sys/dev/nvme/nvme.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (C) 2012-2014 Intel Corporation
   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 AND CONTRIBUTORS ``AS IS'' AND
  17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  26  * SUCH DAMAGE.
  27  */
  28
  29 #include <sys/cdefs.h>
  30 __FBSDID("$FreeBSD$");
  31
  32 #include <sys/param.h>
  33 #include <sys/bus.h>
  34 #include <sys/conf.h>
  35 #include <sys/module.h>
  36
  37 #include <vm/uma.h>
  38
  39 #include <dev/pci/pcireg.h>
  40 #include <dev/pci/pcivar.h>
  41
  42 #include "nvme_private.h"
  43
  44 struct nvme_consumer {
  45         uint32_t                id;
  46         nvme_cons_ns_fn_t       ns_fn;
  47         nvme_cons_ctrlr_fn_t    ctrlr_fn;
  48         nvme_cons_async_fn_t    async_fn;
  49         nvme_cons_fail_fn_t     fail_fn;
  50 };
  51
  52 struct nvme_consumer nvme_consumer[NVME_MAX_CONSUMERS];
  53 #define INVALID_CONSUMER_ID     0xFFFF
  54
  55 uma_zone_t      nvme_request_zone;
  56 int32_t         nvme_retry_count;
  57
  58 MALLOC_DEFINE(M_NVME, "nvme", "nvme(4) memory allocations");
  59
  60 static int    nvme_probe(device_t);
  61 static int    nvme_attach(device_t);
  62 static int    nvme_detach(device_t);
  63 static int    nvme_shutdown(device_t);
  64 static int    nvme_modevent(module_t mod, int type, void *arg);
  65
  66 static devclass_t nvme_devclass;
  67
  68 static device_method_t nvme_pci_methods[] = {
  69         /* Device interface */
  70         DEVMETHOD(device_probe,     nvme_probe),
  71         DEVMETHOD(device_attach,    nvme_attach),
  72         DEVMETHOD(device_detach,    nvme_detach),
  73         DEVMETHOD(device_shutdown,  nvme_shutdown),
  74         { 0, 0 }
  75 };
  76
  77 static driver_t nvme_pci_driver = {
  78         "nvme",
  79         nvme_pci_methods,
  80         sizeof(struct nvme_controller),
  81 };
  82
  83 DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, nvme_modevent, 0);
  84 MODULE_VERSION(nvme, 1);
  85 MODULE_DEPEND(nvme, cam, 1, 1, 1);
  86
  87 static struct _pcsid
  88 {
  89         uint32_t        devid;
  90         int             match_subdevice;
  91         uint16_t        subdevice;
  92         const char      *desc;
  93         uint32_t        quirks;
  94 } pci_ids[] = {
  95         { 0x01118086,           0, 0, "NVMe Controller"  },
  96         { IDT32_PCI_ID,         0, 0, "IDT NVMe Controller (32 channel)"  },
  97         { IDT8_PCI_ID,          0, 0, "IDT NVMe Controller (8 channel)" },
  98         { 0x09538086,           1, 0x3702, "DC P3700 SSD" },
  99         { 0x09538086,           1, 0x3703, "DC P3700 SSD [2.5\" SFF]" },
 100         { 0x09538086,           1, 0x3704, "DC P3500 SSD [Add-in Card]" },
 101         { 0x09538086,           1, 0x3705, "DC P3500 SSD [2.5\" SFF]" },
 102         { 0x09538086,           1, 0x3709, "DC P3600 SSD [Add-in Card]" },
 103         { 0x09538086,           1, 0x370a, "DC P3600 SSD [2.5\" SFF]" },
 104         { 0x00031c58,           0, 0, "HGST SN100",     QUIRK_DELAY_B4_CHK_RDY },
 105         { 0x00231c58,           0, 0, "WDC SN200",      QUIRK_DELAY_B4_CHK_RDY },
 106         { 0x05401c5f,           0, 0, "Memblaze Pblaze4", QUIRK_DELAY_B4_CHK_RDY },
 107         { 0xa821144d,           0, 0, "Samsung PM1725", QUIRK_DELAY_B4_CHK_RDY },
 108         { 0xa822144d,           0, 0, "Samsung PM1725a", QUIRK_DELAY_B4_CHK_RDY },
 109         { 0x01161179,           0, 0, "Toshiba XG5", QUIRK_DISABLE_TIMEOUT },
 110         { 0x00000000,           0, 0, NULL  }
 111 };
 112
 113 static int
 114 nvme_match(uint32_t devid, uint16_t subdevice, struct _pcsid *ep)
 115 {
 116         if (devid != ep->devid)
 117                 return 0;
 118
 119         if (!ep->match_subdevice)
 120                 return 1;
 121
 122         if (subdevice == ep->subdevice)
 123                 return 1;
 124         else
 125                 return 0;
 126 }
 127
 128 static int
 129 nvme_probe (device_t device)
 130 {
 131         struct _pcsid   *ep;
 132         uint32_t        devid;
 133         uint16_t        subdevice;
 134
 135         devid = pci_get_devid(device);
 136         subdevice = pci_get_subdevice(device);
 137         ep = pci_ids;
 138
 139         while (ep->devid) {
 140                 if (nvme_match(devid, subdevice, ep))
 141                         break;
 142                 ++ep;
 143         }
 144
 145         if (ep->desc) {
 146                 device_set_desc(device, ep->desc);
 147                 return (BUS_PROBE_DEFAULT);
 148         }
 149
 150 #if defined(PCIS_STORAGE_NVM)
 151         if (pci_get_class(device)    == PCIC_STORAGE &&
 152             pci_get_subclass(device) == PCIS_STORAGE_NVM &&
 153             pci_get_progif(device)   == PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) {
 154                 device_set_desc(device, "Generic NVMe Device");
 155                 return (BUS_PROBE_GENERIC);
 156         }
 157 #endif
 158
 159         return (ENXIO);
 160 }
 161
 162 static void
 163 nvme_init(void)
 164 {
 165         uint32_t        i;
 166
 167         nvme_request_zone = uma_zcreate("nvme_request",
 168             sizeof(struct nvme_request), NULL, NULL, NULL, NULL, 0, 0);
 169
 170         for (i = 0; i < NVME_MAX_CONSUMERS; i++)
 171                 nvme_consumer[i].id = INVALID_CONSUMER_ID;
 172 }
 173
 174 SYSINIT(nvme_register, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_init, NULL);
 175
 176 static void
 177 nvme_uninit(void)
 178 {
 179         uma_zdestroy(nvme_request_zone);
 180 }
 181
 182 SYSUNINIT(nvme_unregister, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_uninit, NULL);
 183
 184 static void
 185 nvme_load(void)
 186 {
 187 }
 188
 189 static void
 190 nvme_unload(void)
 191 {
 192 }
 193
 194 static int
 195 nvme_shutdown(device_t dev)
 196 {
 197         struct nvme_controller  *ctrlr;
 198
 199         ctrlr = DEVICE2SOFTC(dev);
 200         nvme_ctrlr_shutdown(ctrlr);
 201
 202         return (0);
 203 }
 204
 205 static int
 206 nvme_modevent(module_t mod, int type, void *arg)
 207 {
 208
 209         switch (type) {
 210         case MOD_LOAD:
 211                 nvme_load();
 212                 break;
 213         case MOD_UNLOAD:
 214                 nvme_unload();
 215                 break;
 216         default:
 217                 break;
 218         }
 219
 220         return (0);
 221 }
 222
 223 void
 224 nvme_dump_command(struct nvme_command *cmd)
 225 {
 226
 227         printf(
 228 "opc:%x f:%x cid:%x nsid:%x r2:%x r3:%x mptr:%jx prp1:%jx prp2:%jx cdw:%x %x %x %x %x %x\n",
 229             cmd->opc, cmd->fuse, cmd->cid, le32toh(cmd->nsid),
 230             cmd->rsvd2, cmd->rsvd3,
 231             (uintmax_t)le64toh(cmd->mptr), (uintmax_t)le64toh(cmd->prp1), (uintmax_t)le64toh(cmd->prp2),
 232             le32toh(cmd->cdw10), le32toh(cmd->cdw11), le32toh(cmd->cdw12),
 233             le32toh(cmd->cdw13), le32toh(cmd->cdw14), le32toh(cmd->cdw15));
 234 }
 235
 236 void
 237 nvme_dump_completion(struct nvme_completion *cpl)
 238 {
 239         uint8_t p, sc, sct, m, dnr;
 240         uint16_t status;
 241
 242         status = le16toh(cpl->status);
 243
 244         p = NVME_STATUS_GET_P(status);
 245         sc = NVME_STATUS_GET_SC(status);
 246         sct = NVME_STATUS_GET_SCT(status);
 247         m = NVME_STATUS_GET_M(status);
 248         dnr = NVME_STATUS_GET_DNR(status);
 249
 250         printf("cdw0:%08x sqhd:%04x sqid:%04x "
 251             "cid:%04x p:%x sc:%02x sct:%x m:%x dnr:%x\n",
 252             le32toh(cpl->cdw0), le16toh(cpl->sqhd), le16toh(cpl->sqid),
 253             cpl->cid, p, sc, sct, m, dnr);
 254 }
 255
 256 static int
 257 nvme_attach(device_t dev)
 258 {
 259         struct nvme_controller  *ctrlr = DEVICE2SOFTC(dev);
 260         int                     status;
 261         struct _pcsid           *ep;
 262         uint32_t                devid;
 263         uint16_t                subdevice;
 264
 265         devid = pci_get_devid(dev);
 266         subdevice = pci_get_subdevice(dev);
 267         ep = pci_ids;
 268         while (ep->devid) {
 269                 if (nvme_match(devid, subdevice, ep))
 270                         break;
 271                 ++ep;
 272         }
 273         ctrlr->quirks = ep->quirks;
 274
 275         status = nvme_ctrlr_construct(ctrlr, dev);
 276
 277         if (status != 0) {
 278                 nvme_ctrlr_destruct(ctrlr, dev);
 279                 return (status);
 280         }
 281
 282         /*
 283          * Some drives do not implement the completion timeout feature
 284          * correctly. There's a WAR from the manufacturer to just disable it.
 285          * The driver wouldn't respond correctly to a timeout anyway.
 286          */
 287         if (ep->quirks & QUIRK_DISABLE_TIMEOUT) {
 288                 int ptr;
 289                 uint16_t devctl2;
 290
 291                 status = pci_find_cap(dev, PCIY_EXPRESS, &ptr);
 292                 if (status) {
 293                         device_printf(dev, "Can't locate PCIe capability?");
 294                         return (status);
 295                 }
 296                 devctl2 = pci_read_config(dev, ptr + PCIER_DEVICE_CTL2, sizeof(devctl2));
 297                 devctl2 |= PCIEM_CTL2_COMP_TIMO_DISABLE;
 298                 pci_write_config(dev, ptr + PCIER_DEVICE_CTL2, devctl2, sizeof(devctl2));
 299         }
 300
 301         /*
 302          * Enable busmastering so the completion status messages can
 303          * be busmastered back to the host.
 304          */
 305         pci_enable_busmaster(dev);
 306
 307         /*
 308          * Reset controller twice to ensure we do a transition from cc.en==1
 309          *  to cc.en==0.  This is because we don't really know what status
 310          *  the controller was left in when boot handed off to OS.
 311          */
 312         status = nvme_ctrlr_hw_reset(ctrlr);
 313         if (status != 0) {
 314                 nvme_ctrlr_destruct(ctrlr, dev);
 315                 return (status);
 316         }
 317
 318         status = nvme_ctrlr_hw_reset(ctrlr);
 319         if (status != 0) {
 320                 nvme_ctrlr_destruct(ctrlr, dev);
 321                 return (status);
 322         }
 323
 324         ctrlr->config_hook.ich_func = nvme_ctrlr_start_config_hook;
 325         ctrlr->config_hook.ich_arg = ctrlr;
 326
 327         config_intrhook_establish(&ctrlr->config_hook);
 328
 329         return (0);
 330 }
 331
 332 static int
 333 nvme_detach (device_t dev)
 334 {
 335         struct nvme_controller  *ctrlr = DEVICE2SOFTC(dev);
 336
 337         nvme_ctrlr_destruct(ctrlr, dev);
 338         pci_disable_busmaster(dev);
 339         return (0);
 340 }
 341
 342 static void
 343 nvme_notify(struct nvme_consumer *cons,
 344             struct nvme_controller *ctrlr)
 345 {
 346         struct nvme_namespace   *ns;
 347         void                    *ctrlr_cookie;
 348         int                     cmpset, ns_idx;
 349
 350         /*
 351          * The consumer may register itself after the nvme devices
 352          *  have registered with the kernel, but before the
 353          *  driver has completed initialization.  In that case,
 354          *  return here, and when initialization completes, the
 355          *  controller will make sure the consumer gets notified.
 356          */
 357         if (!ctrlr->is_initialized)
 358                 return;
 359
 360         cmpset = atomic_cmpset_32(&ctrlr->notification_sent, 0, 1);
 361
 362         if (cmpset == 0)
 363                 return;
 364
 365         if (cons->ctrlr_fn != NULL)
 366                 ctrlr_cookie = (*cons->ctrlr_fn)(ctrlr);
 367         else
 368                 ctrlr_cookie = NULL;
 369         ctrlr->cons_cookie[cons->id] = ctrlr_cookie;
 370         if (ctrlr->is_failed) {
 371                 if (cons->fail_fn != NULL)
 372                         (*cons->fail_fn)(ctrlr_cookie);
 373                 /*
 374                  * Do not notify consumers about the namespaces of a
 375                  *  failed controller.
 376                  */
 377                 return;
 378         }
 379         for (ns_idx = 0; ns_idx < min(ctrlr->cdata.nn, NVME_MAX_NAMESPACES); ns_idx++) {
 380                 ns = &ctrlr->ns[ns_idx];
 381                 if (ns->data.nsze == 0)
 382                         continue;
 383                 if (cons->ns_fn != NULL)
 384                         ns->cons_cookie[cons->id] =
 385                             (*cons->ns_fn)(ns, ctrlr_cookie);
 386         }
 387 }
 388
 389 void
 390 nvme_notify_new_controller(struct nvme_controller *ctrlr)
 391 {
 392         int i;
 393
 394         for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
 395                 if (nvme_consumer[i].id != INVALID_CONSUMER_ID) {
 396                         nvme_notify(&nvme_consumer[i], ctrlr);
 397                 }
 398         }
 399 }
 400
 401 static void
 402 nvme_notify_new_consumer(struct nvme_consumer *cons)
 403 {
 404         device_t                *devlist;
 405         struct nvme_controller  *ctrlr;
 406         int                     dev_idx, devcount;
 407
 408         if (devclass_get_devices(nvme_devclass, &devlist, &devcount))
 409                 return;
 410
 411         for (dev_idx = 0; dev_idx < devcount; dev_idx++) {
 412                 ctrlr = DEVICE2SOFTC(devlist[dev_idx]);
 413                 nvme_notify(cons, ctrlr);
 414         }
 415
 416         free(devlist, M_TEMP);
 417 }
 418
 419 void
 420 nvme_notify_async_consumers(struct nvme_controller *ctrlr,
 421                             const struct nvme_completion *async_cpl,
 422                             uint32_t log_page_id, void *log_page_buffer,
 423                             uint32_t log_page_size)
 424 {
 425         struct nvme_consumer    *cons;
 426         uint32_t                i;
 427
 428         for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
 429                 cons = &nvme_consumer[i];
 430                 if (cons->id != INVALID_CONSUMER_ID && cons->async_fn != NULL)
 431                         (*cons->async_fn)(ctrlr->cons_cookie[i], async_cpl,
 432                             log_page_id, log_page_buffer, log_page_size);
 433         }
 434 }
 435
 436 void
 437 nvme_notify_fail_consumers(struct nvme_controller *ctrlr)
 438 {
 439         struct nvme_consumer    *cons;
 440         uint32_t                i;
 441
 442         /*
 443          * This controller failed during initialization (i.e. IDENTIFY
 444          *  command failed or timed out).  Do not notify any nvme
 445          *  consumers of the failure here, since the consumer does not
 446          *  even know about the controller yet.
 447          */
 448         if (!ctrlr->is_initialized)
 449                 return;
 450
 451         for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
 452                 cons = &nvme_consumer[i];
 453                 if (cons->id != INVALID_CONSUMER_ID && cons->fail_fn != NULL)
 454                         cons->fail_fn(ctrlr->cons_cookie[i]);
 455         }
 456 }
 457
 458 void
 459 nvme_notify_ns(struct nvme_controller *ctrlr, int nsid)
 460 {
 461         struct nvme_consumer    *cons;
 462         struct nvme_namespace   *ns = &ctrlr->ns[nsid - 1];
 463         uint32_t                i;
 464
 465         if (!ctrlr->is_initialized)
 466                 return;
 467
 468         for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
 469                 cons = &nvme_consumer[i];
 470                 if (cons->id != INVALID_CONSUMER_ID && cons->ns_fn != NULL)
 471                         ns->cons_cookie[cons->id] =
 472                             (*cons->ns_fn)(ns, ctrlr->cons_cookie[cons->id]);
 473         }
 474 }
 475
 476 struct nvme_consumer *
 477 nvme_register_consumer(nvme_cons_ns_fn_t ns_fn, nvme_cons_ctrlr_fn_t ctrlr_fn,
 478                        nvme_cons_async_fn_t async_fn,
 479                        nvme_cons_fail_fn_t fail_fn)
 480 {
 481         int i;
 482
 483         /*
 484          * TODO: add locking around consumer registration.  Not an issue
 485          *  right now since we only have one nvme consumer - nvd(4).
 486          */
 487         for (i = 0; i < NVME_MAX_CONSUMERS; i++)
 488                 if (nvme_consumer[i].id == INVALID_CONSUMER_ID) {
 489                         nvme_consumer[i].id = i;
 490                         nvme_consumer[i].ns_fn = ns_fn;
 491                         nvme_consumer[i].ctrlr_fn = ctrlr_fn;
 492                         nvme_consumer[i].async_fn = async_fn;
 493                         nvme_consumer[i].fail_fn = fail_fn;
 494
 495                         nvme_notify_new_consumer(&nvme_consumer[i]);
 496                         return (&nvme_consumer[i]);
 497                 }
 498
 499         printf("nvme(4): consumer not registered - no slots available\n");
 500         return (NULL);
 501 }
 502
 503 void
 504 nvme_unregister_consumer(struct nvme_consumer *consumer)
 505 {
 506
 507         consumer->id = INVALID_CONSUMER_ID;
 508 }
 509
 510 void
 511 nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl)
 512 {
 513         struct nvme_completion_poll_status      *status = arg;
 514
 515         /*
 516          * Copy status into the argument passed by the caller, so that
 517          *  the caller can check the status to determine if the
 518          *  the request passed or failed.
 519          */
 520         memcpy(&status->cpl, cpl, sizeof(*cpl));
 521         atomic_store_rel_int(&status->done, 1);
 522 }