xref: /freebsd/sys/dev/nvdimm/nvdimm_spa.c (revision 685dc743)
1 /*-
2  * Copyright (c) 2017, 2018 The FreeBSD Foundation
3  * All rights reserved.
4  * Copyright (c) 2018, 2019 Intel Corporation
5  *
6  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "opt_acpi.h"
33 #include "opt_ddb.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/bio.h>
38 #include <sys/bus.h>
39 #include <sys/conf.h>
40 #include <sys/devicestat.h>
41 #include <sys/disk.h>
42 #include <sys/efi.h>
43 #include <sys/kernel.h>
44 #include <sys/kthread.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/module.h>
49 #include <sys/rwlock.h>
50 #include <sys/sglist.h>
51 #include <sys/uio.h>
52 #include <sys/uuid.h>
53 #include <geom/geom.h>
54 #include <geom/geom_int.h>
55 #include <machine/vmparam.h>
56 #include <vm/vm.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
60 #include <contrib/dev/acpica/include/acpi.h>
61 #include <contrib/dev/acpica/include/accommon.h>
62 #include <contrib/dev/acpica/include/acuuid.h>
63 #include <dev/acpica/acpivar.h>
64 #include <dev/nvdimm/nvdimm_var.h>
65 
66 #define UUID_INITIALIZER_VOLATILE_MEMORY \
67     {0x7305944f,0xfdda,0x44e3,0xb1,0x6c,{0x3f,0x22,0xd2,0x52,0xe5,0xd0}}
68 #define UUID_INITIALIZER_PERSISTENT_MEMORY \
69     {0x66f0d379,0xb4f3,0x4074,0xac,0x43,{0x0d,0x33,0x18,0xb7,0x8c,0xdb}}
70 #define UUID_INITIALIZER_CONTROL_REGION \
71     {0x92f701f6,0x13b4,0x405d,0x91,0x0b,{0x29,0x93,0x67,0xe8,0x23,0x4c}}
72 #define UUID_INITIALIZER_DATA_REGION \
73     {0x91af0530,0x5d86,0x470e,0xa6,0xb0,{0x0a,0x2d,0xb9,0x40,0x82,0x49}}
74 #define UUID_INITIALIZER_VOLATILE_VIRTUAL_DISK \
75     {0x77ab535a,0x45fc,0x624b,0x55,0x60,{0xf7,0xb2,0x81,0xd1,0xf9,0x6e}}
76 #define UUID_INITIALIZER_VOLATILE_VIRTUAL_CD \
77     {0x3d5abd30,0x4175,0x87ce,0x6d,0x64,{0xd2,0xad,0xe5,0x23,0xc4,0xbb}}
78 #define UUID_INITIALIZER_PERSISTENT_VIRTUAL_DISK \
79     {0x5cea02c9,0x4d07,0x69d3,0x26,0x9f,{0x44,0x96,0xfb,0xe0,0x96,0xf9}}
80 #define UUID_INITIALIZER_PERSISTENT_VIRTUAL_CD \
81     {0x08018188,0x42cd,0xbb48,0x10,0x0f,{0x53,0x87,0xd5,0x3d,0xed,0x3d}}
82 
83 static struct nvdimm_SPA_uuid_list_elm {
84 	const char		*u_name;
85 	struct uuid		u_id;
86 	const bool		u_usr_acc;
87 } nvdimm_SPA_uuid_list[] = {
88 	[SPA_TYPE_VOLATILE_MEMORY] = {
89 		.u_name =	"VOLA MEM ",
90 		.u_id =		UUID_INITIALIZER_VOLATILE_MEMORY,
91 		.u_usr_acc =	true,
92 	},
93 	[SPA_TYPE_PERSISTENT_MEMORY] = {
94 		.u_name =	"PERS MEM",
95 		.u_id =		UUID_INITIALIZER_PERSISTENT_MEMORY,
96 		.u_usr_acc =	true,
97 	},
98 	[SPA_TYPE_CONTROL_REGION] = {
99 		.u_name =	"CTRL RG ",
100 		.u_id =		UUID_INITIALIZER_CONTROL_REGION,
101 		.u_usr_acc =	false,
102 	},
103 	[SPA_TYPE_DATA_REGION] = {
104 		.u_name =	"DATA RG ",
105 		.u_id =		UUID_INITIALIZER_DATA_REGION,
106 		.u_usr_acc =	true,
107 	},
108 	[SPA_TYPE_VOLATILE_VIRTUAL_DISK] = {
109 		.u_name =	"VIRT DSK",
110 		.u_id =		UUID_INITIALIZER_VOLATILE_VIRTUAL_DISK,
111 		.u_usr_acc =	true,
112 	},
113 	[SPA_TYPE_VOLATILE_VIRTUAL_CD] = {
114 		.u_name =	"VIRT CD ",
115 		.u_id =		UUID_INITIALIZER_VOLATILE_VIRTUAL_CD,
116 		.u_usr_acc =	true,
117 	},
118 	[SPA_TYPE_PERSISTENT_VIRTUAL_DISK] = {
119 		.u_name =	"PV DSK  ",
120 		.u_id =		UUID_INITIALIZER_PERSISTENT_VIRTUAL_DISK,
121 		.u_usr_acc =	true,
122 	},
123 	[SPA_TYPE_PERSISTENT_VIRTUAL_CD] = {
124 		.u_name =	"PV CD   ",
125 		.u_id =		UUID_INITIALIZER_PERSISTENT_VIRTUAL_CD,
126 		.u_usr_acc =	true,
127 	},
128 };
129 
130 enum SPA_mapping_type
nvdimm_spa_type_from_name(const char * name)131 nvdimm_spa_type_from_name(const char *name)
132 {
133 	int j;
134 
135 	for (j = 0; j < nitems(nvdimm_SPA_uuid_list); j++) {
136 		if (strcmp(name, nvdimm_SPA_uuid_list[j].u_name) != 0)
137 			continue;
138 		return (j);
139 	}
140 	return (SPA_TYPE_UNKNOWN);
141 }
142 
143 enum SPA_mapping_type
nvdimm_spa_type_from_uuid(struct uuid * uuid)144 nvdimm_spa_type_from_uuid(struct uuid *uuid)
145 {
146 	int j;
147 
148 	for (j = 0; j < nitems(nvdimm_SPA_uuid_list); j++) {
149 		if (uuidcmp(uuid, &nvdimm_SPA_uuid_list[j].u_id) != 0)
150 			continue;
151 		return (j);
152 	}
153 	return (SPA_TYPE_UNKNOWN);
154 }
155 
156 bool
nvdimm_spa_type_user_accessible(enum SPA_mapping_type spa_type)157 nvdimm_spa_type_user_accessible(enum SPA_mapping_type spa_type)
158 {
159 
160 	if ((int)spa_type < 0 || spa_type >= nitems(nvdimm_SPA_uuid_list))
161 		return (false);
162 	return (nvdimm_SPA_uuid_list[spa_type].u_usr_acc);
163 }
164 
165 static vm_memattr_t
nvdimm_spa_memattr(uint64_t efi_mem_flags)166 nvdimm_spa_memattr(uint64_t efi_mem_flags)
167 {
168 	vm_memattr_t mode;
169 
170 	if ((efi_mem_flags & EFI_MD_ATTR_WB) != 0)
171 		mode = VM_MEMATTR_WRITE_BACK;
172 	else if ((efi_mem_flags & EFI_MD_ATTR_WT) != 0)
173 		mode = VM_MEMATTR_WRITE_THROUGH;
174 	else if ((efi_mem_flags & EFI_MD_ATTR_WC) != 0)
175 		mode = VM_MEMATTR_WRITE_COMBINING;
176 	else if ((efi_mem_flags & EFI_MD_ATTR_WP) != 0)
177 		mode = VM_MEMATTR_WRITE_PROTECTED;
178 	else if ((efi_mem_flags & EFI_MD_ATTR_UC) != 0)
179 		mode = VM_MEMATTR_UNCACHEABLE;
180 	else {
181 		if (bootverbose)
182 			printf("SPA mapping attr %#lx unsupported\n",
183 			    efi_mem_flags);
184 		mode = VM_MEMATTR_UNCACHEABLE;
185 	}
186 	return (mode);
187 }
188 
189 static int
nvdimm_spa_uio(struct nvdimm_spa_dev * dev,struct uio * uio)190 nvdimm_spa_uio(struct nvdimm_spa_dev *dev, struct uio *uio)
191 {
192 	struct vm_page m, *ma;
193 	off_t off;
194 	vm_memattr_t mattr;
195 	int error, n;
196 
197 	error = 0;
198 	if (dev->spa_kva == NULL) {
199 		mattr = dev->spa_memattr;
200 		bzero(&m, sizeof(m));
201 		vm_page_initfake(&m, 0, mattr);
202 		ma = &m;
203 		while (uio->uio_resid > 0) {
204 			if (uio->uio_offset >= dev->spa_len)
205 				break;
206 			off = dev->spa_phys_base + uio->uio_offset;
207 			vm_page_updatefake(&m, trunc_page(off), mattr);
208 			n = PAGE_SIZE;
209 			if (n > uio->uio_resid)
210 				n = uio->uio_resid;
211 			error = uiomove_fromphys(&ma, off & PAGE_MASK, n, uio);
212 			if (error != 0)
213 				break;
214 		}
215 	} else {
216 		while (uio->uio_resid > 0) {
217 			if (uio->uio_offset >= dev->spa_len)
218 				break;
219 			n = INT_MAX;
220 			if (n > uio->uio_resid)
221 				n = uio->uio_resid;
222 			if (uio->uio_offset + n > dev->spa_len)
223 				n = dev->spa_len - uio->uio_offset;
224 			error = uiomove((char *)dev->spa_kva + uio->uio_offset,
225 			    n, uio);
226 			if (error != 0)
227 				break;
228 		}
229 	}
230 	return (error);
231 }
232 
233 static int
nvdimm_spa_rw(struct cdev * dev,struct uio * uio,int ioflag)234 nvdimm_spa_rw(struct cdev *dev, struct uio *uio, int ioflag)
235 {
236 
237 	return (nvdimm_spa_uio(dev->si_drv1, uio));
238 }
239 
240 static int
nvdimm_spa_ioctl(struct cdev * cdev,u_long cmd,caddr_t data,int fflag,struct thread * td)241 nvdimm_spa_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
242     struct thread *td)
243 {
244 	struct nvdimm_spa_dev *dev;
245 	int error;
246 
247 	dev = cdev->si_drv1;
248 	error = 0;
249 	switch (cmd) {
250 	case DIOCGSECTORSIZE:
251 		*(u_int *)data = DEV_BSIZE;
252 		break;
253 	case DIOCGMEDIASIZE:
254 		*(off_t *)data = dev->spa_len;
255 		break;
256 	default:
257 		error = ENOTTY;
258 		break;
259 	}
260 	return (error);
261 }
262 
263 static int
nvdimm_spa_mmap_single(struct cdev * cdev,vm_ooffset_t * offset,vm_size_t size,vm_object_t * objp,int nprot)264 nvdimm_spa_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t size,
265     vm_object_t *objp, int nprot)
266 {
267 	struct nvdimm_spa_dev *dev;
268 
269 	dev = cdev->si_drv1;
270 	if (dev->spa_obj == NULL)
271 		return (ENXIO);
272 	if (*offset >= dev->spa_len || *offset + size < *offset ||
273 	    *offset + size > dev->spa_len)
274 		return (EINVAL);
275 	vm_object_reference(dev->spa_obj);
276 	*objp = dev->spa_obj;
277 	return (0);
278 }
279 
280 static struct cdevsw spa_cdevsw = {
281 	.d_version =	D_VERSION,
282 	.d_flags =	D_DISK,
283 	.d_name =	"nvdimm_spa",
284 	.d_read =	nvdimm_spa_rw,
285 	.d_write =	nvdimm_spa_rw,
286 	.d_ioctl =	nvdimm_spa_ioctl,
287 	.d_mmap_single = nvdimm_spa_mmap_single,
288 };
289 
290 static void
nvdimm_spa_g_all_unmapped(struct nvdimm_spa_dev * dev,struct bio * bp,int rw)291 nvdimm_spa_g_all_unmapped(struct nvdimm_spa_dev *dev, struct bio *bp, int rw)
292 {
293 	struct vm_page maa[bp->bio_ma_n];
294 	vm_page_t ma[bp->bio_ma_n];
295 	vm_memattr_t mattr;
296 	int i;
297 
298 	mattr = dev->spa_memattr;
299 	for (i = 0; i < nitems(ma); i++) {
300 		bzero(&maa[i], sizeof(maa[i]));
301 		vm_page_initfake(&maa[i], dev->spa_phys_base +
302 		    trunc_page(bp->bio_offset) + PAGE_SIZE * i, mattr);
303 		ma[i] = &maa[i];
304 	}
305 	if (rw == BIO_READ)
306 		pmap_copy_pages(ma, bp->bio_offset & PAGE_MASK, bp->bio_ma,
307 		    bp->bio_ma_offset, bp->bio_length);
308 	else
309 		pmap_copy_pages(bp->bio_ma, bp->bio_ma_offset, ma,
310 		    bp->bio_offset & PAGE_MASK, bp->bio_length);
311 }
312 
313 static void
nvdimm_spa_g_thread(void * arg)314 nvdimm_spa_g_thread(void *arg)
315 {
316 	struct g_spa *sc;
317 	struct bio *bp;
318 	struct uio auio;
319 	struct iovec aiovec;
320 	int error;
321 
322 	sc = arg;
323 	for (;;) {
324 		mtx_lock(&sc->spa_g_mtx);
325 		for (;;) {
326 			bp = bioq_takefirst(&sc->spa_g_queue);
327 			if (bp != NULL)
328 				break;
329 			msleep(&sc->spa_g_queue, &sc->spa_g_mtx, PRIBIO,
330 			    "spa_g", 0);
331 			if (!sc->spa_g_proc_run) {
332 				sc->spa_g_proc_exiting = true;
333 				wakeup(&sc->spa_g_queue);
334 				mtx_unlock(&sc->spa_g_mtx);
335 				kproc_exit(0);
336 			}
337 			continue;
338 		}
339 		mtx_unlock(&sc->spa_g_mtx);
340 		if (bp->bio_cmd != BIO_READ && bp->bio_cmd != BIO_WRITE &&
341 		    bp->bio_cmd != BIO_FLUSH) {
342 			error = EOPNOTSUPP;
343 			goto completed;
344 		}
345 
346 		error = 0;
347 		if (bp->bio_cmd == BIO_FLUSH) {
348 			if (sc->dev->spa_kva != NULL) {
349 				pmap_large_map_wb(sc->dev->spa_kva,
350 				    sc->dev->spa_len);
351 			} else {
352 				pmap_flush_cache_phys_range(
353 				    (vm_paddr_t)sc->dev->spa_phys_base,
354 				    (vm_paddr_t)sc->dev->spa_phys_base +
355 				    sc->dev->spa_len, sc->dev->spa_memattr);
356 			}
357 			/*
358 			 * XXX flush IMC
359 			 */
360 			goto completed;
361 		}
362 
363 		if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
364 			if (sc->dev->spa_kva != NULL) {
365 				aiovec.iov_base = (char *)sc->dev->spa_kva +
366 				    bp->bio_offset;
367 				aiovec.iov_len = bp->bio_length;
368 				auio.uio_iov = &aiovec;
369 				auio.uio_iovcnt = 1;
370 				auio.uio_resid = bp->bio_length;
371 				auio.uio_offset = bp->bio_offset;
372 				auio.uio_segflg = UIO_SYSSPACE;
373 				auio.uio_rw = bp->bio_cmd == BIO_READ ?
374 				    UIO_WRITE : UIO_READ;
375 				auio.uio_td = curthread;
376 				error = uiomove_fromphys(bp->bio_ma,
377 				    bp->bio_ma_offset, bp->bio_length, &auio);
378 				bp->bio_resid = auio.uio_resid;
379 			} else {
380 				nvdimm_spa_g_all_unmapped(sc->dev, bp,
381 				    bp->bio_cmd);
382 				bp->bio_resid = bp->bio_length;
383 				error = 0;
384 			}
385 		} else {
386 			aiovec.iov_base = bp->bio_data;
387 			aiovec.iov_len = bp->bio_length;
388 			auio.uio_iov = &aiovec;
389 			auio.uio_iovcnt = 1;
390 			auio.uio_resid = bp->bio_length;
391 			auio.uio_offset = bp->bio_offset;
392 			auio.uio_segflg = UIO_SYSSPACE;
393 			auio.uio_rw = bp->bio_cmd == BIO_READ ? UIO_READ :
394 			    UIO_WRITE;
395 			auio.uio_td = curthread;
396 			error = nvdimm_spa_uio(sc->dev, &auio);
397 			bp->bio_resid = auio.uio_resid;
398 		}
399 		bp->bio_bcount = bp->bio_length;
400 		devstat_end_transaction_bio(sc->spa_g_devstat, bp);
401 completed:
402 		bp->bio_completed = bp->bio_length;
403 		g_io_deliver(bp, error);
404 	}
405 }
406 
407 static void
nvdimm_spa_g_start(struct bio * bp)408 nvdimm_spa_g_start(struct bio *bp)
409 {
410 	struct g_spa *sc;
411 
412 	sc = bp->bio_to->geom->softc;
413 	if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
414 		devstat_start_transaction_bio(sc->spa_g_devstat, bp);
415 	}
416 	mtx_lock(&sc->spa_g_mtx);
417 	bioq_disksort(&sc->spa_g_queue, bp);
418 	wakeup(&sc->spa_g_queue);
419 	mtx_unlock(&sc->spa_g_mtx);
420 }
421 
422 static int
nvdimm_spa_g_access(struct g_provider * pp,int r,int w,int e)423 nvdimm_spa_g_access(struct g_provider *pp, int r, int w, int e)
424 {
425 
426 	return (0);
427 }
428 
429 static struct g_geom * nvdimm_spa_g_create(struct nvdimm_spa_dev *dev,
430     const char *name);
431 static g_ctl_destroy_geom_t nvdimm_spa_g_destroy_geom;
432 
433 struct g_class nvdimm_spa_g_class = {
434 	.name =		"SPA",
435 	.version =	G_VERSION,
436 	.start =	nvdimm_spa_g_start,
437 	.access =	nvdimm_spa_g_access,
438 	.destroy_geom =	nvdimm_spa_g_destroy_geom,
439 };
440 DECLARE_GEOM_CLASS(nvdimm_spa_g_class, g_spa);
441 
442 int
nvdimm_spa_init(struct SPA_mapping * spa,ACPI_NFIT_SYSTEM_ADDRESS * nfitaddr,enum SPA_mapping_type spa_type)443 nvdimm_spa_init(struct SPA_mapping *spa, ACPI_NFIT_SYSTEM_ADDRESS *nfitaddr,
444     enum SPA_mapping_type spa_type)
445 {
446 	char *name;
447 	int error;
448 
449 	spa->spa_type = spa_type;
450 	spa->spa_nfit_idx = nfitaddr->RangeIndex;
451 	spa->dev.spa_domain =
452 	    ((nfitaddr->Flags & ACPI_NFIT_PROXIMITY_VALID) != 0) ?
453 	    nfitaddr->ProximityDomain : -1;
454 	spa->dev.spa_phys_base = nfitaddr->Address;
455 	spa->dev.spa_len = nfitaddr->Length;
456 	spa->dev.spa_efi_mem_flags = nfitaddr->MemoryMapping;
457 	if (bootverbose) {
458 		printf("NVDIMM SPA%d base %#016jx len %#016jx %s fl %#jx\n",
459 		    spa->spa_nfit_idx,
460 		    (uintmax_t)spa->dev.spa_phys_base,
461 		    (uintmax_t)spa->dev.spa_len,
462 		    nvdimm_SPA_uuid_list[spa_type].u_name,
463 		    spa->dev.spa_efi_mem_flags);
464 	}
465 	spa->dev.spa_memattr = nvdimm_spa_memattr(nfitaddr->MemoryMapping);
466 	if (!nvdimm_SPA_uuid_list[spa_type].u_usr_acc)
467 		return (0);
468 
469 	asprintf(&name, M_NVDIMM, "spa%d", spa->spa_nfit_idx);
470 	error = nvdimm_spa_dev_init(&spa->dev, name, spa->spa_nfit_idx);
471 	free(name, M_NVDIMM);
472 	return (error);
473 }
474 
475 int
nvdimm_spa_dev_init(struct nvdimm_spa_dev * dev,const char * name,int unit)476 nvdimm_spa_dev_init(struct nvdimm_spa_dev *dev, const char *name, int unit)
477 {
478 	struct make_dev_args mda;
479 	struct sglist *spa_sg;
480 	char *devname;
481 	int error, error1;
482 
483 	error1 = pmap_large_map(dev->spa_phys_base, dev->spa_len,
484 	    &dev->spa_kva, dev->spa_memattr);
485 	if (error1 != 0) {
486 		printf("NVDIMM %s cannot map into KVA, error %d\n", name,
487 		    error1);
488 		dev->spa_kva = NULL;
489 	}
490 
491 	spa_sg = sglist_alloc(1, M_WAITOK);
492 	error = sglist_append_phys(spa_sg, dev->spa_phys_base,
493 	    dev->spa_len);
494 	if (error == 0) {
495 		dev->spa_obj = vm_pager_allocate(OBJT_SG, spa_sg, dev->spa_len,
496 		    VM_PROT_ALL, 0, NULL);
497 		if (dev->spa_obj == NULL) {
498 			printf("NVDIMM %s failed to alloc vm object", name);
499 			sglist_free(spa_sg);
500 		}
501 	} else {
502 		printf("NVDIMM %s failed to init sglist, error %d", name,
503 		    error);
504 		sglist_free(spa_sg);
505 	}
506 
507 	make_dev_args_init(&mda);
508 	mda.mda_flags = MAKEDEV_WAITOK | MAKEDEV_CHECKNAME;
509 	mda.mda_devsw = &spa_cdevsw;
510 	mda.mda_cr = NULL;
511 	mda.mda_uid = UID_ROOT;
512 	mda.mda_gid = GID_OPERATOR;
513 	mda.mda_mode = 0660;
514 	mda.mda_si_drv1 = dev;
515 	mda.mda_unit = unit;
516 	asprintf(&devname, M_NVDIMM, "nvdimm_%s", name);
517 	error = make_dev_s(&mda, &dev->spa_dev, "%s", devname);
518 	free(devname, M_NVDIMM);
519 	if (error != 0) {
520 		printf("NVDIMM %s cannot create devfs node, error %d\n", name,
521 		    error);
522 		if (error1 == 0)
523 			error1 = error;
524 	}
525 	dev->spa_g = nvdimm_spa_g_create(dev, name);
526 	if (dev->spa_g == NULL && error1 == 0)
527 		error1 = ENXIO;
528 	return (error1);
529 }
530 
531 static struct g_geom *
nvdimm_spa_g_create(struct nvdimm_spa_dev * dev,const char * name)532 nvdimm_spa_g_create(struct nvdimm_spa_dev *dev, const char *name)
533 {
534 	struct g_geom *gp;
535 	struct g_spa *sc;
536 	int error;
537 
538 	gp = NULL;
539 	sc = malloc(sizeof(struct g_spa), M_NVDIMM, M_WAITOK | M_ZERO);
540 	sc->dev = dev;
541 	bioq_init(&sc->spa_g_queue);
542 	mtx_init(&sc->spa_g_mtx, "spag", NULL, MTX_DEF);
543 	sc->spa_g_proc_run = true;
544 	sc->spa_g_proc_exiting = false;
545 	error = kproc_create(nvdimm_spa_g_thread, sc, &sc->spa_g_proc, 0, 0,
546 	    "g_spa");
547 	if (error != 0) {
548 		mtx_destroy(&sc->spa_g_mtx);
549 		free(sc, M_NVDIMM);
550 		printf("NVDIMM %s cannot create geom worker, error %d\n", name,
551 		    error);
552 	} else {
553 		g_topology_lock();
554 		gp = g_new_geomf(&nvdimm_spa_g_class, "%s", name);
555 		gp->softc = sc;
556 		sc->spa_p = g_new_providerf(gp, "%s", name);
557 		sc->spa_p->mediasize = dev->spa_len;
558 		sc->spa_p->sectorsize = DEV_BSIZE;
559 		sc->spa_p->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE |
560 		    G_PF_ACCEPT_UNMAPPED;
561 		g_error_provider(sc->spa_p, 0);
562 		sc->spa_g_devstat = devstat_new_entry("spa", -1, DEV_BSIZE,
563 		    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT,
564 		    DEVSTAT_PRIORITY_MAX);
565 		g_topology_unlock();
566 	}
567 	return (gp);
568 }
569 
570 void
nvdimm_spa_fini(struct SPA_mapping * spa)571 nvdimm_spa_fini(struct SPA_mapping *spa)
572 {
573 
574 	nvdimm_spa_dev_fini(&spa->dev);
575 }
576 
577 void
nvdimm_spa_dev_fini(struct nvdimm_spa_dev * dev)578 nvdimm_spa_dev_fini(struct nvdimm_spa_dev *dev)
579 {
580 
581 	if (dev->spa_g != NULL) {
582 		g_topology_lock();
583 		nvdimm_spa_g_destroy_geom(NULL, dev->spa_g->class, dev->spa_g);
584 		g_topology_unlock();
585 	}
586 	if (dev->spa_dev != NULL) {
587 		destroy_dev(dev->spa_dev);
588 		dev->spa_dev = NULL;
589 	}
590 	vm_object_deallocate(dev->spa_obj);
591 	if (dev->spa_kva != NULL) {
592 		pmap_large_unmap(dev->spa_kva, dev->spa_len);
593 		dev->spa_kva = NULL;
594 	}
595 }
596 
597 static int
nvdimm_spa_g_destroy_geom(struct gctl_req * req,struct g_class * cp,struct g_geom * gp)598 nvdimm_spa_g_destroy_geom(struct gctl_req *req, struct g_class *cp,
599     struct g_geom *gp)
600 {
601 	struct g_spa *sc;
602 
603 	sc = gp->softc;
604 	mtx_lock(&sc->spa_g_mtx);
605 	sc->spa_g_proc_run = false;
606 	wakeup(&sc->spa_g_queue);
607 	while (!sc->spa_g_proc_exiting)
608 		msleep(&sc->spa_g_queue, &sc->spa_g_mtx, PRIBIO, "spa_e", 0);
609 	mtx_unlock(&sc->spa_g_mtx);
610 	g_topology_assert();
611 	g_wither_geom(gp, ENXIO);
612 	sc->spa_p = NULL;
613 	if (sc->spa_g_devstat != NULL) {
614 		devstat_remove_entry(sc->spa_g_devstat);
615 		sc->spa_g_devstat = NULL;
616 	}
617 	mtx_destroy(&sc->spa_g_mtx);
618 	free(sc, M_NVDIMM);
619 	return (0);
620 }
621