1 /* 2 * Persistent Memory Driver 3 * 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>. 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> 25 #include <linux/slab.h> 26 #include <linux/pmem.h> 27 #include <linux/nd.h> 28 #include "nd.h" 29 30 struct pmem_device { 31 struct request_queue *pmem_queue; 32 struct gendisk *pmem_disk; 33 34 /* One contiguous memory region per device */ 35 phys_addr_t phys_addr; 36 void __pmem *virt_addr; 37 size_t size; 38 }; 39 40 static int pmem_major; 41 42 static void pmem_do_bvec(struct pmem_device *pmem, struct page *page, 43 unsigned int len, unsigned int off, int rw, 44 sector_t sector) 45 { 46 void *mem = kmap_atomic(page); 47 size_t pmem_off = sector << 9; 48 void __pmem *pmem_addr = pmem->virt_addr + pmem_off; 49 50 if (rw == READ) { 51 memcpy_from_pmem(mem + off, pmem_addr, len); 52 flush_dcache_page(page); 53 } else { 54 flush_dcache_page(page); 55 memcpy_to_pmem(pmem_addr, mem + off, len); 56 } 57 58 kunmap_atomic(mem); 59 } 60 61 static void pmem_make_request(struct request_queue *q, struct bio *bio) 62 { 63 bool do_acct; 64 unsigned long start; 65 struct bio_vec bvec; 66 struct bvec_iter iter; 67 struct block_device *bdev = bio->bi_bdev; 68 struct pmem_device *pmem = bdev->bd_disk->private_data; 69 70 do_acct = nd_iostat_start(bio, &start); 71 bio_for_each_segment(bvec, bio, iter) 72 pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len, bvec.bv_offset, 73 bio_data_dir(bio), iter.bi_sector); 74 if (do_acct) 75 nd_iostat_end(bio, start); 76 77 if (bio_data_dir(bio)) 78 wmb_pmem(); 79 80 bio_endio(bio, 0); 81 } 82 83 static int pmem_rw_page(struct block_device *bdev, sector_t sector, 84 struct page *page, int rw) 85 { 86 struct pmem_device *pmem = bdev->bd_disk->private_data; 87 88 pmem_do_bvec(pmem, page, PAGE_CACHE_SIZE, 0, rw, sector); 89 page_endio(page, rw & WRITE, 0); 90 91 return 0; 92 } 93 94 static long pmem_direct_access(struct block_device *bdev, sector_t sector, 95 void **kaddr, unsigned long *pfn, long size) 96 { 97 struct pmem_device *pmem = bdev->bd_disk->private_data; 98 size_t offset = sector << 9; 99 100 if (!pmem) 101 return -ENODEV; 102 103 /* FIXME convert DAX to comprehend that this mapping has a lifetime */ 104 *kaddr = (void __force *) pmem->virt_addr + offset; 105 *pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT; 106 107 return pmem->size - offset; 108 } 109 110 static const struct block_device_operations pmem_fops = { 111 .owner = THIS_MODULE, 112 .rw_page = pmem_rw_page, 113 .direct_access = pmem_direct_access, 114 .revalidate_disk = nvdimm_revalidate_disk, 115 }; 116 117 static struct pmem_device *pmem_alloc(struct device *dev, 118 struct resource *res, int id) 119 { 120 struct pmem_device *pmem; 121 122 pmem = kzalloc(sizeof(*pmem), GFP_KERNEL); 123 if (!pmem) 124 return ERR_PTR(-ENOMEM); 125 126 pmem->phys_addr = res->start; 127 pmem->size = resource_size(res); 128 if (!arch_has_pmem_api()) 129 dev_warn(dev, "unable to guarantee persistence of writes\n"); 130 131 if (!request_mem_region(pmem->phys_addr, pmem->size, dev_name(dev))) { 132 dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n", 133 &pmem->phys_addr, pmem->size); 134 kfree(pmem); 135 return ERR_PTR(-EBUSY); 136 } 137 138 pmem->virt_addr = memremap_pmem(pmem->phys_addr, pmem->size); 139 if (!pmem->virt_addr) { 140 release_mem_region(pmem->phys_addr, pmem->size); 141 kfree(pmem); 142 return ERR_PTR(-ENXIO); 143 } 144 145 return pmem; 146 } 147 148 static void pmem_detach_disk(struct pmem_device *pmem) 149 { 150 del_gendisk(pmem->pmem_disk); 151 put_disk(pmem->pmem_disk); 152 blk_cleanup_queue(pmem->pmem_queue); 153 } 154 155 static int pmem_attach_disk(struct nd_namespace_common *ndns, 156 struct pmem_device *pmem) 157 { 158 struct gendisk *disk; 159 160 pmem->pmem_queue = blk_alloc_queue(GFP_KERNEL); 161 if (!pmem->pmem_queue) 162 return -ENOMEM; 163 164 blk_queue_make_request(pmem->pmem_queue, pmem_make_request); 165 blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX); 166 blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY); 167 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue); 168 169 disk = alloc_disk(0); 170 if (!disk) { 171 blk_cleanup_queue(pmem->pmem_queue); 172 return -ENOMEM; 173 } 174 175 disk->major = pmem_major; 176 disk->first_minor = 0; 177 disk->fops = &pmem_fops; 178 disk->private_data = pmem; 179 disk->queue = pmem->pmem_queue; 180 disk->flags = GENHD_FL_EXT_DEVT; 181 nvdimm_namespace_disk_name(ndns, disk->disk_name); 182 disk->driverfs_dev = &ndns->dev; 183 set_capacity(disk, pmem->size >> 9); 184 pmem->pmem_disk = disk; 185 186 add_disk(disk); 187 revalidate_disk(disk); 188 189 return 0; 190 } 191 192 static int pmem_rw_bytes(struct nd_namespace_common *ndns, 193 resource_size_t offset, void *buf, size_t size, int rw) 194 { 195 struct pmem_device *pmem = dev_get_drvdata(ndns->claim); 196 197 if (unlikely(offset + size > pmem->size)) { 198 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n"); 199 return -EFAULT; 200 } 201 202 if (rw == READ) 203 memcpy_from_pmem(buf, pmem->virt_addr + offset, size); 204 else { 205 memcpy_to_pmem(pmem->virt_addr + offset, buf, size); 206 wmb_pmem(); 207 } 208 209 return 0; 210 } 211 212 static void pmem_free(struct pmem_device *pmem) 213 { 214 memunmap_pmem(pmem->virt_addr); 215 release_mem_region(pmem->phys_addr, pmem->size); 216 kfree(pmem); 217 } 218 219 static int nd_pmem_probe(struct device *dev) 220 { 221 struct nd_region *nd_region = to_nd_region(dev->parent); 222 struct nd_namespace_common *ndns; 223 struct nd_namespace_io *nsio; 224 struct pmem_device *pmem; 225 int rc; 226 227 ndns = nvdimm_namespace_common_probe(dev); 228 if (IS_ERR(ndns)) 229 return PTR_ERR(ndns); 230 231 nsio = to_nd_namespace_io(&ndns->dev); 232 pmem = pmem_alloc(dev, &nsio->res, nd_region->id); 233 if (IS_ERR(pmem)) 234 return PTR_ERR(pmem); 235 236 dev_set_drvdata(dev, pmem); 237 ndns->rw_bytes = pmem_rw_bytes; 238 if (is_nd_btt(dev)) 239 rc = nvdimm_namespace_attach_btt(ndns); 240 else if (nd_btt_probe(ndns, pmem) == 0) { 241 /* we'll come back as btt-pmem */ 242 rc = -ENXIO; 243 } else 244 rc = pmem_attach_disk(ndns, pmem); 245 if (rc) 246 pmem_free(pmem); 247 return rc; 248 } 249 250 static int nd_pmem_remove(struct device *dev) 251 { 252 struct pmem_device *pmem = dev_get_drvdata(dev); 253 254 if (is_nd_btt(dev)) 255 nvdimm_namespace_detach_btt(to_nd_btt(dev)->ndns); 256 else 257 pmem_detach_disk(pmem); 258 pmem_free(pmem); 259 260 return 0; 261 } 262 263 MODULE_ALIAS("pmem"); 264 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO); 265 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM); 266 static struct nd_device_driver nd_pmem_driver = { 267 .probe = nd_pmem_probe, 268 .remove = nd_pmem_remove, 269 .drv = { 270 .name = "nd_pmem", 271 }, 272 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM, 273 }; 274 275 static int __init pmem_init(void) 276 { 277 int error; 278 279 pmem_major = register_blkdev(0, "pmem"); 280 if (pmem_major < 0) 281 return pmem_major; 282 283 error = nd_driver_register(&nd_pmem_driver); 284 if (error) { 285 unregister_blkdev(pmem_major, "pmem"); 286 return error; 287 } 288 289 return 0; 290 } 291 module_init(pmem_init); 292 293 static void pmem_exit(void) 294 { 295 driver_unregister(&nd_pmem_driver.drv); 296 unregister_blkdev(pmem_major, "pmem"); 297 } 298 module_exit(pmem_exit); 299 300 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>"); 301 MODULE_LICENSE("GPL v2"); 302