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> |
32ab0a3f | 32 | #include "pfn.h" |
9f53f9fa | 33 | #include "nd.h" |
9e853f23 RZ |
34 | |
35 | struct pmem_device { | |
36 | struct request_queue *pmem_queue; | |
37 | struct gendisk *pmem_disk; | |
32ab0a3f | 38 | struct nd_namespace_common *ndns; |
9e853f23 RZ |
39 | |
40 | /* One contiguous memory region per device */ | |
41 | phys_addr_t phys_addr; | |
32ab0a3f DW |
42 | /* when non-zero this device is hosting a 'pfn' instance */ |
43 | phys_addr_t data_offset; | |
c4544205 | 44 | u64 pfn_flags; |
61031952 | 45 | void __pmem *virt_addr; |
cfe30b87 | 46 | /* immutable base size of the namespace */ |
9e853f23 | 47 | size_t size; |
cfe30b87 DW |
48 | /* trim size when namespace capacity has been section aligned */ |
49 | u32 pfn_pad; | |
b95f5f43 | 50 | struct badblocks bb; |
9e853f23 RZ |
51 | }; |
52 | ||
e10624f8 DW |
53 | static bool is_bad_pmem(struct badblocks *bb, sector_t sector, unsigned int len) |
54 | { | |
55 | if (bb->count) { | |
56 | sector_t first_bad; | |
57 | int num_bad; | |
58 | ||
59 | return !!badblocks_check(bb, sector, len / 512, &first_bad, | |
60 | &num_bad); | |
61 | } | |
62 | ||
63 | return false; | |
64 | } | |
65 | ||
59e64739 DW |
66 | static void pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset, |
67 | unsigned int len) | |
68 | { | |
69 | struct device *dev = disk_to_dev(pmem->pmem_disk); | |
70 | sector_t sector; | |
71 | long cleared; | |
72 | ||
73 | sector = (offset - pmem->data_offset) / 512; | |
74 | cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len); | |
75 | ||
76 | if (cleared > 0 && cleared / 512) { | |
77 | dev_dbg(dev, "%s: %llx clear %ld sector%s\n", | |
78 | __func__, (unsigned long long) sector, | |
79 | cleared / 512, cleared / 512 > 1 ? "s" : ""); | |
80 | badblocks_clear(&pmem->bb, sector, cleared / 512); | |
81 | } | |
82 | invalidate_pmem(pmem->virt_addr + offset, len); | |
83 | } | |
84 | ||
e10624f8 | 85 | static int pmem_do_bvec(struct pmem_device *pmem, struct page *page, |
9e853f23 RZ |
86 | unsigned int len, unsigned int off, int rw, |
87 | sector_t sector) | |
88 | { | |
b5ebc8ec | 89 | int rc = 0; |
59e64739 | 90 | bool bad_pmem = false; |
9e853f23 | 91 | void *mem = kmap_atomic(page); |
32ab0a3f | 92 | phys_addr_t pmem_off = sector * 512 + pmem->data_offset; |
61031952 | 93 | void __pmem *pmem_addr = pmem->virt_addr + pmem_off; |
9e853f23 | 94 | |
59e64739 DW |
95 | if (unlikely(is_bad_pmem(&pmem->bb, sector, len))) |
96 | bad_pmem = true; | |
97 | ||
9e853f23 | 98 | if (rw == READ) { |
59e64739 | 99 | if (unlikely(bad_pmem)) |
b5ebc8ec DW |
100 | rc = -EIO; |
101 | else { | |
fc0c2028 | 102 | rc = memcpy_from_pmem(mem + off, pmem_addr, len); |
b5ebc8ec DW |
103 | flush_dcache_page(page); |
104 | } | |
9e853f23 RZ |
105 | } else { |
106 | flush_dcache_page(page); | |
61031952 | 107 | memcpy_to_pmem(pmem_addr, mem + off, len); |
59e64739 DW |
108 | if (unlikely(bad_pmem)) { |
109 | pmem_clear_poison(pmem, pmem_off, len); | |
110 | memcpy_to_pmem(pmem_addr, mem + off, len); | |
111 | } | |
9e853f23 RZ |
112 | } |
113 | ||
114 | kunmap_atomic(mem); | |
b5ebc8ec | 115 | return rc; |
9e853f23 RZ |
116 | } |
117 | ||
dece1635 | 118 | static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio) |
9e853f23 | 119 | { |
e10624f8 | 120 | int rc = 0; |
f0dc089c DW |
121 | bool do_acct; |
122 | unsigned long start; | |
9e853f23 | 123 | struct bio_vec bvec; |
9e853f23 | 124 | struct bvec_iter iter; |
edc870e5 DW |
125 | struct block_device *bdev = bio->bi_bdev; |
126 | struct pmem_device *pmem = bdev->bd_disk->private_data; | |
9e853f23 | 127 | |
f0dc089c | 128 | do_acct = nd_iostat_start(bio, &start); |
e10624f8 DW |
129 | bio_for_each_segment(bvec, bio, iter) { |
130 | rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, | |
131 | bvec.bv_offset, bio_data_dir(bio), | |
132 | iter.bi_sector); | |
133 | if (rc) { | |
134 | bio->bi_error = rc; | |
135 | break; | |
136 | } | |
137 | } | |
f0dc089c DW |
138 | if (do_acct) |
139 | nd_iostat_end(bio, start); | |
61031952 RZ |
140 | |
141 | if (bio_data_dir(bio)) | |
142 | wmb_pmem(); | |
143 | ||
4246a0b6 | 144 | bio_endio(bio); |
dece1635 | 145 | return BLK_QC_T_NONE; |
9e853f23 RZ |
146 | } |
147 | ||
148 | static int pmem_rw_page(struct block_device *bdev, sector_t sector, | |
149 | struct page *page, int rw) | |
150 | { | |
151 | struct pmem_device *pmem = bdev->bd_disk->private_data; | |
e10624f8 | 152 | int rc; |
9e853f23 | 153 | |
09cbfeaf | 154 | rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, rw, sector); |
ba8fe0f8 RZ |
155 | if (rw & WRITE) |
156 | wmb_pmem(); | |
9e853f23 | 157 | |
e10624f8 DW |
158 | /* |
159 | * The ->rw_page interface is subtle and tricky. The core | |
160 | * retries on any error, so we can only invoke page_endio() in | |
161 | * the successful completion case. Otherwise, we'll see crashes | |
162 | * caused by double completion. | |
163 | */ | |
164 | if (rc == 0) | |
165 | page_endio(page, rw & WRITE, 0); | |
166 | ||
167 | return rc; | |
9e853f23 RZ |
168 | } |
169 | ||
170 | static long pmem_direct_access(struct block_device *bdev, sector_t sector, | |
34c0fd54 | 171 | void __pmem **kaddr, pfn_t *pfn) |
9e853f23 RZ |
172 | { |
173 | struct pmem_device *pmem = bdev->bd_disk->private_data; | |
32ab0a3f | 174 | resource_size_t offset = sector * 512 + pmem->data_offset; |
589e75d1 | 175 | |
e2e05394 | 176 | *kaddr = pmem->virt_addr + offset; |
34c0fd54 | 177 | *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags); |
9e853f23 | 178 | |
cfe30b87 | 179 | return pmem->size - pmem->pfn_pad - offset; |
9e853f23 RZ |
180 | } |
181 | ||
182 | static const struct block_device_operations pmem_fops = { | |
183 | .owner = THIS_MODULE, | |
184 | .rw_page = pmem_rw_page, | |
185 | .direct_access = pmem_direct_access, | |
58138820 | 186 | .revalidate_disk = nvdimm_revalidate_disk, |
9e853f23 RZ |
187 | }; |
188 | ||
9f53f9fa DW |
189 | static struct pmem_device *pmem_alloc(struct device *dev, |
190 | struct resource *res, int id) | |
9e853f23 RZ |
191 | { |
192 | struct pmem_device *pmem; | |
468ded03 | 193 | struct request_queue *q; |
9e853f23 | 194 | |
708ab62b | 195 | pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL); |
9e853f23 | 196 | if (!pmem) |
8c2f7e86 | 197 | return ERR_PTR(-ENOMEM); |
9e853f23 RZ |
198 | |
199 | pmem->phys_addr = res->start; | |
200 | pmem->size = resource_size(res); | |
96601adb | 201 | if (!arch_has_wmb_pmem()) |
61031952 | 202 | dev_warn(dev, "unable to guarantee persistence of writes\n"); |
9e853f23 | 203 | |
708ab62b CH |
204 | if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size, |
205 | dev_name(dev))) { | |
9f53f9fa DW |
206 | dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n", |
207 | &pmem->phys_addr, pmem->size); | |
8c2f7e86 | 208 | return ERR_PTR(-EBUSY); |
9e853f23 RZ |
209 | } |
210 | ||
468ded03 DW |
211 | q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev)); |
212 | if (!q) | |
213 | return ERR_PTR(-ENOMEM); | |
214 | ||
34c0fd54 DW |
215 | pmem->pfn_flags = PFN_DEV; |
216 | if (pmem_should_map_pages(dev)) { | |
4b94ffdc | 217 | pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res, |
5c2c2587 | 218 | &q->q_usage_counter, NULL); |
34c0fd54 DW |
219 | pmem->pfn_flags |= PFN_MAP; |
220 | } else | |
a639315d DW |
221 | pmem->virt_addr = (void __pmem *) devm_memremap(dev, |
222 | pmem->phys_addr, pmem->size, | |
223 | ARCH_MEMREMAP_PMEM); | |
b36f4761 | 224 | |
468ded03 DW |
225 | if (IS_ERR(pmem->virt_addr)) { |
226 | blk_cleanup_queue(q); | |
b36f4761 | 227 | return (void __force *) pmem->virt_addr; |
468ded03 | 228 | } |
8c2f7e86 | 229 | |
468ded03 | 230 | pmem->pmem_queue = q; |
8c2f7e86 DW |
231 | return pmem; |
232 | } | |
233 | ||
234 | static void pmem_detach_disk(struct pmem_device *pmem) | |
235 | { | |
32ab0a3f DW |
236 | if (!pmem->pmem_disk) |
237 | return; | |
238 | ||
8c2f7e86 DW |
239 | del_gendisk(pmem->pmem_disk); |
240 | put_disk(pmem->pmem_disk); | |
241 | blk_cleanup_queue(pmem->pmem_queue); | |
242 | } | |
243 | ||
32ab0a3f DW |
244 | static int pmem_attach_disk(struct device *dev, |
245 | struct nd_namespace_common *ndns, struct pmem_device *pmem) | |
8c2f7e86 | 246 | { |
a3901802 | 247 | struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); |
538ea4aa | 248 | int nid = dev_to_node(dev); |
a3901802 | 249 | struct resource bb_res; |
8c2f7e86 | 250 | struct gendisk *disk; |
9e853f23 | 251 | |
9e853f23 | 252 | blk_queue_make_request(pmem->pmem_queue, pmem_make_request); |
6b47496a | 253 | blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE); |
43d3fa3a | 254 | blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX); |
9e853f23 | 255 | blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY); |
0f51c4fa | 256 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue); |
9e853f23 | 257 | |
538ea4aa | 258 | disk = alloc_disk_node(0, nid); |
8c2f7e86 DW |
259 | if (!disk) { |
260 | blk_cleanup_queue(pmem->pmem_queue); | |
261 | return -ENOMEM; | |
262 | } | |
9e853f23 | 263 | |
9e853f23 RZ |
264 | disk->fops = &pmem_fops; |
265 | disk->private_data = pmem; | |
266 | disk->queue = pmem->pmem_queue; | |
267 | disk->flags = GENHD_FL_EXT_DEVT; | |
5212e11f | 268 | nvdimm_namespace_disk_name(ndns, disk->disk_name); |
32ab0a3f | 269 | disk->driverfs_dev = dev; |
cfe30b87 DW |
270 | set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset) |
271 | / 512); | |
9e853f23 | 272 | pmem->pmem_disk = disk; |
710d69cc | 273 | devm_exit_badblocks(dev, &pmem->bb); |
b95f5f43 DW |
274 | if (devm_init_badblocks(dev, &pmem->bb)) |
275 | return -ENOMEM; | |
a3901802 DW |
276 | bb_res.start = nsio->res.start + pmem->data_offset; |
277 | bb_res.end = nsio->res.end; | |
278 | if (is_nd_pfn(dev)) { | |
279 | struct nd_pfn *nd_pfn = to_nd_pfn(dev); | |
280 | struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb; | |
281 | ||
282 | bb_res.start += __le32_to_cpu(pfn_sb->start_pad); | |
283 | bb_res.end -= __le32_to_cpu(pfn_sb->end_trunc); | |
284 | } | |
285 | nvdimm_badblocks_populate(to_nd_region(dev->parent), &pmem->bb, | |
286 | &bb_res); | |
57f7f317 | 287 | disk->bb = &pmem->bb; |
9e853f23 | 288 | add_disk(disk); |
58138820 | 289 | revalidate_disk(disk); |
9e853f23 | 290 | |
8c2f7e86 DW |
291 | return 0; |
292 | } | |
9e853f23 | 293 | |
8c2f7e86 DW |
294 | static int pmem_rw_bytes(struct nd_namespace_common *ndns, |
295 | resource_size_t offset, void *buf, size_t size, int rw) | |
296 | { | |
297 | struct pmem_device *pmem = dev_get_drvdata(ndns->claim); | |
298 | ||
299 | if (unlikely(offset + size > pmem->size)) { | |
300 | dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n"); | |
301 | return -EFAULT; | |
302 | } | |
303 | ||
710d69cc DW |
304 | if (rw == READ) { |
305 | unsigned int sz_align = ALIGN(size + (offset & (512 - 1)), 512); | |
306 | ||
307 | if (unlikely(is_bad_pmem(&pmem->bb, offset / 512, sz_align))) | |
308 | return -EIO; | |
fc0c2028 | 309 | return memcpy_from_pmem(buf, pmem->virt_addr + offset, size); |
710d69cc | 310 | } else { |
61031952 RZ |
311 | memcpy_to_pmem(pmem->virt_addr + offset, buf, size); |
312 | wmb_pmem(); | |
313 | } | |
8c2f7e86 DW |
314 | |
315 | return 0; | |
316 | } | |
317 | ||
32ab0a3f DW |
318 | static int nd_pfn_init(struct nd_pfn *nd_pfn) |
319 | { | |
320 | struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL); | |
321 | struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev); | |
322 | struct nd_namespace_common *ndns = nd_pfn->ndns; | |
cfe30b87 DW |
323 | u32 start_pad = 0, end_trunc = 0; |
324 | resource_size_t start, size; | |
325 | struct nd_namespace_io *nsio; | |
32ab0a3f DW |
326 | struct nd_region *nd_region; |
327 | unsigned long npfns; | |
328 | phys_addr_t offset; | |
329 | u64 checksum; | |
330 | int rc; | |
331 | ||
332 | if (!pfn_sb) | |
333 | return -ENOMEM; | |
334 | ||
335 | nd_pfn->pfn_sb = pfn_sb; | |
336 | rc = nd_pfn_validate(nd_pfn); | |
3fa96268 DW |
337 | if (rc == -ENODEV) |
338 | /* no info block, do init */; | |
339 | else | |
32ab0a3f DW |
340 | return rc; |
341 | ||
32ab0a3f DW |
342 | nd_region = to_nd_region(nd_pfn->dev.parent); |
343 | if (nd_region->ro) { | |
344 | dev_info(&nd_pfn->dev, | |
345 | "%s is read-only, unable to init metadata\n", | |
346 | dev_name(&nd_region->dev)); | |
347 | goto err; | |
348 | } | |
349 | ||
350 | memset(pfn_sb, 0, sizeof(*pfn_sb)); | |
cfe30b87 DW |
351 | |
352 | /* | |
353 | * Check if pmem collides with 'System RAM' when section aligned and | |
354 | * trim it accordingly | |
355 | */ | |
356 | nsio = to_nd_namespace_io(&ndns->dev); | |
357 | start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start); | |
358 | size = resource_size(&nsio->res); | |
359 | if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM, | |
360 | IORES_DESC_NONE) == REGION_MIXED) { | |
361 | ||
362 | start = nsio->res.start; | |
363 | start_pad = PHYS_SECTION_ALIGN_UP(start) - start; | |
364 | } | |
365 | ||
366 | start = nsio->res.start; | |
367 | size = PHYS_SECTION_ALIGN_UP(start + size) - start; | |
368 | if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM, | |
369 | IORES_DESC_NONE) == REGION_MIXED) { | |
370 | size = resource_size(&nsio->res); | |
371 | end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size); | |
372 | } | |
373 | ||
374 | if (start_pad + end_trunc) | |
375 | dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n", | |
376 | dev_name(&ndns->dev), start_pad + end_trunc); | |
377 | ||
32ab0a3f DW |
378 | /* |
379 | * Note, we use 64 here for the standard size of struct page, | |
380 | * debugging options may cause it to be larger in which case the | |
381 | * implementation will limit the pfns advertised through | |
382 | * ->direct_access() to those that are included in the memmap. | |
383 | */ | |
cfe30b87 DW |
384 | start += start_pad; |
385 | npfns = (pmem->size - start_pad - end_trunc - SZ_8K) / SZ_4K; | |
32ab0a3f | 386 | if (nd_pfn->mode == PFN_MODE_PMEM) |
cfe30b87 DW |
387 | offset = ALIGN(start + SZ_8K + 64 * npfns, nd_pfn->align) |
388 | - start; | |
32ab0a3f | 389 | else if (nd_pfn->mode == PFN_MODE_RAM) |
cfe30b87 | 390 | offset = ALIGN(start + SZ_8K, nd_pfn->align) - start; |
32ab0a3f DW |
391 | else |
392 | goto err; | |
393 | ||
cfe30b87 DW |
394 | if (offset + start_pad + end_trunc >= pmem->size) { |
395 | dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n", | |
396 | dev_name(&ndns->dev)); | |
397 | goto err; | |
398 | } | |
399 | ||
400 | npfns = (pmem->size - offset - start_pad - end_trunc) / SZ_4K; | |
32ab0a3f DW |
401 | pfn_sb->mode = cpu_to_le32(nd_pfn->mode); |
402 | pfn_sb->dataoff = cpu_to_le64(offset); | |
403 | pfn_sb->npfns = cpu_to_le64(npfns); | |
404 | memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN); | |
405 | memcpy(pfn_sb->uuid, nd_pfn->uuid, 16); | |
a34d5e8a | 406 | memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16); |
32ab0a3f | 407 | pfn_sb->version_major = cpu_to_le16(1); |
cfe30b87 DW |
408 | pfn_sb->version_minor = cpu_to_le16(1); |
409 | pfn_sb->start_pad = cpu_to_le32(start_pad); | |
410 | pfn_sb->end_trunc = cpu_to_le32(end_trunc); | |
32ab0a3f DW |
411 | checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb); |
412 | pfn_sb->checksum = cpu_to_le64(checksum); | |
413 | ||
414 | rc = nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb)); | |
415 | if (rc) | |
416 | goto err; | |
417 | ||
418 | return 0; | |
419 | err: | |
420 | nd_pfn->pfn_sb = NULL; | |
421 | kfree(pfn_sb); | |
422 | return -ENXIO; | |
423 | } | |
424 | ||
425 | static int nvdimm_namespace_detach_pfn(struct nd_namespace_common *ndns) | |
426 | { | |
427 | struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim); | |
428 | struct pmem_device *pmem; | |
429 | ||
430 | /* free pmem disk */ | |
431 | pmem = dev_get_drvdata(&nd_pfn->dev); | |
432 | pmem_detach_disk(pmem); | |
433 | ||
434 | /* release nd_pfn resources */ | |
435 | kfree(nd_pfn->pfn_sb); | |
436 | nd_pfn->pfn_sb = NULL; | |
437 | ||
438 | return 0; | |
439 | } | |
440 | ||
d9cbe09d DW |
441 | /* |
442 | * We hotplug memory at section granularity, pad the reserved area from | |
443 | * the previous section base to the namespace base address. | |
444 | */ | |
445 | static unsigned long init_altmap_base(resource_size_t base) | |
446 | { | |
45f68802 | 447 | unsigned long base_pfn = PHYS_PFN(base); |
d9cbe09d DW |
448 | |
449 | return PFN_SECTION_ALIGN_DOWN(base_pfn); | |
450 | } | |
451 | ||
452 | static unsigned long init_altmap_reserve(resource_size_t base) | |
453 | { | |
45f68802 DW |
454 | unsigned long reserve = PHYS_PFN(SZ_8K); |
455 | unsigned long base_pfn = PHYS_PFN(base); | |
d9cbe09d DW |
456 | |
457 | reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn); | |
458 | return reserve; | |
459 | } | |
460 | ||
cfe30b87 | 461 | static int __nvdimm_namespace_attach_pfn(struct nd_pfn *nd_pfn) |
9e853f23 | 462 | { |
32ab0a3f | 463 | int rc; |
cfe30b87 DW |
464 | struct resource res; |
465 | struct request_queue *q; | |
466 | struct pmem_device *pmem; | |
467 | struct vmem_altmap *altmap; | |
468 | struct device *dev = &nd_pfn->dev; | |
469 | struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb; | |
470 | struct nd_namespace_common *ndns = nd_pfn->ndns; | |
471 | u32 start_pad = __le32_to_cpu(pfn_sb->start_pad); | |
472 | u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc); | |
473 | struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); | |
474 | resource_size_t base = nsio->res.start + start_pad; | |
d2c0f041 | 475 | struct vmem_altmap __altmap = { |
cfe30b87 DW |
476 | .base_pfn = init_altmap_base(base), |
477 | .reserve = init_altmap_reserve(base), | |
d2c0f041 | 478 | }; |
32ab0a3f | 479 | |
cfe30b87 DW |
480 | pmem = dev_get_drvdata(dev); |
481 | pmem->data_offset = le64_to_cpu(pfn_sb->dataoff); | |
482 | pmem->pfn_pad = start_pad + end_trunc; | |
32ab0a3f DW |
483 | nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode); |
484 | if (nd_pfn->mode == PFN_MODE_RAM) { | |
cfe30b87 | 485 | if (pmem->data_offset < SZ_8K) |
32ab0a3f DW |
486 | return -EINVAL; |
487 | nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns); | |
488 | altmap = NULL; | |
d2c0f041 | 489 | } else if (nd_pfn->mode == PFN_MODE_PMEM) { |
cfe30b87 | 490 | nd_pfn->npfns = (pmem->size - pmem->pfn_pad - pmem->data_offset) |
d2c0f041 DW |
491 | / PAGE_SIZE; |
492 | if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns) | |
493 | dev_info(&nd_pfn->dev, | |
494 | "number of pfns truncated from %lld to %ld\n", | |
495 | le64_to_cpu(nd_pfn->pfn_sb->npfns), | |
496 | nd_pfn->npfns); | |
497 | altmap = & __altmap; | |
45f68802 | 498 | altmap->free = PHYS_PFN(pmem->data_offset - SZ_8K); |
d2c0f041 | 499 | altmap->alloc = 0; |
32ab0a3f DW |
500 | } else { |
501 | rc = -ENXIO; | |
502 | goto err; | |
503 | } | |
504 | ||
505 | /* establish pfn range for lookup, and switch to direct map */ | |
5c2c2587 | 506 | q = pmem->pmem_queue; |
cfe30b87 DW |
507 | memcpy(&res, &nsio->res, sizeof(res)); |
508 | res.start += start_pad; | |
509 | res.end -= end_trunc; | |
a639315d | 510 | devm_memunmap(dev, (void __force *) pmem->virt_addr); |
cfe30b87 | 511 | pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &res, |
5c2c2587 | 512 | &q->q_usage_counter, altmap); |
34c0fd54 | 513 | pmem->pfn_flags |= PFN_MAP; |
32ab0a3f DW |
514 | if (IS_ERR(pmem->virt_addr)) { |
515 | rc = PTR_ERR(pmem->virt_addr); | |
516 | goto err; | |
517 | } | |
518 | ||
519 | /* attach pmem disk in "pfn-mode" */ | |
32ab0a3f DW |
520 | rc = pmem_attach_disk(dev, ndns, pmem); |
521 | if (rc) | |
522 | goto err; | |
523 | ||
524 | return rc; | |
525 | err: | |
526 | nvdimm_namespace_detach_pfn(ndns); | |
527 | return rc; | |
cfe30b87 DW |
528 | |
529 | } | |
530 | ||
531 | static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns) | |
532 | { | |
533 | struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim); | |
534 | int rc; | |
535 | ||
536 | if (!nd_pfn->uuid || !nd_pfn->ndns) | |
537 | return -ENODEV; | |
538 | ||
539 | rc = nd_pfn_init(nd_pfn); | |
540 | if (rc) | |
541 | return rc; | |
542 | /* we need a valid pfn_sb before we can init a vmem_altmap */ | |
543 | return __nvdimm_namespace_attach_pfn(nd_pfn); | |
9e853f23 RZ |
544 | } |
545 | ||
9f53f9fa | 546 | static int nd_pmem_probe(struct device *dev) |
9e853f23 | 547 | { |
9f53f9fa | 548 | struct nd_region *nd_region = to_nd_region(dev->parent); |
8c2f7e86 DW |
549 | struct nd_namespace_common *ndns; |
550 | struct nd_namespace_io *nsio; | |
9e853f23 | 551 | struct pmem_device *pmem; |
9e853f23 | 552 | |
8c2f7e86 DW |
553 | ndns = nvdimm_namespace_common_probe(dev); |
554 | if (IS_ERR(ndns)) | |
555 | return PTR_ERR(ndns); | |
bf9bccc1 | 556 | |
8c2f7e86 | 557 | nsio = to_nd_namespace_io(&ndns->dev); |
9f53f9fa | 558 | pmem = pmem_alloc(dev, &nsio->res, nd_region->id); |
9e853f23 RZ |
559 | if (IS_ERR(pmem)) |
560 | return PTR_ERR(pmem); | |
561 | ||
32ab0a3f | 562 | pmem->ndns = ndns; |
9f53f9fa | 563 | dev_set_drvdata(dev, pmem); |
8c2f7e86 | 564 | ndns->rw_bytes = pmem_rw_bytes; |
710d69cc DW |
565 | if (devm_init_badblocks(dev, &pmem->bb)) |
566 | return -ENOMEM; | |
a3901802 | 567 | nvdimm_badblocks_populate(nd_region, &pmem->bb, &nsio->res); |
708ab62b | 568 | |
468ded03 DW |
569 | if (is_nd_btt(dev)) { |
570 | /* btt allocates its own request_queue */ | |
571 | blk_cleanup_queue(pmem->pmem_queue); | |
572 | pmem->pmem_queue = NULL; | |
708ab62b | 573 | return nvdimm_namespace_attach_btt(ndns); |
468ded03 | 574 | } |
708ab62b | 575 | |
32ab0a3f DW |
576 | if (is_nd_pfn(dev)) |
577 | return nvdimm_namespace_attach_pfn(ndns); | |
578 | ||
468ded03 DW |
579 | if (nd_btt_probe(ndns, pmem) == 0 || nd_pfn_probe(ndns, pmem) == 0) { |
580 | /* | |
581 | * We'll come back as either btt-pmem, or pfn-pmem, so | |
582 | * drop the queue allocation for now. | |
583 | */ | |
584 | blk_cleanup_queue(pmem->pmem_queue); | |
32ab0a3f DW |
585 | return -ENXIO; |
586 | } | |
587 | ||
588 | return pmem_attach_disk(dev, ndns, pmem); | |
9e853f23 RZ |
589 | } |
590 | ||
9f53f9fa | 591 | static int nd_pmem_remove(struct device *dev) |
9e853f23 | 592 | { |
9f53f9fa | 593 | struct pmem_device *pmem = dev_get_drvdata(dev); |
9e853f23 | 594 | |
8c2f7e86 | 595 | if (is_nd_btt(dev)) |
32ab0a3f DW |
596 | nvdimm_namespace_detach_btt(pmem->ndns); |
597 | else if (is_nd_pfn(dev)) | |
598 | nvdimm_namespace_detach_pfn(pmem->ndns); | |
8c2f7e86 DW |
599 | else |
600 | pmem_detach_disk(pmem); | |
8c2f7e86 | 601 | |
9e853f23 RZ |
602 | return 0; |
603 | } | |
604 | ||
71999466 DW |
605 | static void nd_pmem_notify(struct device *dev, enum nvdimm_event event) |
606 | { | |
607 | struct pmem_device *pmem = dev_get_drvdata(dev); | |
608 | struct nd_namespace_common *ndns = pmem->ndns; | |
a3901802 DW |
609 | struct nd_region *nd_region = to_nd_region(dev->parent); |
610 | struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev); | |
611 | struct resource res = { | |
612 | .start = nsio->res.start + pmem->data_offset, | |
613 | .end = nsio->res.end, | |
614 | }; | |
71999466 DW |
615 | |
616 | if (event != NVDIMM_REVALIDATE_POISON) | |
617 | return; | |
618 | ||
a3901802 DW |
619 | if (is_nd_pfn(dev)) { |
620 | struct nd_pfn *nd_pfn = to_nd_pfn(dev); | |
621 | struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb; | |
622 | ||
623 | res.start += __le32_to_cpu(pfn_sb->start_pad); | |
624 | res.end -= __le32_to_cpu(pfn_sb->end_trunc); | |
625 | } | |
626 | ||
627 | nvdimm_badblocks_populate(nd_region, &pmem->bb, &res); | |
71999466 DW |
628 | } |
629 | ||
9f53f9fa DW |
630 | MODULE_ALIAS("pmem"); |
631 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); | |
bf9bccc1 | 632 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); |
9f53f9fa DW |
633 | static struct nd_device_driver nd_pmem_driver = { |
634 | .probe = nd_pmem_probe, | |
635 | .remove = nd_pmem_remove, | |
71999466 | 636 | .notify = nd_pmem_notify, |
9f53f9fa DW |
637 | .drv = { |
638 | .name = "nd_pmem", | |
9e853f23 | 639 | }, |
bf9bccc1 | 640 | .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, |
9e853f23 RZ |
641 | }; |
642 | ||
643 | static int __init pmem_init(void) | |
644 | { | |
55155291 | 645 | return nd_driver_register(&nd_pmem_driver); |
9e853f23 RZ |
646 | } |
647 | module_init(pmem_init); | |
648 | ||
649 | static void pmem_exit(void) | |
650 | { | |
9f53f9fa | 651 | driver_unregister(&nd_pmem_driver.drv); |
9e853f23 RZ |
652 | } |
653 | module_exit(pmem_exit); | |
654 | ||
655 | MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); | |
656 | MODULE_LICENSE("GPL v2"); |