xref: /freebsd/sys/dev/nvdimm/nvdimm.c (revision 9768746b) 
1 /*-
2  * Copyright (c) 2017 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 __FBSDID("$FreeBSD$");
33 
34 #include "opt_acpi.h"
35 #include "opt_ddb.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/bio.h>
40 #include <sys/bitstring.h>
41 #include <sys/bus.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/sbuf.h>
47 #include <sys/sysctl.h>
48 #include <sys/uuid.h>
49 
50 #include <contrib/dev/acpica/include/acpi.h>
51 #include <contrib/dev/acpica/include/accommon.h>
52 #include <contrib/dev/acpica/include/acuuid.h>
53 #include <dev/acpica/acpivar.h>
54 
55 #include <dev/nvdimm/nvdimm_var.h>
56 
57 #define _COMPONENT	ACPI_OEM
58 ACPI_MODULE_NAME("NVDIMM")
59 
60 static struct uuid intel_nvdimm_dsm_uuid =
61     {0x4309AC30,0x0D11,0x11E4,0x91,0x91,{0x08,0x00,0x20,0x0C,0x9A,0x66}};
62 #define INTEL_NVDIMM_DSM_REV 1
63 #define INTEL_NVDIMM_DSM_GET_LABEL_SIZE 4
64 #define INTEL_NVDIMM_DSM_GET_LABEL_DATA 5
65 
66 MALLOC_DEFINE(M_NVDIMM, "nvdimm", "NVDIMM driver memory");
67 
68 static int
69 read_label_area_size(struct nvdimm_dev *nv)
70 {
71 	ACPI_OBJECT *result_buffer;
72 	ACPI_HANDLE handle;
73 	ACPI_STATUS status;
74 	ACPI_BUFFER result;
75 	uint32_t *out;
76 	int error;
77 
78 	handle = nvdimm_root_get_acpi_handle(nv->nv_dev);
79 	if (handle == NULL)
80 		return (ENODEV);
81 	result.Length = ACPI_ALLOCATE_BUFFER;
82 	result.Pointer = NULL;
83 	status = acpi_EvaluateDSM(handle, (uint8_t *)&intel_nvdimm_dsm_uuid,
84 	    INTEL_NVDIMM_DSM_REV, INTEL_NVDIMM_DSM_GET_LABEL_SIZE, NULL,
85 	    &result);
86 	error = ENXIO;
87 	if (ACPI_SUCCESS(status) && result.Pointer != NULL &&
88 	    result.Length >= sizeof(ACPI_OBJECT)) {
89 		result_buffer = result.Pointer;
90 		if (result_buffer->Type == ACPI_TYPE_BUFFER &&
91 		    result_buffer->Buffer.Length >= 12) {
92 			out = (uint32_t *)result_buffer->Buffer.Pointer;
93 			nv->label_area_size = out[1];
94 			nv->max_label_xfer = out[2];
95 			error = 0;
96 		}
97 	}
98 	if (result.Pointer != NULL)
99 		AcpiOsFree(result.Pointer);
100 	return (error);
101 }
102 
103 static int
104 read_label_area(struct nvdimm_dev *nv, uint8_t *dest, off_t offset,
105     off_t length)
106 {
107 	ACPI_BUFFER result;
108 	ACPI_HANDLE handle;
109 	ACPI_OBJECT params_pkg, params_buf, *result_buf;
110 	ACPI_STATUS status;
111 	uint32_t params[2];
112 	off_t to_read;
113 	int error;
114 
115 	error = 0;
116 	handle = nvdimm_root_get_acpi_handle(nv->nv_dev);
117 	if (offset < 0 || length <= 0 ||
118 	    offset + length > nv->label_area_size ||
119 	    handle == NULL)
120 		return (ENODEV);
121 	params_pkg.Type = ACPI_TYPE_PACKAGE;
122 	params_pkg.Package.Count = 1;
123 	params_pkg.Package.Elements = &params_buf;
124 	params_buf.Type = ACPI_TYPE_BUFFER;
125 	params_buf.Buffer.Length = sizeof(params);
126 	params_buf.Buffer.Pointer = (UINT8 *)params;
127 	while (length > 0) {
128 		to_read = MIN(length, nv->max_label_xfer);
129 		params[0] = offset;
130 		params[1] = to_read;
131 		result.Length = ACPI_ALLOCATE_BUFFER;
132 		result.Pointer = NULL;
133 		status = acpi_EvaluateDSM(handle,
134 		    (uint8_t *)&intel_nvdimm_dsm_uuid, INTEL_NVDIMM_DSM_REV,
135 		    INTEL_NVDIMM_DSM_GET_LABEL_DATA, &params_pkg, &result);
136 		if (ACPI_FAILURE(status) ||
137 		    result.Length < sizeof(ACPI_OBJECT) ||
138 		    result.Pointer == NULL) {
139 			error = ENXIO;
140 			break;
141 		}
142 		result_buf = (ACPI_OBJECT *)result.Pointer;
143 		if (result_buf->Type != ACPI_TYPE_BUFFER ||
144 		    result_buf->Buffer.Pointer == NULL ||
145 		    result_buf->Buffer.Length != 4 + to_read ||
146 		    ((uint16_t *)result_buf->Buffer.Pointer)[0] != 0) {
147 			error = ENXIO;
148 			break;
149 		}
150 		bcopy(result_buf->Buffer.Pointer + 4, dest, to_read);
151 		dest += to_read;
152 		offset += to_read;
153 		length -= to_read;
154 		if (result.Pointer != NULL) {
155 			AcpiOsFree(result.Pointer);
156 			result.Pointer = NULL;
157 		}
158 	}
159 	if (result.Pointer != NULL)
160 		AcpiOsFree(result.Pointer);
161 	return (error);
162 }
163 
164 static uint64_t
165 fletcher64(const void *data, size_t length)
166 {
167 	size_t i;
168 	uint32_t a, b;
169 	const uint32_t *d;
170 
171 	a = 0;
172 	b = 0;
173 	d = (const uint32_t *)data;
174 	length = length / sizeof(uint32_t);
175 	for (i = 0; i < length; i++) {
176 		a += d[i];
177 		b += a;
178 	}
179 	return ((uint64_t)b << 32 | a);
180 }
181 
182 static bool
183 label_index_is_valid(struct nvdimm_label_index *index, uint32_t max_labels,
184     size_t size, size_t offset)
185 {
186 	uint64_t checksum;
187 
188 	index = (struct nvdimm_label_index *)((uint8_t *)index + size * offset);
189 	if (strcmp(index->signature, NVDIMM_INDEX_BLOCK_SIGNATURE) != 0)
190 		return false;
191 	checksum = index->checksum;
192 	index->checksum = 0;
193 	if (checksum != fletcher64(index, size) ||
194 	    index->this_offset != size * offset || index->this_size != size ||
195 	    index->other_offset != size * (offset == 0 ? 1 : 0) ||
196 	    index->seq == 0 || index->seq > 3 || index->slot_cnt > max_labels ||
197 	    index->label_size != 1)
198 		return false;
199 	return true;
200 }
201 
202 static int
203 read_label(struct nvdimm_dev *nv, int num)
204 {
205 	struct nvdimm_label_entry *entry, *i, *next;
206 	uint64_t checksum;
207 	off_t offset;
208 	int error;
209 
210 	offset = nv->label_index->label_offset +
211 	    num * (128 << nv->label_index->label_size);
212 	entry = malloc(sizeof(*entry), M_NVDIMM, M_WAITOK);
213 	error = read_label_area(nv, (uint8_t *)&entry->label, offset,
214 	    sizeof(struct nvdimm_label));
215 	if (error != 0) {
216 		free(entry, M_NVDIMM);
217 		return (error);
218 	}
219 	checksum = entry->label.checksum;
220 	entry->label.checksum = 0;
221 	if (checksum != fletcher64(&entry->label, sizeof(entry->label)) ||
222 	    entry->label.slot != num) {
223 		free(entry, M_NVDIMM);
224 		return (ENXIO);
225 	}
226 
227 	/* Insertion ordered by dimm_phys_addr */
228 	if (SLIST_EMPTY(&nv->labels) ||
229 	    entry->label.dimm_phys_addr <=
230 	    SLIST_FIRST(&nv->labels)->label.dimm_phys_addr) {
231 		SLIST_INSERT_HEAD(&nv->labels, entry, link);
232 		return (0);
233 	}
234 	SLIST_FOREACH_SAFE(i, &nv->labels, link, next) {
235 		if (next == NULL ||
236 		    entry->label.dimm_phys_addr <= next->label.dimm_phys_addr) {
237 			SLIST_INSERT_AFTER(i, entry, link);
238 			return (0);
239 		}
240 	}
241 	__assert_unreachable();
242 }
243 
244 static int
245 read_labels(struct nvdimm_dev *nv)
246 {
247 	struct nvdimm_label_index *indices, *index1;
248 	size_t index_size, num_labels;
249 	int error, n;
250 	bool index_0_valid, index_1_valid;
251 
252 	for (index_size = 256; ; index_size += 256) {
253 		num_labels = 8 * (index_size -
254 		    sizeof(struct nvdimm_label_index));
255 		if (index_size + num_labels * sizeof(struct nvdimm_label) >=
256 		    nv->label_area_size)
257 			break;
258 	}
259 	num_labels = (nv->label_area_size - index_size) /
260 	    sizeof(struct nvdimm_label);
261 	indices = malloc(2 * index_size, M_NVDIMM, M_WAITOK);
262 	index1 = (void *)((uint8_t *)indices + index_size);
263 	error = read_label_area(nv, (void *)indices, 0, 2 * index_size);
264 	if (error != 0) {
265 		free(indices, M_NVDIMM);
266 		return (error);
267 	}
268 	index_0_valid = label_index_is_valid(indices, num_labels, index_size,
269 	    0);
270 	index_1_valid = label_index_is_valid(indices, num_labels, index_size,
271 	    1);
272 	if (!index_0_valid && !index_1_valid) {
273 		free(indices, M_NVDIMM);
274 		return (ENXIO);
275 	}
276 	if (index_0_valid && index_1_valid) {
277 		if (((int)indices->seq - (int)index1->seq + 3) % 3 == 1) {
278 			/* index 0 was more recently updated */
279 			index_1_valid = false;
280 		} else {
281 			/*
282 			 * either index 1 was more recently updated,
283 			 * or the sequence numbers are equal, in which
284 			 * case the specification says the block with
285 			 * the higher offset is to be treated as valid
286 			 */
287 			index_0_valid = false;
288 		}
289 	}
290 	nv->label_index = malloc(index_size, M_NVDIMM, M_WAITOK);
291 	bcopy(index_0_valid ? indices : index1, nv->label_index, index_size);
292 	free(indices, M_NVDIMM);
293 	bit_ffc_at((bitstr_t *)nv->label_index->free, 0,
294 	    nv->label_index->slot_cnt, &n);
295 	while (n >= 0) {
296 		read_label(nv, n);
297 		bit_ffc_at((bitstr_t *)nv->label_index->free, n + 1,
298 		    nv->label_index->slot_cnt, &n);
299 	}
300 	return (0);
301 }
302 
303 static int
304 nvdimm_probe(device_t dev)
305 {
306 
307 	return (BUS_PROBE_NOWILDCARD);
308 }
309 
310 static int
311 nvdimm_attach(device_t dev)
312 {
313 	struct nvdimm_dev *nv;
314 	struct sysctl_ctx_list *ctx;
315 	struct sysctl_oid *oid;
316 	struct sysctl_oid_list *children;
317 	struct sbuf *sb;
318 	ACPI_TABLE_NFIT *nfitbl;
319 	ACPI_STATUS status;
320 	ACPI_NFIT_MEMORY_MAP **maps;
321 	int error, i, num_maps;
322 	uint16_t flags;
323 
324 	nv = device_get_softc(dev);
325 	ctx = device_get_sysctl_ctx(dev);
326 	oid = device_get_sysctl_tree(dev);
327 	children = SYSCTL_CHILDREN(oid);
328 	MPASS(nvdimm_root_get_acpi_handle(dev) != NULL);
329 	nv->nv_dev = dev;
330 	nv->nv_handle = nvdimm_root_get_device_handle(dev);
331 
332 	status = AcpiGetTable(ACPI_SIG_NFIT, 1, (ACPI_TABLE_HEADER **)&nfitbl);
333 	if (ACPI_FAILURE(status)) {
334 		if (bootverbose)
335 			device_printf(dev, "cannot get NFIT\n");
336 		return (ENXIO);
337 	}
338 	acpi_nfit_get_flush_addrs(nfitbl, nv->nv_handle, &nv->nv_flush_addr,
339 	    &nv->nv_flush_addr_cnt);
340 
341 	/*
342 	 * Each NVDIMM should have at least one memory map associated with it.
343 	 * If any of the maps have one of the error flags set, reflect that in
344 	 * the overall status.
345 	 */
346 	acpi_nfit_get_memory_maps_by_dimm(nfitbl, nv->nv_handle, &maps,
347 	    &num_maps);
348 	if (num_maps == 0) {
349 		free(nv->nv_flush_addr, M_NVDIMM);
350 		free(maps, M_NVDIMM);
351 		device_printf(dev, "cannot find memory map\n");
352 		return (ENXIO);
353 	}
354 	flags = 0;
355 	for (i = 0; i < num_maps; i++) {
356 		flags |= maps[i]->Flags;
357 	}
358 	free(maps, M_NVDIMM);
359 
360 	/* sbuf_new_auto(9) is M_WAITOK; no need to check for NULL. */
361 	sb = sbuf_new_auto();
362 	(void) sbuf_printf(sb, "0x%b", flags,
363 	    "\20"
364 	    "\001SAVE_FAILED"
365 	    "\002RESTORE_FAILED"
366 	    "\003FLUSH_FAILED"
367 	    "\004NOT_ARMED"
368 	    "\005HEALTH_OBSERVED"
369 	    "\006HEALTH_ENABLED"
370 	    "\007MAP_FAILED");
371 	error = sbuf_finish(sb);
372 	if (error != 0) {
373 		sbuf_delete(sb);
374 		free(nv->nv_flush_addr, M_NVDIMM);
375 		device_printf(dev, "cannot convert flags to string\n");
376 		return (error);
377 	}
378 	/* strdup(9) is M_WAITOK; no need to check for NULL. */
379 	nv->nv_flags_str = strdup(sbuf_data(sb), M_NVDIMM);
380 	sbuf_delete(sb);
381 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "flags",
382 	    CTLFLAG_RD | CTLFLAG_MPSAFE, nv->nv_flags_str, 0,
383 	    "NVDIMM State Flags");
384 	/*
385 	 * Anything other than HEALTH_ENABLED indicates a fault condition of
386 	 * some kind, so log if that's seen.
387 	 */
388 	if ((flags & ~ACPI_NFIT_MEM_HEALTH_ENABLED) != 0)
389 		device_printf(dev, "flags: %s\n", nv->nv_flags_str);
390 
391 	AcpiPutTable(&nfitbl->Header);
392 	error = read_label_area_size(nv);
393 	if (error == 0) {
394 		/*
395 		 * Ignoring errors reading labels. Not all NVDIMMs
396 		 * support labels and namespaces.
397 		 */
398 		read_labels(nv);
399 	}
400 	return (0);
401 }
402 
403 static int
404 nvdimm_detach(device_t dev)
405 {
406 	struct nvdimm_dev *nv;
407 	struct nvdimm_label_entry *label, *next;
408 
409 	nv = device_get_softc(dev);
410 	free(nv->nv_flags_str, M_NVDIMM);
411 	free(nv->nv_flush_addr, M_NVDIMM);
412 	free(nv->label_index, M_NVDIMM);
413 	SLIST_FOREACH_SAFE(label, &nv->labels, link, next) {
414 		SLIST_REMOVE_HEAD(&nv->labels, link);
415 		free(label, M_NVDIMM);
416 	}
417 	return (0);
418 }
419 
420 static int
421 nvdimm_suspend(device_t dev)
422 {
423 
424 	return (0);
425 }
426 
427 static int
428 nvdimm_resume(device_t dev)
429 {
430 
431 	return (0);
432 }
433 
434 static device_method_t nvdimm_methods[] = {
435 	DEVMETHOD(device_probe, nvdimm_probe),
436 	DEVMETHOD(device_attach, nvdimm_attach),
437 	DEVMETHOD(device_detach, nvdimm_detach),
438 	DEVMETHOD(device_suspend, nvdimm_suspend),
439 	DEVMETHOD(device_resume, nvdimm_resume),
440 	DEVMETHOD_END
441 };
442 
443 static driver_t	nvdimm_driver = {
444 	"nvdimm",
445 	nvdimm_methods,
446 	sizeof(struct nvdimm_dev),
447 };
448 
449 struct nvdimm_dev *
450 nvdimm_find_by_handle(nfit_handle_t nv_handle)
451 {
452 	struct nvdimm_dev *res;
453 	device_t *dimms;
454 	int i, error, num_dimms;
455 
456 	res = NULL;
457 	error = devclass_get_devices(devclass_find(nvdimm_driver.name), &dimms,
458 	    &num_dimms);
459 	if (error != 0)
460 		return (NULL);
461 	for (i = 0; i < num_dimms; i++) {
462 		if (nvdimm_root_get_device_handle(dimms[i]) == nv_handle) {
463 			res = device_get_softc(dimms[i]);
464 			break;
465 		}
466 	}
467 	free(dimms, M_TEMP);
468 	return (res);
469 }
470 
471 DRIVER_MODULE(nvdimm, nvdimm_acpi_root, nvdimm_driver, NULL, NULL);
472 MODULE_DEPEND(nvdimm, acpi, 1, 1, 1);
473