xref: /freebsd/sys/dev/nvme/nvme_pci.c (revision 840aca288042eaf625a23908e807abdfde0bc21d)
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 
38 #include <dev/pci/pcireg.h>
39 #include <dev/pci/pcivar.h>
40 
41 #include "nvme_private.h"
42 
43 static int    nvme_pci_probe(device_t);
44 static int    nvme_pci_attach(device_t);
45 static int    nvme_pci_detach(device_t);
46 
47 static void nvme_ctrlr_setup_interrupts(struct nvme_controller *ctrlr);
48 
49 static device_method_t nvme_pci_methods[] = {
50 	/* Device interface */
51 	DEVMETHOD(device_probe,     nvme_pci_probe),
52 	DEVMETHOD(device_attach,    nvme_pci_attach),
53 	DEVMETHOD(device_detach,    nvme_pci_detach),
54 	DEVMETHOD(device_shutdown,  nvme_shutdown),
55 	{ 0, 0 }
56 };
57 
58 static driver_t nvme_pci_driver = {
59 	"nvme",
60 	nvme_pci_methods,
61 	sizeof(struct nvme_controller),
62 };
63 
64 DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, NULL, 0);
65 
66 static struct _pcsid
67 {
68 	uint32_t	devid;
69 	int		match_subdevice;
70 	uint16_t	subdevice;
71 	const char	*desc;
72 	uint32_t	quirks;
73 } pci_ids[] = {
74 	{ 0x01118086,		0, 0, "NVMe Controller"  },
75 	{ IDT32_PCI_ID,		0, 0, "IDT NVMe Controller (32 channel)"  },
76 	{ IDT8_PCI_ID,		0, 0, "IDT NVMe Controller (8 channel)" },
77 	{ 0x09538086,		1, 0x3702, "DC P3700 SSD" },
78 	{ 0x09538086,		1, 0x3703, "DC P3700 SSD [2.5\" SFF]" },
79 	{ 0x09538086,		1, 0x3704, "DC P3500 SSD [Add-in Card]" },
80 	{ 0x09538086,		1, 0x3705, "DC P3500 SSD [2.5\" SFF]" },
81 	{ 0x09538086,		1, 0x3709, "DC P3600 SSD [Add-in Card]" },
82 	{ 0x09538086,		1, 0x370a, "DC P3600 SSD [2.5\" SFF]" },
83 	{ 0x00031c58,		0, 0, "HGST SN100",	QUIRK_DELAY_B4_CHK_RDY },
84 	{ 0x00231c58,		0, 0, "WDC SN200",	QUIRK_DELAY_B4_CHK_RDY },
85 	{ 0x05401c5f,		0, 0, "Memblaze Pblaze4", QUIRK_DELAY_B4_CHK_RDY },
86 	{ 0xa821144d,		0, 0, "Samsung PM1725", QUIRK_DELAY_B4_CHK_RDY },
87 	{ 0xa822144d,		0, 0, "Samsung PM1725a", QUIRK_DELAY_B4_CHK_RDY },
88 	{ 0x00000000,		0, 0, NULL  }
89 };
90 
91 
92 static int
93 nvme_match(uint32_t devid, uint16_t subdevice, struct _pcsid *ep)
94 {
95 	if (devid != ep->devid)
96 		return 0;
97 
98 	if (!ep->match_subdevice)
99 		return 1;
100 
101 	if (subdevice == ep->subdevice)
102 		return 1;
103 	else
104 		return 0;
105 }
106 
107 static int
108 nvme_pci_probe (device_t device)
109 {
110 	struct nvme_controller *ctrlr = DEVICE2SOFTC(device);
111 	struct _pcsid	*ep;
112 	uint32_t	devid;
113 	uint16_t	subdevice;
114 
115 	devid = pci_get_devid(device);
116 	subdevice = pci_get_subdevice(device);
117 	ep = pci_ids;
118 
119 	while (ep->devid) {
120 		if (nvme_match(devid, subdevice, ep))
121 			break;
122 		++ep;
123 	}
124 	if (ep->devid)
125 		ctrlr->quirks = ep->quirks;
126 
127 	if (ep->desc) {
128 		device_set_desc(device, ep->desc);
129 		return (BUS_PROBE_DEFAULT);
130 	}
131 
132 #if defined(PCIS_STORAGE_NVM)
133 	if (pci_get_class(device)    == PCIC_STORAGE &&
134 	    pci_get_subclass(device) == PCIS_STORAGE_NVM &&
135 	    pci_get_progif(device)   == PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) {
136 		device_set_desc(device, "Generic NVMe Device");
137 		return (BUS_PROBE_GENERIC);
138 	}
139 #endif
140 
141 	return (ENXIO);
142 }
143 
144 static int
145 nvme_ctrlr_allocate_bar(struct nvme_controller *ctrlr)
146 {
147 
148 	ctrlr->resource_id = PCIR_BAR(0);
149 
150 	ctrlr->resource = bus_alloc_resource_any(ctrlr->dev, SYS_RES_MEMORY,
151 	    &ctrlr->resource_id, RF_ACTIVE);
152 
153 	if(ctrlr->resource == NULL) {
154 		nvme_printf(ctrlr, "unable to allocate pci resource\n");
155 		return (ENOMEM);
156 	}
157 
158 	ctrlr->bus_tag = rman_get_bustag(ctrlr->resource);
159 	ctrlr->bus_handle = rman_get_bushandle(ctrlr->resource);
160 	ctrlr->regs = (struct nvme_registers *)ctrlr->bus_handle;
161 
162 	/*
163 	 * The NVMe spec allows for the MSI-X table to be placed behind
164 	 *  BAR 4/5, separate from the control/doorbell registers.  Always
165 	 *  try to map this bar, because it must be mapped prior to calling
166 	 *  pci_alloc_msix().  If the table isn't behind BAR 4/5,
167 	 *  bus_alloc_resource() will just return NULL which is OK.
168 	 */
169 	ctrlr->bar4_resource_id = PCIR_BAR(4);
170 	ctrlr->bar4_resource = bus_alloc_resource_any(ctrlr->dev, SYS_RES_MEMORY,
171 	    &ctrlr->bar4_resource_id, RF_ACTIVE);
172 
173 	return (0);
174 }
175 
176 static int
177 nvme_pci_attach(device_t dev)
178 {
179 	struct nvme_controller*ctrlr = DEVICE2SOFTC(dev);
180 	int status;
181 
182 	ctrlr->dev = dev;
183 	status = nvme_ctrlr_allocate_bar(ctrlr);
184 	if (status != 0)
185 		goto bad;
186 	pci_enable_busmaster(dev);
187 	nvme_ctrlr_setup_interrupts(ctrlr);
188 	return nvme_attach(dev);
189 bad:
190 	if (ctrlr->resource != NULL) {
191 		bus_release_resource(dev, SYS_RES_MEMORY,
192 		    ctrlr->resource_id, ctrlr->resource);
193 	}
194 
195 	if (ctrlr->bar4_resource != NULL) {
196 		bus_release_resource(dev, SYS_RES_MEMORY,
197 		    ctrlr->bar4_resource_id, ctrlr->bar4_resource);
198 	}
199 
200 	if (ctrlr->tag)
201 		bus_teardown_intr(dev, ctrlr->res, ctrlr->tag);
202 
203 	if (ctrlr->res)
204 		bus_release_resource(dev, SYS_RES_IRQ,
205 		    rman_get_rid(ctrlr->res), ctrlr->res);
206 
207 	if (ctrlr->msix_enabled)
208 		pci_release_msi(dev);
209 
210 	return status;
211 }
212 
213 static int
214 nvme_pci_detach(device_t dev)
215 {
216 	struct nvme_controller*ctrlr = DEVICE2SOFTC(dev);
217 	int rv;
218 
219 	rv = nvme_detach(dev);
220 	if (ctrlr->msix_enabled)
221 		pci_release_msi(dev);
222 	pci_disable_busmaster(dev);
223 	return (rv);
224 }
225 
226 static int
227 nvme_ctrlr_configure_intx(struct nvme_controller *ctrlr)
228 {
229 
230 	ctrlr->msix_enabled = 0;
231 	ctrlr->num_io_queues = 1;
232 	ctrlr->num_cpus_per_ioq = mp_ncpus;
233 	ctrlr->rid = 0;
234 	ctrlr->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ,
235 	    &ctrlr->rid, RF_SHAREABLE | RF_ACTIVE);
236 
237 	if (ctrlr->res == NULL) {
238 		nvme_printf(ctrlr, "unable to allocate shared IRQ\n");
239 		return (ENOMEM);
240 	}
241 
242 	bus_setup_intr(ctrlr->dev, ctrlr->res,
243 	    INTR_TYPE_MISC | INTR_MPSAFE, NULL, nvme_ctrlr_intx_handler,
244 	    ctrlr, &ctrlr->tag);
245 
246 	if (ctrlr->tag == NULL) {
247 		nvme_printf(ctrlr, "unable to setup intx handler\n");
248 		return (ENOMEM);
249 	}
250 
251 	return (0);
252 }
253 
254 static void
255 nvme_ctrlr_setup_interrupts(struct nvme_controller *ctrlr)
256 {
257 	device_t	dev;
258 	int		per_cpu_io_queues;
259 	int		min_cpus_per_ioq;
260 	int		num_vectors_requested, num_vectors_allocated;
261 	int		num_vectors_available;
262 
263 	dev = ctrlr->dev;
264 	min_cpus_per_ioq = 1;
265 	TUNABLE_INT_FETCH("hw.nvme.min_cpus_per_ioq", &min_cpus_per_ioq);
266 
267 	if (min_cpus_per_ioq < 1) {
268 		min_cpus_per_ioq = 1;
269 	} else if (min_cpus_per_ioq > mp_ncpus) {
270 		min_cpus_per_ioq = mp_ncpus;
271 	}
272 
273 	per_cpu_io_queues = 1;
274 	TUNABLE_INT_FETCH("hw.nvme.per_cpu_io_queues", &per_cpu_io_queues);
275 
276 	if (per_cpu_io_queues == 0) {
277 		min_cpus_per_ioq = mp_ncpus;
278 	}
279 
280 	ctrlr->force_intx = 0;
281 	TUNABLE_INT_FETCH("hw.nvme.force_intx", &ctrlr->force_intx);
282 
283 	/*
284 	 * FreeBSD currently cannot allocate more than about 190 vectors at
285 	 *  boot, meaning that systems with high core count and many devices
286 	 *  requesting per-CPU interrupt vectors will not get their full
287 	 *  allotment.  So first, try to allocate as many as we may need to
288 	 *  understand what is available, then immediately release them.
289 	 *  Then figure out how many of those we will actually use, based on
290 	 *  assigning an equal number of cores to each I/O queue.
291 	 */
292 
293 	/* One vector for per core I/O queue, plus one vector for admin queue. */
294 	num_vectors_available = min(pci_msix_count(dev), mp_ncpus + 1);
295 	if (pci_alloc_msix(dev, &num_vectors_available) != 0) {
296 		num_vectors_available = 0;
297 	}
298 	pci_release_msi(dev);
299 
300 	if (ctrlr->force_intx || num_vectors_available < 2) {
301 		nvme_ctrlr_configure_intx(ctrlr);
302 		return;
303 	}
304 
305 	/*
306 	 * Do not use all vectors for I/O queues - one must be saved for the
307 	 *  admin queue.
308 	 */
309 	ctrlr->num_cpus_per_ioq = max(min_cpus_per_ioq,
310 	    howmany(mp_ncpus, num_vectors_available - 1));
311 
312 	ctrlr->num_io_queues = howmany(mp_ncpus, ctrlr->num_cpus_per_ioq);
313 	num_vectors_requested = ctrlr->num_io_queues + 1;
314 	num_vectors_allocated = num_vectors_requested;
315 
316 	/*
317 	 * Now just allocate the number of vectors we need.  This should
318 	 *  succeed, since we previously called pci_alloc_msix()
319 	 *  successfully returning at least this many vectors, but just to
320 	 *  be safe, if something goes wrong just revert to INTx.
321 	 */
322 	if (pci_alloc_msix(dev, &num_vectors_allocated) != 0) {
323 		nvme_ctrlr_configure_intx(ctrlr);
324 		return;
325 	}
326 
327 	if (num_vectors_allocated < num_vectors_requested) {
328 		pci_release_msi(dev);
329 		nvme_ctrlr_configure_intx(ctrlr);
330 		return;
331 	}
332 
333 	ctrlr->msix_enabled = 1;
334 }
335