1 // SPDX-License-Identifier: GPL-2.0 2 3 #define pr_fmt(fmt) "papr-scm: " fmt 4 5 #include <linux/of.h> 6 #include <linux/kernel.h> 7 #include <linux/module.h> 8 #include <linux/ioport.h> 9 #include <linux/slab.h> 10 #include <linux/ndctl.h> 11 #include <linux/sched.h> 12 #include <linux/libnvdimm.h> 13 #include <linux/platform_device.h> 14 15 #include <asm/plpar_wrappers.h> 16 17 #define BIND_ANY_ADDR (~0ul) 18 19 #define PAPR_SCM_DIMM_CMD_MASK \ 20 ((1ul << ND_CMD_GET_CONFIG_SIZE) | \ 21 (1ul << ND_CMD_GET_CONFIG_DATA) | \ 22 (1ul << ND_CMD_SET_CONFIG_DATA)) 23 24 struct papr_scm_priv { 25 struct platform_device *pdev; 26 struct device_node *dn; 27 uint32_t drc_index; 28 uint64_t blocks; 29 uint64_t block_size; 30 int metadata_size; 31 bool is_volatile; 32 33 uint64_t bound_addr; 34 35 struct nvdimm_bus_descriptor bus_desc; 36 struct nvdimm_bus *bus; 37 struct nvdimm *nvdimm; 38 struct resource res; 39 struct nd_region *region; 40 struct nd_interleave_set nd_set; 41 }; 42 43 static int drc_pmem_bind(struct papr_scm_priv *p) 44 { 45 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 46 uint64_t rc, token; 47 uint64_t saved = 0; 48 49 /* 50 * When the hypervisor cannot map all the requested memory in a single 51 * hcall it returns H_BUSY and we call again with the token until 52 * we get H_SUCCESS. Aborting the retry loop before getting H_SUCCESS 53 * leave the system in an undefined state, so we wait. 54 */ 55 token = 0; 56 57 do { 58 rc = plpar_hcall(H_SCM_BIND_MEM, ret, p->drc_index, 0, 59 p->blocks, BIND_ANY_ADDR, token); 60 token = ret[0]; 61 if (!saved) 62 saved = ret[1]; 63 cond_resched(); 64 } while (rc == H_BUSY); 65 66 if (rc) { 67 dev_err(&p->pdev->dev, "bind err: %lld\n", rc); 68 return -ENXIO; 69 } 70 71 p->bound_addr = saved; 72 73 dev_dbg(&p->pdev->dev, "bound drc %x to %pR\n", p->drc_index, &p->res); 74 75 return 0; 76 } 77 78 static int drc_pmem_unbind(struct papr_scm_priv *p) 79 { 80 unsigned long ret[PLPAR_HCALL_BUFSIZE]; 81 uint64_t rc, token; 82 83 token = 0; 84 85 /* NB: unbind has the same retry requirements mentioned above */ 86 do { 87 rc = plpar_hcall(H_SCM_UNBIND_MEM, ret, p->drc_index, 88 p->bound_addr, p->blocks, token); 89 token = ret[0]; 90 cond_resched(); 91 } while (rc == H_BUSY); 92 93 if (rc) 94 dev_err(&p->pdev->dev, "unbind error: %lld\n", rc); 95 96 return !!rc; 97 } 98 99 static int papr_scm_meta_get(struct papr_scm_priv *p, 100 struct nd_cmd_get_config_data_hdr *hdr) 101 { 102 unsigned long data[PLPAR_HCALL_BUFSIZE]; 103 unsigned long offset, data_offset; 104 int len, read; 105 int64_t ret; 106 107 if ((hdr->in_offset + hdr->in_length) >= p->metadata_size) 108 return -EINVAL; 109 110 for (len = hdr->in_length; len; len -= read) { 111 112 data_offset = hdr->in_length - len; 113 offset = hdr->in_offset + data_offset; 114 115 if (len >= 8) 116 read = 8; 117 else if (len >= 4) 118 read = 4; 119 else if (len >= 2) 120 read = 2; 121 else 122 read = 1; 123 124 ret = plpar_hcall(H_SCM_READ_METADATA, data, p->drc_index, 125 offset, read); 126 127 if (ret == H_PARAMETER) /* bad DRC index */ 128 return -ENODEV; 129 if (ret) 130 return -EINVAL; /* other invalid parameter */ 131 132 switch (read) { 133 case 8: 134 *(uint64_t *)(hdr->out_buf + data_offset) = be64_to_cpu(data[0]); 135 break; 136 case 4: 137 *(uint32_t *)(hdr->out_buf + data_offset) = be32_to_cpu(data[0] & 0xffffffff); 138 break; 139 140 case 2: 141 *(uint16_t *)(hdr->out_buf + data_offset) = be16_to_cpu(data[0] & 0xffff); 142 break; 143 144 case 1: 145 *(uint8_t *)(hdr->out_buf + data_offset) = (data[0] & 0xff); 146 break; 147 } 148 } 149 return 0; 150 } 151 152 static int papr_scm_meta_set(struct papr_scm_priv *p, 153 struct nd_cmd_set_config_hdr *hdr) 154 { 155 unsigned long offset, data_offset; 156 int len, wrote; 157 unsigned long data; 158 __be64 data_be; 159 int64_t ret; 160 161 if ((hdr->in_offset + hdr->in_length) >= p->metadata_size) 162 return -EINVAL; 163 164 for (len = hdr->in_length; len; len -= wrote) { 165 166 data_offset = hdr->in_length - len; 167 offset = hdr->in_offset + data_offset; 168 169 if (len >= 8) { 170 data = *(uint64_t *)(hdr->in_buf + data_offset); 171 data_be = cpu_to_be64(data); 172 wrote = 8; 173 } else if (len >= 4) { 174 data = *(uint32_t *)(hdr->in_buf + data_offset); 175 data &= 0xffffffff; 176 data_be = cpu_to_be32(data); 177 wrote = 4; 178 } else if (len >= 2) { 179 data = *(uint16_t *)(hdr->in_buf + data_offset); 180 data &= 0xffff; 181 data_be = cpu_to_be16(data); 182 wrote = 2; 183 } else { 184 data_be = *(uint8_t *)(hdr->in_buf + data_offset); 185 data_be &= 0xff; 186 wrote = 1; 187 } 188 189 ret = plpar_hcall_norets(H_SCM_WRITE_METADATA, p->drc_index, 190 offset, data_be, wrote); 191 if (ret == H_PARAMETER) /* bad DRC index */ 192 return -ENODEV; 193 if (ret) 194 return -EINVAL; /* other invalid parameter */ 195 } 196 197 return 0; 198 } 199 200 int papr_scm_ndctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm, 201 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc) 202 { 203 struct nd_cmd_get_config_size *get_size_hdr; 204 struct papr_scm_priv *p; 205 206 /* Only dimm-specific calls are supported atm */ 207 if (!nvdimm) 208 return -EINVAL; 209 210 p = nvdimm_provider_data(nvdimm); 211 212 switch (cmd) { 213 case ND_CMD_GET_CONFIG_SIZE: 214 get_size_hdr = buf; 215 216 get_size_hdr->status = 0; 217 get_size_hdr->max_xfer = 8; 218 get_size_hdr->config_size = p->metadata_size; 219 *cmd_rc = 0; 220 break; 221 222 case ND_CMD_GET_CONFIG_DATA: 223 *cmd_rc = papr_scm_meta_get(p, buf); 224 break; 225 226 case ND_CMD_SET_CONFIG_DATA: 227 *cmd_rc = papr_scm_meta_set(p, buf); 228 break; 229 230 default: 231 return -EINVAL; 232 } 233 234 dev_dbg(&p->pdev->dev, "returned with cmd_rc = %d\n", *cmd_rc); 235 236 return 0; 237 } 238 239 static const struct attribute_group *region_attr_groups[] = { 240 &nd_region_attribute_group, 241 &nd_device_attribute_group, 242 &nd_mapping_attribute_group, 243 &nd_numa_attribute_group, 244 NULL, 245 }; 246 247 static const struct attribute_group *bus_attr_groups[] = { 248 &nvdimm_bus_attribute_group, 249 NULL, 250 }; 251 252 static const struct attribute_group *papr_scm_dimm_groups[] = { 253 &nvdimm_attribute_group, 254 &nd_device_attribute_group, 255 NULL, 256 }; 257 258 static int papr_scm_nvdimm_init(struct papr_scm_priv *p) 259 { 260 struct device *dev = &p->pdev->dev; 261 struct nd_mapping_desc mapping; 262 struct nd_region_desc ndr_desc; 263 unsigned long dimm_flags; 264 265 p->bus_desc.ndctl = papr_scm_ndctl; 266 p->bus_desc.module = THIS_MODULE; 267 p->bus_desc.of_node = p->pdev->dev.of_node; 268 p->bus_desc.attr_groups = bus_attr_groups; 269 p->bus_desc.provider_name = kstrdup(p->pdev->name, GFP_KERNEL); 270 271 if (!p->bus_desc.provider_name) 272 return -ENOMEM; 273 274 p->bus = nvdimm_bus_register(NULL, &p->bus_desc); 275 if (!p->bus) { 276 dev_err(dev, "Error creating nvdimm bus %pOF\n", p->dn); 277 return -ENXIO; 278 } 279 280 dimm_flags = 0; 281 set_bit(NDD_ALIASING, &dimm_flags); 282 283 p->nvdimm = nvdimm_create(p->bus, p, papr_scm_dimm_groups, 284 dimm_flags, PAPR_SCM_DIMM_CMD_MASK, 0, NULL); 285 if (!p->nvdimm) { 286 dev_err(dev, "Error creating DIMM object for %pOF\n", p->dn); 287 goto err; 288 } 289 290 if (nvdimm_bus_check_dimm_count(p->bus, 1)) 291 goto err; 292 293 /* now add the region */ 294 295 memset(&mapping, 0, sizeof(mapping)); 296 mapping.nvdimm = p->nvdimm; 297 mapping.start = 0; 298 mapping.size = p->blocks * p->block_size; // XXX: potential overflow? 299 300 memset(&ndr_desc, 0, sizeof(ndr_desc)); 301 ndr_desc.attr_groups = region_attr_groups; 302 ndr_desc.numa_node = dev_to_node(&p->pdev->dev); 303 ndr_desc.target_node = ndr_desc.numa_node; 304 ndr_desc.res = &p->res; 305 ndr_desc.of_node = p->dn; 306 ndr_desc.provider_data = p; 307 ndr_desc.mapping = &mapping; 308 ndr_desc.num_mappings = 1; 309 ndr_desc.nd_set = &p->nd_set; 310 set_bit(ND_REGION_PAGEMAP, &ndr_desc.flags); 311 312 if (p->is_volatile) 313 p->region = nvdimm_volatile_region_create(p->bus, &ndr_desc); 314 else 315 p->region = nvdimm_pmem_region_create(p->bus, &ndr_desc); 316 if (!p->region) { 317 dev_err(dev, "Error registering region %pR from %pOF\n", 318 ndr_desc.res, p->dn); 319 goto err; 320 } 321 322 return 0; 323 324 err: nvdimm_bus_unregister(p->bus); 325 kfree(p->bus_desc.provider_name); 326 return -ENXIO; 327 } 328 329 static int papr_scm_probe(struct platform_device *pdev) 330 { 331 struct device_node *dn = pdev->dev.of_node; 332 u32 drc_index, metadata_size; 333 u64 blocks, block_size; 334 struct papr_scm_priv *p; 335 const char *uuid_str; 336 u64 uuid[2]; 337 int rc; 338 339 /* check we have all the required DT properties */ 340 if (of_property_read_u32(dn, "ibm,my-drc-index", &drc_index)) { 341 dev_err(&pdev->dev, "%pOF: missing drc-index!\n", dn); 342 return -ENODEV; 343 } 344 345 if (of_property_read_u64(dn, "ibm,block-size", &block_size)) { 346 dev_err(&pdev->dev, "%pOF: missing block-size!\n", dn); 347 return -ENODEV; 348 } 349 350 if (of_property_read_u64(dn, "ibm,number-of-blocks", &blocks)) { 351 dev_err(&pdev->dev, "%pOF: missing number-of-blocks!\n", dn); 352 return -ENODEV; 353 } 354 355 if (of_property_read_string(dn, "ibm,unit-guid", &uuid_str)) { 356 dev_err(&pdev->dev, "%pOF: missing unit-guid!\n", dn); 357 return -ENODEV; 358 } 359 360 361 p = kzalloc(sizeof(*p), GFP_KERNEL); 362 if (!p) 363 return -ENOMEM; 364 365 /* optional DT properties */ 366 of_property_read_u32(dn, "ibm,metadata-size", &metadata_size); 367 368 p->dn = dn; 369 p->drc_index = drc_index; 370 p->block_size = block_size; 371 p->blocks = blocks; 372 p->is_volatile = !of_property_read_bool(dn, "ibm,cache-flush-required"); 373 374 /* We just need to ensure that set cookies are unique across */ 375 uuid_parse(uuid_str, (uuid_t *) uuid); 376 /* 377 * cookie1 and cookie2 are not really little endian 378 * we store a little endian representation of the 379 * uuid str so that we can compare this with the label 380 * area cookie irrespective of the endian config with which 381 * the kernel is built. 382 */ 383 p->nd_set.cookie1 = cpu_to_le64(uuid[0]); 384 p->nd_set.cookie2 = cpu_to_le64(uuid[1]); 385 386 /* might be zero */ 387 p->metadata_size = metadata_size; 388 p->pdev = pdev; 389 390 /* request the hypervisor to bind this region to somewhere in memory */ 391 rc = drc_pmem_bind(p); 392 if (rc) 393 goto err; 394 395 /* setup the resource for the newly bound range */ 396 p->res.start = p->bound_addr; 397 p->res.end = p->bound_addr + p->blocks * p->block_size - 1; 398 p->res.name = pdev->name; 399 p->res.flags = IORESOURCE_MEM; 400 401 rc = papr_scm_nvdimm_init(p); 402 if (rc) 403 goto err2; 404 405 platform_set_drvdata(pdev, p); 406 407 return 0; 408 409 err2: drc_pmem_unbind(p); 410 err: kfree(p); 411 return rc; 412 } 413 414 static int papr_scm_remove(struct platform_device *pdev) 415 { 416 struct papr_scm_priv *p = platform_get_drvdata(pdev); 417 418 nvdimm_bus_unregister(p->bus); 419 drc_pmem_unbind(p); 420 kfree(p); 421 422 return 0; 423 } 424 425 static const struct of_device_id papr_scm_match[] = { 426 { .compatible = "ibm,pmemory" }, 427 { }, 428 }; 429 430 static struct platform_driver papr_scm_driver = { 431 .probe = papr_scm_probe, 432 .remove = papr_scm_remove, 433 .driver = { 434 .name = "papr_scm", 435 .owner = THIS_MODULE, 436 .of_match_table = papr_scm_match, 437 }, 438 }; 439 440 module_platform_driver(papr_scm_driver); 441 MODULE_DEVICE_TABLE(of, papr_scm_match); 442 MODULE_LICENSE("GPL"); 443 MODULE_AUTHOR("IBM Corporation"); 444