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/memory_hotplug.h>
25 #include <linux/moduleparam.h>
26 #include <linux/vmalloc.h>
27 #include <linux/slab.h>
28 #include <linux/pmem.h>
34 struct request_queue *pmem_queue;
35 struct gendisk *pmem_disk;
36 struct nd_namespace_common *ndns;
38 /* One contiguous memory region per device */
39 phys_addr_t phys_addr;
40 /* when non-zero this device is hosting a 'pfn' instance */
41 phys_addr_t data_offset;
42 void __pmem *virt_addr;
46 static int pmem_major;
48 static void pmem_do_bvec(struct pmem_device *pmem, struct page *page,
49 unsigned int len, unsigned int off, int rw,
52 void *mem = kmap_atomic(page);
53 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
54 void __pmem *pmem_addr = pmem->virt_addr + pmem_off;
57 memcpy_from_pmem(mem + off, pmem_addr, len);
58 flush_dcache_page(page);
60 flush_dcache_page(page);
61 memcpy_to_pmem(pmem_addr, mem + off, len);
67 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
72 struct bvec_iter iter;
73 struct block_device *bdev = bio->bi_bdev;
74 struct pmem_device *pmem = bdev->bd_disk->private_data;
76 do_acct = nd_iostat_start(bio, &start);
77 bio_for_each_segment(bvec, bio, iter)
78 pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, bvec.bv_offset,
79 bio_data_dir(bio), iter.bi_sector);
81 nd_iostat_end(bio, start);
83 if (bio_data_dir(bio))
90 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
91 struct page *page, int rw)
93 struct pmem_device *pmem = bdev->bd_disk->private_data;
95 pmem_do_bvec(pmem, page, PAGE_CACHE_SIZE, 0, rw, sector);
98 page_endio(page, rw & WRITE, 0);
103 static long pmem_direct_access(struct block_device *bdev, sector_t sector,
104 void __pmem **kaddr, unsigned long *pfn)
106 struct pmem_device *pmem = bdev->bd_disk->private_data;
107 resource_size_t offset = sector * 512 + pmem->data_offset;
108 resource_size_t size;
110 if (pmem->data_offset) {
112 * Limit the direct_access() size to what is covered by
115 size = (pmem->size - offset) & ~ND_PFN_MASK;
117 size = pmem->size - offset;
119 /* FIXME convert DAX to comprehend that this mapping has a lifetime */
120 *kaddr = pmem->virt_addr + offset;
121 *pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT;
126 static const struct block_device_operations pmem_fops = {
127 .owner = THIS_MODULE,
128 .rw_page = pmem_rw_page,
129 .direct_access = pmem_direct_access,
130 .revalidate_disk = nvdimm_revalidate_disk,
133 static struct pmem_device *pmem_alloc(struct device *dev,
134 struct resource *res, int id)
136 struct pmem_device *pmem;
138 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
140 return ERR_PTR(-ENOMEM);
142 pmem->phys_addr = res->start;
143 pmem->size = resource_size(res);
144 if (!arch_has_wmb_pmem())
145 dev_warn(dev, "unable to guarantee persistence of writes\n");
147 if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size,
149 dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n",
150 &pmem->phys_addr, pmem->size);
151 return ERR_PTR(-EBUSY);
154 if (pmem_should_map_pages(dev))
155 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res);
157 pmem->virt_addr = (void __pmem *) devm_memremap(dev,
158 pmem->phys_addr, pmem->size,
161 if (IS_ERR(pmem->virt_addr))
162 return (void __force *) pmem->virt_addr;
167 static void pmem_detach_disk(struct pmem_device *pmem)
169 if (!pmem->pmem_disk)
172 del_gendisk(pmem->pmem_disk);
173 put_disk(pmem->pmem_disk);
174 blk_cleanup_queue(pmem->pmem_queue);
177 static int pmem_attach_disk(struct device *dev,
178 struct nd_namespace_common *ndns, struct pmem_device *pmem)
180 int nid = dev_to_node(dev);
181 struct gendisk *disk;
183 pmem->pmem_queue = blk_alloc_queue_node(GFP_KERNEL, nid);
184 if (!pmem->pmem_queue)
187 blk_queue_make_request(pmem->pmem_queue, pmem_make_request);
188 blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE);
189 blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX);
190 blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY);
191 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue);
193 disk = alloc_disk_node(0, nid);
195 blk_cleanup_queue(pmem->pmem_queue);
199 disk->major = pmem_major;
200 disk->first_minor = 0;
201 disk->fops = &pmem_fops;
202 disk->private_data = pmem;
203 disk->queue = pmem->pmem_queue;
204 disk->flags = GENHD_FL_EXT_DEVT;
205 nvdimm_namespace_disk_name(ndns, disk->disk_name);
206 disk->driverfs_dev = dev;
207 set_capacity(disk, (pmem->size - pmem->data_offset) / 512);
208 pmem->pmem_disk = disk;
211 revalidate_disk(disk);
216 static int pmem_rw_bytes(struct nd_namespace_common *ndns,
217 resource_size_t offset, void *buf, size_t size, int rw)
219 struct pmem_device *pmem = dev_get_drvdata(ndns->claim);
221 if (unlikely(offset + size > pmem->size)) {
222 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
227 memcpy_from_pmem(buf, pmem->virt_addr + offset, size);
229 memcpy_to_pmem(pmem->virt_addr + offset, buf, size);
236 static int nd_pfn_init(struct nd_pfn *nd_pfn)
238 struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL);
239 struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev);
240 struct nd_namespace_common *ndns = nd_pfn->ndns;
241 struct nd_region *nd_region;
250 nd_pfn->pfn_sb = pfn_sb;
251 rc = nd_pfn_validate(nd_pfn);
252 if (rc == 0 || rc == -EBUSY)
255 /* section alignment for simple hotplug */
256 if (nvdimm_namespace_capacity(ndns) < ND_PFN_ALIGN
257 || pmem->phys_addr & ND_PFN_MASK)
260 nd_region = to_nd_region(nd_pfn->dev.parent);
262 dev_info(&nd_pfn->dev,
263 "%s is read-only, unable to init metadata\n",
264 dev_name(&nd_region->dev));
268 memset(pfn_sb, 0, sizeof(*pfn_sb));
269 npfns = (pmem->size - SZ_8K) / SZ_4K;
271 * Note, we use 64 here for the standard size of struct page,
272 * debugging options may cause it to be larger in which case the
273 * implementation will limit the pfns advertised through
274 * ->direct_access() to those that are included in the memmap.
276 if (nd_pfn->mode == PFN_MODE_PMEM)
277 offset = ALIGN(SZ_8K + 64 * npfns, PMD_SIZE);
278 else if (nd_pfn->mode == PFN_MODE_RAM)
283 npfns = (pmem->size - offset) / SZ_4K;
284 pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
285 pfn_sb->dataoff = cpu_to_le64(offset);
286 pfn_sb->npfns = cpu_to_le64(npfns);
287 memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN);
288 memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
289 pfn_sb->version_major = cpu_to_le16(1);
290 checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
291 pfn_sb->checksum = cpu_to_le64(checksum);
293 rc = nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb));
299 nd_pfn->pfn_sb = NULL;
304 static int nvdimm_namespace_detach_pfn(struct nd_namespace_common *ndns)
306 struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
307 struct pmem_device *pmem;
310 pmem = dev_get_drvdata(&nd_pfn->dev);
311 pmem_detach_disk(pmem);
313 /* release nd_pfn resources */
314 kfree(nd_pfn->pfn_sb);
315 nd_pfn->pfn_sb = NULL;
320 static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
322 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
323 struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
324 struct device *dev = &nd_pfn->dev;
325 struct vmem_altmap *altmap;
326 struct nd_region *nd_region;
327 struct nd_pfn_sb *pfn_sb;
328 struct pmem_device *pmem;
332 if (!nd_pfn->uuid || !nd_pfn->ndns)
335 nd_region = to_nd_region(dev->parent);
336 rc = nd_pfn_init(nd_pfn);
340 if (PAGE_SIZE != SZ_4K) {
341 dev_err(dev, "only supported on systems with 4K PAGE_SIZE\n");
344 if (nsio->res.start & ND_PFN_MASK) {
345 dev_err(dev, "%s not memory hotplug section aligned\n",
346 dev_name(&ndns->dev));
350 pfn_sb = nd_pfn->pfn_sb;
351 offset = le64_to_cpu(pfn_sb->dataoff);
352 nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode);
353 if (nd_pfn->mode == PFN_MODE_RAM) {
356 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
363 /* establish pfn range for lookup, and switch to direct map */
364 pmem = dev_get_drvdata(dev);
365 devm_memunmap(dev, (void __force *) pmem->virt_addr);
366 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &nsio->res);
367 if (IS_ERR(pmem->virt_addr)) {
368 rc = PTR_ERR(pmem->virt_addr);
372 /* attach pmem disk in "pfn-mode" */
373 pmem->data_offset = offset;
374 rc = pmem_attach_disk(dev, ndns, pmem);
380 nvdimm_namespace_detach_pfn(ndns);
384 static int nd_pmem_probe(struct device *dev)
386 struct nd_region *nd_region = to_nd_region(dev->parent);
387 struct nd_namespace_common *ndns;
388 struct nd_namespace_io *nsio;
389 struct pmem_device *pmem;
391 ndns = nvdimm_namespace_common_probe(dev);
393 return PTR_ERR(ndns);
395 nsio = to_nd_namespace_io(&ndns->dev);
396 pmem = pmem_alloc(dev, &nsio->res, nd_region->id);
398 return PTR_ERR(pmem);
401 dev_set_drvdata(dev, pmem);
402 ndns->rw_bytes = pmem_rw_bytes;
405 return nvdimm_namespace_attach_btt(ndns);
408 return nvdimm_namespace_attach_pfn(ndns);
410 if (nd_btt_probe(ndns, pmem) == 0) {
411 /* we'll come back as btt-pmem */
415 if (nd_pfn_probe(ndns, pmem) == 0) {
416 /* we'll come back as pfn-pmem */
420 return pmem_attach_disk(dev, ndns, pmem);
423 static int nd_pmem_remove(struct device *dev)
425 struct pmem_device *pmem = dev_get_drvdata(dev);
428 nvdimm_namespace_detach_btt(pmem->ndns);
429 else if (is_nd_pfn(dev))
430 nvdimm_namespace_detach_pfn(pmem->ndns);
432 pmem_detach_disk(pmem);
437 MODULE_ALIAS("pmem");
438 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
439 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
440 static struct nd_device_driver nd_pmem_driver = {
441 .probe = nd_pmem_probe,
442 .remove = nd_pmem_remove,
446 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
449 static int __init pmem_init(void)
453 pmem_major = register_blkdev(0, "pmem");
457 error = nd_driver_register(&nd_pmem_driver);
459 unregister_blkdev(pmem_major, "pmem");
465 module_init(pmem_init);
467 static void pmem_exit(void)
469 driver_unregister(&nd_pmem_driver.drv);
470 unregister_blkdev(pmem_major, "pmem");
472 module_exit(pmem_exit);
474 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
475 MODULE_LICENSE("GPL v2");