nvme.c revision 253476
1/*- 2 * Copyright (C) 2012-2013 Intel Corporation 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/dev/nvme/nvme.c 253476 2013-07-19 21:40:57Z jimharris $"); 29 30#include <sys/param.h> 31#include <sys/bus.h> 32#include <sys/conf.h> 33#include <sys/module.h> 34 35#include <vm/uma.h> 36 37#include <dev/pci/pcireg.h> 38#include <dev/pci/pcivar.h> 39 40#include "nvme_private.h" 41 42struct nvme_consumer { 43 uint32_t id; 44 nvme_cons_ns_fn_t ns_fn; 45 nvme_cons_ctrlr_fn_t ctrlr_fn; 46 nvme_cons_async_fn_t async_fn; 47 nvme_cons_fail_fn_t fail_fn; 48}; 49 50struct nvme_consumer nvme_consumer[NVME_MAX_CONSUMERS]; 51#define INVALID_CONSUMER_ID 0xFFFF 52 53uma_zone_t nvme_request_zone; 54int32_t nvme_retry_count; 55 56MALLOC_DEFINE(M_NVME, "nvme", "nvme(4) memory allocations"); 57 58static int nvme_probe(device_t); 59static int nvme_attach(device_t); 60static int nvme_detach(device_t); 61static int nvme_modevent(module_t mod, int type, void *arg); 62 63static devclass_t nvme_devclass; 64 65static device_method_t nvme_pci_methods[] = { 66 /* Device interface */ 67 DEVMETHOD(device_probe, nvme_probe), 68 DEVMETHOD(device_attach, nvme_attach), 69 DEVMETHOD(device_detach, nvme_detach), 70 { 0, 0 } 71}; 72 73static driver_t nvme_pci_driver = { 74 "nvme", 75 nvme_pci_methods, 76 sizeof(struct nvme_controller), 77}; 78 79DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, nvme_modevent, 0); 80MODULE_VERSION(nvme, 1); 81 82static struct _pcsid 83{ 84 u_int32_t type; 85 const char *desc; 86} pci_ids[] = { 87 { 0x01118086, "NVMe Controller" }, 88 { CHATHAM_PCI_ID, "Chatham Prototype NVMe Controller" }, 89 { IDT32_PCI_ID, "IDT NVMe Controller (32 channel)" }, 90 { IDT8_PCI_ID, "IDT NVMe Controller (8 channel)" }, 91 { 0x00000000, NULL } 92}; 93 94static int 95nvme_probe (device_t device) 96{ 97 struct _pcsid *ep; 98 u_int32_t type; 99 100 type = pci_get_devid(device); 101 ep = pci_ids; 102 103 while (ep->type && ep->type != type) 104 ++ep; 105 106 if (ep->desc) { 107 device_set_desc(device, ep->desc); 108 return (BUS_PROBE_DEFAULT); 109 } 110 111#if defined(PCIS_STORAGE_NVM) 112 if (pci_get_class(device) == PCIC_STORAGE && 113 pci_get_subclass(device) == PCIS_STORAGE_NVM && 114 pci_get_progif(device) == PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) { 115 device_set_desc(device, "Generic NVMe Device"); 116 return (BUS_PROBE_GENERIC); 117 } 118#endif 119 120 return (ENXIO); 121} 122 123static void 124nvme_init(void) 125{ 126 uint32_t i; 127 128 nvme_request_zone = uma_zcreate("nvme_request", 129 sizeof(struct nvme_request), NULL, NULL, NULL, NULL, 0, 0); 130 131 for (i = 0; i < NVME_MAX_CONSUMERS; i++) 132 nvme_consumer[i].id = INVALID_CONSUMER_ID; 133} 134 135SYSINIT(nvme_register, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_init, NULL); 136 137static void 138nvme_uninit(void) 139{ 140 uma_zdestroy(nvme_request_zone); 141} 142 143SYSUNINIT(nvme_unregister, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_uninit, NULL); 144 145static void 146nvme_load(void) 147{ 148} 149 150static void 151nvme_unload(void) 152{ 153} 154 155static void 156nvme_shutdown(void) 157{ 158 device_t *devlist; 159 struct nvme_controller *ctrlr; 160 union cc_register cc; 161 union csts_register csts; 162 int dev, devcount; 163 164 if (devclass_get_devices(nvme_devclass, &devlist, &devcount)) 165 return; 166 167 for (dev = 0; dev < devcount; dev++) { 168 /* 169 * Only notify controller of shutdown when a real shutdown is 170 * in process, not when a module unload occurs. It seems at 171 * least some controllers (Chatham at least) don't let you 172 * re-enable the controller after shutdown notification has 173 * been received. 174 */ 175 ctrlr = DEVICE2SOFTC(devlist[dev]); 176 cc.raw = nvme_mmio_read_4(ctrlr, cc); 177 cc.bits.shn = NVME_SHN_NORMAL; 178 nvme_mmio_write_4(ctrlr, cc, cc.raw); 179 csts.raw = nvme_mmio_read_4(ctrlr, csts); 180 while (csts.bits.shst != NVME_SHST_COMPLETE) { 181 DELAY(5); 182 csts.raw = nvme_mmio_read_4(ctrlr, csts); 183 } 184 } 185 186 free(devlist, M_TEMP); 187} 188 189static int 190nvme_modevent(module_t mod, int type, void *arg) 191{ 192 193 switch (type) { 194 case MOD_LOAD: 195 nvme_load(); 196 break; 197 case MOD_UNLOAD: 198 nvme_unload(); 199 break; 200 case MOD_SHUTDOWN: 201 nvme_shutdown(); 202 break; 203 default: 204 break; 205 } 206 207 return (0); 208} 209 210void 211nvme_dump_command(struct nvme_command *cmd) 212{ 213 printf( 214"opc:%x f:%x r1:%x cid:%x nsid:%x r2:%x r3:%x mptr:%jx prp1:%jx prp2:%jx cdw:%x %x %x %x %x %x\n", 215 cmd->opc, cmd->fuse, cmd->rsvd1, cmd->cid, cmd->nsid, 216 cmd->rsvd2, cmd->rsvd3, 217 (uintmax_t)cmd->mptr, (uintmax_t)cmd->prp1, (uintmax_t)cmd->prp2, 218 cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14, 219 cmd->cdw15); 220} 221 222void 223nvme_dump_completion(struct nvme_completion *cpl) 224{ 225 printf("cdw0:%08x sqhd:%04x sqid:%04x " 226 "cid:%04x p:%x sc:%02x sct:%x m:%x dnr:%x\n", 227 cpl->cdw0, cpl->sqhd, cpl->sqid, 228 cpl->cid, cpl->status.p, cpl->status.sc, cpl->status.sct, 229 cpl->status.m, cpl->status.dnr); 230} 231 232static int 233nvme_attach(device_t dev) 234{ 235 struct nvme_controller *ctrlr = DEVICE2SOFTC(dev); 236 int status; 237 238 status = nvme_ctrlr_construct(ctrlr, dev); 239 240 if (status != 0) 241 return (status); 242 243 /* 244 * Reset controller twice to ensure we do a transition from cc.en==1 245 * to cc.en==0. This is because we don't really know what status 246 * the controller was left in when boot handed off to OS. 247 */ 248 status = nvme_ctrlr_hw_reset(ctrlr); 249 if (status != 0) 250 return (status); 251 252 status = nvme_ctrlr_hw_reset(ctrlr); 253 if (status != 0) 254 return (status); 255 256 nvme_sysctl_initialize_ctrlr(ctrlr); 257 258 pci_enable_busmaster(dev); 259 260 ctrlr->config_hook.ich_func = nvme_ctrlr_start_config_hook; 261 ctrlr->config_hook.ich_arg = ctrlr; 262 263 config_intrhook_establish(&ctrlr->config_hook); 264 265 return (0); 266} 267 268static int 269nvme_detach (device_t dev) 270{ 271 struct nvme_controller *ctrlr = DEVICE2SOFTC(dev); 272 273 nvme_ctrlr_destruct(ctrlr, dev); 274 pci_disable_busmaster(dev); 275 return (0); 276} 277 278static void 279nvme_notify_consumer(struct nvme_consumer *cons) 280{ 281 device_t *devlist; 282 struct nvme_controller *ctrlr; 283 struct nvme_namespace *ns; 284 void *ctrlr_cookie; 285 int dev_idx, ns_idx, devcount; 286 287 if (devclass_get_devices(nvme_devclass, &devlist, &devcount)) 288 return; 289 290 for (dev_idx = 0; dev_idx < devcount; dev_idx++) { 291 ctrlr = DEVICE2SOFTC(devlist[dev_idx]); 292 if (cons->ctrlr_fn != NULL) 293 ctrlr_cookie = (*cons->ctrlr_fn)(ctrlr); 294 else 295 ctrlr_cookie = NULL; 296 ctrlr->cons_cookie[cons->id] = ctrlr_cookie; 297 for (ns_idx = 0; ns_idx < ctrlr->cdata.nn; ns_idx++) { 298 ns = &ctrlr->ns[ns_idx]; 299 if (cons->ns_fn != NULL) 300 ns->cons_cookie[cons->id] = 301 (*cons->ns_fn)(ns, ctrlr_cookie); 302 } 303 } 304 305 free(devlist, M_TEMP); 306} 307 308void 309nvme_notify_async_consumers(struct nvme_controller *ctrlr, 310 const struct nvme_completion *async_cpl, 311 uint32_t log_page_id, void *log_page_buffer, 312 uint32_t log_page_size) 313{ 314 struct nvme_consumer *cons; 315 uint32_t i; 316 317 for (i = 0; i < NVME_MAX_CONSUMERS; i++) { 318 cons = &nvme_consumer[i]; 319 if (cons->id != INVALID_CONSUMER_ID && cons->async_fn != NULL) 320 (*cons->async_fn)(ctrlr->cons_cookie[i], async_cpl, 321 log_page_id, log_page_buffer, log_page_size); 322 } 323} 324 325void 326nvme_notify_fail_consumers(struct nvme_controller *ctrlr) 327{ 328 struct nvme_consumer *cons; 329 uint32_t i; 330 331 for (i = 0; i < NVME_MAX_CONSUMERS; i++) { 332 cons = &nvme_consumer[i]; 333 if (cons->id != INVALID_CONSUMER_ID && cons->fail_fn != NULL) 334 cons->fail_fn(ctrlr->cons_cookie[i]); 335 } 336} 337 338struct nvme_consumer * 339nvme_register_consumer(nvme_cons_ns_fn_t ns_fn, nvme_cons_ctrlr_fn_t ctrlr_fn, 340 nvme_cons_async_fn_t async_fn, 341 nvme_cons_fail_fn_t fail_fn) 342{ 343 int i; 344 345 /* 346 * TODO: add locking around consumer registration. Not an issue 347 * right now since we only have one nvme consumer - nvd(4). 348 */ 349 for (i = 0; i < NVME_MAX_CONSUMERS; i++) 350 if (nvme_consumer[i].id == INVALID_CONSUMER_ID) { 351 nvme_consumer[i].id = i; 352 nvme_consumer[i].ns_fn = ns_fn; 353 nvme_consumer[i].ctrlr_fn = ctrlr_fn; 354 nvme_consumer[i].async_fn = async_fn; 355 nvme_consumer[i].fail_fn = fail_fn; 356 357 nvme_notify_consumer(&nvme_consumer[i]); 358 return (&nvme_consumer[i]); 359 } 360 361 printf("nvme(4): consumer not registered - no slots available\n"); 362 return (NULL); 363} 364 365void 366nvme_unregister_consumer(struct nvme_consumer *consumer) 367{ 368 369 consumer->id = INVALID_CONSUMER_ID; 370} 371 372void 373nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl) 374{ 375 struct nvme_completion_poll_status *status = arg; 376 377 /* 378 * Copy status into the argument passed by the caller, so that 379 * the caller can check the status to determine if the 380 * the request passed or failed. 381 */ 382 memcpy(&status->cpl, cpl, sizeof(*cpl)); 383 wmb(); 384 status->done = TRUE; 385} 386