344 | } | |
345 | if (nsio->res.start & ND_PFN_MASK) { | |
346 | dev_err(dev, "%s not memory hotplug section aligned\n", | |
347 | dev_name(&ndns->dev)); | |
348 | return -ENXIO; | |
349 | } | |
350 | ||
351 | pfn_sb = nd_pfn->pfn_sb; | |
352 | offset = le64_to_cpu(pfn_sb->dataoff); | |
353 | nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode); | |
354 | if (nd_pfn->mode == PFN_MODE_RAM) { | |
355 | if (offset != SZ_8K) | |
356 | return -EINVAL; | |
357 | nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns); | |
358 | altmap = NULL; | |
359 | } else { | |
360 | rc = -ENXIO; | |
361 | goto err; | |
362 | } | |
363 | ||
364 | /* establish pfn range for lookup, and switch to direct map */ | |
365 | pmem = dev_get_drvdata(dev); | |
366 | memunmap_pmem(dev, pmem->virt_addr); | |
367 | pmem->virt_addr = (void __pmem *)devm_memremap_pages(dev, &nsio->res); | |
368 | if (IS_ERR(pmem->virt_addr)) { | |
369 | rc = PTR_ERR(pmem->virt_addr); | |
370 | goto err; | |
371 | } | |
372 | ||
373 | /* attach pmem disk in "pfn-mode" */ | |
374 | pmem->data_offset = offset; | |
375 | rc = pmem_attach_disk(dev, ndns, pmem); | |
376 | if (rc) | |
377 | goto err; | |
378 | ||
379 | return rc; | |
380 | err: | |
381 | nvdimm_namespace_detach_pfn(ndns); | |
382 | return rc; | |
9e853f23 RZ |
383 | } |
384 | ||
9f53f9fa | 385 | static int nd_pmem_probe(struct device *dev) |
9e853f23 | 386 | { |
9f53f9fa | 387 | struct nd_region *nd_region = to_nd_region(dev->parent); |
8c2f7e86 DW |
388 | struct nd_namespace_common *ndns; |
389 | struct nd_namespace_io *nsio; | |
9e853f23 | 390 | struct pmem_device *pmem; |
9e853f23 | 391 | |
8c2f7e86 DW |
392 | ndns = nvdimm_namespace_common_probe(dev); |
393 | if (IS_ERR(ndns)) | |
394 | return PTR_ERR(ndns); | |
bf9bccc1 | 395 | |
8c2f7e86 | 396 | nsio = to_nd_namespace_io(&ndns->dev); |
9f53f9fa | 397 | pmem = pmem_alloc(dev, &nsio->res, nd_region->id); |
9e853f23 RZ |
398 | if (IS_ERR(pmem)) |
399 | return PTR_ERR(pmem); | |
400 | ||
32ab0a3f | 401 | pmem->ndns = ndns; |
9f53f9fa | 402 | dev_set_drvdata(dev, pmem); |
8c2f7e86 | 403 | ndns->rw_bytes = pmem_rw_bytes; |
708ab62b | 404 | |
8c2f7e86 | 405 | if (is_nd_btt(dev)) |
708ab62b CH |
406 | return nvdimm_namespace_attach_btt(ndns); |
407 | ||
32ab0a3f DW |
408 | if (is_nd_pfn(dev)) |
409 | return nvdimm_namespace_attach_pfn(ndns); | |
410 | ||
411 | if (nd_btt_probe(ndns, pmem) == 0) { | |
8c2f7e86 | 412 | /* we'll come back as btt-pmem */ |
708ab62b | 413 | return -ENXIO; |
32ab0a3f DW |
414 | } |
415 | ||
416 | if (nd_pfn_probe(ndns, pmem) == 0) { | |
417 | /* we'll come back as pfn-pmem */ | |
418 | return -ENXIO; | |
419 | } | |
420 | ||
421 | return pmem_attach_disk(dev, ndns, pmem); | |
9e853f23 RZ |
422 | } |
423 | ||
9f53f9fa | 424 | static int nd_pmem_remove(struct device *dev) |
9e853f23 | 425 | { |
9f53f9fa | 426 | struct pmem_device *pmem = dev_get_drvdata(dev); |
9e853f23 | 427 | |
8c2f7e86 | 428 | if (is_nd_btt(dev)) |
32ab0a3f DW |
429 | nvdimm_namespace_detach_btt(pmem->ndns); |
430 | else if (is_nd_pfn(dev)) | |
431 | nvdimm_namespace_detach_pfn(pmem->ndns); | |
8c2f7e86 DW |
432 | else |
433 | pmem_detach_disk(pmem); | |
8c2f7e86 | 434 | |
9e853f23 RZ |
435 | return 0; |
436 | } | |
437 | ||
9f53f9fa DW |
438 | MODULE_ALIAS("pmem"); |
439 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); | |
bf9bccc1 | 440 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); |
9f53f9fa DW |
441 | static struct nd_device_driver nd_pmem_driver = { |
442 | .probe = nd_pmem_probe, | |
443 | .remove = nd_pmem_remove, | |
444 | .drv = { | |
445 | .name = "nd_pmem", | |
9e853f23 | 446 | }, |
bf9bccc1 | 447 | .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, |
9e853f23 RZ |
448 | }; |
449 | ||
450 | static int __init pmem_init(void) | |
451 | { | |
452 | int error; | |
453 | ||
454 | pmem_major = register_blkdev(0, "pmem"); | |
455 | if (pmem_major < 0) | |
456 | return pmem_major; | |
457 | ||
9f53f9fa DW |
458 | error = nd_driver_register(&nd_pmem_driver); |
459 | if (error) { | |
9e853f23 | 460 | unregister_blkdev(pmem_major, "pmem"); |
9f53f9fa DW |
461 | return error; |
462 | } | |
463 | ||
464 | return 0; | |
9e853f23 RZ |
465 | } |
466 | module_init(pmem_init); | |
467 | ||
468 | static void pmem_exit(void) | |
469 | { | |
9f53f9fa | 470 | driver_unregister(&nd_pmem_driver.drv); |
9e853f23 RZ |
471 | unregister_blkdev(pmem_major, "pmem"); |
472 | } | |
473 | module_exit(pmem_exit); | |
474 | ||
475 | MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); | |
476 | MODULE_LICENSE("GPL v2"); |