ld_nvme.c revision 1.17
1 /* $NetBSD: ld_nvme.c,v 1.17 2017/08/09 16:44:40 mlelstv Exp $ */ 2 3 /*- 4 * Copyright (C) 2016 NONAKA Kimihiro <nonaka (at) netbsd.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __KERNEL_RCSID(0, "$NetBSD: ld_nvme.c,v 1.17 2017/08/09 16:44:40 mlelstv Exp $"); 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/kernel.h> 34 #include <sys/device.h> 35 #include <sys/buf.h> 36 #include <sys/bufq.h> 37 #include <sys/disk.h> 38 #include <sys/kmem.h> 39 #include <sys/module.h> 40 41 #include <dev/ldvar.h> 42 #include <dev/ic/nvmereg.h> 43 #include <dev/ic/nvmevar.h> 44 45 #include "ioconf.h" 46 47 struct ld_nvme_softc { 48 struct ld_softc sc_ld; 49 struct nvme_softc *sc_nvme; 50 51 uint16_t sc_nsid; 52 53 /* getcache handling */ 54 kmutex_t sc_getcache_lock; 55 kcondvar_t sc_getcache_cv; 56 kcondvar_t sc_getcache_ready_cv; 57 bool sc_getcache_waiting; 58 bool sc_getcache_ready; 59 int sc_getcache_result; 60 }; 61 62 static int ld_nvme_match(device_t, cfdata_t, void *); 63 static void ld_nvme_attach(device_t, device_t, void *); 64 static int ld_nvme_detach(device_t, int); 65 66 CFATTACH_DECL_NEW(ld_nvme, sizeof(struct ld_nvme_softc), 67 ld_nvme_match, ld_nvme_attach, ld_nvme_detach, NULL); 68 69 static int ld_nvme_start(struct ld_softc *, struct buf *); 70 static int ld_nvme_dump(struct ld_softc *, void *, int, int); 71 static int ld_nvme_flush(struct ld_softc *, bool); 72 static int ld_nvme_getcache(struct ld_softc *, int *); 73 static int ld_nvme_ioctl(struct ld_softc *, u_long, void *, int32_t, bool); 74 75 static void ld_nvme_biodone(void *, struct buf *, uint16_t, uint32_t); 76 static void ld_nvme_syncdone(void *, struct buf *, uint16_t, uint32_t); 77 static void ld_nvme_getcache_done(void *, struct buf *, uint16_t, uint32_t); 78 79 static int 80 ld_nvme_match(device_t parent, cfdata_t match, void *aux) 81 { 82 struct nvme_attach_args *naa = aux; 83 84 if (naa->naa_nsid == 0) 85 return 0; 86 87 return 1; 88 } 89 90 static void 91 ld_nvme_attach(device_t parent, device_t self, void *aux) 92 { 93 struct ld_nvme_softc *sc = device_private(self); 94 struct ld_softc *ld = &sc->sc_ld; 95 struct nvme_softc *nsc = device_private(parent); 96 struct nvme_attach_args *naa = aux; 97 struct nvme_namespace *ns; 98 struct nvm_namespace_format *f; 99 uint64_t nsze; 100 int error; 101 102 ld->sc_dv = self; 103 sc->sc_nvme = nsc; 104 sc->sc_nsid = naa->naa_nsid; 105 106 mutex_init(&sc->sc_getcache_lock, MUTEX_DEFAULT, IPL_SOFTBIO); 107 cv_init(&sc->sc_getcache_cv, "nvmegcq"); 108 cv_init(&sc->sc_getcache_ready_cv, "nvmegcr"); 109 110 aprint_naive("\n"); 111 aprint_normal("\n"); 112 113 error = nvme_ns_identify(sc->sc_nvme, sc->sc_nsid); 114 if (error) { 115 aprint_error_dev(self, "couldn't identify namespace\n"); 116 return; 117 } 118 119 ns = nvme_ns_get(sc->sc_nvme, sc->sc_nsid); 120 KASSERT(ns); 121 nsze = lemtoh64(&ns->ident->nsze); 122 f = &ns->ident->lbaf[NVME_ID_NS_FLBAS(ns->ident->flbas)]; 123 124 ld->sc_secsize = 1 << f->lbads; 125 ld->sc_secperunit = nsze; 126 ld->sc_maxxfer = naa->naa_maxphys; 127 ld->sc_maxqueuecnt = naa->naa_qentries; 128 ld->sc_start = ld_nvme_start; 129 ld->sc_dump = ld_nvme_dump; 130 ld->sc_ioctl = ld_nvme_ioctl; 131 ld->sc_flags = LDF_ENABLED | LDF_NO_RND | LDF_MPSAFE; 132 ldattach(ld, "fcfs"); 133 } 134 135 static int 136 ld_nvme_detach(device_t self, int flags) 137 { 138 struct ld_nvme_softc *sc = device_private(self); 139 struct ld_softc *ld = &sc->sc_ld; 140 int rv; 141 142 if ((rv = ldbegindetach(ld, flags)) != 0) 143 return rv; 144 ldenddetach(ld); 145 146 nvme_ns_free(sc->sc_nvme, sc->sc_nsid); 147 148 return 0; 149 } 150 151 static int 152 ld_nvme_start(struct ld_softc *ld, struct buf *bp) 153 { 154 struct ld_nvme_softc *sc = device_private(ld->sc_dv); 155 int flags = BUF_ISWRITE(bp) ? 0 : NVME_NS_CTX_F_READ; 156 157 if (bp->b_flags & B_MEDIA_FUA) 158 flags |= NVME_NS_CTX_F_FUA; 159 160 return nvme_ns_dobio(sc->sc_nvme, sc->sc_nsid, sc, 161 bp, bp->b_data, bp->b_bcount, 162 sc->sc_ld.sc_secsize, bp->b_rawblkno, 163 flags, 164 ld_nvme_biodone); 165 } 166 167 static int 168 ld_nvme_dump(struct ld_softc *ld, void *data, int blkno, int blkcnt) 169 { 170 struct ld_nvme_softc *sc = device_private(ld->sc_dv); 171 172 return nvme_ns_dobio(sc->sc_nvme, sc->sc_nsid, sc, 173 NULL, data, blkcnt * ld->sc_secsize, 174 sc->sc_ld.sc_secsize, blkno, 175 NVME_NS_CTX_F_POLL, 176 ld_nvme_biodone); 177 } 178 179 static void 180 ld_nvme_biodone(void *xc, struct buf *bp, uint16_t cmd_status, uint32_t cdw0) 181 { 182 struct ld_nvme_softc *sc = xc; 183 uint16_t status = NVME_CQE_SC(cmd_status); 184 185 if (bp != NULL) { 186 if (status != NVME_CQE_SC_SUCCESS) { 187 bp->b_error = EIO; 188 bp->b_resid = bp->b_bcount; 189 aprint_error_dev(sc->sc_ld.sc_dv, "I/O error\n"); 190 } else { 191 bp->b_resid = 0; 192 } 193 lddone(&sc->sc_ld, bp); 194 } else { 195 if (status != NVME_CQE_SC_SUCCESS) { 196 aprint_error_dev(sc->sc_ld.sc_dv, "I/O error\n"); 197 } 198 } 199 } 200 201 static int 202 ld_nvme_flush(struct ld_softc *ld, bool poll) 203 { 204 struct ld_nvme_softc *sc = device_private(ld->sc_dv); 205 206 if (!nvme_has_volatile_write_cache(sc->sc_nvme)) { 207 /* cache not present, no value in trying to flush it */ 208 return 0; 209 } 210 211 return nvme_ns_sync(sc->sc_nvme, sc->sc_nsid, sc, 212 poll ? NVME_NS_CTX_F_POLL : 0, 213 ld_nvme_syncdone); 214 } 215 216 static void 217 ld_nvme_syncdone(void *xc, struct buf *bp, uint16_t cmd_status, uint32_t cdw0) 218 { 219 /* nothing to do */ 220 } 221 222 static int 223 ld_nvme_getcache(struct ld_softc *ld, int *addr) 224 { 225 int error; 226 struct ld_nvme_softc *sc = device_private(ld->sc_dv); 227 228 /* 229 * DPO not supported, Dataset Management (DSM) field doesn't specify 230 * the same semantics. 231 */ 232 *addr = DKCACHE_FUA; 233 234 if (!nvme_has_volatile_write_cache(sc->sc_nvme)) { 235 /* cache simply not present */ 236 return 0; 237 } 238 239 /* 240 * This is admin queue request. The queue is relatively limited in size, 241 * and this is not performance critical call, so have at most one pending 242 * cache request at a time to avoid spurious EWOULDBLOCK failures. 243 */ 244 mutex_enter(&sc->sc_getcache_lock); 245 while (sc->sc_getcache_waiting) { 246 error = cv_wait_sig(&sc->sc_getcache_cv, &sc->sc_getcache_lock); 247 if (error) 248 goto out; 249 } 250 sc->sc_getcache_waiting = true; 251 sc->sc_getcache_ready = false; 252 mutex_exit(&sc->sc_getcache_lock); 253 254 error = nvme_admin_getcache(sc->sc_nvme, sc, ld_nvme_getcache_done); 255 if (error) { 256 mutex_enter(&sc->sc_getcache_lock); 257 goto out; 258 } 259 260 mutex_enter(&sc->sc_getcache_lock); 261 while (!sc->sc_getcache_ready) { 262 error = cv_wait_sig(&sc->sc_getcache_ready_cv, 263 &sc->sc_getcache_lock); 264 if (error) 265 goto out; 266 } 267 268 KDASSERT(sc->sc_getcache_ready); 269 270 if (sc->sc_getcache_result >= 0) 271 *addr |= sc->sc_getcache_result; 272 else 273 error = EINVAL; 274 275 out: 276 sc->sc_getcache_waiting = false; 277 278 /* wake one of eventual waiters */ 279 cv_signal(&sc->sc_getcache_cv); 280 281 mutex_exit(&sc->sc_getcache_lock); 282 283 return error; 284 } 285 286 static void 287 ld_nvme_getcache_done(void *xc, struct buf *bp, uint16_t cmd_status, uint32_t cdw0) 288 { 289 struct ld_nvme_softc *sc = xc; 290 uint16_t status = NVME_CQE_SC(cmd_status); 291 int result; 292 293 if (status == NVME_CQE_SC_SUCCESS) { 294 result = 0; 295 296 if (cdw0 & NVME_CQE_CDW0_VWC_WCE) 297 result |= DKCACHE_WRITE; 298 299 /* 300 * If volatile write cache is present, the flag shall also be 301 * settable. 302 */ 303 result |= DKCACHE_WCHANGE; 304 } else { 305 result = -1; 306 } 307 308 mutex_enter(&sc->sc_getcache_lock); 309 sc->sc_getcache_result = result; 310 sc->sc_getcache_ready = true; 311 312 /* wake up the waiter */ 313 cv_signal(&sc->sc_getcache_ready_cv); 314 315 mutex_exit(&sc->sc_getcache_lock); 316 } 317 318 static int 319 ld_nvme_ioctl(struct ld_softc *ld, u_long cmd, void *addr, int32_t flag, bool poll) 320 { 321 int error; 322 323 switch (cmd) { 324 case DIOCCACHESYNC: 325 error = ld_nvme_flush(ld, poll); 326 break; 327 328 case DIOCGCACHE: 329 error = ld_nvme_getcache(ld, (int *)addr); 330 break; 331 332 default: 333 error = EPASSTHROUGH; 334 break; 335 } 336 337 return error; 338 } 339 340 MODULE(MODULE_CLASS_DRIVER, ld_nvme, "ld,nvme"); 341 342 #ifdef _MODULE 343 /* 344 * XXX Don't allow ioconf.c to redefine the "struct cfdriver ld_cd" 345 * XXX it will be defined in the common-code module 346 */ 347 #undef CFDRIVER_DECL 348 #define CFDRIVER_DECL(name, class, attr) 349 #include "ioconf.c" 350 #endif 351 352 static int 353 ld_nvme_modcmd(modcmd_t cmd, void *opaque) 354 { 355 #ifdef _MODULE 356 /* 357 * We ignore the cfdriver_vec[] that ioconf provides, since 358 * the cfdrivers are attached already. 359 */ 360 static struct cfdriver * const no_cfdriver_vec[] = { NULL }; 361 #endif 362 int error = 0; 363 364 #ifdef _MODULE 365 switch (cmd) { 366 case MODULE_CMD_INIT: 367 error = config_init_component(no_cfdriver_vec, 368 cfattach_ioconf_ld_nvme, cfdata_ioconf_ld_nvme); 369 break; 370 case MODULE_CMD_FINI: 371 error = config_fini_component(no_cfdriver_vec, 372 cfattach_ioconf_ld_nvme, cfdata_ioconf_ld_nvme); 373 break; 374 default: 375 error = ENOTTY; 376 break; 377 } 378 #endif 379 380 return error; 381 } 382
Indexes created Thu Oct 02 07:10:07 GMT 2025