xref: /freebsd/sys/dev/nvme/nvme.c (revision e9b1dc32c9bd2ebae5f9e140bfa0e0321bc366b5)
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 	{ 0x00000000,		0, 0, NULL  }
110 };
111 
112 static int
113 nvme_match(uint32_t devid, uint16_t subdevice, struct _pcsid *ep)
114 {
115 	if (devid != ep->devid)
116 		return 0;
117 
118 	if (!ep->match_subdevice)
119 		return 1;
120 
121 	if (subdevice == ep->subdevice)
122 		return 1;
123 	else
124 		return 0;
125 }
126 
127 static int
128 nvme_probe (device_t device)
129 {
130 	struct _pcsid	*ep;
131 	uint32_t	devid;
132 	uint16_t	subdevice;
133 
134 	devid = pci_get_devid(device);
135 	subdevice = pci_get_subdevice(device);
136 	ep = pci_ids;
137 
138 	while (ep->devid) {
139 		if (nvme_match(devid, subdevice, ep))
140 			break;
141 		++ep;
142 	}
143 
144 	if (ep->desc) {
145 		device_set_desc(device, ep->desc);
146 		return (BUS_PROBE_DEFAULT);
147 	}
148 
149 #if defined(PCIS_STORAGE_NVM)
150 	if (pci_get_class(device)    == PCIC_STORAGE &&
151 	    pci_get_subclass(device) == PCIS_STORAGE_NVM &&
152 	    pci_get_progif(device)   == PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) {
153 		device_set_desc(device, "Generic NVMe Device");
154 		return (BUS_PROBE_GENERIC);
155 	}
156 #endif
157 
158 	return (ENXIO);
159 }
160 
161 static void
162 nvme_init(void)
163 {
164 	uint32_t	i;
165 
166 	nvme_request_zone = uma_zcreate("nvme_request",
167 	    sizeof(struct nvme_request), NULL, NULL, NULL, NULL, 0, 0);
168 
169 	for (i = 0; i < NVME_MAX_CONSUMERS; i++)
170 		nvme_consumer[i].id = INVALID_CONSUMER_ID;
171 }
172 
173 SYSINIT(nvme_register, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_init, NULL);
174 
175 static void
176 nvme_uninit(void)
177 {
178 	uma_zdestroy(nvme_request_zone);
179 }
180 
181 SYSUNINIT(nvme_unregister, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_uninit, NULL);
182 
183 static void
184 nvme_load(void)
185 {
186 }
187 
188 static void
189 nvme_unload(void)
190 {
191 }
192 
193 static int
194 nvme_shutdown(device_t dev)
195 {
196 	struct nvme_controller	*ctrlr;
197 
198 	ctrlr = DEVICE2SOFTC(dev);
199 	nvme_ctrlr_shutdown(ctrlr);
200 
201 	return (0);
202 }
203 
204 static int
205 nvme_modevent(module_t mod, int type, void *arg)
206 {
207 
208 	switch (type) {
209 	case MOD_LOAD:
210 		nvme_load();
211 		break;
212 	case MOD_UNLOAD:
213 		nvme_unload();
214 		break;
215 	default:
216 		break;
217 	}
218 
219 	return (0);
220 }
221 
222 void
223 nvme_dump_command(struct nvme_command *cmd)
224 {
225 
226 	printf(
227 "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",
228 	    cmd->opc, cmd->fuse, cmd->cid, le32toh(cmd->nsid),
229 	    cmd->rsvd2, cmd->rsvd3,
230 	    (uintmax_t)le64toh(cmd->mptr), (uintmax_t)le64toh(cmd->prp1), (uintmax_t)le64toh(cmd->prp2),
231 	    le32toh(cmd->cdw10), le32toh(cmd->cdw11), le32toh(cmd->cdw12),
232 	    le32toh(cmd->cdw13), le32toh(cmd->cdw14), le32toh(cmd->cdw15));
233 }
234 
235 void
236 nvme_dump_completion(struct nvme_completion *cpl)
237 {
238 	uint8_t p, sc, sct, m, dnr;
239 	uint16_t status;
240 
241 	status = le16toh(cpl->status);
242 
243 	p = NVME_STATUS_GET_P(status);
244 	sc = NVME_STATUS_GET_SC(status);
245 	sct = NVME_STATUS_GET_SCT(status);
246 	m = NVME_STATUS_GET_M(status);
247 	dnr = NVME_STATUS_GET_DNR(status);
248 
249 	printf("cdw0:%08x sqhd:%04x sqid:%04x "
250 	    "cid:%04x p:%x sc:%02x sct:%x m:%x dnr:%x\n",
251 	    le32toh(cpl->cdw0), le16toh(cpl->sqhd), le16toh(cpl->sqid),
252 	    cpl->cid, p, sc, sct, m, dnr);
253 }
254 
255 static int
256 nvme_attach(device_t dev)
257 {
258 	struct nvme_controller	*ctrlr = DEVICE2SOFTC(dev);
259 	int			status;
260 	struct _pcsid		*ep;
261 	uint32_t		devid;
262 	uint16_t		subdevice;
263 
264 	devid = pci_get_devid(dev);
265 	subdevice = pci_get_subdevice(dev);
266 	ep = pci_ids;
267 	while (ep->devid) {
268 		if (nvme_match(devid, subdevice, ep))
269 			break;
270 		++ep;
271 	}
272 	ctrlr->quirks = ep->quirks;
273 
274 	status = nvme_ctrlr_construct(ctrlr, dev);
275 
276 	if (status != 0) {
277 		nvme_ctrlr_destruct(ctrlr, dev);
278 		return (status);
279 	}
280 
281 	/*
282 	 * Enable busmastering so the completion status messages can
283 	 * be busmastered back to the host.
284 	 */
285 	pci_enable_busmaster(dev);
286 
287 	/*
288 	 * Reset controller twice to ensure we do a transition from cc.en==1
289 	 *  to cc.en==0.  This is because we don't really know what status
290 	 *  the controller was left in when boot handed off to OS.
291 	 */
292 	status = nvme_ctrlr_hw_reset(ctrlr);
293 	if (status != 0) {
294 		nvme_ctrlr_destruct(ctrlr, dev);
295 		return (status);
296 	}
297 
298 	status = nvme_ctrlr_hw_reset(ctrlr);
299 	if (status != 0) {
300 		nvme_ctrlr_destruct(ctrlr, dev);
301 		return (status);
302 	}
303 
304 	ctrlr->config_hook.ich_func = nvme_ctrlr_start_config_hook;
305 	ctrlr->config_hook.ich_arg = ctrlr;
306 
307 	config_intrhook_establish(&ctrlr->config_hook);
308 
309 	return (0);
310 }
311 
312 static int
313 nvme_detach (device_t dev)
314 {
315 	struct nvme_controller	*ctrlr = DEVICE2SOFTC(dev);
316 
317 	nvme_ctrlr_destruct(ctrlr, dev);
318 	pci_disable_busmaster(dev);
319 	return (0);
320 }
321 
322 static void
323 nvme_notify(struct nvme_consumer *cons,
324 	    struct nvme_controller *ctrlr)
325 {
326 	struct nvme_namespace	*ns;
327 	void			*ctrlr_cookie;
328 	int			cmpset, ns_idx;
329 
330 	/*
331 	 * The consumer may register itself after the nvme devices
332 	 *  have registered with the kernel, but before the
333 	 *  driver has completed initialization.  In that case,
334 	 *  return here, and when initialization completes, the
335 	 *  controller will make sure the consumer gets notified.
336 	 */
337 	if (!ctrlr->is_initialized)
338 		return;
339 
340 	cmpset = atomic_cmpset_32(&ctrlr->notification_sent, 0, 1);
341 
342 	if (cmpset == 0)
343 		return;
344 
345 	if (cons->ctrlr_fn != NULL)
346 		ctrlr_cookie = (*cons->ctrlr_fn)(ctrlr);
347 	else
348 		ctrlr_cookie = NULL;
349 	ctrlr->cons_cookie[cons->id] = ctrlr_cookie;
350 	if (ctrlr->is_failed) {
351 		if (cons->fail_fn != NULL)
352 			(*cons->fail_fn)(ctrlr_cookie);
353 		/*
354 		 * Do not notify consumers about the namespaces of a
355 		 *  failed controller.
356 		 */
357 		return;
358 	}
359 	for (ns_idx = 0; ns_idx < min(ctrlr->cdata.nn, NVME_MAX_NAMESPACES); ns_idx++) {
360 		ns = &ctrlr->ns[ns_idx];
361 		if (ns->data.nsze == 0)
362 			continue;
363 		if (cons->ns_fn != NULL)
364 			ns->cons_cookie[cons->id] =
365 			    (*cons->ns_fn)(ns, ctrlr_cookie);
366 	}
367 }
368 
369 void
370 nvme_notify_new_controller(struct nvme_controller *ctrlr)
371 {
372 	int i;
373 
374 	for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
375 		if (nvme_consumer[i].id != INVALID_CONSUMER_ID) {
376 			nvme_notify(&nvme_consumer[i], ctrlr);
377 		}
378 	}
379 }
380 
381 static void
382 nvme_notify_new_consumer(struct nvme_consumer *cons)
383 {
384 	device_t		*devlist;
385 	struct nvme_controller	*ctrlr;
386 	int			dev_idx, devcount;
387 
388 	if (devclass_get_devices(nvme_devclass, &devlist, &devcount))
389 		return;
390 
391 	for (dev_idx = 0; dev_idx < devcount; dev_idx++) {
392 		ctrlr = DEVICE2SOFTC(devlist[dev_idx]);
393 		nvme_notify(cons, ctrlr);
394 	}
395 
396 	free(devlist, M_TEMP);
397 }
398 
399 void
400 nvme_notify_async_consumers(struct nvme_controller *ctrlr,
401 			    const struct nvme_completion *async_cpl,
402 			    uint32_t log_page_id, void *log_page_buffer,
403 			    uint32_t log_page_size)
404 {
405 	struct nvme_consumer	*cons;
406 	uint32_t		i;
407 
408 	for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
409 		cons = &nvme_consumer[i];
410 		if (cons->id != INVALID_CONSUMER_ID && cons->async_fn != NULL)
411 			(*cons->async_fn)(ctrlr->cons_cookie[i], async_cpl,
412 			    log_page_id, log_page_buffer, log_page_size);
413 	}
414 }
415 
416 void
417 nvme_notify_fail_consumers(struct nvme_controller *ctrlr)
418 {
419 	struct nvme_consumer	*cons;
420 	uint32_t		i;
421 
422 	/*
423 	 * This controller failed during initialization (i.e. IDENTIFY
424 	 *  command failed or timed out).  Do not notify any nvme
425 	 *  consumers of the failure here, since the consumer does not
426 	 *  even know about the controller yet.
427 	 */
428 	if (!ctrlr->is_initialized)
429 		return;
430 
431 	for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
432 		cons = &nvme_consumer[i];
433 		if (cons->id != INVALID_CONSUMER_ID && cons->fail_fn != NULL)
434 			cons->fail_fn(ctrlr->cons_cookie[i]);
435 	}
436 }
437 
438 void
439 nvme_notify_ns(struct nvme_controller *ctrlr, int nsid)
440 {
441 	struct nvme_consumer	*cons;
442 	struct nvme_namespace	*ns = &ctrlr->ns[nsid - 1];
443 	uint32_t		i;
444 
445 	if (!ctrlr->is_initialized)
446 		return;
447 
448 	for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
449 		cons = &nvme_consumer[i];
450 		if (cons->id != INVALID_CONSUMER_ID && cons->ns_fn != NULL)
451 			ns->cons_cookie[cons->id] =
452 			    (*cons->ns_fn)(ns, ctrlr->cons_cookie[cons->id]);
453 	}
454 }
455 
456 struct nvme_consumer *
457 nvme_register_consumer(nvme_cons_ns_fn_t ns_fn, nvme_cons_ctrlr_fn_t ctrlr_fn,
458 		       nvme_cons_async_fn_t async_fn,
459 		       nvme_cons_fail_fn_t fail_fn)
460 {
461 	int i;
462 
463 	/*
464 	 * TODO: add locking around consumer registration.  Not an issue
465 	 *  right now since we only have one nvme consumer - nvd(4).
466 	 */
467 	for (i = 0; i < NVME_MAX_CONSUMERS; i++)
468 		if (nvme_consumer[i].id == INVALID_CONSUMER_ID) {
469 			nvme_consumer[i].id = i;
470 			nvme_consumer[i].ns_fn = ns_fn;
471 			nvme_consumer[i].ctrlr_fn = ctrlr_fn;
472 			nvme_consumer[i].async_fn = async_fn;
473 			nvme_consumer[i].fail_fn = fail_fn;
474 
475 			nvme_notify_new_consumer(&nvme_consumer[i]);
476 			return (&nvme_consumer[i]);
477 		}
478 
479 	printf("nvme(4): consumer not registered - no slots available\n");
480 	return (NULL);
481 }
482 
483 void
484 nvme_unregister_consumer(struct nvme_consumer *consumer)
485 {
486 
487 	consumer->id = INVALID_CONSUMER_ID;
488 }
489 
490 void
491 nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl)
492 {
493 	struct nvme_completion_poll_status	*status = arg;
494 
495 	/*
496 	 * Copy status into the argument passed by the caller, so that
497 	 *  the caller can check the status to determine if the
498 	 *  the request passed or failed.
499 	 */
500 	memcpy(&status->cpl, cpl, sizeof(*cpl));
501 	atomic_store_rel_int(&status->done, 1);
502 }
503