]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/dev/nvme/nvme_sim.c
MFV r323110: 8558 lwp_create() returns EAGAIN on system with more than 80K ZFS filesy...
[FreeBSD/FreeBSD.git] / sys / dev / nvme / nvme_sim.c
1 /*-
2  * Copyright (c) 2016 Netflix, Inc
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  *    without modification, immediately at the beginning of the file.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 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/ioccom.h>
36 #include <sys/malloc.h>
37 #include <sys/proc.h>
38 #include <sys/smp.h>
39
40 #include <cam/cam.h>
41 #include <cam/cam_ccb.h>
42 #include <cam/cam_sim.h>
43 #include <cam/cam_xpt_sim.h>
44 #include <cam/cam_xpt_internal.h>       // Yes, this is wrong.
45 #include <cam/cam_debug.h>
46
47 #include "nvme_private.h"
48
49 #define ccb_accb_ptr spriv_ptr0
50 #define ccb_ctrlr_ptr spriv_ptr1
51 static void     nvme_sim_action(struct cam_sim *sim, union ccb *ccb);
52 static void     nvme_sim_poll(struct cam_sim *sim);
53
54 #define sim2softc(sim)  ((struct nvme_sim_softc *)cam_sim_softc(sim))
55 #define sim2ns(sim)     (sim2softc(sim)->s_ns)
56 #define sim2ctrlr(sim)  (sim2softc(sim)->s_ctrlr)
57
58 struct nvme_sim_softc
59 {
60         struct nvme_controller  *s_ctrlr;
61         struct nvme_namespace   *s_ns;
62         struct cam_sim          *s_sim;
63         struct cam_path         *s_path;
64 };
65
66 static void
67 nvme_sim_nvmeio_done(void *ccb_arg, const struct nvme_completion *cpl)
68 {
69         union ccb *ccb = (union ccb *)ccb_arg;
70
71         /*
72          * Let the periph know the completion, and let it sort out what
73          * it means. Make our best guess, though for the status code.
74          */
75         memcpy(&ccb->nvmeio.cpl, cpl, sizeof(*cpl));
76         if (nvme_completion_is_error(cpl))
77                 ccb->ccb_h.status = CAM_REQ_CMP_ERR;
78         else
79                 ccb->ccb_h.status = CAM_REQ_CMP;
80         xpt_done(ccb);
81 }
82
83 static void
84 nvme_sim_nvmeio(struct cam_sim *sim, union ccb *ccb)
85 {
86         struct ccb_nvmeio       *nvmeio = &ccb->nvmeio;
87         struct nvme_request     *req;
88         void                    *payload;
89         uint32_t                size;
90         struct nvme_controller *ctrlr;
91
92         ctrlr = sim2ctrlr(sim);
93         payload = nvmeio->data_ptr;
94         size = nvmeio->dxfer_len;
95         /* SG LIST ??? */
96         if ((nvmeio->ccb_h.flags & CAM_DATA_MASK) == CAM_DATA_BIO)
97                 req = nvme_allocate_request_bio((struct bio *)payload,
98                     nvme_sim_nvmeio_done, ccb);
99         else if ((nvmeio->ccb_h.flags & CAM_DATA_SG) == CAM_DATA_SG)
100                 req = nvme_allocate_request_ccb(ccb, nvme_sim_nvmeio_done, ccb);
101         else if (payload == NULL)
102                 req = nvme_allocate_request_null(nvme_sim_nvmeio_done, ccb);
103         else
104                 req = nvme_allocate_request_vaddr(payload, size,
105                     nvme_sim_nvmeio_done, ccb);
106
107         if (req == NULL) {
108                 nvmeio->ccb_h.status = CAM_RESRC_UNAVAIL;
109                 xpt_done(ccb);
110                 return;
111         }
112
113         memcpy(&req->cmd, &ccb->nvmeio.cmd, sizeof(ccb->nvmeio.cmd));
114
115         if (ccb->ccb_h.func_code == XPT_NVME_IO)
116                 nvme_ctrlr_submit_io_request(ctrlr, req);
117         else
118                 nvme_ctrlr_submit_admin_request(ctrlr, req);
119
120         ccb->ccb_h.status |= CAM_SIM_QUEUED;
121 }
122
123 static void
124 nvme_sim_action(struct cam_sim *sim, union ccb *ccb)
125 {
126         struct nvme_controller *ctrlr;
127         struct nvme_namespace *ns;
128
129         CAM_DEBUG(ccb->ccb_h.path, CAM_DEBUG_TRACE,
130             ("nvme_sim_action: func= %#x\n",
131                 ccb->ccb_h.func_code));
132
133         /*
134          * XXX when we support multiple namespaces in the base driver we'll need
135          * to revisit how all this gets stored and saved in the periph driver's
136          * reserved areas. Right now we store all three in the softc of the sim.
137          */
138         ns = sim2ns(sim);
139         ctrlr = sim2ctrlr(sim);
140
141         mtx_assert(&ctrlr->lock, MA_OWNED);
142
143         switch (ccb->ccb_h.func_code) {
144         case XPT_CALC_GEOMETRY:         /* Calculate Geometry Totally nuts ? XXX */
145                 /* 
146                  * Only meaningful for old-school SCSI disks since only the SCSI
147                  * da driver generates them. Reject all these that slip through.
148                  */
149                 /*FALLTHROUGH*/
150         case XPT_ABORT:                 /* Abort the specified CCB */
151                 ccb->ccb_h.status = CAM_REQ_INVALID;
152                 break;
153         case XPT_SET_TRAN_SETTINGS:
154                 /*
155                  * NVMe doesn't really have different transfer settings, but
156                  * other parts of CAM think failure here is a big deal.
157                  */
158                 ccb->ccb_h.status = CAM_REQ_CMP;
159                 break;
160         case XPT_PATH_INQ:              /* Path routing inquiry */
161         {
162                 struct ccb_pathinq *cpi = &ccb->cpi;
163
164                 /*
165                  * NVMe may have multiple LUNs on the same path. Current generation
166                  * of NVMe devives support only a single name space. Multiple name
167                  * space drives are coming, but it's unclear how we should report
168                  * them up the stack.
169                  */
170                 cpi->version_num = 1;
171                 cpi->hba_inquiry = 0;
172                 cpi->target_sprt = 0;
173                 cpi->hba_misc =  PIM_UNMAPPED /* | PIM_NOSCAN */;
174                 cpi->hba_eng_cnt = 0;
175                 cpi->max_target = 0;
176                 cpi->max_lun = ctrlr->cdata.nn;
177                 cpi->maxio = nvme_ns_get_max_io_xfer_size(ns);
178                 cpi->initiator_id = 0;
179                 cpi->bus_id = cam_sim_bus(sim);
180                 cpi->base_transfer_speed = 4000000;     /* 4 GB/s 4 lanes pcie 3 */
181                 strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
182                 strlcpy(cpi->hba_vid, "NVMe", HBA_IDLEN);
183                 strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
184                 cpi->unit_number = cam_sim_unit(sim);
185                 cpi->transport = XPORT_NVME;            /* XXX XPORT_PCIE ? */
186                 cpi->transport_version = 1;             /* XXX Get PCIe spec ? */
187                 cpi->protocol = PROTO_NVME;
188                 cpi->protocol_version = NVME_REV_1;     /* Groks all 1.x NVMe cards */
189                 cpi->xport_specific.nvme.nsid = ns->id;
190                 cpi->ccb_h.status = CAM_REQ_CMP;
191                 break;
192         }
193         case XPT_GET_TRAN_SETTINGS:     /* Get transport settings */
194         {
195                 struct ccb_trans_settings       *cts;
196                 struct ccb_trans_settings_nvme  *nvmep;
197                 struct ccb_trans_settings_nvme  *nvmex;
198
199                 cts = &ccb->cts;
200                 nvmex = &cts->xport_specific.nvme;
201                 nvmep = &cts->proto_specific.nvme;
202
203                 nvmex->valid = CTS_NVME_VALID_SPEC;
204                 nvmex->spec_major = 1;                  /* XXX read from card */
205                 nvmex->spec_minor = 2;
206                 nvmex->spec_tiny = 0;
207
208                 nvmep->valid = CTS_NVME_VALID_SPEC;
209                 nvmep->spec_major = 1;                  /* XXX read from card */
210                 nvmep->spec_minor = 2;
211                 nvmep->spec_tiny = 0;
212                 cts->transport = XPORT_NVME;
213                 cts->protocol = PROTO_NVME;
214                 cts->ccb_h.status = CAM_REQ_CMP;
215                 break;
216         }
217         case XPT_TERM_IO:               /* Terminate the I/O process */
218                 /*
219                  * every driver handles this, but nothing generates it. Assume
220                  * it's OK to just say 'that worked'.
221                  */
222                 /*FALLTHROUGH*/
223         case XPT_RESET_DEV:             /* Bus Device Reset the specified device */
224         case XPT_RESET_BUS:             /* Reset the specified bus */
225                 /*
226                  * NVMe doesn't really support physically resetting the bus. It's part
227                  * of the bus scanning dance, so return sucess to tell the process to
228                  * proceed.
229                  */
230                 ccb->ccb_h.status = CAM_REQ_CMP;
231                 break;
232         case XPT_NVME_IO:               /* Execute the requested I/O operation */
233         case XPT_NVME_ADMIN:            /* or Admin operation */
234                 nvme_sim_nvmeio(sim, ccb);
235                 return;                 /* no done */
236         default:
237                 ccb->ccb_h.status = CAM_REQ_INVALID;
238                 break;
239         }
240         xpt_done(ccb);
241 }
242
243 static void
244 nvme_sim_poll(struct cam_sim *sim)
245 {
246
247         nvme_ctrlr_intx_handler(sim2ctrlr(sim));
248 }
249
250 static void *
251 nvme_sim_new_controller(struct nvme_controller *ctrlr)
252 {
253         struct cam_devq *devq;
254         int max_trans;
255         int unit;
256         struct nvme_sim_softc *sc = NULL;
257
258         max_trans = ctrlr->max_hw_pend_io;
259         unit = device_get_unit(ctrlr->dev);
260         devq = cam_simq_alloc(max_trans);
261         if (devq == NULL)
262                 return NULL;
263
264         sc = malloc(sizeof(*sc), M_NVME, M_ZERO | M_WAITOK);
265
266         sc->s_ctrlr = ctrlr;
267
268         sc->s_sim = cam_sim_alloc(nvme_sim_action, nvme_sim_poll,
269             "nvme", sc, unit, &ctrlr->lock, max_trans, max_trans, devq);
270         if (sc->s_sim == NULL) {
271                 printf("Failed to allocate a sim\n");
272                 cam_simq_free(devq);
273                 free(sc, M_NVME);
274                 return NULL;
275         }
276
277         return sc;
278 }
279
280 static void
281 nvme_sim_rescan_target(struct nvme_controller *ctrlr, struct cam_path *path)
282 {
283         union ccb *ccb;
284
285         ccb = xpt_alloc_ccb_nowait();
286         if (ccb == NULL) {
287                 printf("unable to alloc CCB for rescan\n");
288                 return;
289         }
290
291         if (xpt_clone_path(&ccb->ccb_h.path, path) != CAM_REQ_CMP) {
292                 printf("unable to copy path for rescan\n");
293                 xpt_free_ccb(ccb);
294                 return;
295         }
296
297         xpt_rescan(ccb);
298 }
299         
300 static void *
301 nvme_sim_new_ns(struct nvme_namespace *ns, void *sc_arg)
302 {
303         struct nvme_sim_softc *sc = sc_arg;
304         struct nvme_controller *ctrlr = sc->s_ctrlr;
305         int i;
306
307         sc->s_ns = ns;
308
309         /*
310          * XXX this is creating one bus per ns, but it should be one
311          * XXX target per controller, and one LUN per namespace.
312          * XXX Current drives only support one NS, so there's time
313          * XXX to fix it later when new drives arrive.
314          *
315          * XXX I'm pretty sure the xpt_bus_register() call below is
316          * XXX like super lame and it really belongs in the sim_new_ctrlr
317          * XXX callback. Then the create_path below would be pretty close
318          * XXX to being right. Except we should be per-ns not per-ctrlr
319          * XXX data.
320          */
321
322         mtx_lock(&ctrlr->lock);
323 /* Create bus */
324
325         /*
326          * XXX do I need to lock ctrlr->lock ? 
327          * XXX do I need to lock the path?
328          * ata and scsi seem to in their code, but their discovery is
329          * somewhat more asynchronous. We're only every called one at a
330          * time, and nothing is in parallel.
331          */
332
333         i = 0;
334         if (xpt_bus_register(sc->s_sim, ctrlr->dev, 0) != CAM_SUCCESS)
335                 goto error;
336         i++;
337         if (xpt_create_path(&sc->s_path, /*periph*/NULL, cam_sim_path(sc->s_sim),
338             1, ns->id) != CAM_REQ_CMP)
339                 goto error;
340         i++;
341
342         sc->s_path->device->nvme_data = nvme_ns_get_data(ns);
343         sc->s_path->device->nvme_cdata = nvme_ctrlr_get_data(ns->ctrlr);
344
345 /* Scan bus */
346         nvme_sim_rescan_target(ctrlr, sc->s_path);
347
348         mtx_unlock(&ctrlr->lock);
349
350         return ns;
351
352 error:
353         switch (i) {
354         case 2:
355                 xpt_free_path(sc->s_path);
356         case 1:
357                 xpt_bus_deregister(cam_sim_path(sc->s_sim));
358         case 0:
359                 cam_sim_free(sc->s_sim, /*free_devq*/TRUE);
360         }
361         mtx_unlock(&ctrlr->lock);
362         return NULL;
363 }
364
365 static void
366 nvme_sim_controller_fail(void *ctrlr_arg)
367 {
368         /* XXX cleanup XXX */
369 }
370
371 struct nvme_consumer *consumer_cookie;
372
373 static void
374 nvme_sim_init(void)
375 {
376         if (nvme_use_nvd)
377                 return;
378
379         consumer_cookie = nvme_register_consumer(nvme_sim_new_ns,
380             nvme_sim_new_controller, NULL, nvme_sim_controller_fail);
381 }
382
383 SYSINIT(nvme_sim_register, SI_SUB_DRIVERS, SI_ORDER_ANY,
384     nvme_sim_init, NULL);
385
386 static void
387 nvme_sim_uninit(void)
388 {
389         if (nvme_use_nvd)
390                 return;
391         /* XXX Cleanup */
392
393         nvme_unregister_consumer(consumer_cookie);
394 }
395
396 SYSUNINIT(nvme_sim_unregister, SI_SUB_DRIVERS, SI_ORDER_ANY,
397     nvme_sim_uninit, NULL);