1 /* $NetBSD: nvme_pci.c,v 1.20 2018/04/18 10:05:59 nonaka Exp $ */ 2 /* $OpenBSD: nvme_pci.c,v 1.3 2016/04/14 11:18:32 dlg Exp $ */ 3 4 /* 5 * Copyright (c) 2014 David Gwynne <dlg@openbsd.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20 /*- 21 * Copyright (C) 2016 NONAKA Kimihiro <nonaka@netbsd.org> 22 * All rights reserved. 23 * 24 * Redistribution and use in source and binary forms, with or without 25 * modification, are permitted provided that the following conditions 26 * are met: 27 * 1. Redistributions of source code must retain the above copyright 28 * notice, this list of conditions and the following disclaimer. 29 * 2. Redistributions in binary form must reproduce the above copyright 30 * notice, this list of conditions and the following disclaimer in the 31 * documentation and/or other materials provided with the distribution. 32 * 33 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 34 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 35 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 36 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 37 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 38 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 39 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 40 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 41 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 42 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 43 */ 44 45 #include <sys/cdefs.h> 46 __KERNEL_RCSID(0, "$NetBSD: nvme_pci.c,v 1.20 2018/04/18 10:05:59 nonaka Exp $"); 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/kernel.h> 51 #include <sys/device.h> 52 #include <sys/bitops.h> 53 #include <sys/bus.h> 54 #include <sys/cpu.h> 55 #include <sys/interrupt.h> 56 #include <sys/kmem.h> 57 #include <sys/pmf.h> 58 #include <sys/module.h> 59 60 #include <dev/pci/pcireg.h> 61 #include <dev/pci/pcivar.h> 62 #include <dev/pci/pcidevs.h> 63 64 #include <dev/ic/nvmereg.h> 65 #include <dev/ic/nvmevar.h> 66 67 int nvme_pci_force_intx = 0; 68 int nvme_pci_mpsafe = 1; 69 int nvme_pci_mq = 1; /* INTx: ioq=1, MSI/MSI-X: ioq=ncpu */ 70 71 #define NVME_PCI_BAR 0x10 72 73 struct nvme_pci_softc { 74 struct nvme_softc psc_nvme; 75 76 pci_chipset_tag_t psc_pc; 77 pci_intr_handle_t *psc_intrs; 78 int psc_nintrs; 79 }; 80 81 static int nvme_pci_match(device_t, cfdata_t, void *); 82 static void nvme_pci_attach(device_t, device_t, void *); 83 static int nvme_pci_detach(device_t, int); 84 static int nvme_pci_rescan(device_t, const char *, const int *); 85 86 CFATTACH_DECL3_NEW(nvme_pci, sizeof(struct nvme_pci_softc), 87 nvme_pci_match, nvme_pci_attach, nvme_pci_detach, NULL, nvme_pci_rescan, 88 nvme_childdet, DVF_DETACH_SHUTDOWN); 89 90 static int nvme_pci_intr_establish(struct nvme_softc *, 91 uint16_t, struct nvme_queue *); 92 static int nvme_pci_intr_disestablish(struct nvme_softc *, uint16_t); 93 static int nvme_pci_setup_intr(struct pci_attach_args *, 94 struct nvme_pci_softc *); 95 96 static const struct nvme_pci_quirk { 97 pci_vendor_id_t vendor; 98 pci_product_id_t product; 99 uint32_t quirks; 100 } nvme_pci_quirks[] = { 101 { PCI_VENDOR_HGST, PCI_PRODUCT_HGST_SN100, 102 NVME_QUIRK_DELAY_B4_CHK_RDY }, 103 { PCI_VENDOR_HGST, PCI_PRODUCT_HGST_SN200, 104 NVME_QUIRK_DELAY_B4_CHK_RDY }, 105 { PCI_VENDOR_BEIJING_MEMBLAZE, PCI_PRODUCT_BEIJING_MEMBLAZE_PBLAZE4, 106 NVME_QUIRK_DELAY_B4_CHK_RDY }, 107 { PCI_VENDOR_SAMSUNGELEC3, PCI_PRODUCT_SAMSUNGELEC3_172X, 108 NVME_QUIRK_DELAY_B4_CHK_RDY }, 109 { PCI_VENDOR_SAMSUNGELEC3, PCI_PRODUCT_SAMSUNGELEC3_172XAB, 110 NVME_QUIRK_DELAY_B4_CHK_RDY }, 111 }; 112 113 static const struct nvme_pci_quirk * 114 nvme_pci_lookup_quirk(struct pci_attach_args *pa) 115 { 116 const struct nvme_pci_quirk *q; 117 int i; 118 119 for (i = 0; i < __arraycount(nvme_pci_quirks); i++) { 120 q = &nvme_pci_quirks[i]; 121 122 if (PCI_VENDOR(pa->pa_id) == q->vendor && 123 PCI_PRODUCT(pa->pa_id) == q->product) 124 return q; 125 } 126 return NULL; 127 } 128 129 static int 130 nvme_pci_match(device_t parent, cfdata_t match, void *aux) 131 { 132 struct pci_attach_args *pa = aux; 133 134 if (PCI_CLASS(pa->pa_class) == PCI_CLASS_MASS_STORAGE && 135 PCI_SUBCLASS(pa->pa_class) == PCI_SUBCLASS_MASS_STORAGE_NVM && 136 PCI_INTERFACE(pa->pa_class) == PCI_INTERFACE_NVM_NVME) 137 return 1; 138 139 return 0; 140 } 141 142 static void 143 nvme_pci_attach(device_t parent, device_t self, void *aux) 144 { 145 struct nvme_pci_softc *psc = device_private(self); 146 struct nvme_softc *sc = &psc->psc_nvme; 147 struct pci_attach_args *pa = aux; 148 const struct nvme_pci_quirk *quirk; 149 pcireg_t memtype, reg; 150 bus_addr_t memaddr; 151 int flags, error; 152 int msixoff; 153 154 sc->sc_dev = self; 155 psc->psc_pc = pa->pa_pc; 156 if (pci_dma64_available(pa)) 157 sc->sc_dmat = pa->pa_dmat64; 158 else 159 sc->sc_dmat = pa->pa_dmat; 160 161 pci_aprint_devinfo(pa, NULL); 162 163 reg = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG); 164 if ((reg & PCI_COMMAND_MASTER_ENABLE) == 0) { 165 reg |= PCI_COMMAND_MASTER_ENABLE; 166 pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, reg); 167 } 168 169 /* Map registers */ 170 memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, NVME_PCI_BAR); 171 if (PCI_MAPREG_TYPE(memtype) != PCI_MAPREG_TYPE_MEM) { 172 aprint_error_dev(self, "invalid type (type=0x%x)\n", memtype); 173 return; 174 } 175 sc->sc_iot = pa->pa_memt; 176 error = pci_mapreg_info(pa->pa_pc, pa->pa_tag, PCI_MAPREG_START, 177 memtype, &memaddr, &sc->sc_ios, &flags); 178 if (error) { 179 aprint_error_dev(self, "can't get map info\n"); 180 return; 181 } 182 183 if (pci_get_capability(pa->pa_pc, pa->pa_tag, PCI_CAP_MSIX, &msixoff, 184 NULL)) { 185 pcireg_t msixtbl; 186 uint32_t table_offset; 187 int bir; 188 189 msixtbl = pci_conf_read(pa->pa_pc, pa->pa_tag, 190 msixoff + PCI_MSIX_TBLOFFSET); 191 table_offset = msixtbl & PCI_MSIX_TBLOFFSET_MASK; 192 bir = msixtbl & PCI_MSIX_PBABIR_MASK; 193 if (bir == 0) { 194 sc->sc_ios = table_offset; 195 } 196 } 197 198 error = bus_space_map(sc->sc_iot, memaddr, sc->sc_ios, flags, 199 &sc->sc_ioh); 200 if (error != 0) { 201 aprint_error_dev(self, "can't map mem space (error=%d)\n", 202 error); 203 return; 204 } 205 206 /* Establish interrupts */ 207 if (nvme_pci_setup_intr(pa, psc) != 0) { 208 aprint_error_dev(self, "unable to allocate interrupt\n"); 209 goto unmap; 210 } 211 sc->sc_intr_establish = nvme_pci_intr_establish; 212 sc->sc_intr_disestablish = nvme_pci_intr_disestablish; 213 214 sc->sc_ih = kmem_zalloc(sizeof(*sc->sc_ih) * psc->psc_nintrs, KM_SLEEP); 215 sc->sc_softih = kmem_zalloc( 216 sizeof(*sc->sc_softih) * psc->psc_nintrs, KM_SLEEP); 217 218 quirk = nvme_pci_lookup_quirk(pa); 219 if (quirk != NULL) 220 sc->sc_quirks = quirk->quirks; 221 222 if (nvme_attach(sc) != 0) { 223 /* error printed by nvme_attach() */ 224 goto softintr_free; 225 } 226 227 if (!pmf_device_register(self, NULL, NULL)) 228 aprint_error_dev(self, "couldn't establish power handler\n"); 229 230 SET(sc->sc_flags, NVME_F_ATTACHED); 231 return; 232 233 softintr_free: 234 kmem_free(sc->sc_softih, sizeof(*sc->sc_softih) * psc->psc_nintrs); 235 kmem_free(sc->sc_ih, sizeof(*sc->sc_ih) * psc->psc_nintrs); 236 sc->sc_nq = 0; 237 pci_intr_release(pa->pa_pc, psc->psc_intrs, psc->psc_nintrs); 238 psc->psc_nintrs = 0; 239 unmap: 240 bus_space_unmap(sc->sc_iot, sc->sc_ioh, sc->sc_ios); 241 sc->sc_ios = 0; 242 } 243 244 static int 245 nvme_pci_rescan(device_t self, const char *attr, const int *flags) 246 { 247 248 return nvme_rescan(self, attr, flags); 249 } 250 251 static int 252 nvme_pci_detach(device_t self, int flags) 253 { 254 struct nvme_pci_softc *psc = device_private(self); 255 struct nvme_softc *sc = &psc->psc_nvme; 256 int error; 257 258 if (!ISSET(sc->sc_flags, NVME_F_ATTACHED)) 259 return 0; 260 261 error = nvme_detach(sc, flags); 262 if (error) 263 return error; 264 265 kmem_free(sc->sc_softih, sizeof(*sc->sc_softih) * psc->psc_nintrs); 266 sc->sc_softih = NULL; 267 268 kmem_free(sc->sc_ih, sizeof(*sc->sc_ih) * psc->psc_nintrs); 269 pci_intr_release(psc->psc_pc, psc->psc_intrs, psc->psc_nintrs); 270 bus_space_unmap(sc->sc_iot, sc->sc_ioh, sc->sc_ios); 271 return 0; 272 } 273 274 static int 275 nvme_pci_intr_establish(struct nvme_softc *sc, uint16_t qid, 276 struct nvme_queue *q) 277 { 278 struct nvme_pci_softc *psc = (struct nvme_pci_softc *)sc; 279 char intr_xname[INTRDEVNAMEBUF]; 280 char intrbuf[PCI_INTRSTR_LEN]; 281 const char *intrstr = NULL; 282 int (*ih_func)(void *); 283 void (*ih_func_soft)(void *); 284 void *ih_arg; 285 int error; 286 287 KASSERT(sc->sc_use_mq || qid == NVME_ADMIN_Q); 288 KASSERT(sc->sc_ih[qid] == NULL); 289 290 if (nvme_pci_mpsafe) { 291 pci_intr_setattr(psc->psc_pc, &psc->psc_intrs[qid], 292 PCI_INTR_MPSAFE, true); 293 } 294 295 if (!sc->sc_use_mq) { 296 snprintf(intr_xname, sizeof(intr_xname), "%s", 297 device_xname(sc->sc_dev)); 298 ih_arg = sc; 299 ih_func = nvme_intr; 300 ih_func_soft = nvme_softintr_intx; 301 } else { 302 if (qid == NVME_ADMIN_Q) { 303 snprintf(intr_xname, sizeof(intr_xname), "%s adminq", 304 device_xname(sc->sc_dev)); 305 } else { 306 snprintf(intr_xname, sizeof(intr_xname), "%s ioq%d", 307 device_xname(sc->sc_dev), qid); 308 } 309 ih_arg = q; 310 ih_func = nvme_intr_msi; 311 ih_func_soft = nvme_softintr_msi; 312 } 313 314 /* establish hardware interrupt */ 315 sc->sc_ih[qid] = pci_intr_establish_xname(psc->psc_pc, 316 psc->psc_intrs[qid], IPL_BIO, ih_func, ih_arg, intr_xname); 317 if (sc->sc_ih[qid] == NULL) { 318 aprint_error_dev(sc->sc_dev, 319 "unable to establish %s interrupt\n", intr_xname); 320 return 1; 321 } 322 323 /* establish also the software interrupt */ 324 sc->sc_softih[qid] = softint_establish( 325 SOFTINT_BIO|(nvme_pci_mpsafe ? SOFTINT_MPSAFE : 0), 326 ih_func_soft, q); 327 if (sc->sc_softih[qid] == NULL) { 328 pci_intr_disestablish(psc->psc_pc, sc->sc_ih[qid]); 329 sc->sc_ih[qid] = NULL; 330 331 aprint_error_dev(sc->sc_dev, 332 "unable to establish %s soft interrupt\n", 333 intr_xname); 334 return 1; 335 } 336 337 intrstr = pci_intr_string(psc->psc_pc, psc->psc_intrs[qid], intrbuf, 338 sizeof(intrbuf)); 339 if (!sc->sc_use_mq) { 340 aprint_normal_dev(sc->sc_dev, "interrupting at %s\n", intrstr); 341 } else if (qid == NVME_ADMIN_Q) { 342 aprint_normal_dev(sc->sc_dev, 343 "for admin queue interrupting at %s\n", intrstr); 344 } else if (!nvme_pci_mpsafe) { 345 aprint_normal_dev(sc->sc_dev, 346 "for io queue %d interrupting at %s\n", qid, intrstr); 347 } else { 348 kcpuset_t *affinity; 349 cpuid_t affinity_to; 350 351 kcpuset_create(&affinity, true); 352 affinity_to = (qid - 1) % ncpu; 353 kcpuset_set(affinity, affinity_to); 354 error = interrupt_distribute(sc->sc_ih[qid], affinity, NULL); 355 kcpuset_destroy(affinity); 356 aprint_normal_dev(sc->sc_dev, 357 "for io queue %d interrupting at %s", qid, intrstr); 358 if (error == 0) 359 aprint_normal(" affinity to cpu%lu", affinity_to); 360 aprint_normal("\n"); 361 } 362 return 0; 363 } 364 365 static int 366 nvme_pci_intr_disestablish(struct nvme_softc *sc, uint16_t qid) 367 { 368 struct nvme_pci_softc *psc = (struct nvme_pci_softc *)sc; 369 370 KASSERT(sc->sc_use_mq || qid == NVME_ADMIN_Q); 371 KASSERT(sc->sc_ih[qid] != NULL); 372 373 if (sc->sc_softih) { 374 softint_disestablish(sc->sc_softih[qid]); 375 sc->sc_softih[qid] = NULL; 376 } 377 378 pci_intr_disestablish(psc->psc_pc, sc->sc_ih[qid]); 379 sc->sc_ih[qid] = NULL; 380 381 return 0; 382 } 383 384 static int 385 nvme_pci_setup_intr(struct pci_attach_args *pa, struct nvme_pci_softc *psc) 386 { 387 struct nvme_softc *sc = &psc->psc_nvme; 388 int error; 389 int counts[PCI_INTR_TYPE_SIZE], alloced_counts[PCI_INTR_TYPE_SIZE]; 390 pci_intr_handle_t *ihps; 391 int max_type, intr_type; 392 393 if (nvme_pci_force_intx) { 394 max_type = PCI_INTR_TYPE_INTX; 395 goto force_intx; 396 } 397 398 /* MSI-X */ 399 max_type = PCI_INTR_TYPE_MSIX; 400 counts[PCI_INTR_TYPE_MSIX] = min(pci_msix_count(pa->pa_pc, pa->pa_tag), 401 ncpu + 1); 402 if (counts[PCI_INTR_TYPE_MSIX] > 0) { 403 memset(alloced_counts, 0, sizeof(alloced_counts)); 404 alloced_counts[PCI_INTR_TYPE_MSIX] = counts[PCI_INTR_TYPE_MSIX]; 405 if (pci_intr_alloc(pa, &ihps, alloced_counts, 406 PCI_INTR_TYPE_MSIX)) { 407 counts[PCI_INTR_TYPE_MSIX] = 0; 408 } else { 409 counts[PCI_INTR_TYPE_MSIX] = 410 alloced_counts[PCI_INTR_TYPE_MSIX]; 411 pci_intr_release(pa->pa_pc, ihps, 412 alloced_counts[PCI_INTR_TYPE_MSIX]); 413 } 414 } 415 if (counts[PCI_INTR_TYPE_MSIX] < 2) { 416 counts[PCI_INTR_TYPE_MSIX] = 0; 417 max_type = PCI_INTR_TYPE_MSI; 418 } else if (!nvme_pci_mq || !nvme_pci_mpsafe) { 419 counts[PCI_INTR_TYPE_MSIX] = 2; /* adminq + 1 ioq */ 420 } 421 422 retry_msi: 423 /* MSI */ 424 counts[PCI_INTR_TYPE_MSI] = pci_msi_count(pa->pa_pc, pa->pa_tag); 425 if (counts[PCI_INTR_TYPE_MSI] > 0) { 426 while (counts[PCI_INTR_TYPE_MSI] > ncpu + 1) { 427 if (counts[PCI_INTR_TYPE_MSI] / 2 <= ncpu + 1) 428 break; 429 counts[PCI_INTR_TYPE_MSI] /= 2; 430 } 431 memset(alloced_counts, 0, sizeof(alloced_counts)); 432 alloced_counts[PCI_INTR_TYPE_MSI] = counts[PCI_INTR_TYPE_MSI]; 433 if (pci_intr_alloc(pa, &ihps, alloced_counts, 434 PCI_INTR_TYPE_MSI)) { 435 counts[PCI_INTR_TYPE_MSI] = 0; 436 } else { 437 counts[PCI_INTR_TYPE_MSI] = 438 alloced_counts[PCI_INTR_TYPE_MSI]; 439 pci_intr_release(pa->pa_pc, ihps, 440 alloced_counts[PCI_INTR_TYPE_MSI]); 441 } 442 } 443 if (counts[PCI_INTR_TYPE_MSI] < 1) { 444 counts[PCI_INTR_TYPE_MSI] = 0; 445 if (max_type == PCI_INTR_TYPE_MSI) 446 max_type = PCI_INTR_TYPE_INTX; 447 } else if (!nvme_pci_mq || !nvme_pci_mpsafe) { 448 if (counts[PCI_INTR_TYPE_MSI] > 2) 449 counts[PCI_INTR_TYPE_MSI] = 2; /* adminq + 1 ioq */ 450 } 451 452 force_intx: 453 /* INTx */ 454 counts[PCI_INTR_TYPE_INTX] = 1; 455 456 memcpy(alloced_counts, counts, sizeof(counts)); 457 error = pci_intr_alloc(pa, &ihps, alloced_counts, max_type); 458 if (error) { 459 if (max_type != PCI_INTR_TYPE_INTX) { 460 retry: 461 memset(counts, 0, sizeof(counts)); 462 if (max_type == PCI_INTR_TYPE_MSIX) { 463 max_type = PCI_INTR_TYPE_MSI; 464 goto retry_msi; 465 } else { 466 max_type = PCI_INTR_TYPE_INTX; 467 goto force_intx; 468 } 469 } 470 return error; 471 } 472 473 intr_type = pci_intr_type(pa->pa_pc, ihps[0]); 474 if (alloced_counts[intr_type] < counts[intr_type]) { 475 if (intr_type != PCI_INTR_TYPE_INTX) { 476 pci_intr_release(pa->pa_pc, ihps, 477 alloced_counts[intr_type]); 478 max_type = intr_type; 479 goto retry; 480 } 481 return EBUSY; 482 } 483 484 psc->psc_intrs = ihps; 485 psc->psc_nintrs = alloced_counts[intr_type]; 486 if (intr_type == PCI_INTR_TYPE_MSI) { 487 if (alloced_counts[intr_type] > ncpu + 1) 488 alloced_counts[intr_type] = ncpu + 1; 489 } 490 sc->sc_use_mq = alloced_counts[intr_type] > 1; 491 sc->sc_nq = sc->sc_use_mq ? alloced_counts[intr_type] - 1 : 1; 492 493 return 0; 494 } 495 496 MODULE(MODULE_CLASS_DRIVER, nvme, "pci,dk_subr"); 497 498 #ifdef _MODULE 499 #include "ioconf.c" 500 #endif 501 502 static int 503 nvme_modcmd(modcmd_t cmd, void *opaque) 504 { 505 #ifdef _MODULE 506 devmajor_t cmajor, bmajor; 507 extern const struct cdevsw nvme_cdevsw; 508 #endif 509 int error = 0; 510 511 #ifdef _MODULE 512 switch (cmd) { 513 case MODULE_CMD_INIT: 514 error = config_init_component(cfdriver_ioconf_nvme_pci, 515 cfattach_ioconf_nvme_pci, cfdata_ioconf_nvme_pci); 516 if (error) 517 break; 518 519 bmajor = cmajor = NODEVMAJOR; 520 error = devsw_attach(nvme_cd.cd_name, NULL, &bmajor, 521 &nvme_cdevsw, &cmajor); 522 if (error) { 523 aprint_error("%s: unable to register devsw\n", 524 nvme_cd.cd_name); 525 /* do not abort, just /dev/nvme* will not work */ 526 } 527 break; 528 case MODULE_CMD_FINI: 529 devsw_detach(NULL, &nvme_cdevsw); 530 531 error = config_fini_component(cfdriver_ioconf_nvme_pci, 532 cfattach_ioconf_nvme_pci, cfdata_ioconf_nvme_pci); 533 break; 534 default: 535 break; 536 } 537 #endif 538 return error; 539 } 540