sys/dev/nvme/nvme_pci.c

   1 /*-
   2  * Copyright (C) 2012-2016 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("$FreeBSD$");
  29
  30 #include <sys/param.h>
  31 #include <sys/systm.h>
  32 #include <sys/buf.h>
  33 #include <sys/bus.h>
  34 #include <sys/conf.h>
  35 #include <sys/proc.h>
  36 #include <sys/smp.h>
  37 #include <vm/vm.h>
  38
  39 #include <dev/pci/pcireg.h>
  40 #include <dev/pci/pcivar.h>
  41
  42 #include "nvme_private.h"
  43
  44 static int    nvme_pci_probe(device_t);
  45 static int    nvme_pci_attach(device_t);
  46 static int    nvme_pci_detach(device_t);
  47 static int    nvme_pci_suspend(device_t);
  48 static int    nvme_pci_resume(device_t);
  49
  50 static void nvme_ctrlr_setup_interrupts(struct nvme_controller *ctrlr);
  51
  52 static device_method_t nvme_pci_methods[] = {
  53         /* Device interface */
  54         DEVMETHOD(device_probe,     nvme_pci_probe),
  55         DEVMETHOD(device_attach,    nvme_pci_attach),
  56         DEVMETHOD(device_detach,    nvme_pci_detach),
  57         DEVMETHOD(device_suspend,   nvme_pci_suspend),
  58         DEVMETHOD(device_resume,    nvme_pci_resume),
  59         DEVMETHOD(device_shutdown,  nvme_shutdown),
  60         { 0, 0 }
  61 };
  62
  63 static driver_t nvme_pci_driver = {
  64         "nvme",
  65         nvme_pci_methods,
  66         sizeof(struct nvme_controller),
  67 };
  68
  69 DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, NULL, 0);
  70
  71 static struct _pcsid
  72 {
  73         uint32_t        devid;
  74         int             match_subdevice;
  75         uint16_t        subdevice;
  76         const char      *desc;
  77         uint32_t        quirks;
  78 } pci_ids[] = {
  79         { 0x01118086,           0, 0, "NVMe Controller"  },
  80         { IDT32_PCI_ID,         0, 0, "IDT NVMe Controller (32 channel)"  },
  81         { IDT8_PCI_ID,          0, 0, "IDT NVMe Controller (8 channel)" },
  82         { 0x09538086,           1, 0x3702, "DC P3700 SSD" },
  83         { 0x09538086,           1, 0x3703, "DC P3700 SSD [2.5\" SFF]" },
  84         { 0x09538086,           1, 0x3704, "DC P3500 SSD [Add-in Card]" },
  85         { 0x09538086,           1, 0x3705, "DC P3500 SSD [2.5\" SFF]" },
  86         { 0x09538086,           1, 0x3709, "DC P3600 SSD [Add-in Card]" },
  87         { 0x09538086,           1, 0x370a, "DC P3600 SSD [2.5\" SFF]" },
  88         { 0x00031c58,           0, 0, "HGST SN100",     QUIRK_DELAY_B4_CHK_RDY },
  89         { 0x00231c58,           0, 0, "WDC SN200",      QUIRK_DELAY_B4_CHK_RDY },
  90         { 0x05401c5f,           0, 0, "Memblaze Pblaze4", QUIRK_DELAY_B4_CHK_RDY },
  91         { 0xa821144d,           0, 0, "Samsung PM1725", QUIRK_DELAY_B4_CHK_RDY },
  92         { 0xa822144d,           0, 0, "Samsung PM1725a", QUIRK_DELAY_B4_CHK_RDY },
  93         { 0x00000000,           0, 0, NULL  }
  94 };
  95
  96 static int
  97 nvme_match(uint32_t devid, uint16_t subdevice, struct _pcsid *ep)
  98 {
  99         if (devid != ep->devid)
 100                 return 0;
 101
 102         if (!ep->match_subdevice)
 103                 return 1;
 104
 105         if (subdevice == ep->subdevice)
 106                 return 1;
 107         else
 108                 return 0;
 109 }
 110
 111 static int
 112 nvme_pci_probe (device_t device)
 113 {
 114         struct nvme_controller *ctrlr = DEVICE2SOFTC(device);
 115         struct _pcsid   *ep;
 116         uint32_t        devid;
 117         uint16_t        subdevice;
 118
 119         devid = pci_get_devid(device);
 120         subdevice = pci_get_subdevice(device);
 121         ep = pci_ids;
 122
 123         while (ep->devid) {
 124                 if (nvme_match(devid, subdevice, ep))
 125                         break;
 126                 ++ep;
 127         }
 128         if (ep->devid)
 129                 ctrlr->quirks = ep->quirks;
 130
 131         if (ep->desc) {
 132                 device_set_desc(device, ep->desc);
 133                 return (BUS_PROBE_DEFAULT);
 134         }
 135
 136 #if defined(PCIS_STORAGE_NVM)
 137         if (pci_get_class(device)    == PCIC_STORAGE &&
 138             pci_get_subclass(device) == PCIS_STORAGE_NVM &&
 139             pci_get_progif(device)   == PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) {
 140                 device_set_desc(device, "Generic NVMe Device");
 141                 return (BUS_PROBE_GENERIC);
 142         }
 143 #endif
 144
 145         return (ENXIO);
 146 }
 147
 148 static int
 149 nvme_ctrlr_allocate_bar(struct nvme_controller *ctrlr)
 150 {
 151
 152         ctrlr->resource_id = PCIR_BAR(0);
 153
 154         ctrlr->resource = bus_alloc_resource_any(ctrlr->dev, SYS_RES_MEMORY,
 155             &ctrlr->resource_id, RF_ACTIVE);
 156
 157         if(ctrlr->resource == NULL) {
 158                 nvme_printf(ctrlr, "unable to allocate pci resource\n");
 159                 return (ENOMEM);
 160         }
 161
 162         ctrlr->bus_tag = rman_get_bustag(ctrlr->resource);
 163         ctrlr->bus_handle = rman_get_bushandle(ctrlr->resource);
 164         ctrlr->regs = (struct nvme_registers *)ctrlr->bus_handle;
 165
 166         /*
 167          * The NVMe spec allows for the MSI-X table to be placed behind
 168          *  BAR 4/5, separate from the control/doorbell registers.  Always
 169          *  try to map this bar, because it must be mapped prior to calling
 170          *  pci_alloc_msix().  If the table isn't behind BAR 4/5,
 171          *  bus_alloc_resource() will just return NULL which is OK.
 172          */
 173         ctrlr->bar4_resource_id = PCIR_BAR(4);
 174         ctrlr->bar4_resource = bus_alloc_resource_any(ctrlr->dev, SYS_RES_MEMORY,
 175             &ctrlr->bar4_resource_id, RF_ACTIVE);
 176
 177         return (0);
 178 }
 179
 180 static int
 181 nvme_pci_attach(device_t dev)
 182 {
 183         struct nvme_controller*ctrlr = DEVICE2SOFTC(dev);
 184         int status;
 185
 186         ctrlr->dev = dev;
 187         status = nvme_ctrlr_allocate_bar(ctrlr);
 188         if (status != 0)
 189                 goto bad;
 190         pci_enable_busmaster(dev);
 191         nvme_ctrlr_setup_interrupts(ctrlr);
 192         return nvme_attach(dev);
 193 bad:
 194         if (ctrlr->resource != NULL) {
 195                 bus_release_resource(dev, SYS_RES_MEMORY,
 196                     ctrlr->resource_id, ctrlr->resource);
 197         }
 198
 199         if (ctrlr->bar4_resource != NULL) {
 200                 bus_release_resource(dev, SYS_RES_MEMORY,
 201                     ctrlr->bar4_resource_id, ctrlr->bar4_resource);
 202         }
 203
 204         if (ctrlr->tag)
 205                 bus_teardown_intr(dev, ctrlr->res, ctrlr->tag);
 206
 207         if (ctrlr->res)
 208                 bus_release_resource(dev, SYS_RES_IRQ,
 209                     rman_get_rid(ctrlr->res), ctrlr->res);
 210
 211         if (ctrlr->msix_enabled)
 212                 pci_release_msi(dev);
 213
 214         return status;
 215 }
 216
 217 static int
 218 nvme_pci_detach(device_t dev)
 219 {
 220         struct nvme_controller*ctrlr = DEVICE2SOFTC(dev);
 221         int rv;
 222
 223         rv = nvme_detach(dev);
 224         if (ctrlr->msix_enabled)
 225                 pci_release_msi(dev);
 226         pci_disable_busmaster(dev);
 227         return (rv);
 228 }
 229
 230 static int
 231 nvme_ctrlr_configure_intx(struct nvme_controller *ctrlr)
 232 {
 233
 234         ctrlr->msix_enabled = 0;
 235         ctrlr->num_io_queues = 1;
 236         ctrlr->rid = 0;
 237         ctrlr->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ,
 238             &ctrlr->rid, RF_SHAREABLE | RF_ACTIVE);
 239
 240         if (ctrlr->res == NULL) {
 241                 nvme_printf(ctrlr, "unable to allocate shared IRQ\n");
 242                 return (ENOMEM);
 243         }
 244
 245         if (bus_setup_intr(ctrlr->dev, ctrlr->res,
 246             INTR_TYPE_MISC | INTR_MPSAFE, NULL, nvme_ctrlr_intx_handler,
 247             ctrlr, &ctrlr->tag) != 0) {
 248                 nvme_printf(ctrlr, "unable to setup intx handler\n");
 249                 return (ENOMEM);
 250         }
 251
 252         return (0);
 253 }
 254
 255 static void
 256 nvme_ctrlr_setup_interrupts(struct nvme_controller *ctrlr)
 257 {
 258         device_t        dev;
 259         int             force_intx, num_io_queues, per_cpu_io_queues;
 260         int             min_cpus_per_ioq;
 261         int             num_vectors_requested, num_vectors_allocated;
 262
 263         dev = ctrlr->dev;
 264
 265         force_intx = 0;
 266         TUNABLE_INT_FETCH("hw.nvme.force_intx", &force_intx);
 267         if (force_intx || pci_msix_count(dev) < 2) {
 268                 nvme_ctrlr_configure_intx(ctrlr);
 269                 return;
 270         }
 271
 272         num_io_queues = mp_ncpus;
 273         TUNABLE_INT_FETCH("hw.nvme.num_io_queues", &num_io_queues);
 274         if (num_io_queues < 1 || num_io_queues > mp_ncpus)
 275                 num_io_queues = mp_ncpus;
 276
 277         per_cpu_io_queues = 1;
 278         TUNABLE_INT_FETCH("hw.nvme.per_cpu_io_queues", &per_cpu_io_queues);
 279         if (per_cpu_io_queues == 0)
 280                 num_io_queues = 1;
 281
 282         min_cpus_per_ioq = smp_threads_per_core;
 283         TUNABLE_INT_FETCH("hw.nvme.min_cpus_per_ioq", &min_cpus_per_ioq);
 284         if (min_cpus_per_ioq > 1) {
 285                 num_io_queues = min(num_io_queues,
 286                     max(1, mp_ncpus / min_cpus_per_ioq));
 287         }
 288
 289         num_io_queues = min(num_io_queues, pci_msix_count(dev) - 1);
 290
 291 again:
 292         if (num_io_queues > vm_ndomains)
 293                 num_io_queues -= num_io_queues % vm_ndomains;
 294         /* One vector for per core I/O queue, plus one vector for admin queue. */
 295         num_vectors_requested = num_io_queues + 1;
 296         num_vectors_allocated = num_vectors_requested;
 297         if (pci_alloc_msix(dev, &num_vectors_allocated) != 0) {
 298                 nvme_ctrlr_configure_intx(ctrlr);
 299                 return;
 300         }
 301         if (num_vectors_allocated < 2) {
 302                 pci_release_msi(dev);
 303                 nvme_ctrlr_configure_intx(ctrlr);
 304                 return;
 305         }
 306         if (num_vectors_allocated != num_vectors_requested) {
 307                 pci_release_msi(dev);
 308                 num_io_queues = num_vectors_allocated - 1;
 309                 goto again;
 310         }
 311
 312         ctrlr->msix_enabled = 1;
 313         ctrlr->num_io_queues = num_io_queues;
 314 }
 315
 316 static int
 317 nvme_pci_suspend(device_t dev)
 318 {
 319         struct nvme_controller  *ctrlr;
 320
 321         ctrlr = DEVICE2SOFTC(dev);
 322         return (nvme_ctrlr_suspend(ctrlr));
 323 }
 324
 325 static int
 326 nvme_pci_resume(device_t dev)
 327 {
 328         struct nvme_controller  *ctrlr;
 329
 330         ctrlr = DEVICE2SOFTC(dev);
 331         return (nvme_ctrlr_resume(ctrlr));
 332 }