?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
acpi, nfit: validate commands against the device type
[linux-2.6/btrfs-unstable.git] / drivers / acpi / nfit / core.c
blobff2580e7611d18c6d56c58d50c2cbc3a2d54aa36
1 /*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13 #include <linux/list_sort.h>
14 #include <linux/libnvdimm.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/ndctl.h>
18 #include <linux/sysfs.h>
19 #include <linux/delay.h>
20 #include <linux/list.h>
21 #include <linux/acpi.h>
22 #include <linux/sort.h>
23 #include <linux/io.h>
24 #include <linux/nd.h>
25 #include <asm/cacheflush.h>
26 #include "nfit.h"
27
28 /*
29 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
30 * irrelevant.
31 */
32 #include <linux/io-64-nonatomic-hi-lo.h>
33
34 static bool force_enable_dimms;
35 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
36 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
37
38 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
39 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
40 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
41
42 /* after three payloads of overflow, it's dead jim */
43 static unsigned int scrub_overflow_abort = 3;
44 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
45 MODULE_PARM_DESC(scrub_overflow_abort,
46 "Number of times we overflow ARS results before abort");
47
48 static bool disable_vendor_specific;
49 module_param(disable_vendor_specific, bool, S_IRUGO);
50 MODULE_PARM_DESC(disable_vendor_specific,
51 "Limit commands to the publicly specified set");
52
53 static unsigned long override_dsm_mask;
54 module_param(override_dsm_mask, ulong, S_IRUGO);
55 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
56
57 static int default_dsm_family = -1;
58 module_param(default_dsm_family, int, S_IRUGO);
59 MODULE_PARM_DESC(default_dsm_family,
60 "Try this DSM type first when identifying NVDIMM family");
61
62 LIST_HEAD(acpi_descs);
63 DEFINE_MUTEX(acpi_desc_lock);
64
65 static struct workqueue_struct *nfit_wq;
66
67 struct nfit_table_prev {
68 struct list_head spas;
69 struct list_head memdevs;
70 struct list_head dcrs;
71 struct list_head bdws;
72 struct list_head idts;
73 struct list_head flushes;
74 };
75
76 static guid_t nfit_uuid[NFIT_UUID_MAX];
77
78 const guid_t *to_nfit_uuid(enum nfit_uuids id)
79 {
80 return &nfit_uuid[id];
81 }
82 EXPORT_SYMBOL(to_nfit_uuid);
83
84 static struct acpi_nfit_desc *to_acpi_nfit_desc(
85 struct nvdimm_bus_descriptor *nd_desc)
86 {
87 return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
88 }
89
90 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
91 {
92 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
93
94 /*
95 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
96 * acpi_device.
97 */
98 if (!nd_desc->provider_name
99 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
100 return NULL;
101
102 return to_acpi_device(acpi_desc->dev);
103 }
104
105 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
106 {
107 struct nd_cmd_clear_error *clear_err;
108 struct nd_cmd_ars_status *ars_status;
109 u16 flags;
110
111 switch (cmd) {
112 case ND_CMD_ARS_CAP:
113 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
114 return -ENOTTY;
115
116 /* Command failed */
117 if (status & 0xffff)
118 return -EIO;
119
120 /* No supported scan types for this range */
121 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
122 if ((status >> 16 & flags) == 0)
123 return -ENOTTY;
124 return 0;
125 case ND_CMD_ARS_START:
126 /* ARS is in progress */
127 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
128 return -EBUSY;
129
130 /* Command failed */
131 if (status & 0xffff)
132 return -EIO;
133 return 0;
134 case ND_CMD_ARS_STATUS:
135 ars_status = buf;
136 /* Command failed */
137 if (status & 0xffff)
138 return -EIO;
139 /* Check extended status (Upper two bytes) */
140 if (status == NFIT_ARS_STATUS_DONE)
141 return 0;
142
143 /* ARS is in progress */
144 if (status == NFIT_ARS_STATUS_BUSY)
145 return -EBUSY;
146
147 /* No ARS performed for the current boot */
148 if (status == NFIT_ARS_STATUS_NONE)
149 return -EAGAIN;
150
151 /*
152 * ARS interrupted, either we overflowed or some other
153 * agent wants the scan to stop. If we didn't overflow
154 * then just continue with the returned results.
155 */
156 if (status == NFIT_ARS_STATUS_INTR) {
157 if (ars_status->out_length >= 40 && (ars_status->flags
158 & NFIT_ARS_F_OVERFLOW))
159 return -ENOSPC;
160 return 0;
161 }
162
163 /* Unknown status */
164 if (status >> 16)
165 return -EIO;
166 return 0;
167 case ND_CMD_CLEAR_ERROR:
168 clear_err = buf;
169 if (status & 0xffff)
170 return -EIO;
171 if (!clear_err->cleared)
172 return -EIO;
173 if (clear_err->length > clear_err->cleared)
174 return clear_err->cleared;
175 return 0;
176 default:
177 break;
178 }
179
180 /* all other non-zero status results in an error */
181 if (status)
182 return -EIO;
183 return 0;
184 }
185
186 #define ACPI_LABELS_LOCKED 3
187
188 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
189 u32 status)
190 {
191 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
192
193 switch (cmd) {
194 case ND_CMD_GET_CONFIG_SIZE:
195 /*
196 * In the _LSI, _LSR, _LSW case the locked status is
197 * communicated via the read/write commands
198 */
199 if (nfit_mem->has_lsi)
200 break;
201
202 if (status >> 16 & ND_CONFIG_LOCKED)
203 return -EACCES;
204 break;
205 case ND_CMD_GET_CONFIG_DATA:
206 if (nfit_mem->has_lsr && status == ACPI_LABELS_LOCKED)
207 return -EACCES;
208 break;
209 case ND_CMD_SET_CONFIG_DATA:
210 if (nfit_mem->has_lsw && status == ACPI_LABELS_LOCKED)
211 return -EACCES;
212 break;
213 default:
214 break;
215 }
216
217 /* all other non-zero status results in an error */
218 if (status)
219 return -EIO;
220 return 0;
221 }
222
223 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
224 u32 status)
225 {
226 if (!nvdimm)
227 return xlat_bus_status(buf, cmd, status);
228 return xlat_nvdimm_status(nvdimm, buf, cmd, status);
229 }
230
231 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
232 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
233 {
234 int i;
235 void *dst;
236 size_t size = 0;
237 union acpi_object *buf = NULL;
238
239 if (pkg->type != ACPI_TYPE_PACKAGE) {
240 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
241 pkg->type);
242 goto err;
243 }
244
245 for (i = 0; i < pkg->package.count; i++) {
246 union acpi_object *obj = &pkg->package.elements[i];
247
248 if (obj->type == ACPI_TYPE_INTEGER)
249 size += 4;
250 else if (obj->type == ACPI_TYPE_BUFFER)
251 size += obj->buffer.length;
252 else {
253 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
254 obj->type);
255 goto err;
256 }
257 }
258
259 buf = ACPI_ALLOCATE(sizeof(*buf) + size);
260 if (!buf)
261 goto err;
262
263 dst = buf + 1;
264 buf->type = ACPI_TYPE_BUFFER;
265 buf->buffer.length = size;
266 buf->buffer.pointer = dst;
267 for (i = 0; i < pkg->package.count; i++) {
268 union acpi_object *obj = &pkg->package.elements[i];
269
270 if (obj->type == ACPI_TYPE_INTEGER) {
271 memcpy(dst, &obj->integer.value, 4);
272 dst += 4;
273 } else if (obj->type == ACPI_TYPE_BUFFER) {
274 memcpy(dst, obj->buffer.pointer, obj->buffer.length);
275 dst += obj->buffer.length;
276 }
277 }
278 err:
279 ACPI_FREE(pkg);
280 return buf;
281 }
282
283 static union acpi_object *int_to_buf(union acpi_object *integer)
284 {
285 union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
286 void *dst = NULL;
287
288 if (!buf)
289 goto err;
290
291 if (integer->type != ACPI_TYPE_INTEGER) {
292 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
293 integer->type);
294 goto err;
295 }
296
297 dst = buf + 1;
298 buf->type = ACPI_TYPE_BUFFER;
299 buf->buffer.length = 4;
300 buf->buffer.pointer = dst;
301 memcpy(dst, &integer->integer.value, 4);
302 err:
303 ACPI_FREE(integer);
304 return buf;
305 }
306
307 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
308 u32 len, void *data)
309 {
310 acpi_status rc;
311 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
312 struct acpi_object_list input = {
313 .count = 3,
314 .pointer = (union acpi_object []) {
315 [0] = {
316 .integer.type = ACPI_TYPE_INTEGER,
317 .integer.value = offset,
318 },
319 [1] = {
320 .integer.type = ACPI_TYPE_INTEGER,
321 .integer.value = len,
322 },
323 [2] = {
324 .buffer.type = ACPI_TYPE_BUFFER,
325 .buffer.pointer = data,
326 .buffer.length = len,
327 },
328 },
329 };
330
331 rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
332 if (ACPI_FAILURE(rc))
333 return NULL;
334 return int_to_buf(buf.pointer);
335 }
336
337 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
338 u32 len)
339 {
340 acpi_status rc;
341 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
342 struct acpi_object_list input = {
343 .count = 2,
344 .pointer = (union acpi_object []) {
345 [0] = {
346 .integer.type = ACPI_TYPE_INTEGER,
347 .integer.value = offset,
348 },
349 [1] = {
350 .integer.type = ACPI_TYPE_INTEGER,
351 .integer.value = len,
352 },
353 },
354 };
355
356 rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
357 if (ACPI_FAILURE(rc))
358 return NULL;
359 return pkg_to_buf(buf.pointer);
360 }
361
362 static union acpi_object *acpi_label_info(acpi_handle handle)
363 {
364 acpi_status rc;
365 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
366
367 rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
368 if (ACPI_FAILURE(rc))
369 return NULL;
370 return pkg_to_buf(buf.pointer);
371 }
372
373 static u8 nfit_dsm_revid(unsigned family, unsigned func)
374 {
375 static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
376 [NVDIMM_FAMILY_INTEL] = {
377 [NVDIMM_INTEL_GET_MODES] = 2,
378 [NVDIMM_INTEL_GET_FWINFO] = 2,
379 [NVDIMM_INTEL_START_FWUPDATE] = 2,
380 [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
381 [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
382 [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
383 [NVDIMM_INTEL_SET_THRESHOLD] = 2,
384 [NVDIMM_INTEL_INJECT_ERROR] = 2,
385 },
386 };
387 u8 id;
388
389 if (family > NVDIMM_FAMILY_MAX)
390 return 0;
391 if (func > 31)
392 return 0;
393 id = revid_table[family][func];
394 if (id == 0)
395 return 1; /* default */
396 return id;
397 }
398
399 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
400 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
401 {
402 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
403 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
404 union acpi_object in_obj, in_buf, *out_obj;
405 const struct nd_cmd_desc *desc = NULL;
406 struct device *dev = acpi_desc->dev;
407 struct nd_cmd_pkg *call_pkg = NULL;
408 const char *cmd_name, *dimm_name;
409 unsigned long cmd_mask, dsm_mask;
410 u32 offset, fw_status = 0;
411 acpi_handle handle;
412 unsigned int func;
413 const guid_t *guid;
414 int rc, i;
415
416 func = cmd;
417 if (cmd == ND_CMD_CALL) {
418 call_pkg = buf;
419 func = call_pkg->nd_command;
420
421 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
422 if (call_pkg->nd_reserved2[i])
423 return -EINVAL;
424 }
425
426 if (nvdimm) {
427 struct acpi_device *adev = nfit_mem->adev;
428
429 if (!adev)
430 return -ENOTTY;
431 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
432 return -ENOTTY;
433
434 dimm_name = nvdimm_name(nvdimm);
435 cmd_name = nvdimm_cmd_name(cmd);
436 cmd_mask = nvdimm_cmd_mask(nvdimm);
437 dsm_mask = nfit_mem->dsm_mask;
438 desc = nd_cmd_dimm_desc(cmd);
439 guid = to_nfit_uuid(nfit_mem->family);
440 handle = adev->handle;
441 } else {
442 struct acpi_device *adev = to_acpi_dev(acpi_desc);
443
444 cmd_name = nvdimm_bus_cmd_name(cmd);
445 cmd_mask = nd_desc->cmd_mask;
446 dsm_mask = cmd_mask;
447 if (cmd == ND_CMD_CALL)
448 dsm_mask = nd_desc->bus_dsm_mask;
449 desc = nd_cmd_bus_desc(cmd);
450 guid = to_nfit_uuid(NFIT_DEV_BUS);
451 handle = adev->handle;
452 dimm_name = "bus";
453 }
454
455 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
456 return -ENOTTY;
457
458 if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
459 return -ENOTTY;
460
461 in_obj.type = ACPI_TYPE_PACKAGE;
462 in_obj.package.count = 1;
463 in_obj.package.elements = &in_buf;
464 in_buf.type = ACPI_TYPE_BUFFER;
465 in_buf.buffer.pointer = buf;
466 in_buf.buffer.length = 0;
467
468 /* libnvdimm has already validated the input envelope */
469 for (i = 0; i < desc->in_num; i++)
470 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
471 i, buf);
472
473 if (call_pkg) {
474 /* skip over package wrapper */
475 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
476 in_buf.buffer.length = call_pkg->nd_size_in;
477 }
478
479 dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
480 __func__, dimm_name, cmd, func, in_buf.buffer.length);
481 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
482 in_buf.buffer.pointer,
483 min_t(u32, 256, in_buf.buffer.length), true);
484
485 /* call the BIOS, prefer the named methods over _DSM if available */
486 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE && nfit_mem->has_lsi)
487 out_obj = acpi_label_info(handle);
488 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && nfit_mem->has_lsr) {
489 struct nd_cmd_get_config_data_hdr *p = buf;
490
491 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
492 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
493 && nfit_mem->has_lsw) {
494 struct nd_cmd_set_config_hdr *p = buf;
495
496 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
497 p->in_buf);
498 } else {
499 u8 revid;
500
501 if (nvdimm)
502 revid = nfit_dsm_revid(nfit_mem->family, func);
503 else
504 revid = 1;
505 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
506 }
507
508 if (!out_obj) {
509 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
510 cmd_name);
511 return -EINVAL;
512 }
513
514 if (call_pkg) {
515 call_pkg->nd_fw_size = out_obj->buffer.length;
516 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
517 out_obj->buffer.pointer,
518 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
519
520 ACPI_FREE(out_obj);
521 /*
522 * Need to support FW function w/o known size in advance.
523 * Caller can determine required size based upon nd_fw_size.
524 * If we return an error (like elsewhere) then caller wouldn't
525 * be able to rely upon data returned to make calculation.
526 */
527 return 0;
528 }
529
530 if (out_obj->package.type != ACPI_TYPE_BUFFER) {
531 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
532 __func__, dimm_name, cmd_name, out_obj->type);
533 rc = -EINVAL;
534 goto out;
535 }
536
537 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__, dimm_name,
538 cmd_name, out_obj->buffer.length);
539 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
540 out_obj->buffer.pointer,
541 min_t(u32, 128, out_obj->buffer.length), true);
542
543 for (i = 0, offset = 0; i < desc->out_num; i++) {
544 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
545 (u32 *) out_obj->buffer.pointer,
546 out_obj->buffer.length - offset);
547
548 if (offset + out_size > out_obj->buffer.length) {
549 dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
550 __func__, dimm_name, cmd_name, i);
551 break;
552 }
553
554 if (in_buf.buffer.length + offset + out_size > buf_len) {
555 dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
556 __func__, dimm_name, cmd_name, i);
557 rc = -ENXIO;
558 goto out;
559 }
560 memcpy(buf + in_buf.buffer.length + offset,
561 out_obj->buffer.pointer + offset, out_size);
562 offset += out_size;
563 }
564
565 /*
566 * Set fw_status for all the commands with a known format to be
567 * later interpreted by xlat_status().
568 */
569 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
570 && cmd <= ND_CMD_CLEAR_ERROR)
571 || (nvdimm && cmd >= ND_CMD_SMART
572 && cmd <= ND_CMD_VENDOR)))
573 fw_status = *(u32 *) out_obj->buffer.pointer;
574
575 if (offset + in_buf.buffer.length < buf_len) {
576 if (i >= 1) {
577 /*
578 * status valid, return the number of bytes left
579 * unfilled in the output buffer
580 */
581 rc = buf_len - offset - in_buf.buffer.length;
582 if (cmd_rc)
583 *cmd_rc = xlat_status(nvdimm, buf, cmd,
584 fw_status);
585 } else {
586 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
587 __func__, dimm_name, cmd_name, buf_len,
588 offset);
589 rc = -ENXIO;
590 }
591 } else {
592 rc = 0;
593 if (cmd_rc)
594 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
595 }
596
597 out:
598 ACPI_FREE(out_obj);
599
600 return rc;
601 }
602 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
603
604 static const char *spa_type_name(u16 type)
605 {
606 static const char *to_name[] = {
607 [NFIT_SPA_VOLATILE] = "volatile",
608 [NFIT_SPA_PM] = "pmem",
609 [NFIT_SPA_DCR] = "dimm-control-region",
610 [NFIT_SPA_BDW] = "block-data-window",
611 [NFIT_SPA_VDISK] = "volatile-disk",
612 [NFIT_SPA_VCD] = "volatile-cd",
613 [NFIT_SPA_PDISK] = "persistent-disk",
614 [NFIT_SPA_PCD] = "persistent-cd",
615
616 };
617
618 if (type > NFIT_SPA_PCD)
619 return "unknown";
620
621 return to_name[type];
622 }
623
624 int nfit_spa_type(struct acpi_nfit_system_address *spa)
625 {
626 int i;
627
628 for (i = 0; i < NFIT_UUID_MAX; i++)
629 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
630 return i;
631 return -1;
632 }
633
634 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
635 struct nfit_table_prev *prev,
636 struct acpi_nfit_system_address *spa)
637 {
638 struct device *dev = acpi_desc->dev;
639 struct nfit_spa *nfit_spa;
640
641 if (spa->header.length != sizeof(*spa))
642 return false;
643
644 list_for_each_entry(nfit_spa, &prev->spas, list) {
645 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
646 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
647 return true;
648 }
649 }
650
651 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
652 GFP_KERNEL);
653 if (!nfit_spa)
654 return false;
655 INIT_LIST_HEAD(&nfit_spa->list);
656 memcpy(nfit_spa->spa, spa, sizeof(*spa));
657 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
658 dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
659 spa->range_index,
660 spa_type_name(nfit_spa_type(spa)));
661 return true;
662 }
663
664 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
665 struct nfit_table_prev *prev,
666 struct acpi_nfit_memory_map *memdev)
667 {
668 struct device *dev = acpi_desc->dev;
669 struct nfit_memdev *nfit_memdev;
670
671 if (memdev->header.length != sizeof(*memdev))
672 return false;
673
674 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
675 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
676 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
677 return true;
678 }
679
680 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
681 GFP_KERNEL);
682 if (!nfit_memdev)
683 return false;
684 INIT_LIST_HEAD(&nfit_memdev->list);
685 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
686 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
687 dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
688 __func__, memdev->device_handle, memdev->range_index,
689 memdev->region_index, memdev->flags);
690 return true;
691 }
692
693 /*
694 * An implementation may provide a truncated control region if no block windows
695 * are defined.
696 */
697 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
698 {
699 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
700 window_size))
701 return 0;
702 if (dcr->windows)
703 return sizeof(*dcr);
704 return offsetof(struct acpi_nfit_control_region, window_size);
705 }
706
707 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
708 struct nfit_table_prev *prev,
709 struct acpi_nfit_control_region *dcr)
710 {
711 struct device *dev = acpi_desc->dev;
712 struct nfit_dcr *nfit_dcr;
713
714 if (!sizeof_dcr(dcr))
715 return false;
716
717 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
718 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
719 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
720 return true;
721 }
722
723 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
724 GFP_KERNEL);
725 if (!nfit_dcr)
726 return false;
727 INIT_LIST_HEAD(&nfit_dcr->list);
728 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
729 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
730 dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
731 dcr->region_index, dcr->windows);
732 return true;
733 }
734
735 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
736 struct nfit_table_prev *prev,
737 struct acpi_nfit_data_region *bdw)
738 {
739 struct device *dev = acpi_desc->dev;
740 struct nfit_bdw *nfit_bdw;
741
742 if (bdw->header.length != sizeof(*bdw))
743 return false;
744 list_for_each_entry(nfit_bdw, &prev->bdws, list)
745 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
746 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
747 return true;
748 }
749
750 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
751 GFP_KERNEL);
752 if (!nfit_bdw)
753 return false;
754 INIT_LIST_HEAD(&nfit_bdw->list);
755 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
756 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
757 dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
758 bdw->region_index, bdw->windows);
759 return true;
760 }
761
762 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
763 {
764 if (idt->header.length < sizeof(*idt))
765 return 0;
766 return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
767 }
768
769 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
770 struct nfit_table_prev *prev,
771 struct acpi_nfit_interleave *idt)
772 {
773 struct device *dev = acpi_desc->dev;
774 struct nfit_idt *nfit_idt;
775
776 if (!sizeof_idt(idt))
777 return false;
778
779 list_for_each_entry(nfit_idt, &prev->idts, list) {
780 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
781 continue;
782
783 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
784 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
785 return true;
786 }
787 }
788
789 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
790 GFP_KERNEL);
791 if (!nfit_idt)
792 return false;
793 INIT_LIST_HEAD(&nfit_idt->list);
794 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
795 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
796 dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
797 idt->interleave_index, idt->line_count);
798 return true;
799 }
800
801 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
802 {
803 if (flush->header.length < sizeof(*flush))
804 return 0;
805 return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
806 }
807
808 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
809 struct nfit_table_prev *prev,
810 struct acpi_nfit_flush_address *flush)
811 {
812 struct device *dev = acpi_desc->dev;
813 struct nfit_flush *nfit_flush;
814
815 if (!sizeof_flush(flush))
816 return false;
817
818 list_for_each_entry(nfit_flush, &prev->flushes, list) {
819 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
820 continue;
821
822 if (memcmp(nfit_flush->flush, flush,
823 sizeof_flush(flush)) == 0) {
824 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
825 return true;
826 }
827 }
828
829 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
830 + sizeof_flush(flush), GFP_KERNEL);
831 if (!nfit_flush)
832 return false;
833 INIT_LIST_HEAD(&nfit_flush->list);
834 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
835 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
836 dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
837 flush->device_handle, flush->hint_count);
838 return true;
839 }
840
841 static void *add_table(struct acpi_nfit_desc *acpi_desc,
842 struct nfit_table_prev *prev, void *table, const void *end)
843 {
844 struct device *dev = acpi_desc->dev;
845 struct acpi_nfit_header *hdr;
846 void *err = ERR_PTR(-ENOMEM);
847
848 if (table >= end)
849 return NULL;
850
851 hdr = table;
852 if (!hdr->length) {
853 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
854 hdr->type);
855 return NULL;
856 }
857
858 switch (hdr->type) {
859 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
860 if (!add_spa(acpi_desc, prev, table))
861 return err;
862 break;
863 case ACPI_NFIT_TYPE_MEMORY_MAP:
864 if (!add_memdev(acpi_desc, prev, table))
865 return err;
866 break;
867 case ACPI_NFIT_TYPE_CONTROL_REGION:
868 if (!add_dcr(acpi_desc, prev, table))
869 return err;
870 break;
871 case ACPI_NFIT_TYPE_DATA_REGION:
872 if (!add_bdw(acpi_desc, prev, table))
873 return err;
874 break;
875 case ACPI_NFIT_TYPE_INTERLEAVE:
876 if (!add_idt(acpi_desc, prev, table))
877 return err;
878 break;
879 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
880 if (!add_flush(acpi_desc, prev, table))
881 return err;
882 break;
883 case ACPI_NFIT_TYPE_SMBIOS:
884 dev_dbg(dev, "%s: smbios\n", __func__);
885 break;
886 default:
887 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
888 break;
889 }
890
891 return table + hdr->length;
892 }
893
894 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
895 struct nfit_mem *nfit_mem)
896 {
897 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
898 u16 dcr = nfit_mem->dcr->region_index;
899 struct nfit_spa *nfit_spa;
900
901 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
902 u16 range_index = nfit_spa->spa->range_index;
903 int type = nfit_spa_type(nfit_spa->spa);
904 struct nfit_memdev *nfit_memdev;
905
906 if (type != NFIT_SPA_BDW)
907 continue;
908
909 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
910 if (nfit_memdev->memdev->range_index != range_index)
911 continue;
912 if (nfit_memdev->memdev->device_handle != device_handle)
913 continue;
914 if (nfit_memdev->memdev->region_index != dcr)
915 continue;
916
917 nfit_mem->spa_bdw = nfit_spa->spa;
918 return;
919 }
920 }
921
922 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
923 nfit_mem->spa_dcr->range_index);
924 nfit_mem->bdw = NULL;
925 }
926
927 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
928 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
929 {
930 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
931 struct nfit_memdev *nfit_memdev;
932 struct nfit_bdw *nfit_bdw;
933 struct nfit_idt *nfit_idt;
934 u16 idt_idx, range_index;
935
936 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
937 if (nfit_bdw->bdw->region_index != dcr)
938 continue;
939 nfit_mem->bdw = nfit_bdw->bdw;
940 break;
941 }
942
943 if (!nfit_mem->bdw)
944 return;
945
946 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
947
948 if (!nfit_mem->spa_bdw)
949 return;
950
951 range_index = nfit_mem->spa_bdw->range_index;
952 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
953 if (nfit_memdev->memdev->range_index != range_index ||
954 nfit_memdev->memdev->region_index != dcr)
955 continue;
956 nfit_mem->memdev_bdw = nfit_memdev->memdev;
957 idt_idx = nfit_memdev->memdev->interleave_index;
958 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
959 if (nfit_idt->idt->interleave_index != idt_idx)
960 continue;
961 nfit_mem->idt_bdw = nfit_idt->idt;
962 break;
963 }
964 break;
965 }
966 }
967
968 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
969 struct acpi_nfit_system_address *spa)
970 {
971 struct nfit_mem *nfit_mem, *found;
972 struct nfit_memdev *nfit_memdev;
973 int type = spa ? nfit_spa_type(spa) : 0;
974
975 switch (type) {
976 case NFIT_SPA_DCR:
977 case NFIT_SPA_PM:
978 break;
979 default:
980 if (spa)
981 return 0;
982 }
983
984 /*
985 * This loop runs in two modes, when a dimm is mapped the loop
986 * adds memdev associations to an existing dimm, or creates a
987 * dimm. In the unmapped dimm case this loop sweeps for memdev
988 * instances with an invalid / zero range_index and adds those
989 * dimms without spa associations.
990 */
991 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
992 struct nfit_flush *nfit_flush;
993 struct nfit_dcr *nfit_dcr;
994 u32 device_handle;
995 u16 dcr;
996
997 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
998 continue;
999 if (!spa && nfit_memdev->memdev->range_index)
1000 continue;
1001 found = NULL;
1002 dcr = nfit_memdev->memdev->region_index;
1003 device_handle = nfit_memdev->memdev->device_handle;
1004 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1005 if (__to_nfit_memdev(nfit_mem)->device_handle
1006 == device_handle) {
1007 found = nfit_mem;
1008 break;
1009 }
1010
1011 if (found)
1012 nfit_mem = found;
1013 else {
1014 nfit_mem = devm_kzalloc(acpi_desc->dev,
1015 sizeof(*nfit_mem), GFP_KERNEL);
1016 if (!nfit_mem)
1017 return -ENOMEM;
1018 INIT_LIST_HEAD(&nfit_mem->list);
1019 nfit_mem->acpi_desc = acpi_desc;
1020 list_add(&nfit_mem->list, &acpi_desc->dimms);
1021 }
1022
1023 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1024 if (nfit_dcr->dcr->region_index != dcr)
1025 continue;
1026 /*
1027 * Record the control region for the dimm. For
1028 * the ACPI 6.1 case, where there are separate
1029 * control regions for the pmem vs blk
1030 * interfaces, be sure to record the extended
1031 * blk details.
1032 */
1033 if (!nfit_mem->dcr)
1034 nfit_mem->dcr = nfit_dcr->dcr;
1035 else if (nfit_mem->dcr->windows == 0
1036 && nfit_dcr->dcr->windows)
1037 nfit_mem->dcr = nfit_dcr->dcr;
1038 break;
1039 }
1040
1041 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1042 struct acpi_nfit_flush_address *flush;
1043 u16 i;
1044
1045 if (nfit_flush->flush->device_handle != device_handle)
1046 continue;
1047 nfit_mem->nfit_flush = nfit_flush;
1048 flush = nfit_flush->flush;
1049 nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
1050 flush->hint_count
1051 * sizeof(struct resource), GFP_KERNEL);
1052 if (!nfit_mem->flush_wpq)
1053 return -ENOMEM;
1054 for (i = 0; i < flush->hint_count; i++) {
1055 struct resource *res = &nfit_mem->flush_wpq[i];
1056
1057 res->start = flush->hint_address[i];
1058 res->end = res->start + 8 - 1;
1059 }
1060 break;
1061 }
1062
1063 if (dcr && !nfit_mem->dcr) {
1064 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1065 spa->range_index, dcr);
1066 return -ENODEV;
1067 }
1068
1069 if (type == NFIT_SPA_DCR) {
1070 struct nfit_idt *nfit_idt;
1071 u16 idt_idx;
1072
1073 /* multiple dimms may share a SPA when interleaved */
1074 nfit_mem->spa_dcr = spa;
1075 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1076 idt_idx = nfit_memdev->memdev->interleave_index;
1077 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1078 if (nfit_idt->idt->interleave_index != idt_idx)
1079 continue;
1080 nfit_mem->idt_dcr = nfit_idt->idt;
1081 break;
1082 }
1083 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1084 } else if (type == NFIT_SPA_PM) {
1085 /*
1086 * A single dimm may belong to multiple SPA-PM
1087 * ranges, record at least one in addition to
1088 * any SPA-DCR range.
1089 */
1090 nfit_mem->memdev_pmem = nfit_memdev->memdev;
1091 } else
1092 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1093 }
1094
1095 return 0;
1096 }
1097
1098 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1099 {
1100 struct nfit_mem *a = container_of(_a, typeof(*a), list);
1101 struct nfit_mem *b = container_of(_b, typeof(*b), list);
1102 u32 handleA, handleB;
1103
1104 handleA = __to_nfit_memdev(a)->device_handle;
1105 handleB = __to_nfit_memdev(b)->device_handle;
1106 if (handleA < handleB)
1107 return -1;
1108 else if (handleA > handleB)
1109 return 1;
1110 return 0;
1111 }
1112
1113 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1114 {
1115 struct nfit_spa *nfit_spa;
1116 int rc;
1117
1118
1119 /*
1120 * For each SPA-DCR or SPA-PMEM address range find its
1121 * corresponding MEMDEV(s). From each MEMDEV find the
1122 * corresponding DCR. Then, if we're operating on a SPA-DCR,
1123 * try to find a SPA-BDW and a corresponding BDW that references
1124 * the DCR. Throw it all into an nfit_mem object. Note, that
1125 * BDWs are optional.
1126 */
1127 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1128 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1129 if (rc)
1130 return rc;
1131 }
1132
1133 /*
1134 * If a DIMM has failed to be mapped into SPA there will be no
1135 * SPA entries above. Find and register all the unmapped DIMMs
1136 * for reporting and recovery purposes.
1137 */
1138 rc = __nfit_mem_init(acpi_desc, NULL);
1139 if (rc)
1140 return rc;
1141
1142 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1143
1144 return 0;
1145 }
1146
1147 static ssize_t bus_dsm_mask_show(struct device *dev,
1148 struct device_attribute *attr, char *buf)
1149 {
1150 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1151 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1152
1153 return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1154 }
1155 static struct device_attribute dev_attr_bus_dsm_mask =
1156 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1157
1158 static ssize_t revision_show(struct device *dev,
1159 struct device_attribute *attr, char *buf)
1160 {
1161 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1162 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1163 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1164
1165 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1166 }
1167 static DEVICE_ATTR_RO(revision);
1168
1169 static ssize_t hw_error_scrub_show(struct device *dev,
1170 struct device_attribute *attr, char *buf)
1171 {
1172 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1173 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1174 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1175
1176 return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1177 }
1178
1179 /*
1180 * The 'hw_error_scrub' attribute can have the following values written to it:
1181 * '0': Switch to the default mode where an exception will only insert
1182 * the address of the memory error into the poison and badblocks lists.
1183 * '1': Enable a full scrub to happen if an exception for a memory error is
1184 * received.
1185 */
1186 static ssize_t hw_error_scrub_store(struct device *dev,
1187 struct device_attribute *attr, const char *buf, size_t size)
1188 {
1189 struct nvdimm_bus_descriptor *nd_desc;
1190 ssize_t rc;
1191 long val;
1192
1193 rc = kstrtol(buf, 0, &val);
1194 if (rc)
1195 return rc;
1196
1197 device_lock(dev);
1198 nd_desc = dev_get_drvdata(dev);
1199 if (nd_desc) {
1200 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1201
1202 switch (val) {
1203 case HW_ERROR_SCRUB_ON:
1204 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1205 break;
1206 case HW_ERROR_SCRUB_OFF:
1207 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1208 break;
1209 default:
1210 rc = -EINVAL;
1211 break;
1212 }
1213 }
1214 device_unlock(dev);
1215 if (rc)
1216 return rc;
1217 return size;
1218 }
1219 static DEVICE_ATTR_RW(hw_error_scrub);
1220
1221 /*
1222 * This shows the number of full Address Range Scrubs that have been
1223 * completed since driver load time. Userspace can wait on this using
1224 * select/poll etc. A '+' at the end indicates an ARS is in progress
1225 */
1226 static ssize_t scrub_show(struct device *dev,
1227 struct device_attribute *attr, char *buf)
1228 {
1229 struct nvdimm_bus_descriptor *nd_desc;
1230 ssize_t rc = -ENXIO;
1231
1232 device_lock(dev);
1233 nd_desc = dev_get_drvdata(dev);
1234 if (nd_desc) {
1235 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1236
1237 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1238 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
1239 }
1240 device_unlock(dev);
1241 return rc;
1242 }
1243
1244 static ssize_t scrub_store(struct device *dev,
1245 struct device_attribute *attr, const char *buf, size_t size)
1246 {
1247 struct nvdimm_bus_descriptor *nd_desc;
1248 ssize_t rc;
1249 long val;
1250
1251 rc = kstrtol(buf, 0, &val);
1252 if (rc)
1253 return rc;
1254 if (val != 1)
1255 return -EINVAL;
1256
1257 device_lock(dev);
1258 nd_desc = dev_get_drvdata(dev);
1259 if (nd_desc) {
1260 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1261
1262 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1263 }
1264 device_unlock(dev);
1265 if (rc)
1266 return rc;
1267 return size;
1268 }
1269 static DEVICE_ATTR_RW(scrub);
1270
1271 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1272 {
1273 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1274 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1275 | 1 << ND_CMD_ARS_STATUS;
1276
1277 return (nd_desc->cmd_mask & mask) == mask;
1278 }
1279
1280 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1281 {
1282 struct device *dev = container_of(kobj, struct device, kobj);
1283 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1284
1285 if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1286 return 0;
1287 return a->mode;
1288 }
1289
1290 static struct attribute *acpi_nfit_attributes[] = {
1291 &dev_attr_revision.attr,
1292 &dev_attr_scrub.attr,
1293 &dev_attr_hw_error_scrub.attr,
1294 &dev_attr_bus_dsm_mask.attr,
1295 NULL,
1296 };
1297
1298 static const struct attribute_group acpi_nfit_attribute_group = {
1299 .name = "nfit",
1300 .attrs = acpi_nfit_attributes,
1301 .is_visible = nfit_visible,
1302 };
1303
1304 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1305 &nvdimm_bus_attribute_group,
1306 &acpi_nfit_attribute_group,
1307 NULL,
1308 };
1309
1310 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1311 {
1312 struct nvdimm *nvdimm = to_nvdimm(dev);
1313 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1314
1315 return __to_nfit_memdev(nfit_mem);
1316 }
1317
1318 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1319 {
1320 struct nvdimm *nvdimm = to_nvdimm(dev);
1321 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1322
1323 return nfit_mem->dcr;
1324 }
1325
1326 static ssize_t handle_show(struct device *dev,
1327 struct device_attribute *attr, char *buf)
1328 {
1329 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1330
1331 return sprintf(buf, "%#x\n", memdev->device_handle);
1332 }
1333 static DEVICE_ATTR_RO(handle);
1334
1335 static ssize_t phys_id_show(struct device *dev,
1336 struct device_attribute *attr, char *buf)
1337 {
1338 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1339
1340 return sprintf(buf, "%#x\n", memdev->physical_id);
1341 }
1342 static DEVICE_ATTR_RO(phys_id);
1343
1344 static ssize_t vendor_show(struct device *dev,
1345 struct device_attribute *attr, char *buf)
1346 {
1347 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1348
1349 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1350 }
1351 static DEVICE_ATTR_RO(vendor);
1352
1353 static ssize_t rev_id_show(struct device *dev,
1354 struct device_attribute *attr, char *buf)
1355 {
1356 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1357
1358 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1359 }
1360 static DEVICE_ATTR_RO(rev_id);
1361
1362 static ssize_t device_show(struct device *dev,
1363 struct device_attribute *attr, char *buf)
1364 {
1365 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1366
1367 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1368 }
1369 static DEVICE_ATTR_RO(device);
1370
1371 static ssize_t subsystem_vendor_show(struct device *dev,
1372 struct device_attribute *attr, char *buf)
1373 {
1374 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1375
1376 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1377 }
1378 static DEVICE_ATTR_RO(subsystem_vendor);
1379
1380 static ssize_t subsystem_rev_id_show(struct device *dev,
1381 struct device_attribute *attr, char *buf)
1382 {
1383 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1384
1385 return sprintf(buf, "0x%04x\n",
1386 be16_to_cpu(dcr->subsystem_revision_id));
1387 }
1388 static DEVICE_ATTR_RO(subsystem_rev_id);
1389
1390 static ssize_t subsystem_device_show(struct device *dev,
1391 struct device_attribute *attr, char *buf)
1392 {
1393 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1394
1395 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1396 }
1397 static DEVICE_ATTR_RO(subsystem_device);
1398
1399 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1400 {
1401 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1402 int formats = 0;
1403
1404 if (nfit_mem->memdev_pmem)
1405 formats++;
1406 if (nfit_mem->memdev_bdw)
1407 formats++;
1408 return formats;
1409 }
1410
1411 static ssize_t format_show(struct device *dev,
1412 struct device_attribute *attr, char *buf)
1413 {
1414 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1415
1416 return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1417 }
1418 static DEVICE_ATTR_RO(format);
1419
1420 static ssize_t format1_show(struct device *dev,
1421 struct device_attribute *attr, char *buf)
1422 {
1423 u32 handle;
1424 ssize_t rc = -ENXIO;
1425 struct nfit_mem *nfit_mem;
1426 struct nfit_memdev *nfit_memdev;
1427 struct acpi_nfit_desc *acpi_desc;
1428 struct nvdimm *nvdimm = to_nvdimm(dev);
1429 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1430
1431 nfit_mem = nvdimm_provider_data(nvdimm);
1432 acpi_desc = nfit_mem->acpi_desc;
1433 handle = to_nfit_memdev(dev)->device_handle;
1434
1435 /* assumes DIMMs have at most 2 published interface codes */
1436 mutex_lock(&acpi_desc->init_mutex);
1437 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1438 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1439 struct nfit_dcr *nfit_dcr;
1440
1441 if (memdev->device_handle != handle)
1442 continue;
1443
1444 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1445 if (nfit_dcr->dcr->region_index != memdev->region_index)
1446 continue;
1447 if (nfit_dcr->dcr->code == dcr->code)
1448 continue;
1449 rc = sprintf(buf, "0x%04x\n",
1450 le16_to_cpu(nfit_dcr->dcr->code));
1451 break;
1452 }
1453 if (rc != ENXIO)
1454 break;
1455 }
1456 mutex_unlock(&acpi_desc->init_mutex);
1457 return rc;
1458 }
1459 static DEVICE_ATTR_RO(format1);
1460
1461 static ssize_t formats_show(struct device *dev,
1462 struct device_attribute *attr, char *buf)
1463 {
1464 struct nvdimm *nvdimm = to_nvdimm(dev);
1465
1466 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1467 }
1468 static DEVICE_ATTR_RO(formats);
1469
1470 static ssize_t serial_show(struct device *dev,
1471 struct device_attribute *attr, char *buf)
1472 {
1473 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1474
1475 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1476 }
1477 static DEVICE_ATTR_RO(serial);
1478
1479 static ssize_t family_show(struct device *dev,
1480 struct device_attribute *attr, char *buf)
1481 {
1482 struct nvdimm *nvdimm = to_nvdimm(dev);
1483 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1484
1485 if (nfit_mem->family < 0)
1486 return -ENXIO;
1487 return sprintf(buf, "%d\n", nfit_mem->family);
1488 }
1489 static DEVICE_ATTR_RO(family);
1490
1491 static ssize_t dsm_mask_show(struct device *dev,
1492 struct device_attribute *attr, char *buf)
1493 {
1494 struct nvdimm *nvdimm = to_nvdimm(dev);
1495 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1496
1497 if (nfit_mem->family < 0)
1498 return -ENXIO;
1499 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1500 }
1501 static DEVICE_ATTR_RO(dsm_mask);
1502
1503 static ssize_t flags_show(struct device *dev,
1504 struct device_attribute *attr, char *buf)
1505 {
1506 u16 flags = to_nfit_memdev(dev)->flags;
1507
1508 return sprintf(buf, "%s%s%s%s%s%s%s\n",
1509 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1510 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1511 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1512 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1513 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1514 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1515 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1516 }
1517 static DEVICE_ATTR_RO(flags);
1518
1519 static ssize_t id_show(struct device *dev,
1520 struct device_attribute *attr, char *buf)
1521 {
1522 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1523
1524 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1525 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1526 be16_to_cpu(dcr->vendor_id),
1527 dcr->manufacturing_location,
1528 be16_to_cpu(dcr->manufacturing_date),
1529 be32_to_cpu(dcr->serial_number));
1530 else
1531 return sprintf(buf, "%04x-%08x\n",
1532 be16_to_cpu(dcr->vendor_id),
1533 be32_to_cpu(dcr->serial_number));
1534 }
1535 static DEVICE_ATTR_RO(id);
1536
1537 static struct attribute *acpi_nfit_dimm_attributes[] = {
1538 &dev_attr_handle.attr,
1539 &dev_attr_phys_id.attr,
1540 &dev_attr_vendor.attr,
1541 &dev_attr_device.attr,
1542 &dev_attr_rev_id.attr,
1543 &dev_attr_subsystem_vendor.attr,
1544 &dev_attr_subsystem_device.attr,
1545 &dev_attr_subsystem_rev_id.attr,
1546 &dev_attr_format.attr,
1547 &dev_attr_formats.attr,
1548 &dev_attr_format1.attr,
1549 &dev_attr_serial.attr,
1550 &dev_attr_flags.attr,
1551 &dev_attr_id.attr,
1552 &dev_attr_family.attr,
1553 &dev_attr_dsm_mask.attr,
1554 NULL,
1555 };
1556
1557 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1558 struct attribute *a, int n)
1559 {
1560 struct device *dev = container_of(kobj, struct device, kobj);
1561 struct nvdimm *nvdimm = to_nvdimm(dev);
1562
1563 if (!to_nfit_dcr(dev)) {
1564 /* Without a dcr only the memdev attributes can be surfaced */
1565 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1566 || a == &dev_attr_flags.attr
1567 || a == &dev_attr_family.attr
1568 || a == &dev_attr_dsm_mask.attr)
1569 return a->mode;
1570 return 0;
1571 }
1572
1573 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1574 return 0;
1575 return a->mode;
1576 }
1577
1578 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1579 .name = "nfit",
1580 .attrs = acpi_nfit_dimm_attributes,
1581 .is_visible = acpi_nfit_dimm_attr_visible,
1582 };
1583
1584 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1585 &nvdimm_attribute_group,
1586 &nd_device_attribute_group,
1587 &acpi_nfit_dimm_attribute_group,
1588 NULL,
1589 };
1590
1591 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1592 u32 device_handle)
1593 {
1594 struct nfit_mem *nfit_mem;
1595
1596 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1597 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1598 return nfit_mem->nvdimm;
1599
1600 return NULL;
1601 }
1602
1603 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1604 {
1605 struct nfit_mem *nfit_mem;
1606 struct acpi_nfit_desc *acpi_desc;
1607
1608 dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1609 event);
1610
1611 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1612 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1613 event);
1614 return;
1615 }
1616
1617 acpi_desc = dev_get_drvdata(dev->parent);
1618 if (!acpi_desc)
1619 return;
1620
1621 /*
1622 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1623 * is still valid.
1624 */
1625 nfit_mem = dev_get_drvdata(dev);
1626 if (nfit_mem && nfit_mem->flags_attr)
1627 sysfs_notify_dirent(nfit_mem->flags_attr);
1628 }
1629 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1630
1631 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1632 {
1633 struct acpi_device *adev = data;
1634 struct device *dev = &adev->dev;
1635
1636 device_lock(dev->parent);
1637 __acpi_nvdimm_notify(dev, event);
1638 device_unlock(dev->parent);
1639 }
1640
1641 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1642 struct nfit_mem *nfit_mem, u32 device_handle)
1643 {
1644 struct acpi_device *adev, *adev_dimm;
1645 struct device *dev = acpi_desc->dev;
1646 union acpi_object *obj;
1647 unsigned long dsm_mask;
1648 const guid_t *guid;
1649 int i;
1650 int family = -1;
1651
1652 /* nfit test assumes 1:1 relationship between commands and dsms */
1653 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1654 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1655 adev = to_acpi_dev(acpi_desc);
1656 if (!adev)
1657 return 0;
1658
1659 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1660 nfit_mem->adev = adev_dimm;
1661 if (!adev_dimm) {
1662 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1663 device_handle);
1664 return force_enable_dimms ? 0 : -ENODEV;
1665 }
1666
1667 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1668 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1669 dev_err(dev, "%s: notification registration failed\n",
1670 dev_name(&adev_dimm->dev));
1671 return -ENXIO;
1672 }
1673
1674 /*
1675 * Until standardization materializes we need to consider 4
1676 * different command sets. Note, that checking for function0 (bit0)
1677 * tells us if any commands are reachable through this GUID.
1678 */
1679 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1680 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1681 if (family < 0 || i == default_dsm_family)
1682 family = i;
1683
1684 /* limit the supported commands to those that are publicly documented */
1685 nfit_mem->family = family;
1686 if (override_dsm_mask && !disable_vendor_specific)
1687 dsm_mask = override_dsm_mask;
1688 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1689 dsm_mask = NVDIMM_INTEL_CMDMASK;
1690 if (disable_vendor_specific)
1691 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1692 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1693 dsm_mask = 0x1c3c76;
1694 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1695 dsm_mask = 0x1fe;
1696 if (disable_vendor_specific)
1697 dsm_mask &= ~(1 << 8);
1698 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1699 dsm_mask = 0xffffffff;
1700 } else {
1701 dev_dbg(dev, "unknown dimm command family\n");
1702 nfit_mem->family = -1;
1703 /* DSMs are optional, continue loading the driver... */
1704 return 0;
1705 }
1706
1707 guid = to_nfit_uuid(nfit_mem->family);
1708 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1709 if (acpi_check_dsm(adev_dimm->handle, guid,
1710 nfit_dsm_revid(nfit_mem->family, i),
1711 1ULL << i))
1712 set_bit(i, &nfit_mem->dsm_mask);
1713
1714 obj = acpi_label_info(adev_dimm->handle);
1715 if (obj) {
1716 ACPI_FREE(obj);
1717 nfit_mem->has_lsi = 1;
1718 dev_dbg(dev, "%s: has _LSI\n", dev_name(&adev_dimm->dev));
1719 }
1720
1721 obj = acpi_label_read(adev_dimm->handle, 0, 0);
1722 if (obj) {
1723 ACPI_FREE(obj);
1724 nfit_mem->has_lsr = 1;
1725 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1726 }
1727
1728 obj = acpi_label_write(adev_dimm->handle, 0, 0, NULL);
1729 if (obj) {
1730 ACPI_FREE(obj);
1731 nfit_mem->has_lsw = 1;
1732 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1733 }
1734
1735 return 0;
1736 }
1737
1738 static void shutdown_dimm_notify(void *data)
1739 {
1740 struct acpi_nfit_desc *acpi_desc = data;
1741 struct nfit_mem *nfit_mem;
1742
1743 mutex_lock(&acpi_desc->init_mutex);
1744 /*
1745 * Clear out the nfit_mem->flags_attr and shut down dimm event
1746 * notifications.
1747 */
1748 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1749 struct acpi_device *adev_dimm = nfit_mem->adev;
1750
1751 if (nfit_mem->flags_attr) {
1752 sysfs_put(nfit_mem->flags_attr);
1753 nfit_mem->flags_attr = NULL;
1754 }
1755 if (adev_dimm)
1756 acpi_remove_notify_handler(adev_dimm->handle,
1757 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1758 }
1759 mutex_unlock(&acpi_desc->init_mutex);
1760 }
1761
1762 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1763 {
1764 struct nfit_mem *nfit_mem;
1765 int dimm_count = 0, rc;
1766 struct nvdimm *nvdimm;
1767
1768 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1769 struct acpi_nfit_flush_address *flush;
1770 unsigned long flags = 0, cmd_mask;
1771 struct nfit_memdev *nfit_memdev;
1772 u32 device_handle;
1773 u16 mem_flags;
1774
1775 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1776 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1777 if (nvdimm) {
1778 dimm_count++;
1779 continue;
1780 }
1781
1782 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1783 set_bit(NDD_ALIASING, &flags);
1784
1785 /* collate flags across all memdevs for this dimm */
1786 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1787 struct acpi_nfit_memory_map *dimm_memdev;
1788
1789 dimm_memdev = __to_nfit_memdev(nfit_mem);
1790 if (dimm_memdev->device_handle
1791 != nfit_memdev->memdev->device_handle)
1792 continue;
1793 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1794 }
1795
1796 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1797 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1798 set_bit(NDD_UNARMED, &flags);
1799
1800 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1801 if (rc)
1802 continue;
1803
1804 /*
1805 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1806 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1807 * userspace interface.
1808 */
1809 cmd_mask = 1UL << ND_CMD_CALL;
1810 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1811 /*
1812 * These commands have a 1:1 correspondence
1813 * between DSM payload and libnvdimm ioctl
1814 * payload format.
1815 */
1816 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1817 }
1818
1819 if (nfit_mem->has_lsi)
1820 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1821 if (nfit_mem->has_lsr)
1822 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1823 if (nfit_mem->has_lsw)
1824 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1825
1826 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1827 : NULL;
1828 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1829 acpi_nfit_dimm_attribute_groups,
1830 flags, cmd_mask, flush ? flush->hint_count : 0,
1831 nfit_mem->flush_wpq);
1832 if (!nvdimm)
1833 return -ENOMEM;
1834
1835 nfit_mem->nvdimm = nvdimm;
1836 dimm_count++;
1837
1838 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1839 continue;
1840
1841 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1842 nvdimm_name(nvdimm),
1843 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1844 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1845 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1846 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1847 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1848
1849 }
1850
1851 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1852 if (rc)
1853 return rc;
1854
1855 /*
1856 * Now that dimms are successfully registered, and async registration
1857 * is flushed, attempt to enable event notification.
1858 */
1859 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1860 struct kernfs_node *nfit_kernfs;
1861
1862 nvdimm = nfit_mem->nvdimm;
1863 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1864 if (nfit_kernfs)
1865 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1866 "flags");
1867 sysfs_put(nfit_kernfs);
1868 if (!nfit_mem->flags_attr)
1869 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1870 nvdimm_name(nvdimm));
1871 }
1872
1873 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1874 acpi_desc);
1875 }
1876
1877 /*
1878 * These constants are private because there are no kernel consumers of
1879 * these commands.
1880 */
1881 enum nfit_aux_cmds {
1882 NFIT_CMD_TRANSLATE_SPA = 5,
1883 NFIT_CMD_ARS_INJECT_SET = 7,
1884 NFIT_CMD_ARS_INJECT_CLEAR = 8,
1885 NFIT_CMD_ARS_INJECT_GET = 9,
1886 };
1887
1888 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1889 {
1890 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1891 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1892 struct acpi_device *adev;
1893 unsigned long dsm_mask;
1894 int i;
1895
1896 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1897 nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1898 adev = to_acpi_dev(acpi_desc);
1899 if (!adev)
1900 return;
1901
1902 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1903 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1904 set_bit(i, &nd_desc->cmd_mask);
1905 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1906
1907 dsm_mask =
1908 (1 << ND_CMD_ARS_CAP) |
1909 (1 << ND_CMD_ARS_START) |
1910 (1 << ND_CMD_ARS_STATUS) |
1911 (1 << ND_CMD_CLEAR_ERROR) |
1912 (1 << NFIT_CMD_TRANSLATE_SPA) |
1913 (1 << NFIT_CMD_ARS_INJECT_SET) |
1914 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1915 (1 << NFIT_CMD_ARS_INJECT_GET);
1916 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1917 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1918 set_bit(i, &nd_desc->bus_dsm_mask);
1919 }
1920
1921 static ssize_t range_index_show(struct device *dev,
1922 struct device_attribute *attr, char *buf)
1923 {
1924 struct nd_region *nd_region = to_nd_region(dev);
1925 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1926
1927 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1928 }
1929 static DEVICE_ATTR_RO(range_index);
1930
1931 static ssize_t ecc_unit_size_show(struct device *dev,
1932 struct device_attribute *attr, char *buf)
1933 {
1934 struct nd_region *nd_region = to_nd_region(dev);
1935 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1936
1937 return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1938 }
1939 static DEVICE_ATTR_RO(ecc_unit_size);
1940
1941 static struct attribute *acpi_nfit_region_attributes[] = {
1942 &dev_attr_range_index.attr,
1943 &dev_attr_ecc_unit_size.attr,
1944 NULL,
1945 };
1946
1947 static const struct attribute_group acpi_nfit_region_attribute_group = {
1948 .name = "nfit",
1949 .attrs = acpi_nfit_region_attributes,
1950 };
1951
1952 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1953 &nd_region_attribute_group,
1954 &nd_mapping_attribute_group,
1955 &nd_device_attribute_group,
1956 &nd_numa_attribute_group,
1957 &acpi_nfit_region_attribute_group,
1958 NULL,
1959 };
1960
1961 /* enough info to uniquely specify an interleave set */
1962 struct nfit_set_info {
1963 struct nfit_set_info_map {
1964 u64 region_offset;
1965 u32 serial_number;
1966 u32 pad;
1967 } mapping[0];
1968 };
1969
1970 struct nfit_set_info2 {
1971 struct nfit_set_info_map2 {
1972 u64 region_offset;
1973 u32 serial_number;
1974 u16 vendor_id;
1975 u16 manufacturing_date;
1976 u8 manufacturing_location;
1977 u8 reserved[31];
1978 } mapping[0];
1979 };
1980
1981 static size_t sizeof_nfit_set_info(int num_mappings)
1982 {
1983 return sizeof(struct nfit_set_info)
1984 + num_mappings * sizeof(struct nfit_set_info_map);
1985 }
1986
1987 static size_t sizeof_nfit_set_info2(int num_mappings)
1988 {
1989 return sizeof(struct nfit_set_info2)
1990 + num_mappings * sizeof(struct nfit_set_info_map2);
1991 }
1992
1993 static int cmp_map_compat(const void *m0, const void *m1)
1994 {
1995 const struct nfit_set_info_map *map0 = m0;
1996 const struct nfit_set_info_map *map1 = m1;
1997
1998 return memcmp(&map0->region_offset, &map1->region_offset,
1999 sizeof(u64));
2000 }
2001
2002 static int cmp_map(const void *m0, const void *m1)
2003 {
2004 const struct nfit_set_info_map *map0 = m0;
2005 const struct nfit_set_info_map *map1 = m1;
2006
2007 if (map0->region_offset < map1->region_offset)
2008 return -1;
2009 else if (map0->region_offset > map1->region_offset)
2010 return 1;
2011 return 0;
2012 }
2013
2014 static int cmp_map2(const void *m0, const void *m1)
2015 {
2016 const struct nfit_set_info_map2 *map0 = m0;
2017 const struct nfit_set_info_map2 *map1 = m1;
2018
2019 if (map0->region_offset < map1->region_offset)
2020 return -1;
2021 else if (map0->region_offset > map1->region_offset)
2022 return 1;
2023 return 0;
2024 }
2025
2026 /* Retrieve the nth entry referencing this spa */
2027 static struct acpi_nfit_memory_map *memdev_from_spa(
2028 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2029 {
2030 struct nfit_memdev *nfit_memdev;
2031
2032 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2033 if (nfit_memdev->memdev->range_index == range_index)
2034 if (n-- == 0)
2035 return nfit_memdev->memdev;
2036 return NULL;
2037 }
2038
2039 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2040 struct nd_region_desc *ndr_desc,
2041 struct acpi_nfit_system_address *spa)
2042 {
2043 struct device *dev = acpi_desc->dev;
2044 struct nd_interleave_set *nd_set;
2045 u16 nr = ndr_desc->num_mappings;
2046 struct nfit_set_info2 *info2;
2047 struct nfit_set_info *info;
2048 int i;
2049
2050 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2051 if (!nd_set)
2052 return -ENOMEM;
2053 ndr_desc->nd_set = nd_set;
2054 guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2055
2056 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2057 if (!info)
2058 return -ENOMEM;
2059
2060 info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2061 if (!info2)
2062 return -ENOMEM;
2063
2064 for (i = 0; i < nr; i++) {
2065 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2066 struct nfit_set_info_map *map = &info->mapping[i];
2067 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2068 struct nvdimm *nvdimm = mapping->nvdimm;
2069 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2070 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2071 spa->range_index, i);
2072 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2073
2074 if (!memdev || !nfit_mem->dcr) {
2075 dev_err(dev, "%s: failed to find DCR\n", __func__);
2076 return -ENODEV;
2077 }
2078
2079 map->region_offset = memdev->region_offset;
2080 map->serial_number = dcr->serial_number;
2081
2082 map2->region_offset = memdev->region_offset;
2083 map2->serial_number = dcr->serial_number;
2084 map2->vendor_id = dcr->vendor_id;
2085 map2->manufacturing_date = dcr->manufacturing_date;
2086 map2->manufacturing_location = dcr->manufacturing_location;
2087 }
2088
2089 /* v1.1 namespaces */
2090 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2091 cmp_map, NULL);
2092 nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2093
2094 /* v1.2 namespaces */
2095 sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2096 cmp_map2, NULL);
2097 nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2098
2099 /* support v1.1 namespaces created with the wrong sort order */
2100 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2101 cmp_map_compat, NULL);
2102 nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2103
2104 /* record the result of the sort for the mapping position */
2105 for (i = 0; i < nr; i++) {
2106 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2107 int j;
2108
2109 for (j = 0; j < nr; j++) {
2110 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2111 struct nvdimm *nvdimm = mapping->nvdimm;
2112 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2113 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2114
2115 if (map2->serial_number == dcr->serial_number &&
2116 map2->vendor_id == dcr->vendor_id &&
2117 map2->manufacturing_date == dcr->manufacturing_date &&
2118 map2->manufacturing_location
2119 == dcr->manufacturing_location) {
2120 mapping->position = i;
2121 break;
2122 }
2123 }
2124 }
2125
2126 ndr_desc->nd_set = nd_set;
2127 devm_kfree(dev, info);
2128 devm_kfree(dev, info2);
2129
2130 return 0;
2131 }
2132
2133 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2134 {
2135 struct acpi_nfit_interleave *idt = mmio->idt;
2136 u32 sub_line_offset, line_index, line_offset;
2137 u64 line_no, table_skip_count, table_offset;
2138
2139 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2140 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2141 line_offset = idt->line_offset[line_index]
2142 * mmio->line_size;
2143 table_offset = table_skip_count * mmio->table_size;
2144
2145 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2146 }
2147
2148 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2149 {
2150 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2151 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2152 const u32 STATUS_MASK = 0x80000037;
2153
2154 if (mmio->num_lines)
2155 offset = to_interleave_offset(offset, mmio);
2156
2157 return readl(mmio->addr.base + offset) & STATUS_MASK;
2158 }
2159
2160 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2161 resource_size_t dpa, unsigned int len, unsigned int write)
2162 {
2163 u64 cmd, offset;
2164 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2165
2166 enum {
2167 BCW_OFFSET_MASK = (1ULL << 48)-1,
2168 BCW_LEN_SHIFT = 48,
2169 BCW_LEN_MASK = (1ULL << 8) - 1,
2170 BCW_CMD_SHIFT = 56,
2171 };
2172
2173 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2174 len = len >> L1_CACHE_SHIFT;
2175 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2176 cmd |= ((u64) write) << BCW_CMD_SHIFT;
2177
2178 offset = nfit_blk->cmd_offset + mmio->size * bw;
2179 if (mmio->num_lines)
2180 offset = to_interleave_offset(offset, mmio);
2181
2182 writeq(cmd, mmio->addr.base + offset);
2183 nvdimm_flush(nfit_blk->nd_region);
2184
2185 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2186 readq(mmio->addr.base + offset);
2187 }
2188
2189 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2190 resource_size_t dpa, void *iobuf, size_t len, int rw,
2191 unsigned int lane)
2192 {
2193 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2194 unsigned int copied = 0;
2195 u64 base_offset;
2196 int rc;
2197
2198 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2199 + lane * mmio->size;
2200 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2201 while (len) {
2202 unsigned int c;
2203 u64 offset;
2204
2205 if (mmio->num_lines) {
2206 u32 line_offset;
2207
2208 offset = to_interleave_offset(base_offset + copied,
2209 mmio);
2210 div_u64_rem(offset, mmio->line_size, &line_offset);
2211 c = min_t(size_t, len, mmio->line_size - line_offset);
2212 } else {
2213 offset = base_offset + nfit_blk->bdw_offset;
2214 c = len;
2215 }
2216
2217 if (rw)
2218 memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2219 else {
2220 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2221 arch_invalidate_pmem((void __force *)
2222 mmio->addr.aperture + offset, c);
2223
2224 memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2225 }
2226
2227 copied += c;
2228 len -= c;
2229 }
2230
2231 if (rw)
2232 nvdimm_flush(nfit_blk->nd_region);
2233
2234 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2235 return rc;
2236 }
2237
2238 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2239 resource_size_t dpa, void *iobuf, u64 len, int rw)
2240 {
2241 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2242 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2243 struct nd_region *nd_region = nfit_blk->nd_region;
2244 unsigned int lane, copied = 0;
2245 int rc = 0;
2246
2247 lane = nd_region_acquire_lane(nd_region);
2248 while (len) {
2249 u64 c = min(len, mmio->size);
2250
2251 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2252 iobuf + copied, c, rw, lane);
2253 if (rc)
2254 break;
2255
2256 copied += c;
2257 len -= c;
2258 }
2259 nd_region_release_lane(nd_region, lane);
2260
2261 return rc;
2262 }
2263
2264 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2265 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2266 {
2267 if (idt) {
2268 mmio->num_lines = idt->line_count;
2269 mmio->line_size = idt->line_size;
2270 if (interleave_ways == 0)
2271 return -ENXIO;
2272 mmio->table_size = mmio->num_lines * interleave_ways
2273 * mmio->line_size;
2274 }
2275
2276 return 0;
2277 }
2278
2279 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2280 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2281 {
2282 struct nd_cmd_dimm_flags flags;
2283 int rc;
2284
2285 memset(&flags, 0, sizeof(flags));
2286 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2287 sizeof(flags), NULL);
2288
2289 if (rc >= 0 && flags.status == 0)
2290 nfit_blk->dimm_flags = flags.flags;
2291 else if (rc == -ENOTTY) {
2292 /* fall back to a conservative default */
2293 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2294 rc = 0;
2295 } else
2296 rc = -ENXIO;
2297
2298 return rc;
2299 }
2300
2301 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2302 struct device *dev)
2303 {
2304 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2305 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2306 struct nfit_blk_mmio *mmio;
2307 struct nfit_blk *nfit_blk;
2308 struct nfit_mem *nfit_mem;
2309 struct nvdimm *nvdimm;
2310 int rc;
2311
2312 nvdimm = nd_blk_region_to_dimm(ndbr);
2313 nfit_mem = nvdimm_provider_data(nvdimm);
2314 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2315 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2316 nfit_mem ? "" : " nfit_mem",
2317 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2318 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2319 return -ENXIO;
2320 }
2321
2322 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2323 if (!nfit_blk)
2324 return -ENOMEM;
2325 nd_blk_region_set_provider_data(ndbr, nfit_blk);
2326 nfit_blk->nd_region = to_nd_region(dev);
2327
2328 /* map block aperture memory */
2329 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2330 mmio = &nfit_blk->mmio[BDW];
2331 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2332 nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2333 if (!mmio->addr.base) {
2334 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2335 nvdimm_name(nvdimm));
2336 return -ENOMEM;
2337 }
2338 mmio->size = nfit_mem->bdw->size;
2339 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2340 mmio->idt = nfit_mem->idt_bdw;
2341 mmio->spa = nfit_mem->spa_bdw;
2342 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2343 nfit_mem->memdev_bdw->interleave_ways);
2344 if (rc) {
2345 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2346 __func__, nvdimm_name(nvdimm));
2347 return rc;
2348 }
2349
2350 /* map block control memory */
2351 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2352 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2353 mmio = &nfit_blk->mmio[DCR];
2354 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2355 nfit_mem->spa_dcr->length);
2356 if (!mmio->addr.base) {
2357 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2358 nvdimm_name(nvdimm));
2359 return -ENOMEM;
2360 }
2361 mmio->size = nfit_mem->dcr->window_size;
2362 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2363 mmio->idt = nfit_mem->idt_dcr;
2364 mmio->spa = nfit_mem->spa_dcr;
2365 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2366 nfit_mem->memdev_dcr->interleave_ways);
2367 if (rc) {
2368 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2369 __func__, nvdimm_name(nvdimm));
2370 return rc;
2371 }
2372
2373 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2374 if (rc < 0) {
2375 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2376 __func__, nvdimm_name(nvdimm));
2377 return rc;
2378 }
2379
2380 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2381 dev_warn(dev, "unable to guarantee persistence of writes\n");
2382
2383 if (mmio->line_size == 0)
2384 return 0;
2385
2386 if ((u32) nfit_blk->cmd_offset % mmio->line_size
2387 + 8 > mmio->line_size) {
2388 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2389 return -ENXIO;
2390 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2391 + 8 > mmio->line_size) {
2392 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2393 return -ENXIO;
2394 }
2395
2396 return 0;
2397 }
2398
2399 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2400 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2401 {
2402 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2403 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2404 int cmd_rc, rc;
2405
2406 cmd->address = spa->address;
2407 cmd->length = spa->length;
2408 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2409 sizeof(*cmd), &cmd_rc);
2410 if (rc < 0)
2411 return rc;
2412 return cmd_rc;
2413 }
2414
2415 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2416 {
2417 int rc;
2418 int cmd_rc;
2419 struct nd_cmd_ars_start ars_start;
2420 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2421 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2422
2423 memset(&ars_start, 0, sizeof(ars_start));
2424 ars_start.address = spa->address;
2425 ars_start.length = spa->length;
2426 ars_start.flags = acpi_desc->ars_start_flags;
2427 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2428 ars_start.type = ND_ARS_PERSISTENT;
2429 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2430 ars_start.type = ND_ARS_VOLATILE;
2431 else
2432 return -ENOTTY;
2433
2434 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2435 sizeof(ars_start), &cmd_rc);
2436
2437 if (rc < 0)
2438 return rc;
2439 return cmd_rc;
2440 }
2441
2442 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2443 {
2444 int rc, cmd_rc;
2445 struct nd_cmd_ars_start ars_start;
2446 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2447 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2448
2449 memset(&ars_start, 0, sizeof(ars_start));
2450 ars_start.address = ars_status->restart_address;
2451 ars_start.length = ars_status->restart_length;
2452 ars_start.type = ars_status->type;
2453 ars_start.flags = acpi_desc->ars_start_flags;
2454 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2455 sizeof(ars_start), &cmd_rc);
2456 if (rc < 0)
2457 return rc;
2458 return cmd_rc;
2459 }
2460
2461 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2462 {
2463 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2464 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2465 int rc, cmd_rc;
2466
2467 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2468 acpi_desc->ars_status_size, &cmd_rc);
2469 if (rc < 0)
2470 return rc;
2471 return cmd_rc;
2472 }
2473
2474 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2475 struct nd_cmd_ars_status *ars_status)
2476 {
2477 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2478 int rc;
2479 u32 i;
2480
2481 /*
2482 * First record starts at 44 byte offset from the start of the
2483 * payload.
2484 */
2485 if (ars_status->out_length < 44)
2486 return 0;
2487 for (i = 0; i < ars_status->num_records; i++) {
2488 /* only process full records */
2489 if (ars_status->out_length
2490 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2491 break;
2492 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2493 ars_status->records[i].err_address,
2494 ars_status->records[i].length);
2495 if (rc)
2496 return rc;
2497 }
2498 if (i < ars_status->num_records)
2499 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2500
2501 return 0;
2502 }
2503
2504 static void acpi_nfit_remove_resource(void *data)
2505 {
2506 struct resource *res = data;
2507
2508 remove_resource(res);
2509 }
2510
2511 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2512 struct nd_region_desc *ndr_desc)
2513 {
2514 struct resource *res, *nd_res = ndr_desc->res;
2515 int is_pmem, ret;
2516
2517 /* No operation if the region is already registered as PMEM */
2518 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2519 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2520 if (is_pmem == REGION_INTERSECTS)
2521 return 0;
2522
2523 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2524 if (!res)
2525 return -ENOMEM;
2526
2527 res->name = "Persistent Memory";
2528 res->start = nd_res->start;
2529 res->end = nd_res->end;
2530 res->flags = IORESOURCE_MEM;
2531 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2532
2533 ret = insert_resource(&iomem_resource, res);
2534 if (ret)
2535 return ret;
2536
2537 ret = devm_add_action_or_reset(acpi_desc->dev,
2538 acpi_nfit_remove_resource,
2539 res);
2540 if (ret)
2541 return ret;
2542
2543 return 0;
2544 }
2545
2546 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2547 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2548 struct acpi_nfit_memory_map *memdev,
2549 struct nfit_spa *nfit_spa)
2550 {
2551 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2552 memdev->device_handle);
2553 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2554 struct nd_blk_region_desc *ndbr_desc;
2555 struct nfit_mem *nfit_mem;
2556 int blk_valid = 0, rc;
2557
2558 if (!nvdimm) {
2559 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2560 spa->range_index, memdev->device_handle);
2561 return -ENODEV;
2562 }
2563
2564 mapping->nvdimm = nvdimm;
2565 switch (nfit_spa_type(spa)) {
2566 case NFIT_SPA_PM:
2567 case NFIT_SPA_VOLATILE:
2568 mapping->start = memdev->address;
2569 mapping->size = memdev->region_size;
2570 break;
2571 case NFIT_SPA_DCR:
2572 nfit_mem = nvdimm_provider_data(nvdimm);
2573 if (!nfit_mem || !nfit_mem->bdw) {
2574 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2575 spa->range_index, nvdimm_name(nvdimm));
2576 } else {
2577 mapping->size = nfit_mem->bdw->capacity;
2578 mapping->start = nfit_mem->bdw->start_address;
2579 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2580 blk_valid = 1;
2581 }
2582
2583 ndr_desc->mapping = mapping;
2584 ndr_desc->num_mappings = blk_valid;
2585 ndbr_desc = to_blk_region_desc(ndr_desc);
2586 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2587 ndbr_desc->do_io = acpi_desc->blk_do_io;
2588 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2589 if (rc)
2590 return rc;
2591 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2592 ndr_desc);
2593 if (!nfit_spa->nd_region)
2594 return -ENOMEM;
2595 break;
2596 }
2597
2598 return 0;
2599 }
2600
2601 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2602 {
2603 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2604 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2605 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2606 nfit_spa_type(spa) == NFIT_SPA_PCD);
2607 }
2608
2609 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2610 {
2611 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2612 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2613 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2614 }
2615
2616 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2617 struct nfit_spa *nfit_spa)
2618 {
2619 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2620 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2621 struct nd_blk_region_desc ndbr_desc;
2622 struct nd_region_desc *ndr_desc;
2623 struct nfit_memdev *nfit_memdev;
2624 struct nvdimm_bus *nvdimm_bus;
2625 struct resource res;
2626 int count = 0, rc;
2627
2628 if (nfit_spa->nd_region)
2629 return 0;
2630
2631 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2632 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2633 __func__);
2634 return 0;
2635 }
2636
2637 memset(&res, 0, sizeof(res));
2638 memset(&mappings, 0, sizeof(mappings));
2639 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2640 res.start = spa->address;
2641 res.end = res.start + spa->length - 1;
2642 ndr_desc = &ndbr_desc.ndr_desc;
2643 ndr_desc->res = &res;
2644 ndr_desc->provider_data = nfit_spa;
2645 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2646 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2647 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2648 spa->proximity_domain);
2649 else
2650 ndr_desc->numa_node = NUMA_NO_NODE;
2651
2652 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2653 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2654 struct nd_mapping_desc *mapping;
2655
2656 if (memdev->range_index != spa->range_index)
2657 continue;
2658 if (count >= ND_MAX_MAPPINGS) {
2659 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2660 spa->range_index, ND_MAX_MAPPINGS);
2661 return -ENXIO;
2662 }
2663 mapping = &mappings[count++];
2664 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2665 memdev, nfit_spa);
2666 if (rc)
2667 goto out;
2668 }
2669
2670 ndr_desc->mapping = mappings;
2671 ndr_desc->num_mappings = count;
2672 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2673 if (rc)
2674 goto out;
2675
2676 nvdimm_bus = acpi_desc->nvdimm_bus;
2677 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2678 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2679 if (rc) {
2680 dev_warn(acpi_desc->dev,
2681 "failed to insert pmem resource to iomem: %d\n",
2682 rc);
2683 goto out;
2684 }
2685
2686 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2687 ndr_desc);
2688 if (!nfit_spa->nd_region)
2689 rc = -ENOMEM;
2690 } else if (nfit_spa_is_volatile(spa)) {
2691 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2692 ndr_desc);
2693 if (!nfit_spa->nd_region)
2694 rc = -ENOMEM;
2695 } else if (nfit_spa_is_virtual(spa)) {
2696 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2697 ndr_desc);
2698 if (!nfit_spa->nd_region)
2699 rc = -ENOMEM;
2700 }
2701
2702 out:
2703 if (rc)
2704 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2705 nfit_spa->spa->range_index);
2706 return rc;
2707 }
2708
2709 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2710 u32 max_ars)
2711 {
2712 struct device *dev = acpi_desc->dev;
2713 struct nd_cmd_ars_status *ars_status;
2714
2715 if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2716 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2717 return 0;
2718 }
2719
2720 if (acpi_desc->ars_status)
2721 devm_kfree(dev, acpi_desc->ars_status);
2722 acpi_desc->ars_status = NULL;
2723 ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2724 if (!ars_status)
2725 return -ENOMEM;
2726 acpi_desc->ars_status = ars_status;
2727 acpi_desc->ars_status_size = max_ars;
2728 return 0;
2729 }
2730
2731 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2732 struct nfit_spa *nfit_spa)
2733 {
2734 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2735 int rc;
2736
2737 if (!nfit_spa->max_ars) {
2738 struct nd_cmd_ars_cap ars_cap;
2739
2740 memset(&ars_cap, 0, sizeof(ars_cap));
2741 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2742 if (rc < 0)
2743 return rc;
2744 nfit_spa->max_ars = ars_cap.max_ars_out;
2745 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2746 /* check that the supported scrub types match the spa type */
2747 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2748 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2749 return -ENOTTY;
2750 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2751 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2752 return -ENOTTY;
2753 }
2754
2755 if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2756 return -ENOMEM;
2757
2758 rc = ars_get_status(acpi_desc);
2759 if (rc < 0 && rc != -ENOSPC)
2760 return rc;
2761
2762 if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2763 return -ENOMEM;
2764
2765 return 0;
2766 }
2767
2768 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2769 struct nfit_spa *nfit_spa)
2770 {
2771 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2772 unsigned int overflow_retry = scrub_overflow_abort;
2773 u64 init_ars_start = 0, init_ars_len = 0;
2774 struct device *dev = acpi_desc->dev;
2775 unsigned int tmo = scrub_timeout;
2776 int rc;
2777
2778 if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2779 return;
2780
2781 rc = ars_start(acpi_desc, nfit_spa);
2782 /*
2783 * If we timed out the initial scan we'll still be busy here,
2784 * and will wait another timeout before giving up permanently.
2785 */
2786 if (rc < 0 && rc != -EBUSY)
2787 return;
2788
2789 do {
2790 u64 ars_start, ars_len;
2791
2792 if (acpi_desc->cancel)
2793 break;
2794 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2795 if (rc == -ENOTTY)
2796 break;
2797 if (rc == -EBUSY && !tmo) {
2798 dev_warn(dev, "range %d ars timeout, aborting\n",
2799 spa->range_index);
2800 break;
2801 }
2802
2803 if (rc == -EBUSY) {
2804 /*
2805 * Note, entries may be appended to the list
2806 * while the lock is dropped, but the workqueue
2807 * being active prevents entries being deleted /
2808 * freed.
2809 */
2810 mutex_unlock(&acpi_desc->init_mutex);
2811 ssleep(1);
2812 tmo--;
2813 mutex_lock(&acpi_desc->init_mutex);
2814 continue;
2815 }
2816
2817 /* we got some results, but there are more pending... */
2818 if (rc == -ENOSPC && overflow_retry--) {
2819 if (!init_ars_len) {
2820 init_ars_len = acpi_desc->ars_status->length;
2821 init_ars_start = acpi_desc->ars_status->address;
2822 }
2823 rc = ars_continue(acpi_desc);
2824 }
2825
2826 if (rc < 0) {
2827 dev_warn(dev, "range %d ars continuation failed\n",
2828 spa->range_index);
2829 break;
2830 }
2831
2832 if (init_ars_len) {
2833 ars_start = init_ars_start;
2834 ars_len = init_ars_len;
2835 } else {
2836 ars_start = acpi_desc->ars_status->address;
2837 ars_len = acpi_desc->ars_status->length;
2838 }
2839 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2840 spa->range_index, ars_start, ars_len);
2841 /* notify the region about new poison entries */
2842 nvdimm_region_notify(nfit_spa->nd_region,
2843 NVDIMM_REVALIDATE_POISON);
2844 break;
2845 } while (1);
2846 }
2847
2848 static void acpi_nfit_scrub(struct work_struct *work)
2849 {
2850 struct device *dev;
2851 u64 init_scrub_length = 0;
2852 struct nfit_spa *nfit_spa;
2853 u64 init_scrub_address = 0;
2854 bool init_ars_done = false;
2855 struct acpi_nfit_desc *acpi_desc;
2856 unsigned int tmo = scrub_timeout;
2857 unsigned int overflow_retry = scrub_overflow_abort;
2858
2859 acpi_desc = container_of(work, typeof(*acpi_desc), work);
2860 dev = acpi_desc->dev;
2861
2862 /*
2863 * We scrub in 2 phases. The first phase waits for any platform
2864 * firmware initiated scrubs to complete and then we go search for the
2865 * affected spa regions to mark them scanned. In the second phase we
2866 * initiate a directed scrub for every range that was not scrubbed in
2867 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2868 * the first phase, but really only care about running phase 2, where
2869 * regions can be notified of new poison.
2870 */
2871
2872 /* process platform firmware initiated scrubs */
2873 retry:
2874 mutex_lock(&acpi_desc->init_mutex);
2875 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2876 struct nd_cmd_ars_status *ars_status;
2877 struct acpi_nfit_system_address *spa;
2878 u64 ars_start, ars_len;
2879 int rc;
2880
2881 if (acpi_desc->cancel)
2882 break;
2883
2884 if (nfit_spa->nd_region)
2885 continue;
2886
2887 if (init_ars_done) {
2888 /*
2889 * No need to re-query, we're now just
2890 * reconciling all the ranges covered by the
2891 * initial scrub
2892 */
2893 rc = 0;
2894 } else
2895 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2896
2897 if (rc == -ENOTTY) {
2898 /* no ars capability, just register spa and move on */
2899 acpi_nfit_register_region(acpi_desc, nfit_spa);
2900 continue;
2901 }
2902
2903 if (rc == -EBUSY && !tmo) {
2904 /* fallthrough to directed scrub in phase 2 */
2905 dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2906 break;
2907 } else if (rc == -EBUSY) {
2908 mutex_unlock(&acpi_desc->init_mutex);
2909 ssleep(1);
2910 tmo--;
2911 goto retry;
2912 }
2913
2914 /* we got some results, but there are more pending... */
2915 if (rc == -ENOSPC && overflow_retry--) {
2916 ars_status = acpi_desc->ars_status;
2917 /*
2918 * Record the original scrub range, so that we
2919 * can recall all the ranges impacted by the
2920 * initial scrub.
2921 */
2922 if (!init_scrub_length) {
2923 init_scrub_length = ars_status->length;
2924 init_scrub_address = ars_status->address;
2925 }
2926 rc = ars_continue(acpi_desc);
2927 if (rc == 0) {
2928 mutex_unlock(&acpi_desc->init_mutex);
2929 goto retry;
2930 }
2931 }
2932
2933 if (rc < 0) {
2934 /*
2935 * Initial scrub failed, we'll give it one more
2936 * try below...
2937 */
2938 break;
2939 }
2940
2941 /* We got some final results, record completed ranges */
2942 ars_status = acpi_desc->ars_status;
2943 if (init_scrub_length) {
2944 ars_start = init_scrub_address;
2945 ars_len = ars_start + init_scrub_length;
2946 } else {
2947 ars_start = ars_status->address;
2948 ars_len = ars_status->length;
2949 }
2950 spa = nfit_spa->spa;
2951
2952 if (!init_ars_done) {
2953 init_ars_done = true;
2954 dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2955 ars_start, ars_len);
2956 }
2957 if (ars_start <= spa->address && ars_start + ars_len
2958 >= spa->address + spa->length)
2959 acpi_nfit_register_region(acpi_desc, nfit_spa);
2960 }
2961
2962 /*
2963 * For all the ranges not covered by an initial scrub we still
2964 * want to see if there are errors, but it's ok to discover them
2965 * asynchronously.
2966 */
2967 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2968 /*
2969 * Flag all the ranges that still need scrubbing, but
2970 * register them now to make data available.
2971 */
2972 if (!nfit_spa->nd_region) {
2973 nfit_spa->ars_required = 1;
2974 acpi_nfit_register_region(acpi_desc, nfit_spa);
2975 }
2976 }
2977 acpi_desc->init_complete = 1;
2978
2979 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2980 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2981 acpi_desc->scrub_count++;
2982 acpi_desc->ars_start_flags = 0;
2983 if (acpi_desc->scrub_count_state)
2984 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2985 mutex_unlock(&acpi_desc->init_mutex);
2986 }
2987
2988 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2989 {
2990 struct nfit_spa *nfit_spa;
2991 int rc;
2992
2993 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2994 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2995 /* BLK regions don't need to wait for ars results */
2996 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2997 if (rc)
2998 return rc;
2999 }
3000
3001 acpi_desc->ars_start_flags = 0;
3002 if (!acpi_desc->cancel)
3003 queue_work(nfit_wq, &acpi_desc->work);
3004 return 0;
3005 }
3006
3007 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3008 struct nfit_table_prev *prev)
3009 {
3010 struct device *dev = acpi_desc->dev;
3011
3012 if (!list_empty(&prev->spas) ||
3013 !list_empty(&prev->memdevs) ||
3014 !list_empty(&prev->dcrs) ||
3015 !list_empty(&prev->bdws) ||
3016 !list_empty(&prev->idts) ||
3017 !list_empty(&prev->flushes)) {
3018 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3019 return -ENXIO;
3020 }
3021 return 0;
3022 }
3023
3024 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3025 {
3026 struct device *dev = acpi_desc->dev;
3027 struct kernfs_node *nfit;
3028 struct device *bus_dev;
3029
3030 if (!ars_supported(acpi_desc->nvdimm_bus))
3031 return 0;
3032
3033 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3034 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3035 if (!nfit) {
3036 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3037 return -ENODEV;
3038 }
3039 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3040 sysfs_put(nfit);
3041 if (!acpi_desc->scrub_count_state) {
3042 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3043 return -ENODEV;
3044 }
3045
3046 return 0;
3047 }
3048
3049 static void acpi_nfit_unregister(void *data)
3050 {
3051 struct acpi_nfit_desc *acpi_desc = data;
3052
3053 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3054 }
3055
3056 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3057 {
3058 struct device *dev = acpi_desc->dev;
3059 struct nfit_table_prev prev;
3060 const void *end;
3061 int rc;
3062
3063 if (!acpi_desc->nvdimm_bus) {
3064 acpi_nfit_init_dsms(acpi_desc);
3065
3066 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3067 &acpi_desc->nd_desc);
3068 if (!acpi_desc->nvdimm_bus)
3069 return -ENOMEM;
3070
3071 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3072 acpi_desc);
3073 if (rc)
3074 return rc;
3075
3076 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3077 if (rc)
3078 return rc;
3079
3080 /* register this acpi_desc for mce notifications */
3081 mutex_lock(&acpi_desc_lock);
3082 list_add_tail(&acpi_desc->list, &acpi_descs);
3083 mutex_unlock(&acpi_desc_lock);
3084 }
3085
3086 mutex_lock(&acpi_desc->init_mutex);
3087
3088 INIT_LIST_HEAD(&prev.spas);
3089 INIT_LIST_HEAD(&prev.memdevs);
3090 INIT_LIST_HEAD(&prev.dcrs);
3091 INIT_LIST_HEAD(&prev.bdws);
3092 INIT_LIST_HEAD(&prev.idts);
3093 INIT_LIST_HEAD(&prev.flushes);
3094
3095 list_cut_position(&prev.spas, &acpi_desc->spas,
3096 acpi_desc->spas.prev);
3097 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3098 acpi_desc->memdevs.prev);
3099 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3100 acpi_desc->dcrs.prev);
3101 list_cut_position(&prev.bdws, &acpi_desc->bdws,
3102 acpi_desc->bdws.prev);
3103 list_cut_position(&prev.idts, &acpi_desc->idts,
3104 acpi_desc->idts.prev);
3105 list_cut_position(&prev.flushes, &acpi_desc->flushes,
3106 acpi_desc->flushes.prev);
3107
3108 end = data + sz;
3109 while (!IS_ERR_OR_NULL(data))
3110 data = add_table(acpi_desc, &prev, data, end);
3111
3112 if (IS_ERR(data)) {
3113 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
3114 PTR_ERR(data));
3115 rc = PTR_ERR(data);
3116 goto out_unlock;
3117 }
3118
3119 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3120 if (rc)
3121 goto out_unlock;
3122
3123 rc = nfit_mem_init(acpi_desc);
3124 if (rc)
3125 goto out_unlock;
3126
3127 rc = acpi_nfit_register_dimms(acpi_desc);
3128 if (rc)
3129 goto out_unlock;
3130
3131 rc = acpi_nfit_register_regions(acpi_desc);
3132
3133 out_unlock:
3134 mutex_unlock(&acpi_desc->init_mutex);
3135 return rc;
3136 }
3137 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3138
3139 struct acpi_nfit_flush_work {
3140 struct work_struct work;
3141 struct completion cmp;
3142 };
3143
3144 static void flush_probe(struct work_struct *work)
3145 {
3146 struct acpi_nfit_flush_work *flush;
3147
3148 flush = container_of(work, typeof(*flush), work);
3149 complete(&flush->cmp);
3150 }
3151
3152 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3153 {
3154 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3155 struct device *dev = acpi_desc->dev;
3156 struct acpi_nfit_flush_work flush;
3157 int rc;
3158
3159 /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3160 device_lock(dev);
3161 device_unlock(dev);
3162
3163 /* bounce the init_mutex to make init_complete valid */
3164 mutex_lock(&acpi_desc->init_mutex);
3165 if (acpi_desc->cancel || acpi_desc->init_complete) {
3166 mutex_unlock(&acpi_desc->init_mutex);
3167 return 0;
3168 }
3169
3170 /*
3171 * Scrub work could take 10s of seconds, userspace may give up so we
3172 * need to be interruptible while waiting.
3173 */
3174 INIT_WORK_ONSTACK(&flush.work, flush_probe);
3175 init_completion(&flush.cmp);
3176 queue_work(nfit_wq, &flush.work);
3177 mutex_unlock(&acpi_desc->init_mutex);
3178
3179 rc = wait_for_completion_interruptible(&flush.cmp);
3180 cancel_work_sync(&flush.work);
3181 return rc;
3182 }
3183
3184 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3185 struct nvdimm *nvdimm, unsigned int cmd)
3186 {
3187 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3188
3189 if (nvdimm)
3190 return 0;
3191 if (cmd != ND_CMD_ARS_START)
3192 return 0;
3193
3194 /*
3195 * The kernel and userspace may race to initiate a scrub, but
3196 * the scrub thread is prepared to lose that initial race. It
3197 * just needs guarantees that any ars it initiates are not
3198 * interrupted by any intervening start reqeusts from userspace.
3199 */
3200 if (work_busy(&acpi_desc->work))
3201 return -EBUSY;
3202
3203 return 0;
3204 }
3205
3206 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
3207 {
3208 struct device *dev = acpi_desc->dev;
3209 struct nfit_spa *nfit_spa;
3210
3211 if (work_busy(&acpi_desc->work))
3212 return -EBUSY;
3213
3214 mutex_lock(&acpi_desc->init_mutex);
3215 if (acpi_desc->cancel) {
3216 mutex_unlock(&acpi_desc->init_mutex);
3217 return 0;
3218 }
3219
3220 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3221 struct acpi_nfit_system_address *spa = nfit_spa->spa;
3222
3223 if (nfit_spa_type(spa) != NFIT_SPA_PM)
3224 continue;
3225
3226 nfit_spa->ars_required = 1;
3227 }
3228 acpi_desc->ars_start_flags = flags;
3229 queue_work(nfit_wq, &acpi_desc->work);
3230 dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
3231 mutex_unlock(&acpi_desc->init_mutex);
3232
3233 return 0;
3234 }
3235
3236 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3237 {
3238 struct nvdimm_bus_descriptor *nd_desc;
3239
3240 dev_set_drvdata(dev, acpi_desc);
3241 acpi_desc->dev = dev;
3242 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3243 nd_desc = &acpi_desc->nd_desc;
3244 nd_desc->provider_name = "ACPI.NFIT";
3245 nd_desc->module = THIS_MODULE;
3246 nd_desc->ndctl = acpi_nfit_ctl;
3247 nd_desc->flush_probe = acpi_nfit_flush_probe;
3248 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3249 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3250
3251 INIT_LIST_HEAD(&acpi_desc->spas);
3252 INIT_LIST_HEAD(&acpi_desc->dcrs);
3253 INIT_LIST_HEAD(&acpi_desc->bdws);
3254 INIT_LIST_HEAD(&acpi_desc->idts);
3255 INIT_LIST_HEAD(&acpi_desc->flushes);
3256 INIT_LIST_HEAD(&acpi_desc->memdevs);
3257 INIT_LIST_HEAD(&acpi_desc->dimms);
3258 INIT_LIST_HEAD(&acpi_desc->list);
3259 mutex_init(&acpi_desc->init_mutex);
3260 INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3261 }
3262 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3263
3264 static void acpi_nfit_put_table(void *table)
3265 {
3266 acpi_put_table(table);
3267 }
3268
3269 void acpi_nfit_shutdown(void *data)
3270 {
3271 struct acpi_nfit_desc *acpi_desc = data;
3272 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3273
3274 /*
3275 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3276 * race teardown
3277 */
3278 mutex_lock(&acpi_desc_lock);
3279 list_del(&acpi_desc->list);
3280 mutex_unlock(&acpi_desc_lock);
3281
3282 mutex_lock(&acpi_desc->init_mutex);
3283 acpi_desc->cancel = 1;
3284 mutex_unlock(&acpi_desc->init_mutex);
3285
3286 /*
3287 * Bounce the nvdimm bus lock to make sure any in-flight
3288 * acpi_nfit_ars_rescan() submissions have had a chance to
3289 * either submit or see ->cancel set.
3290 */
3291 device_lock(bus_dev);
3292 device_unlock(bus_dev);
3293
3294 flush_workqueue(nfit_wq);
3295 }
3296 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3297
3298 static int acpi_nfit_add(struct acpi_device *adev)
3299 {
3300 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3301 struct acpi_nfit_desc *acpi_desc;
3302 struct device *dev = &adev->dev;
3303 struct acpi_table_header *tbl;
3304 acpi_status status = AE_OK;
3305 acpi_size sz;
3306 int rc = 0;
3307
3308 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3309 if (ACPI_FAILURE(status)) {
3310 /* This is ok, we could have an nvdimm hotplugged later */
3311 dev_dbg(dev, "failed to find NFIT at startup\n");
3312 return 0;
3313 }
3314
3315 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3316 if (rc)
3317 return rc;
3318 sz = tbl->length;
3319
3320 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3321 if (!acpi_desc)
3322 return -ENOMEM;
3323 acpi_nfit_desc_init(acpi_desc, &adev->dev);
3324
3325 /* Save the acpi header for exporting the revision via sysfs */
3326 acpi_desc->acpi_header = *tbl;
3327
3328 /* Evaluate _FIT and override with that if present */
3329 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3330 if (ACPI_SUCCESS(status) && buf.length > 0) {
3331 union acpi_object *obj = buf.pointer;
3332
3333 if (obj->type == ACPI_TYPE_BUFFER)
3334 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3335 obj->buffer.length);
3336 else
3337 dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
3338 __func__, (int) obj->type);
3339 kfree(buf.pointer);
3340 } else
3341 /* skip over the lead-in header table */
3342 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3343 + sizeof(struct acpi_table_nfit),
3344 sz - sizeof(struct acpi_table_nfit));
3345
3346 if (rc)
3347 return rc;
3348 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3349 }
3350
3351 static int acpi_nfit_remove(struct acpi_device *adev)
3352 {
3353 /* see acpi_nfit_unregister */
3354 return 0;
3355 }
3356
3357 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3358 {
3359 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3360 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3361 union acpi_object *obj;
3362 acpi_status status;
3363 int ret;
3364
3365 if (!dev->driver) {
3366 /* dev->driver may be null if we're being removed */
3367 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
3368 return;
3369 }
3370
3371 if (!acpi_desc) {
3372 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3373 if (!acpi_desc)
3374 return;
3375 acpi_nfit_desc_init(acpi_desc, dev);
3376 } else {
3377 /*
3378 * Finish previous registration before considering new
3379 * regions.
3380 */
3381 flush_workqueue(nfit_wq);
3382 }
3383
3384 /* Evaluate _FIT */
3385 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3386 if (ACPI_FAILURE(status)) {
3387 dev_err(dev, "failed to evaluate _FIT\n");
3388 return;
3389 }
3390
3391 obj = buf.pointer;
3392 if (obj->type == ACPI_TYPE_BUFFER) {
3393 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3394 obj->buffer.length);
3395 if (ret)
3396 dev_err(dev, "failed to merge updated NFIT\n");
3397 } else
3398 dev_err(dev, "Invalid _FIT\n");
3399 kfree(buf.pointer);
3400 }
3401
3402 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3403 {
3404 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3405 u8 flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3406 0 : ND_ARS_RETURN_PREV_DATA;
3407
3408 acpi_nfit_ars_rescan(acpi_desc, flags);
3409 }
3410
3411 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3412 {
3413 dev_dbg(dev, "%s: event: 0x%x\n", __func__, event);
3414
3415 switch (event) {
3416 case NFIT_NOTIFY_UPDATE:
3417 return acpi_nfit_update_notify(dev, handle);
3418 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3419 return acpi_nfit_uc_error_notify(dev, handle);
3420 default:
3421 return;
3422 }
3423 }
3424 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3425
3426 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3427 {
3428 device_lock(&adev->dev);
3429 __acpi_nfit_notify(&adev->dev, adev->handle, event);
3430 device_unlock(&adev->dev);
3431 }
3432
3433 static const struct acpi_device_id acpi_nfit_ids[] = {
3434 { "ACPI0012", 0 },
3435 { "", 0 },
3436 };
3437 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3438
3439 static struct acpi_driver acpi_nfit_driver = {
3440 .name = KBUILD_MODNAME,
3441 .ids = acpi_nfit_ids,
3442 .ops = {
3443 .add = acpi_nfit_add,
3444 .remove = acpi_nfit_remove,
3445 .notify = acpi_nfit_notify,
3446 },
3447 };
3448
3449 static __init int nfit_init(void)
3450 {
3451 int ret;
3452
3453 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3454 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3455 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3456 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3457 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3458 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3459 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3460
3461 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3462 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3463 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3464 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3465 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3466 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3467 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3468 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3469 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3470 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3471 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3472 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3473 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3474
3475 nfit_wq = create_singlethread_workqueue("nfit");
3476 if (!nfit_wq)
3477 return -ENOMEM;
3478
3479 nfit_mce_register();
3480 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3481 if (ret) {
3482 nfit_mce_unregister();
3483 destroy_workqueue(nfit_wq);
3484 }
3485
3486 return ret;
3487
3488 }
3489
3490 static __exit void nfit_exit(void)
3491 {
3492 nfit_mce_unregister();
3493 acpi_bus_unregister_driver(&acpi_nfit_driver);
3494 destroy_workqueue(nfit_wq);
3495 WARN_ON(!list_empty(&acpi_descs));
3496 }
3497
3498 module_init(nfit_init);
3499 module_exit(nfit_exit);
3500 MODULE_LICENSE("GPL v2");
3501 MODULE_AUTHOR("Intel Corporation");
(deprecated) Btrfs devel tree