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