xref: /freebsd/sys/dev/nvme/nvme.c (revision e2b9900498a0c21109db8c7873978c7b5a8e9864)
1 /*-
2  * Copyright (C) 2012 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/bus.h>
32 #include <sys/conf.h>
33 #include <sys/module.h>
34 
35 #include <vm/uma.h>
36 
37 #include <dev/pci/pcireg.h>
38 #include <dev/pci/pcivar.h>
39 
40 #include "nvme_private.h"
41 
42 struct nvme_consumer {
43 	uint32_t		id;
44 	nvme_cons_ns_fn_t	ns_fn;
45 	nvme_cons_ctrlr_fn_t	ctrlr_fn;
46 	nvme_cons_async_fn_t	async_fn;
47 	nvme_cons_fail_fn_t	fail_fn;
48 };
49 
50 struct nvme_consumer nvme_consumer[NVME_MAX_CONSUMERS];
51 #define	INVALID_CONSUMER_ID	0xFFFF
52 
53 uma_zone_t	nvme_request_zone;
54 int32_t		nvme_retry_count;
55 
56 MALLOC_DEFINE(M_NVME, "nvme", "nvme(4) memory allocations");
57 
58 static int    nvme_probe(device_t);
59 static int    nvme_attach(device_t);
60 static int    nvme_detach(device_t);
61 static int    nvme_modevent(module_t mod, int type, void *arg);
62 
63 static devclass_t nvme_devclass;
64 
65 static device_method_t nvme_pci_methods[] = {
66 	/* Device interface */
67 	DEVMETHOD(device_probe,     nvme_probe),
68 	DEVMETHOD(device_attach,    nvme_attach),
69 	DEVMETHOD(device_detach,    nvme_detach),
70 	{ 0, 0 }
71 };
72 
73 static driver_t nvme_pci_driver = {
74 	"nvme",
75 	nvme_pci_methods,
76 	sizeof(struct nvme_controller),
77 };
78 
79 DRIVER_MODULE(nvme, pci, nvme_pci_driver, nvme_devclass, nvme_modevent, 0);
80 MODULE_VERSION(nvme, 1);
81 
82 static struct _pcsid
83 {
84 	u_int32_t   type;
85 	const char  *desc;
86 } pci_ids[] = {
87 	{ 0x01118086,		"NVMe Controller"  },
88 	{ CHATHAM_PCI_ID,	"Chatham Prototype NVMe Controller"  },
89 	{ IDT32_PCI_ID,		"IDT NVMe Controller (32 channel)"  },
90 	{ IDT8_PCI_ID,		"IDT NVMe Controller (8 channel)" },
91 	{ 0x00000000,		NULL  }
92 };
93 
94 static int
95 nvme_probe (device_t device)
96 {
97 	struct _pcsid	*ep;
98 	u_int32_t	type;
99 
100 	type = pci_get_devid(device);
101 	ep = pci_ids;
102 
103 	while (ep->type && ep->type != type)
104 		++ep;
105 
106 	if (ep->desc) {
107 		device_set_desc(device, ep->desc);
108 		return (BUS_PROBE_DEFAULT);
109 	}
110 
111 #if defined(PCIS_STORAGE_NVM)
112 	if (pci_get_class(device)    == PCIC_STORAGE &&
113 	    pci_get_subclass(device) == PCIS_STORAGE_NVM &&
114 	    pci_get_progif(device)   == PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0) {
115 		device_set_desc(device, "Generic NVMe Device");
116 		return (BUS_PROBE_GENERIC);
117 	}
118 #endif
119 
120 	return (ENXIO);
121 }
122 
123 static void
124 nvme_init(void)
125 {
126 	uint32_t	i;
127 
128 	nvme_request_zone = uma_zcreate("nvme_request",
129 	    sizeof(struct nvme_request), NULL, NULL, NULL, NULL, 0, 0);
130 
131 	for (i = 0; i < NVME_MAX_CONSUMERS; i++)
132 		nvme_consumer[i].id = INVALID_CONSUMER_ID;
133 }
134 
135 SYSINIT(nvme_register, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_init, NULL);
136 
137 static void
138 nvme_uninit(void)
139 {
140 	uma_zdestroy(nvme_request_zone);
141 }
142 
143 SYSUNINIT(nvme_unregister, SI_SUB_DRIVERS, SI_ORDER_SECOND, nvme_uninit, NULL);
144 
145 static void
146 nvme_load(void)
147 {
148 }
149 
150 static void
151 nvme_unload(void)
152 {
153 }
154 
155 static void
156 nvme_shutdown(void)
157 {
158 	device_t		*devlist;
159 	struct nvme_controller	*ctrlr;
160 	union cc_register	cc;
161 	union csts_register	csts;
162 	int			dev, devcount;
163 
164 	if (devclass_get_devices(nvme_devclass, &devlist, &devcount))
165 		return;
166 
167 	for (dev = 0; dev < devcount; dev++) {
168 		/*
169 		 * Only notify controller of shutdown when a real shutdown is
170 		 *  in process, not when a module unload occurs.  It seems at
171 		 *  least some controllers (Chatham at least) don't let you
172 		 *  re-enable the controller after shutdown notification has
173 		 *  been received.
174 		 */
175 		ctrlr = DEVICE2SOFTC(devlist[dev]);
176 		cc.raw = nvme_mmio_read_4(ctrlr, cc);
177 		cc.bits.shn = NVME_SHN_NORMAL;
178 		nvme_mmio_write_4(ctrlr, cc, cc.raw);
179 		csts.raw = nvme_mmio_read_4(ctrlr, csts);
180 		while (csts.bits.shst != NVME_SHST_COMPLETE) {
181 			DELAY(5);
182 			csts.raw = nvme_mmio_read_4(ctrlr, csts);
183 		}
184 	}
185 
186 	free(devlist, M_TEMP);
187 }
188 
189 static int
190 nvme_modevent(module_t mod, int type, void *arg)
191 {
192 
193 	switch (type) {
194 	case MOD_LOAD:
195 		nvme_load();
196 		break;
197 	case MOD_UNLOAD:
198 		nvme_unload();
199 		break;
200 	case MOD_SHUTDOWN:
201 		nvme_shutdown();
202 		break;
203 	default:
204 		break;
205 	}
206 
207 	return (0);
208 }
209 
210 void
211 nvme_dump_command(struct nvme_command *cmd)
212 {
213 	printf(
214 "opc:%x f:%x r1:%x cid:%x nsid:%x r2:%x r3:%x mptr:%jx prp1:%jx prp2:%jx cdw:%x %x %x %x %x %x\n",
215 	    cmd->opc, cmd->fuse, cmd->rsvd1, cmd->cid, cmd->nsid,
216 	    cmd->rsvd2, cmd->rsvd3,
217 	    (uintmax_t)cmd->mptr, (uintmax_t)cmd->prp1, (uintmax_t)cmd->prp2,
218 	    cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14,
219 	    cmd->cdw15);
220 }
221 
222 void
223 nvme_dump_completion(struct nvme_completion *cpl)
224 {
225 	printf("cdw0:%08x sqhd:%04x sqid:%04x "
226 	    "cid:%04x p:%x sc:%02x sct:%x m:%x dnr:%x\n",
227 	    cpl->cdw0, cpl->sqhd, cpl->sqid,
228 	    cpl->cid, cpl->status.p, cpl->status.sc, cpl->status.sct,
229 	    cpl->status.m, cpl->status.dnr);
230 }
231 
232 void
233 nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
234 {
235 	struct nvme_tracker 	*tr = arg;
236 	uint32_t		cur_nseg;
237 
238 	/*
239 	 * If the mapping operation failed, return immediately.  The caller
240 	 *  is responsible for detecting the error status and failing the
241 	 *  tracker manually.
242 	 */
243 	if (error != 0)
244 		return;
245 
246 	/*
247 	 * Note that we specified PAGE_SIZE for alignment and max
248 	 *  segment size when creating the bus dma tags.  So here
249 	 *  we can safely just transfer each segment to its
250 	 *  associated PRP entry.
251 	 */
252 	tr->req->cmd.prp1 = seg[0].ds_addr;
253 
254 	if (nseg == 2) {
255 		tr->req->cmd.prp2 = seg[1].ds_addr;
256 	} else if (nseg > 2) {
257 		cur_nseg = 1;
258 		tr->req->cmd.prp2 = (uint64_t)tr->prp_bus_addr;
259 		while (cur_nseg < nseg) {
260 			tr->prp[cur_nseg-1] =
261 			    (uint64_t)seg[cur_nseg].ds_addr;
262 			cur_nseg++;
263 		}
264 	}
265 
266 	nvme_qpair_submit_tracker(tr->qpair, tr);
267 }
268 
269 static int
270 nvme_attach(device_t dev)
271 {
272 	struct nvme_controller	*ctrlr = DEVICE2SOFTC(dev);
273 	int			status;
274 
275 	status = nvme_ctrlr_construct(ctrlr, dev);
276 
277 	if (status != 0)
278 		return (status);
279 
280 	/*
281 	 * Reset controller twice to ensure we do a transition from cc.en==1
282 	 *  to cc.en==0.  This is because we don't really know what status
283 	 *  the controller was left in when boot handed off to OS.
284 	 */
285 	status = nvme_ctrlr_hw_reset(ctrlr);
286 	if (status != 0)
287 		return (status);
288 
289 	status = nvme_ctrlr_hw_reset(ctrlr);
290 	if (status != 0)
291 		return (status);
292 
293 	nvme_sysctl_initialize_ctrlr(ctrlr);
294 
295 	ctrlr->config_hook.ich_func = nvme_ctrlr_start_config_hook;
296 	ctrlr->config_hook.ich_arg = ctrlr;
297 
298 	config_intrhook_establish(&ctrlr->config_hook);
299 
300 	return (0);
301 }
302 
303 static int
304 nvme_detach (device_t dev)
305 {
306 	struct nvme_controller	*ctrlr = DEVICE2SOFTC(dev);
307 
308 	nvme_ctrlr_destruct(ctrlr, dev);
309 	return (0);
310 }
311 
312 static void
313 nvme_notify_consumer(struct nvme_consumer *cons)
314 {
315 	device_t		*devlist;
316 	struct nvme_controller	*ctrlr;
317 	struct nvme_namespace	*ns;
318 	void			*ctrlr_cookie;
319 	int			dev_idx, ns_idx, devcount;
320 
321 	if (devclass_get_devices(nvme_devclass, &devlist, &devcount))
322 		return;
323 
324 	for (dev_idx = 0; dev_idx < devcount; dev_idx++) {
325 		ctrlr = DEVICE2SOFTC(devlist[dev_idx]);
326 		if (cons->ctrlr_fn != NULL)
327 			ctrlr_cookie = (*cons->ctrlr_fn)(ctrlr);
328 		else
329 			ctrlr_cookie = NULL;
330 		ctrlr->cons_cookie[cons->id] = ctrlr_cookie;
331 		for (ns_idx = 0; ns_idx < ctrlr->cdata.nn; ns_idx++) {
332 			ns = &ctrlr->ns[ns_idx];
333 			if (cons->ns_fn != NULL)
334 				ns->cons_cookie[cons->id] =
335 				    (*cons->ns_fn)(ns, ctrlr_cookie);
336 		}
337 	}
338 
339 	free(devlist, M_TEMP);
340 }
341 
342 void
343 nvme_notify_async_consumers(struct nvme_controller *ctrlr,
344 			    const struct nvme_completion *async_cpl,
345 			    uint32_t log_page_id, void *log_page_buffer,
346 			    uint32_t log_page_size)
347 {
348 	struct nvme_consumer	*cons;
349 	uint32_t		i;
350 
351 	for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
352 		cons = &nvme_consumer[i];
353 		if (cons->id != INVALID_CONSUMER_ID && cons->async_fn != NULL)
354 			(*cons->async_fn)(ctrlr->cons_cookie[i], async_cpl,
355 			    log_page_id, log_page_buffer, log_page_size);
356 	}
357 }
358 
359 void
360 nvme_notify_fail_consumers(struct nvme_controller *ctrlr)
361 {
362 	struct nvme_consumer	*cons;
363 	uint32_t		i;
364 
365 	for (i = 0; i < NVME_MAX_CONSUMERS; i++) {
366 		cons = &nvme_consumer[i];
367 		if (cons->id != INVALID_CONSUMER_ID && cons->fail_fn != NULL)
368 			cons->fail_fn(ctrlr->cons_cookie[i]);
369 	}
370 }
371 
372 struct nvme_consumer *
373 nvme_register_consumer(nvme_cons_ns_fn_t ns_fn, nvme_cons_ctrlr_fn_t ctrlr_fn,
374 		       nvme_cons_async_fn_t async_fn,
375 		       nvme_cons_fail_fn_t fail_fn)
376 {
377 	int i;
378 
379 	/*
380 	 * TODO: add locking around consumer registration.  Not an issue
381 	 *  right now since we only have one nvme consumer - nvd(4).
382 	 */
383 	for (i = 0; i < NVME_MAX_CONSUMERS; i++)
384 		if (nvme_consumer[i].id == INVALID_CONSUMER_ID) {
385 			nvme_consumer[i].id = i;
386 			nvme_consumer[i].ns_fn = ns_fn;
387 			nvme_consumer[i].ctrlr_fn = ctrlr_fn;
388 			nvme_consumer[i].async_fn = async_fn;
389 			nvme_consumer[i].fail_fn = fail_fn;
390 
391 			nvme_notify_consumer(&nvme_consumer[i]);
392 			return (&nvme_consumer[i]);
393 		}
394 
395 	printf("nvme(4): consumer not registered - no slots available\n");
396 	return (NULL);
397 }
398 
399 void
400 nvme_unregister_consumer(struct nvme_consumer *consumer)
401 {
402 
403 	consumer->id = INVALID_CONSUMER_ID;
404 }
405 
406 void
407 nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl)
408 {
409 	struct nvme_completion_poll_status	*status = arg;
410 
411 	/*
412 	 * Copy status into the argument passed by the caller, so that
413 	 *  the caller can check the status to determine if the
414 	 *  the request passed or failed.
415 	 */
416 	memcpy(&status->cpl, cpl, sizeof(*cpl));
417 	wmb();
418 	status->done = TRUE;
419 }
420 
421