]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/acpi/nfit/core.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
nfit, address-range-scrub: fix scrub in-progress reporting
[mirror_ubuntu-bionic-kernel.git] / drivers / acpi / nfit / core.c
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 bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
842 struct acpi_nfit_capabilities *pcap)
843 {
844 struct device *dev = acpi_desc->dev;
845 u32 mask;
846
847 mask = (1 << (pcap->highest_capability + 1)) - 1;
848 acpi_desc->platform_cap = pcap->capabilities & mask;
849 dev_dbg(dev, "%s: cap: %#x\n", __func__, acpi_desc->platform_cap);
850 return true;
851 }
852
853 static void *add_table(struct acpi_nfit_desc *acpi_desc,
854 struct nfit_table_prev *prev, void *table, const void *end)
855 {
856 struct device *dev = acpi_desc->dev;
857 struct acpi_nfit_header *hdr;
858 void *err = ERR_PTR(-ENOMEM);
859
860 if (table >= end)
861 return NULL;
862
863 hdr = table;
864 if (!hdr->length) {
865 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
866 hdr->type);
867 return NULL;
868 }
869
870 switch (hdr->type) {
871 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
872 if (!add_spa(acpi_desc, prev, table))
873 return err;
874 break;
875 case ACPI_NFIT_TYPE_MEMORY_MAP:
876 if (!add_memdev(acpi_desc, prev, table))
877 return err;
878 break;
879 case ACPI_NFIT_TYPE_CONTROL_REGION:
880 if (!add_dcr(acpi_desc, prev, table))
881 return err;
882 break;
883 case ACPI_NFIT_TYPE_DATA_REGION:
884 if (!add_bdw(acpi_desc, prev, table))
885 return err;
886 break;
887 case ACPI_NFIT_TYPE_INTERLEAVE:
888 if (!add_idt(acpi_desc, prev, table))
889 return err;
890 break;
891 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
892 if (!add_flush(acpi_desc, prev, table))
893 return err;
894 break;
895 case ACPI_NFIT_TYPE_SMBIOS:
896 dev_dbg(dev, "%s: smbios\n", __func__);
897 break;
898 case ACPI_NFIT_TYPE_CAPABILITIES:
899 if (!add_platform_cap(acpi_desc, table))
900 return err;
901 break;
902 default:
903 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
904 break;
905 }
906
907 return table + hdr->length;
908 }
909
910 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
911 struct nfit_mem *nfit_mem)
912 {
913 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
914 u16 dcr = nfit_mem->dcr->region_index;
915 struct nfit_spa *nfit_spa;
916
917 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
918 u16 range_index = nfit_spa->spa->range_index;
919 int type = nfit_spa_type(nfit_spa->spa);
920 struct nfit_memdev *nfit_memdev;
921
922 if (type != NFIT_SPA_BDW)
923 continue;
924
925 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
926 if (nfit_memdev->memdev->range_index != range_index)
927 continue;
928 if (nfit_memdev->memdev->device_handle != device_handle)
929 continue;
930 if (nfit_memdev->memdev->region_index != dcr)
931 continue;
932
933 nfit_mem->spa_bdw = nfit_spa->spa;
934 return;
935 }
936 }
937
938 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
939 nfit_mem->spa_dcr->range_index);
940 nfit_mem->bdw = NULL;
941 }
942
943 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
944 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
945 {
946 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
947 struct nfit_memdev *nfit_memdev;
948 struct nfit_bdw *nfit_bdw;
949 struct nfit_idt *nfit_idt;
950 u16 idt_idx, range_index;
951
952 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
953 if (nfit_bdw->bdw->region_index != dcr)
954 continue;
955 nfit_mem->bdw = nfit_bdw->bdw;
956 break;
957 }
958
959 if (!nfit_mem->bdw)
960 return;
961
962 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
963
964 if (!nfit_mem->spa_bdw)
965 return;
966
967 range_index = nfit_mem->spa_bdw->range_index;
968 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
969 if (nfit_memdev->memdev->range_index != range_index ||
970 nfit_memdev->memdev->region_index != dcr)
971 continue;
972 nfit_mem->memdev_bdw = nfit_memdev->memdev;
973 idt_idx = nfit_memdev->memdev->interleave_index;
974 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
975 if (nfit_idt->idt->interleave_index != idt_idx)
976 continue;
977 nfit_mem->idt_bdw = nfit_idt->idt;
978 break;
979 }
980 break;
981 }
982 }
983
984 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
985 struct acpi_nfit_system_address *spa)
986 {
987 struct nfit_mem *nfit_mem, *found;
988 struct nfit_memdev *nfit_memdev;
989 int type = spa ? nfit_spa_type(spa) : 0;
990
991 switch (type) {
992 case NFIT_SPA_DCR:
993 case NFIT_SPA_PM:
994 break;
995 default:
996 if (spa)
997 return 0;
998 }
999
1000 /*
1001 * This loop runs in two modes, when a dimm is mapped the loop
1002 * adds memdev associations to an existing dimm, or creates a
1003 * dimm. In the unmapped dimm case this loop sweeps for memdev
1004 * instances with an invalid / zero range_index and adds those
1005 * dimms without spa associations.
1006 */
1007 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1008 struct nfit_flush *nfit_flush;
1009 struct nfit_dcr *nfit_dcr;
1010 u32 device_handle;
1011 u16 dcr;
1012
1013 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1014 continue;
1015 if (!spa && nfit_memdev->memdev->range_index)
1016 continue;
1017 found = NULL;
1018 dcr = nfit_memdev->memdev->region_index;
1019 device_handle = nfit_memdev->memdev->device_handle;
1020 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1021 if (__to_nfit_memdev(nfit_mem)->device_handle
1022 == device_handle) {
1023 found = nfit_mem;
1024 break;
1025 }
1026
1027 if (found)
1028 nfit_mem = found;
1029 else {
1030 nfit_mem = devm_kzalloc(acpi_desc->dev,
1031 sizeof(*nfit_mem), GFP_KERNEL);
1032 if (!nfit_mem)
1033 return -ENOMEM;
1034 INIT_LIST_HEAD(&nfit_mem->list);
1035 nfit_mem->acpi_desc = acpi_desc;
1036 list_add(&nfit_mem->list, &acpi_desc->dimms);
1037 }
1038
1039 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1040 if (nfit_dcr->dcr->region_index != dcr)
1041 continue;
1042 /*
1043 * Record the control region for the dimm. For
1044 * the ACPI 6.1 case, where there are separate
1045 * control regions for the pmem vs blk
1046 * interfaces, be sure to record the extended
1047 * blk details.
1048 */
1049 if (!nfit_mem->dcr)
1050 nfit_mem->dcr = nfit_dcr->dcr;
1051 else if (nfit_mem->dcr->windows == 0
1052 && nfit_dcr->dcr->windows)
1053 nfit_mem->dcr = nfit_dcr->dcr;
1054 break;
1055 }
1056
1057 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1058 struct acpi_nfit_flush_address *flush;
1059 u16 i;
1060
1061 if (nfit_flush->flush->device_handle != device_handle)
1062 continue;
1063 nfit_mem->nfit_flush = nfit_flush;
1064 flush = nfit_flush->flush;
1065 nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
1066 flush->hint_count
1067 * sizeof(struct resource), GFP_KERNEL);
1068 if (!nfit_mem->flush_wpq)
1069 return -ENOMEM;
1070 for (i = 0; i < flush->hint_count; i++) {
1071 struct resource *res = &nfit_mem->flush_wpq[i];
1072
1073 res->start = flush->hint_address[i];
1074 res->end = res->start + 8 - 1;
1075 }
1076 break;
1077 }
1078
1079 if (dcr && !nfit_mem->dcr) {
1080 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1081 spa->range_index, dcr);
1082 return -ENODEV;
1083 }
1084
1085 if (type == NFIT_SPA_DCR) {
1086 struct nfit_idt *nfit_idt;
1087 u16 idt_idx;
1088
1089 /* multiple dimms may share a SPA when interleaved */
1090 nfit_mem->spa_dcr = spa;
1091 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1092 idt_idx = nfit_memdev->memdev->interleave_index;
1093 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1094 if (nfit_idt->idt->interleave_index != idt_idx)
1095 continue;
1096 nfit_mem->idt_dcr = nfit_idt->idt;
1097 break;
1098 }
1099 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1100 } else if (type == NFIT_SPA_PM) {
1101 /*
1102 * A single dimm may belong to multiple SPA-PM
1103 * ranges, record at least one in addition to
1104 * any SPA-DCR range.
1105 */
1106 nfit_mem->memdev_pmem = nfit_memdev->memdev;
1107 } else
1108 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1109 }
1110
1111 return 0;
1112 }
1113
1114 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1115 {
1116 struct nfit_mem *a = container_of(_a, typeof(*a), list);
1117 struct nfit_mem *b = container_of(_b, typeof(*b), list);
1118 u32 handleA, handleB;
1119
1120 handleA = __to_nfit_memdev(a)->device_handle;
1121 handleB = __to_nfit_memdev(b)->device_handle;
1122 if (handleA < handleB)
1123 return -1;
1124 else if (handleA > handleB)
1125 return 1;
1126 return 0;
1127 }
1128
1129 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1130 {
1131 struct nfit_spa *nfit_spa;
1132 int rc;
1133
1134
1135 /*
1136 * For each SPA-DCR or SPA-PMEM address range find its
1137 * corresponding MEMDEV(s). From each MEMDEV find the
1138 * corresponding DCR. Then, if we're operating on a SPA-DCR,
1139 * try to find a SPA-BDW and a corresponding BDW that references
1140 * the DCR. Throw it all into an nfit_mem object. Note, that
1141 * BDWs are optional.
1142 */
1143 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1144 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1145 if (rc)
1146 return rc;
1147 }
1148
1149 /*
1150 * If a DIMM has failed to be mapped into SPA there will be no
1151 * SPA entries above. Find and register all the unmapped DIMMs
1152 * for reporting and recovery purposes.
1153 */
1154 rc = __nfit_mem_init(acpi_desc, NULL);
1155 if (rc)
1156 return rc;
1157
1158 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1159
1160 return 0;
1161 }
1162
1163 static ssize_t bus_dsm_mask_show(struct device *dev,
1164 struct device_attribute *attr, char *buf)
1165 {
1166 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1167 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1168
1169 return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1170 }
1171 static struct device_attribute dev_attr_bus_dsm_mask =
1172 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1173
1174 static ssize_t revision_show(struct device *dev,
1175 struct device_attribute *attr, char *buf)
1176 {
1177 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1178 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1179 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1180
1181 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1182 }
1183 static DEVICE_ATTR_RO(revision);
1184
1185 static ssize_t hw_error_scrub_show(struct device *dev,
1186 struct device_attribute *attr, char *buf)
1187 {
1188 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1189 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1190 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1191
1192 return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1193 }
1194
1195 /*
1196 * The 'hw_error_scrub' attribute can have the following values written to it:
1197 * '0': Switch to the default mode where an exception will only insert
1198 * the address of the memory error into the poison and badblocks lists.
1199 * '1': Enable a full scrub to happen if an exception for a memory error is
1200 * received.
1201 */
1202 static ssize_t hw_error_scrub_store(struct device *dev,
1203 struct device_attribute *attr, const char *buf, size_t size)
1204 {
1205 struct nvdimm_bus_descriptor *nd_desc;
1206 ssize_t rc;
1207 long val;
1208
1209 rc = kstrtol(buf, 0, &val);
1210 if (rc)
1211 return rc;
1212
1213 device_lock(dev);
1214 nd_desc = dev_get_drvdata(dev);
1215 if (nd_desc) {
1216 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1217
1218 switch (val) {
1219 case HW_ERROR_SCRUB_ON:
1220 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1221 break;
1222 case HW_ERROR_SCRUB_OFF:
1223 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1224 break;
1225 default:
1226 rc = -EINVAL;
1227 break;
1228 }
1229 }
1230 device_unlock(dev);
1231 if (rc)
1232 return rc;
1233 return size;
1234 }
1235 static DEVICE_ATTR_RW(hw_error_scrub);
1236
1237 /*
1238 * This shows the number of full Address Range Scrubs that have been
1239 * completed since driver load time. Userspace can wait on this using
1240 * select/poll etc. A '+' at the end indicates an ARS is in progress
1241 */
1242 static ssize_t scrub_show(struct device *dev,
1243 struct device_attribute *attr, char *buf)
1244 {
1245 struct nvdimm_bus_descriptor *nd_desc;
1246 ssize_t rc = -ENXIO;
1247
1248 device_lock(dev);
1249 nd_desc = dev_get_drvdata(dev);
1250 if (nd_desc) {
1251 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1252
1253 mutex_lock(&acpi_desc->init_mutex);
1254 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1255 work_busy(&acpi_desc->work)
1256 && !acpi_desc->cancel ? "+\n" : "\n");
1257 mutex_unlock(&acpi_desc->init_mutex);
1258 }
1259 device_unlock(dev);
1260 return rc;
1261 }
1262
1263 static ssize_t scrub_store(struct device *dev,
1264 struct device_attribute *attr, const char *buf, size_t size)
1265 {
1266 struct nvdimm_bus_descriptor *nd_desc;
1267 ssize_t rc;
1268 long val;
1269
1270 rc = kstrtol(buf, 0, &val);
1271 if (rc)
1272 return rc;
1273 if (val != 1)
1274 return -EINVAL;
1275
1276 device_lock(dev);
1277 nd_desc = dev_get_drvdata(dev);
1278 if (nd_desc) {
1279 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1280
1281 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1282 }
1283 device_unlock(dev);
1284 if (rc)
1285 return rc;
1286 return size;
1287 }
1288 static DEVICE_ATTR_RW(scrub);
1289
1290 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1291 {
1292 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1293 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1294 | 1 << ND_CMD_ARS_STATUS;
1295
1296 return (nd_desc->cmd_mask & mask) == mask;
1297 }
1298
1299 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1300 {
1301 struct device *dev = container_of(kobj, struct device, kobj);
1302 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1303
1304 if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1305 return 0;
1306 return a->mode;
1307 }
1308
1309 static struct attribute *acpi_nfit_attributes[] = {
1310 &dev_attr_revision.attr,
1311 &dev_attr_scrub.attr,
1312 &dev_attr_hw_error_scrub.attr,
1313 &dev_attr_bus_dsm_mask.attr,
1314 NULL,
1315 };
1316
1317 static const struct attribute_group acpi_nfit_attribute_group = {
1318 .name = "nfit",
1319 .attrs = acpi_nfit_attributes,
1320 .is_visible = nfit_visible,
1321 };
1322
1323 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1324 &nvdimm_bus_attribute_group,
1325 &acpi_nfit_attribute_group,
1326 NULL,
1327 };
1328
1329 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1330 {
1331 struct nvdimm *nvdimm = to_nvdimm(dev);
1332 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1333
1334 return __to_nfit_memdev(nfit_mem);
1335 }
1336
1337 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1338 {
1339 struct nvdimm *nvdimm = to_nvdimm(dev);
1340 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1341
1342 return nfit_mem->dcr;
1343 }
1344
1345 static ssize_t handle_show(struct device *dev,
1346 struct device_attribute *attr, char *buf)
1347 {
1348 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1349
1350 return sprintf(buf, "%#x\n", memdev->device_handle);
1351 }
1352 static DEVICE_ATTR_RO(handle);
1353
1354 static ssize_t phys_id_show(struct device *dev,
1355 struct device_attribute *attr, char *buf)
1356 {
1357 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1358
1359 return sprintf(buf, "%#x\n", memdev->physical_id);
1360 }
1361 static DEVICE_ATTR_RO(phys_id);
1362
1363 static ssize_t vendor_show(struct device *dev,
1364 struct device_attribute *attr, char *buf)
1365 {
1366 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1367
1368 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1369 }
1370 static DEVICE_ATTR_RO(vendor);
1371
1372 static ssize_t rev_id_show(struct device *dev,
1373 struct device_attribute *attr, char *buf)
1374 {
1375 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1376
1377 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1378 }
1379 static DEVICE_ATTR_RO(rev_id);
1380
1381 static ssize_t device_show(struct device *dev,
1382 struct device_attribute *attr, char *buf)
1383 {
1384 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1385
1386 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1387 }
1388 static DEVICE_ATTR_RO(device);
1389
1390 static ssize_t subsystem_vendor_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_vendor_id));
1396 }
1397 static DEVICE_ATTR_RO(subsystem_vendor);
1398
1399 static ssize_t subsystem_rev_id_show(struct device *dev,
1400 struct device_attribute *attr, char *buf)
1401 {
1402 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1403
1404 return sprintf(buf, "0x%04x\n",
1405 be16_to_cpu(dcr->subsystem_revision_id));
1406 }
1407 static DEVICE_ATTR_RO(subsystem_rev_id);
1408
1409 static ssize_t subsystem_device_show(struct device *dev,
1410 struct device_attribute *attr, char *buf)
1411 {
1412 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1413
1414 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1415 }
1416 static DEVICE_ATTR_RO(subsystem_device);
1417
1418 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1419 {
1420 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1421 int formats = 0;
1422
1423 if (nfit_mem->memdev_pmem)
1424 formats++;
1425 if (nfit_mem->memdev_bdw)
1426 formats++;
1427 return formats;
1428 }
1429
1430 static ssize_t format_show(struct device *dev,
1431 struct device_attribute *attr, char *buf)
1432 {
1433 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1434
1435 return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1436 }
1437 static DEVICE_ATTR_RO(format);
1438
1439 static ssize_t format1_show(struct device *dev,
1440 struct device_attribute *attr, char *buf)
1441 {
1442 u32 handle;
1443 ssize_t rc = -ENXIO;
1444 struct nfit_mem *nfit_mem;
1445 struct nfit_memdev *nfit_memdev;
1446 struct acpi_nfit_desc *acpi_desc;
1447 struct nvdimm *nvdimm = to_nvdimm(dev);
1448 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1449
1450 nfit_mem = nvdimm_provider_data(nvdimm);
1451 acpi_desc = nfit_mem->acpi_desc;
1452 handle = to_nfit_memdev(dev)->device_handle;
1453
1454 /* assumes DIMMs have at most 2 published interface codes */
1455 mutex_lock(&acpi_desc->init_mutex);
1456 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1457 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1458 struct nfit_dcr *nfit_dcr;
1459
1460 if (memdev->device_handle != handle)
1461 continue;
1462
1463 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1464 if (nfit_dcr->dcr->region_index != memdev->region_index)
1465 continue;
1466 if (nfit_dcr->dcr->code == dcr->code)
1467 continue;
1468 rc = sprintf(buf, "0x%04x\n",
1469 le16_to_cpu(nfit_dcr->dcr->code));
1470 break;
1471 }
1472 if (rc != ENXIO)
1473 break;
1474 }
1475 mutex_unlock(&acpi_desc->init_mutex);
1476 return rc;
1477 }
1478 static DEVICE_ATTR_RO(format1);
1479
1480 static ssize_t formats_show(struct device *dev,
1481 struct device_attribute *attr, char *buf)
1482 {
1483 struct nvdimm *nvdimm = to_nvdimm(dev);
1484
1485 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1486 }
1487 static DEVICE_ATTR_RO(formats);
1488
1489 static ssize_t serial_show(struct device *dev,
1490 struct device_attribute *attr, char *buf)
1491 {
1492 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1493
1494 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1495 }
1496 static DEVICE_ATTR_RO(serial);
1497
1498 static ssize_t family_show(struct device *dev,
1499 struct device_attribute *attr, char *buf)
1500 {
1501 struct nvdimm *nvdimm = to_nvdimm(dev);
1502 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1503
1504 if (nfit_mem->family < 0)
1505 return -ENXIO;
1506 return sprintf(buf, "%d\n", nfit_mem->family);
1507 }
1508 static DEVICE_ATTR_RO(family);
1509
1510 static ssize_t dsm_mask_show(struct device *dev,
1511 struct device_attribute *attr, char *buf)
1512 {
1513 struct nvdimm *nvdimm = to_nvdimm(dev);
1514 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1515
1516 if (nfit_mem->family < 0)
1517 return -ENXIO;
1518 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1519 }
1520 static DEVICE_ATTR_RO(dsm_mask);
1521
1522 static ssize_t flags_show(struct device *dev,
1523 struct device_attribute *attr, char *buf)
1524 {
1525 u16 flags = to_nfit_memdev(dev)->flags;
1526
1527 return sprintf(buf, "%s%s%s%s%s%s%s\n",
1528 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1529 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1530 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1531 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1532 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1533 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1534 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1535 }
1536 static DEVICE_ATTR_RO(flags);
1537
1538 static ssize_t id_show(struct device *dev,
1539 struct device_attribute *attr, char *buf)
1540 {
1541 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1542
1543 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1544 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1545 be16_to_cpu(dcr->vendor_id),
1546 dcr->manufacturing_location,
1547 be16_to_cpu(dcr->manufacturing_date),
1548 be32_to_cpu(dcr->serial_number));
1549 else
1550 return sprintf(buf, "%04x-%08x\n",
1551 be16_to_cpu(dcr->vendor_id),
1552 be32_to_cpu(dcr->serial_number));
1553 }
1554 static DEVICE_ATTR_RO(id);
1555
1556 static struct attribute *acpi_nfit_dimm_attributes[] = {
1557 &dev_attr_handle.attr,
1558 &dev_attr_phys_id.attr,
1559 &dev_attr_vendor.attr,
1560 &dev_attr_device.attr,
1561 &dev_attr_rev_id.attr,
1562 &dev_attr_subsystem_vendor.attr,
1563 &dev_attr_subsystem_device.attr,
1564 &dev_attr_subsystem_rev_id.attr,
1565 &dev_attr_format.attr,
1566 &dev_attr_formats.attr,
1567 &dev_attr_format1.attr,
1568 &dev_attr_serial.attr,
1569 &dev_attr_flags.attr,
1570 &dev_attr_id.attr,
1571 &dev_attr_family.attr,
1572 &dev_attr_dsm_mask.attr,
1573 NULL,
1574 };
1575
1576 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1577 struct attribute *a, int n)
1578 {
1579 struct device *dev = container_of(kobj, struct device, kobj);
1580 struct nvdimm *nvdimm = to_nvdimm(dev);
1581
1582 if (!to_nfit_dcr(dev)) {
1583 /* Without a dcr only the memdev attributes can be surfaced */
1584 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1585 || a == &dev_attr_flags.attr
1586 || a == &dev_attr_family.attr
1587 || a == &dev_attr_dsm_mask.attr)
1588 return a->mode;
1589 return 0;
1590 }
1591
1592 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1593 return 0;
1594 return a->mode;
1595 }
1596
1597 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1598 .name = "nfit",
1599 .attrs = acpi_nfit_dimm_attributes,
1600 .is_visible = acpi_nfit_dimm_attr_visible,
1601 };
1602
1603 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1604 &nvdimm_attribute_group,
1605 &nd_device_attribute_group,
1606 &acpi_nfit_dimm_attribute_group,
1607 NULL,
1608 };
1609
1610 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1611 u32 device_handle)
1612 {
1613 struct nfit_mem *nfit_mem;
1614
1615 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1616 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1617 return nfit_mem->nvdimm;
1618
1619 return NULL;
1620 }
1621
1622 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1623 {
1624 struct nfit_mem *nfit_mem;
1625 struct acpi_nfit_desc *acpi_desc;
1626
1627 dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1628 event);
1629
1630 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1631 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1632 event);
1633 return;
1634 }
1635
1636 acpi_desc = dev_get_drvdata(dev->parent);
1637 if (!acpi_desc)
1638 return;
1639
1640 /*
1641 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1642 * is still valid.
1643 */
1644 nfit_mem = dev_get_drvdata(dev);
1645 if (nfit_mem && nfit_mem->flags_attr)
1646 sysfs_notify_dirent(nfit_mem->flags_attr);
1647 }
1648 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1649
1650 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1651 {
1652 struct acpi_device *adev = data;
1653 struct device *dev = &adev->dev;
1654
1655 device_lock(dev->parent);
1656 __acpi_nvdimm_notify(dev, event);
1657 device_unlock(dev->parent);
1658 }
1659
1660 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1661 struct nfit_mem *nfit_mem, u32 device_handle)
1662 {
1663 struct acpi_device *adev, *adev_dimm;
1664 struct device *dev = acpi_desc->dev;
1665 union acpi_object *obj;
1666 unsigned long dsm_mask;
1667 const guid_t *guid;
1668 int i;
1669 int family = -1;
1670
1671 /* nfit test assumes 1:1 relationship between commands and dsms */
1672 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1673 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1674 adev = to_acpi_dev(acpi_desc);
1675 if (!adev)
1676 return 0;
1677
1678 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1679 nfit_mem->adev = adev_dimm;
1680 if (!adev_dimm) {
1681 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1682 device_handle);
1683 return force_enable_dimms ? 0 : -ENODEV;
1684 }
1685
1686 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1687 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1688 dev_err(dev, "%s: notification registration failed\n",
1689 dev_name(&adev_dimm->dev));
1690 return -ENXIO;
1691 }
1692 /*
1693 * Record nfit_mem for the notification path to track back to
1694 * the nfit sysfs attributes for this dimm device object.
1695 */
1696 dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1697
1698 /*
1699 * Until standardization materializes we need to consider 4
1700 * different command sets. Note, that checking for function0 (bit0)
1701 * tells us if any commands are reachable through this GUID.
1702 */
1703 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1704 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1705 if (family < 0 || i == default_dsm_family)
1706 family = i;
1707
1708 /* limit the supported commands to those that are publicly documented */
1709 nfit_mem->family = family;
1710 if (override_dsm_mask && !disable_vendor_specific)
1711 dsm_mask = override_dsm_mask;
1712 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1713 dsm_mask = NVDIMM_INTEL_CMDMASK;
1714 if (disable_vendor_specific)
1715 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1716 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1717 dsm_mask = 0x1c3c76;
1718 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1719 dsm_mask = 0x1fe;
1720 if (disable_vendor_specific)
1721 dsm_mask &= ~(1 << 8);
1722 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1723 dsm_mask = 0xffffffff;
1724 } else {
1725 dev_dbg(dev, "unknown dimm command family\n");
1726 nfit_mem->family = -1;
1727 /* DSMs are optional, continue loading the driver... */
1728 return 0;
1729 }
1730
1731 guid = to_nfit_uuid(nfit_mem->family);
1732 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1733 if (acpi_check_dsm(adev_dimm->handle, guid,
1734 nfit_dsm_revid(nfit_mem->family, i),
1735 1ULL << i))
1736 set_bit(i, &nfit_mem->dsm_mask);
1737
1738 obj = acpi_label_info(adev_dimm->handle);
1739 if (obj) {
1740 ACPI_FREE(obj);
1741 nfit_mem->has_lsi = 1;
1742 dev_dbg(dev, "%s: has _LSI\n", dev_name(&adev_dimm->dev));
1743 }
1744
1745 obj = acpi_label_read(adev_dimm->handle, 0, 0);
1746 if (obj) {
1747 ACPI_FREE(obj);
1748 nfit_mem->has_lsr = 1;
1749 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1750 }
1751
1752 obj = acpi_label_write(adev_dimm->handle, 0, 0, NULL);
1753 if (obj) {
1754 ACPI_FREE(obj);
1755 nfit_mem->has_lsw = 1;
1756 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1757 }
1758
1759 return 0;
1760 }
1761
1762 static void shutdown_dimm_notify(void *data)
1763 {
1764 struct acpi_nfit_desc *acpi_desc = data;
1765 struct nfit_mem *nfit_mem;
1766
1767 mutex_lock(&acpi_desc->init_mutex);
1768 /*
1769 * Clear out the nfit_mem->flags_attr and shut down dimm event
1770 * notifications.
1771 */
1772 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1773 struct acpi_device *adev_dimm = nfit_mem->adev;
1774
1775 if (nfit_mem->flags_attr) {
1776 sysfs_put(nfit_mem->flags_attr);
1777 nfit_mem->flags_attr = NULL;
1778 }
1779 if (adev_dimm) {
1780 acpi_remove_notify_handler(adev_dimm->handle,
1781 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1782 dev_set_drvdata(&adev_dimm->dev, NULL);
1783 }
1784 }
1785 mutex_unlock(&acpi_desc->init_mutex);
1786 }
1787
1788 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1789 {
1790 struct nfit_mem *nfit_mem;
1791 int dimm_count = 0, rc;
1792 struct nvdimm *nvdimm;
1793
1794 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1795 struct acpi_nfit_flush_address *flush;
1796 unsigned long flags = 0, cmd_mask;
1797 struct nfit_memdev *nfit_memdev;
1798 u32 device_handle;
1799 u16 mem_flags;
1800
1801 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1802 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1803 if (nvdimm) {
1804 dimm_count++;
1805 continue;
1806 }
1807
1808 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1809 set_bit(NDD_ALIASING, &flags);
1810
1811 /* collate flags across all memdevs for this dimm */
1812 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1813 struct acpi_nfit_memory_map *dimm_memdev;
1814
1815 dimm_memdev = __to_nfit_memdev(nfit_mem);
1816 if (dimm_memdev->device_handle
1817 != nfit_memdev->memdev->device_handle)
1818 continue;
1819 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1820 }
1821
1822 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1823 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1824 set_bit(NDD_UNARMED, &flags);
1825
1826 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1827 if (rc)
1828 continue;
1829
1830 /*
1831 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1832 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1833 * userspace interface.
1834 */
1835 cmd_mask = 1UL << ND_CMD_CALL;
1836 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1837 /*
1838 * These commands have a 1:1 correspondence
1839 * between DSM payload and libnvdimm ioctl
1840 * payload format.
1841 */
1842 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1843 }
1844
1845 if (nfit_mem->has_lsi)
1846 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1847 if (nfit_mem->has_lsr)
1848 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1849 if (nfit_mem->has_lsw)
1850 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1851
1852 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1853 : NULL;
1854 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1855 acpi_nfit_dimm_attribute_groups,
1856 flags, cmd_mask, flush ? flush->hint_count : 0,
1857 nfit_mem->flush_wpq);
1858 if (!nvdimm)
1859 return -ENOMEM;
1860
1861 nfit_mem->nvdimm = nvdimm;
1862 dimm_count++;
1863
1864 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1865 continue;
1866
1867 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1868 nvdimm_name(nvdimm),
1869 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1870 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1871 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1872 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1873 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1874
1875 }
1876
1877 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1878 if (rc)
1879 return rc;
1880
1881 /*
1882 * Now that dimms are successfully registered, and async registration
1883 * is flushed, attempt to enable event notification.
1884 */
1885 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1886 struct kernfs_node *nfit_kernfs;
1887
1888 nvdimm = nfit_mem->nvdimm;
1889 if (!nvdimm)
1890 continue;
1891
1892 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1893 if (nfit_kernfs)
1894 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1895 "flags");
1896 sysfs_put(nfit_kernfs);
1897 if (!nfit_mem->flags_attr)
1898 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1899 nvdimm_name(nvdimm));
1900 }
1901
1902 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1903 acpi_desc);
1904 }
1905
1906 /*
1907 * These constants are private because there are no kernel consumers of
1908 * these commands.
1909 */
1910 enum nfit_aux_cmds {
1911 NFIT_CMD_TRANSLATE_SPA = 5,
1912 NFIT_CMD_ARS_INJECT_SET = 7,
1913 NFIT_CMD_ARS_INJECT_CLEAR = 8,
1914 NFIT_CMD_ARS_INJECT_GET = 9,
1915 };
1916
1917 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1918 {
1919 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1920 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1921 struct acpi_device *adev;
1922 unsigned long dsm_mask;
1923 int i;
1924
1925 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1926 nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1927 adev = to_acpi_dev(acpi_desc);
1928 if (!adev)
1929 return;
1930
1931 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1932 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1933 set_bit(i, &nd_desc->cmd_mask);
1934 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1935
1936 dsm_mask =
1937 (1 << ND_CMD_ARS_CAP) |
1938 (1 << ND_CMD_ARS_START) |
1939 (1 << ND_CMD_ARS_STATUS) |
1940 (1 << ND_CMD_CLEAR_ERROR) |
1941 (1 << NFIT_CMD_TRANSLATE_SPA) |
1942 (1 << NFIT_CMD_ARS_INJECT_SET) |
1943 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1944 (1 << NFIT_CMD_ARS_INJECT_GET);
1945 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1946 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1947 set_bit(i, &nd_desc->bus_dsm_mask);
1948 }
1949
1950 static ssize_t range_index_show(struct device *dev,
1951 struct device_attribute *attr, char *buf)
1952 {
1953 struct nd_region *nd_region = to_nd_region(dev);
1954 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1955
1956 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1957 }
1958 static DEVICE_ATTR_RO(range_index);
1959
1960 static ssize_t ecc_unit_size_show(struct device *dev,
1961 struct device_attribute *attr, char *buf)
1962 {
1963 struct nd_region *nd_region = to_nd_region(dev);
1964 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1965
1966 return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1967 }
1968 static DEVICE_ATTR_RO(ecc_unit_size);
1969
1970 static struct attribute *acpi_nfit_region_attributes[] = {
1971 &dev_attr_range_index.attr,
1972 &dev_attr_ecc_unit_size.attr,
1973 NULL,
1974 };
1975
1976 static const struct attribute_group acpi_nfit_region_attribute_group = {
1977 .name = "nfit",
1978 .attrs = acpi_nfit_region_attributes,
1979 };
1980
1981 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1982 &nd_region_attribute_group,
1983 &nd_mapping_attribute_group,
1984 &nd_device_attribute_group,
1985 &nd_numa_attribute_group,
1986 &acpi_nfit_region_attribute_group,
1987 NULL,
1988 };
1989
1990 /* enough info to uniquely specify an interleave set */
1991 struct nfit_set_info {
1992 struct nfit_set_info_map {
1993 u64 region_offset;
1994 u32 serial_number;
1995 u32 pad;
1996 } mapping[0];
1997 };
1998
1999 struct nfit_set_info2 {
2000 struct nfit_set_info_map2 {
2001 u64 region_offset;
2002 u32 serial_number;
2003 u16 vendor_id;
2004 u16 manufacturing_date;
2005 u8 manufacturing_location;
2006 u8 reserved[31];
2007 } mapping[0];
2008 };
2009
2010 static size_t sizeof_nfit_set_info(int num_mappings)
2011 {
2012 return sizeof(struct nfit_set_info)
2013 + num_mappings * sizeof(struct nfit_set_info_map);
2014 }
2015
2016 static size_t sizeof_nfit_set_info2(int num_mappings)
2017 {
2018 return sizeof(struct nfit_set_info2)
2019 + num_mappings * sizeof(struct nfit_set_info_map2);
2020 }
2021
2022 static int cmp_map_compat(const void *m0, const void *m1)
2023 {
2024 const struct nfit_set_info_map *map0 = m0;
2025 const struct nfit_set_info_map *map1 = m1;
2026
2027 return memcmp(&map0->region_offset, &map1->region_offset,
2028 sizeof(u64));
2029 }
2030
2031 static int cmp_map(const void *m0, const void *m1)
2032 {
2033 const struct nfit_set_info_map *map0 = m0;
2034 const struct nfit_set_info_map *map1 = m1;
2035
2036 if (map0->region_offset < map1->region_offset)
2037 return -1;
2038 else if (map0->region_offset > map1->region_offset)
2039 return 1;
2040 return 0;
2041 }
2042
2043 static int cmp_map2(const void *m0, const void *m1)
2044 {
2045 const struct nfit_set_info_map2 *map0 = m0;
2046 const struct nfit_set_info_map2 *map1 = m1;
2047
2048 if (map0->region_offset < map1->region_offset)
2049 return -1;
2050 else if (map0->region_offset > map1->region_offset)
2051 return 1;
2052 return 0;
2053 }
2054
2055 /* Retrieve the nth entry referencing this spa */
2056 static struct acpi_nfit_memory_map *memdev_from_spa(
2057 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2058 {
2059 struct nfit_memdev *nfit_memdev;
2060
2061 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2062 if (nfit_memdev->memdev->range_index == range_index)
2063 if (n-- == 0)
2064 return nfit_memdev->memdev;
2065 return NULL;
2066 }
2067
2068 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2069 struct nd_region_desc *ndr_desc,
2070 struct acpi_nfit_system_address *spa)
2071 {
2072 struct device *dev = acpi_desc->dev;
2073 struct nd_interleave_set *nd_set;
2074 u16 nr = ndr_desc->num_mappings;
2075 struct nfit_set_info2 *info2;
2076 struct nfit_set_info *info;
2077 int i;
2078
2079 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2080 if (!nd_set)
2081 return -ENOMEM;
2082 ndr_desc->nd_set = nd_set;
2083 guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2084
2085 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2086 if (!info)
2087 return -ENOMEM;
2088
2089 info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2090 if (!info2)
2091 return -ENOMEM;
2092
2093 for (i = 0; i < nr; i++) {
2094 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2095 struct nfit_set_info_map *map = &info->mapping[i];
2096 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2097 struct nvdimm *nvdimm = mapping->nvdimm;
2098 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2099 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2100 spa->range_index, i);
2101 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2102
2103 if (!memdev || !nfit_mem->dcr) {
2104 dev_err(dev, "%s: failed to find DCR\n", __func__);
2105 return -ENODEV;
2106 }
2107
2108 map->region_offset = memdev->region_offset;
2109 map->serial_number = dcr->serial_number;
2110
2111 map2->region_offset = memdev->region_offset;
2112 map2->serial_number = dcr->serial_number;
2113 map2->vendor_id = dcr->vendor_id;
2114 map2->manufacturing_date = dcr->manufacturing_date;
2115 map2->manufacturing_location = dcr->manufacturing_location;
2116 }
2117
2118 /* v1.1 namespaces */
2119 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2120 cmp_map, NULL);
2121 nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2122
2123 /* v1.2 namespaces */
2124 sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2125 cmp_map2, NULL);
2126 nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2127
2128 /* support v1.1 namespaces created with the wrong sort order */
2129 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2130 cmp_map_compat, NULL);
2131 nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2132
2133 /* record the result of the sort for the mapping position */
2134 for (i = 0; i < nr; i++) {
2135 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2136 int j;
2137
2138 for (j = 0; j < nr; j++) {
2139 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2140 struct nvdimm *nvdimm = mapping->nvdimm;
2141 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2142 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2143
2144 if (map2->serial_number == dcr->serial_number &&
2145 map2->vendor_id == dcr->vendor_id &&
2146 map2->manufacturing_date == dcr->manufacturing_date &&
2147 map2->manufacturing_location
2148 == dcr->manufacturing_location) {
2149 mapping->position = i;
2150 break;
2151 }
2152 }
2153 }
2154
2155 ndr_desc->nd_set = nd_set;
2156 devm_kfree(dev, info);
2157 devm_kfree(dev, info2);
2158
2159 return 0;
2160 }
2161
2162 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2163 {
2164 struct acpi_nfit_interleave *idt = mmio->idt;
2165 u32 sub_line_offset, line_index, line_offset;
2166 u64 line_no, table_skip_count, table_offset;
2167
2168 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2169 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2170 line_offset = idt->line_offset[line_index]
2171 * mmio->line_size;
2172 table_offset = table_skip_count * mmio->table_size;
2173
2174 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2175 }
2176
2177 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2178 {
2179 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2180 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2181 const u32 STATUS_MASK = 0x80000037;
2182
2183 if (mmio->num_lines)
2184 offset = to_interleave_offset(offset, mmio);
2185
2186 return readl(mmio->addr.base + offset) & STATUS_MASK;
2187 }
2188
2189 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2190 resource_size_t dpa, unsigned int len, unsigned int write)
2191 {
2192 u64 cmd, offset;
2193 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2194
2195 enum {
2196 BCW_OFFSET_MASK = (1ULL << 48)-1,
2197 BCW_LEN_SHIFT = 48,
2198 BCW_LEN_MASK = (1ULL << 8) - 1,
2199 BCW_CMD_SHIFT = 56,
2200 };
2201
2202 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2203 len = len >> L1_CACHE_SHIFT;
2204 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2205 cmd |= ((u64) write) << BCW_CMD_SHIFT;
2206
2207 offset = nfit_blk->cmd_offset + mmio->size * bw;
2208 if (mmio->num_lines)
2209 offset = to_interleave_offset(offset, mmio);
2210
2211 writeq(cmd, mmio->addr.base + offset);
2212 nvdimm_flush(nfit_blk->nd_region);
2213
2214 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2215 readq(mmio->addr.base + offset);
2216 }
2217
2218 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2219 resource_size_t dpa, void *iobuf, size_t len, int rw,
2220 unsigned int lane)
2221 {
2222 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2223 unsigned int copied = 0;
2224 u64 base_offset;
2225 int rc;
2226
2227 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2228 + lane * mmio->size;
2229 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2230 while (len) {
2231 unsigned int c;
2232 u64 offset;
2233
2234 if (mmio->num_lines) {
2235 u32 line_offset;
2236
2237 offset = to_interleave_offset(base_offset + copied,
2238 mmio);
2239 div_u64_rem(offset, mmio->line_size, &line_offset);
2240 c = min_t(size_t, len, mmio->line_size - line_offset);
2241 } else {
2242 offset = base_offset + nfit_blk->bdw_offset;
2243 c = len;
2244 }
2245
2246 if (rw)
2247 memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2248 else {
2249 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2250 arch_invalidate_pmem((void __force *)
2251 mmio->addr.aperture + offset, c);
2252
2253 memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2254 }
2255
2256 copied += c;
2257 len -= c;
2258 }
2259
2260 if (rw)
2261 nvdimm_flush(nfit_blk->nd_region);
2262
2263 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2264 return rc;
2265 }
2266
2267 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2268 resource_size_t dpa, void *iobuf, u64 len, int rw)
2269 {
2270 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2271 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2272 struct nd_region *nd_region = nfit_blk->nd_region;
2273 unsigned int lane, copied = 0;
2274 int rc = 0;
2275
2276 lane = nd_region_acquire_lane(nd_region);
2277 while (len) {
2278 u64 c = min(len, mmio->size);
2279
2280 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2281 iobuf + copied, c, rw, lane);
2282 if (rc)
2283 break;
2284
2285 copied += c;
2286 len -= c;
2287 }
2288 nd_region_release_lane(nd_region, lane);
2289
2290 return rc;
2291 }
2292
2293 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2294 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2295 {
2296 if (idt) {
2297 mmio->num_lines = idt->line_count;
2298 mmio->line_size = idt->line_size;
2299 if (interleave_ways == 0)
2300 return -ENXIO;
2301 mmio->table_size = mmio->num_lines * interleave_ways
2302 * mmio->line_size;
2303 }
2304
2305 return 0;
2306 }
2307
2308 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2309 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2310 {
2311 struct nd_cmd_dimm_flags flags;
2312 int rc;
2313
2314 memset(&flags, 0, sizeof(flags));
2315 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2316 sizeof(flags), NULL);
2317
2318 if (rc >= 0 && flags.status == 0)
2319 nfit_blk->dimm_flags = flags.flags;
2320 else if (rc == -ENOTTY) {
2321 /* fall back to a conservative default */
2322 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2323 rc = 0;
2324 } else
2325 rc = -ENXIO;
2326
2327 return rc;
2328 }
2329
2330 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2331 struct device *dev)
2332 {
2333 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2334 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2335 struct nfit_blk_mmio *mmio;
2336 struct nfit_blk *nfit_blk;
2337 struct nfit_mem *nfit_mem;
2338 struct nvdimm *nvdimm;
2339 int rc;
2340
2341 nvdimm = nd_blk_region_to_dimm(ndbr);
2342 nfit_mem = nvdimm_provider_data(nvdimm);
2343 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2344 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2345 nfit_mem ? "" : " nfit_mem",
2346 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2347 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2348 return -ENXIO;
2349 }
2350
2351 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2352 if (!nfit_blk)
2353 return -ENOMEM;
2354 nd_blk_region_set_provider_data(ndbr, nfit_blk);
2355 nfit_blk->nd_region = to_nd_region(dev);
2356
2357 /* map block aperture memory */
2358 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2359 mmio = &nfit_blk->mmio[BDW];
2360 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2361 nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2362 if (!mmio->addr.base) {
2363 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2364 nvdimm_name(nvdimm));
2365 return -ENOMEM;
2366 }
2367 mmio->size = nfit_mem->bdw->size;
2368 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2369 mmio->idt = nfit_mem->idt_bdw;
2370 mmio->spa = nfit_mem->spa_bdw;
2371 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2372 nfit_mem->memdev_bdw->interleave_ways);
2373 if (rc) {
2374 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2375 __func__, nvdimm_name(nvdimm));
2376 return rc;
2377 }
2378
2379 /* map block control memory */
2380 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2381 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2382 mmio = &nfit_blk->mmio[DCR];
2383 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2384 nfit_mem->spa_dcr->length);
2385 if (!mmio->addr.base) {
2386 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2387 nvdimm_name(nvdimm));
2388 return -ENOMEM;
2389 }
2390 mmio->size = nfit_mem->dcr->window_size;
2391 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2392 mmio->idt = nfit_mem->idt_dcr;
2393 mmio->spa = nfit_mem->spa_dcr;
2394 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2395 nfit_mem->memdev_dcr->interleave_ways);
2396 if (rc) {
2397 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2398 __func__, nvdimm_name(nvdimm));
2399 return rc;
2400 }
2401
2402 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2403 if (rc < 0) {
2404 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2405 __func__, nvdimm_name(nvdimm));
2406 return rc;
2407 }
2408
2409 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2410 dev_warn(dev, "unable to guarantee persistence of writes\n");
2411
2412 if (mmio->line_size == 0)
2413 return 0;
2414
2415 if ((u32) nfit_blk->cmd_offset % mmio->line_size
2416 + 8 > mmio->line_size) {
2417 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2418 return -ENXIO;
2419 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2420 + 8 > mmio->line_size) {
2421 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2422 return -ENXIO;
2423 }
2424
2425 return 0;
2426 }
2427
2428 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2429 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2430 {
2431 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2432 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2433 int cmd_rc, rc;
2434
2435 cmd->address = spa->address;
2436 cmd->length = spa->length;
2437 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2438 sizeof(*cmd), &cmd_rc);
2439 if (rc < 0)
2440 return rc;
2441 return cmd_rc;
2442 }
2443
2444 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2445 {
2446 int rc;
2447 int cmd_rc;
2448 struct nd_cmd_ars_start ars_start;
2449 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2450 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2451
2452 memset(&ars_start, 0, sizeof(ars_start));
2453 ars_start.address = spa->address;
2454 ars_start.length = spa->length;
2455 ars_start.flags = acpi_desc->ars_start_flags;
2456 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2457 ars_start.type = ND_ARS_PERSISTENT;
2458 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2459 ars_start.type = ND_ARS_VOLATILE;
2460 else
2461 return -ENOTTY;
2462
2463 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2464 sizeof(ars_start), &cmd_rc);
2465
2466 if (rc < 0)
2467 return rc;
2468 return cmd_rc;
2469 }
2470
2471 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2472 {
2473 int rc, cmd_rc;
2474 struct nd_cmd_ars_start ars_start;
2475 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2476 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2477
2478 memset(&ars_start, 0, sizeof(ars_start));
2479 ars_start.address = ars_status->restart_address;
2480 ars_start.length = ars_status->restart_length;
2481 ars_start.type = ars_status->type;
2482 ars_start.flags = acpi_desc->ars_start_flags;
2483 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2484 sizeof(ars_start), &cmd_rc);
2485 if (rc < 0)
2486 return rc;
2487 return cmd_rc;
2488 }
2489
2490 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2491 {
2492 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2493 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2494 int rc, cmd_rc;
2495
2496 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2497 acpi_desc->ars_status_size, &cmd_rc);
2498 if (rc < 0)
2499 return rc;
2500 return cmd_rc;
2501 }
2502
2503 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2504 struct nd_cmd_ars_status *ars_status)
2505 {
2506 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2507 int rc;
2508 u32 i;
2509
2510 /*
2511 * First record starts at 44 byte offset from the start of the
2512 * payload.
2513 */
2514 if (ars_status->out_length < 44)
2515 return 0;
2516 for (i = 0; i < ars_status->num_records; i++) {
2517 /* only process full records */
2518 if (ars_status->out_length
2519 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2520 break;
2521 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2522 ars_status->records[i].err_address,
2523 ars_status->records[i].length);
2524 if (rc)
2525 return rc;
2526 }
2527 if (i < ars_status->num_records)
2528 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2529
2530 return 0;
2531 }
2532
2533 static void acpi_nfit_remove_resource(void *data)
2534 {
2535 struct resource *res = data;
2536
2537 remove_resource(res);
2538 }
2539
2540 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2541 struct nd_region_desc *ndr_desc)
2542 {
2543 struct resource *res, *nd_res = ndr_desc->res;
2544 int is_pmem, ret;
2545
2546 /* No operation if the region is already registered as PMEM */
2547 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2548 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2549 if (is_pmem == REGION_INTERSECTS)
2550 return 0;
2551
2552 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2553 if (!res)
2554 return -ENOMEM;
2555
2556 res->name = "Persistent Memory";
2557 res->start = nd_res->start;
2558 res->end = nd_res->end;
2559 res->flags = IORESOURCE_MEM;
2560 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2561
2562 ret = insert_resource(&iomem_resource, res);
2563 if (ret)
2564 return ret;
2565
2566 ret = devm_add_action_or_reset(acpi_desc->dev,
2567 acpi_nfit_remove_resource,
2568 res);
2569 if (ret)
2570 return ret;
2571
2572 return 0;
2573 }
2574
2575 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2576 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2577 struct acpi_nfit_memory_map *memdev,
2578 struct nfit_spa *nfit_spa)
2579 {
2580 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2581 memdev->device_handle);
2582 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2583 struct nd_blk_region_desc *ndbr_desc;
2584 struct nfit_mem *nfit_mem;
2585 int blk_valid = 0, rc;
2586
2587 if (!nvdimm) {
2588 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2589 spa->range_index, memdev->device_handle);
2590 return -ENODEV;
2591 }
2592
2593 mapping->nvdimm = nvdimm;
2594 switch (nfit_spa_type(spa)) {
2595 case NFIT_SPA_PM:
2596 case NFIT_SPA_VOLATILE:
2597 mapping->start = memdev->address;
2598 mapping->size = memdev->region_size;
2599 break;
2600 case NFIT_SPA_DCR:
2601 nfit_mem = nvdimm_provider_data(nvdimm);
2602 if (!nfit_mem || !nfit_mem->bdw) {
2603 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2604 spa->range_index, nvdimm_name(nvdimm));
2605 } else {
2606 mapping->size = nfit_mem->bdw->capacity;
2607 mapping->start = nfit_mem->bdw->start_address;
2608 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2609 blk_valid = 1;
2610 }
2611
2612 ndr_desc->mapping = mapping;
2613 ndr_desc->num_mappings = blk_valid;
2614 ndbr_desc = to_blk_region_desc(ndr_desc);
2615 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2616 ndbr_desc->do_io = acpi_desc->blk_do_io;
2617 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2618 if (rc)
2619 return rc;
2620 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2621 ndr_desc);
2622 if (!nfit_spa->nd_region)
2623 return -ENOMEM;
2624 break;
2625 }
2626
2627 return 0;
2628 }
2629
2630 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2631 {
2632 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2633 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2634 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2635 nfit_spa_type(spa) == NFIT_SPA_PCD);
2636 }
2637
2638 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2639 {
2640 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2641 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2642 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2643 }
2644
2645 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2646 struct nfit_spa *nfit_spa)
2647 {
2648 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2649 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2650 struct nd_blk_region_desc ndbr_desc;
2651 struct nd_region_desc *ndr_desc;
2652 struct nfit_memdev *nfit_memdev;
2653 struct nvdimm_bus *nvdimm_bus;
2654 struct resource res;
2655 int count = 0, rc;
2656
2657 if (nfit_spa->nd_region)
2658 return 0;
2659
2660 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2661 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2662 __func__);
2663 return 0;
2664 }
2665
2666 memset(&res, 0, sizeof(res));
2667 memset(&mappings, 0, sizeof(mappings));
2668 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2669 res.start = spa->address;
2670 res.end = res.start + spa->length - 1;
2671 ndr_desc = &ndbr_desc.ndr_desc;
2672 ndr_desc->res = &res;
2673 ndr_desc->provider_data = nfit_spa;
2674 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2675 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2676 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2677 spa->proximity_domain);
2678 else
2679 ndr_desc->numa_node = NUMA_NO_NODE;
2680
2681 /*
2682 * Persistence domain bits are hierarchical, if
2683 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2684 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2685 */
2686 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2687 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2688 else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2689 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2690
2691 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2692 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2693 struct nd_mapping_desc *mapping;
2694
2695 if (memdev->range_index != spa->range_index)
2696 continue;
2697 if (count >= ND_MAX_MAPPINGS) {
2698 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2699 spa->range_index, ND_MAX_MAPPINGS);
2700 return -ENXIO;
2701 }
2702 mapping = &mappings[count++];
2703 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2704 memdev, nfit_spa);
2705 if (rc)
2706 goto out;
2707 }
2708
2709 ndr_desc->mapping = mappings;
2710 ndr_desc->num_mappings = count;
2711 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2712 if (rc)
2713 goto out;
2714
2715 nvdimm_bus = acpi_desc->nvdimm_bus;
2716 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2717 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2718 if (rc) {
2719 dev_warn(acpi_desc->dev,
2720 "failed to insert pmem resource to iomem: %d\n",
2721 rc);
2722 goto out;
2723 }
2724
2725 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2726 ndr_desc);
2727 if (!nfit_spa->nd_region)
2728 rc = -ENOMEM;
2729 } else if (nfit_spa_is_volatile(spa)) {
2730 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2731 ndr_desc);
2732 if (!nfit_spa->nd_region)
2733 rc = -ENOMEM;
2734 } else if (nfit_spa_is_virtual(spa)) {
2735 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2736 ndr_desc);
2737 if (!nfit_spa->nd_region)
2738 rc = -ENOMEM;
2739 }
2740
2741 out:
2742 if (rc)
2743 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2744 nfit_spa->spa->range_index);
2745 return rc;
2746 }
2747
2748 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2749 u32 max_ars)
2750 {
2751 struct device *dev = acpi_desc->dev;
2752 struct nd_cmd_ars_status *ars_status;
2753
2754 if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2755 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2756 return 0;
2757 }
2758
2759 if (acpi_desc->ars_status)
2760 devm_kfree(dev, acpi_desc->ars_status);
2761 acpi_desc->ars_status = NULL;
2762 ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2763 if (!ars_status)
2764 return -ENOMEM;
2765 acpi_desc->ars_status = ars_status;
2766 acpi_desc->ars_status_size = max_ars;
2767 return 0;
2768 }
2769
2770 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2771 struct nfit_spa *nfit_spa)
2772 {
2773 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2774 int rc;
2775
2776 if (!nfit_spa->max_ars) {
2777 struct nd_cmd_ars_cap ars_cap;
2778
2779 memset(&ars_cap, 0, sizeof(ars_cap));
2780 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2781 if (rc < 0)
2782 return rc;
2783 nfit_spa->max_ars = ars_cap.max_ars_out;
2784 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2785 /* check that the supported scrub types match the spa type */
2786 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2787 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2788 return -ENOTTY;
2789 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2790 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2791 return -ENOTTY;
2792 }
2793
2794 if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2795 return -ENOMEM;
2796
2797 rc = ars_get_status(acpi_desc);
2798 if (rc < 0 && rc != -ENOSPC)
2799 return rc;
2800
2801 if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2802 return -ENOMEM;
2803
2804 return 0;
2805 }
2806
2807 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2808 struct nfit_spa *nfit_spa)
2809 {
2810 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2811 unsigned int overflow_retry = scrub_overflow_abort;
2812 u64 init_ars_start = 0, init_ars_len = 0;
2813 struct device *dev = acpi_desc->dev;
2814 unsigned int tmo = scrub_timeout;
2815 int rc;
2816
2817 if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2818 return;
2819
2820 rc = ars_start(acpi_desc, nfit_spa);
2821 /*
2822 * If we timed out the initial scan we'll still be busy here,
2823 * and will wait another timeout before giving up permanently.
2824 */
2825 if (rc < 0 && rc != -EBUSY)
2826 return;
2827
2828 do {
2829 u64 ars_start, ars_len;
2830
2831 if (acpi_desc->cancel)
2832 break;
2833 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2834 if (rc == -ENOTTY)
2835 break;
2836 if (rc == -EBUSY && !tmo) {
2837 dev_warn(dev, "range %d ars timeout, aborting\n",
2838 spa->range_index);
2839 break;
2840 }
2841
2842 if (rc == -EBUSY) {
2843 /*
2844 * Note, entries may be appended to the list
2845 * while the lock is dropped, but the workqueue
2846 * being active prevents entries being deleted /
2847 * freed.
2848 */
2849 mutex_unlock(&acpi_desc->init_mutex);
2850 ssleep(1);
2851 tmo--;
2852 mutex_lock(&acpi_desc->init_mutex);
2853 continue;
2854 }
2855
2856 /* we got some results, but there are more pending... */
2857 if (rc == -ENOSPC && overflow_retry--) {
2858 if (!init_ars_len) {
2859 init_ars_len = acpi_desc->ars_status->length;
2860 init_ars_start = acpi_desc->ars_status->address;
2861 }
2862 rc = ars_continue(acpi_desc);
2863 }
2864
2865 if (rc < 0) {
2866 dev_warn(dev, "range %d ars continuation failed\n",
2867 spa->range_index);
2868 break;
2869 }
2870
2871 if (init_ars_len) {
2872 ars_start = init_ars_start;
2873 ars_len = init_ars_len;
2874 } else {
2875 ars_start = acpi_desc->ars_status->address;
2876 ars_len = acpi_desc->ars_status->length;
2877 }
2878 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2879 spa->range_index, ars_start, ars_len);
2880 /* notify the region about new poison entries */
2881 nvdimm_region_notify(nfit_spa->nd_region,
2882 NVDIMM_REVALIDATE_POISON);
2883 break;
2884 } while (1);
2885 }
2886
2887 static void acpi_nfit_scrub(struct work_struct *work)
2888 {
2889 struct device *dev;
2890 u64 init_scrub_length = 0;
2891 struct nfit_spa *nfit_spa;
2892 u64 init_scrub_address = 0;
2893 bool init_ars_done = false;
2894 struct acpi_nfit_desc *acpi_desc;
2895 unsigned int tmo = scrub_timeout;
2896 unsigned int overflow_retry = scrub_overflow_abort;
2897
2898 acpi_desc = container_of(work, typeof(*acpi_desc), work);
2899 dev = acpi_desc->dev;
2900
2901 /*
2902 * We scrub in 2 phases. The first phase waits for any platform
2903 * firmware initiated scrubs to complete and then we go search for the
2904 * affected spa regions to mark them scanned. In the second phase we
2905 * initiate a directed scrub for every range that was not scrubbed in
2906 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2907 * the first phase, but really only care about running phase 2, where
2908 * regions can be notified of new poison.
2909 */
2910
2911 /* process platform firmware initiated scrubs */
2912 retry:
2913 mutex_lock(&acpi_desc->init_mutex);
2914 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2915 struct nd_cmd_ars_status *ars_status;
2916 struct acpi_nfit_system_address *spa;
2917 u64 ars_start, ars_len;
2918 int rc;
2919
2920 if (acpi_desc->cancel)
2921 break;
2922
2923 if (nfit_spa->nd_region)
2924 continue;
2925
2926 if (init_ars_done) {
2927 /*
2928 * No need to re-query, we're now just
2929 * reconciling all the ranges covered by the
2930 * initial scrub
2931 */
2932 rc = 0;
2933 } else
2934 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2935
2936 if (rc == -ENOTTY) {
2937 /* no ars capability, just register spa and move on */
2938 acpi_nfit_register_region(acpi_desc, nfit_spa);
2939 continue;
2940 }
2941
2942 if (rc == -EBUSY && !tmo) {
2943 /* fallthrough to directed scrub in phase 2 */
2944 dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2945 break;
2946 } else if (rc == -EBUSY) {
2947 mutex_unlock(&acpi_desc->init_mutex);
2948 ssleep(1);
2949 tmo--;
2950 goto retry;
2951 }
2952
2953 /* we got some results, but there are more pending... */
2954 if (rc == -ENOSPC && overflow_retry--) {
2955 ars_status = acpi_desc->ars_status;
2956 /*
2957 * Record the original scrub range, so that we
2958 * can recall all the ranges impacted by the
2959 * initial scrub.
2960 */
2961 if (!init_scrub_length) {
2962 init_scrub_length = ars_status->length;
2963 init_scrub_address = ars_status->address;
2964 }
2965 rc = ars_continue(acpi_desc);
2966 if (rc == 0) {
2967 mutex_unlock(&acpi_desc->init_mutex);
2968 goto retry;
2969 }
2970 }
2971
2972 if (rc < 0) {
2973 /*
2974 * Initial scrub failed, we'll give it one more
2975 * try below...
2976 */
2977 break;
2978 }
2979
2980 /* We got some final results, record completed ranges */
2981 ars_status = acpi_desc->ars_status;
2982 if (init_scrub_length) {
2983 ars_start = init_scrub_address;
2984 ars_len = ars_start + init_scrub_length;
2985 } else {
2986 ars_start = ars_status->address;
2987 ars_len = ars_status->length;
2988 }
2989 spa = nfit_spa->spa;
2990
2991 if (!init_ars_done) {
2992 init_ars_done = true;
2993 dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2994 ars_start, ars_len);
2995 }
2996 if (ars_start <= spa->address && ars_start + ars_len
2997 >= spa->address + spa->length)
2998 acpi_nfit_register_region(acpi_desc, nfit_spa);
2999 }
3000
3001 /*
3002 * For all the ranges not covered by an initial scrub we still
3003 * want to see if there are errors, but it's ok to discover them
3004 * asynchronously.
3005 */
3006 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3007 /*
3008 * Flag all the ranges that still need scrubbing, but
3009 * register them now to make data available.
3010 */
3011 if (!nfit_spa->nd_region) {
3012 nfit_spa->ars_required = 1;
3013 acpi_nfit_register_region(acpi_desc, nfit_spa);
3014 }
3015 }
3016 acpi_desc->init_complete = 1;
3017
3018 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3019 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
3020 acpi_desc->scrub_count++;
3021 acpi_desc->ars_start_flags = 0;
3022 if (acpi_desc->scrub_count_state)
3023 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3024 mutex_unlock(&acpi_desc->init_mutex);
3025 }
3026
3027 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3028 {
3029 struct nfit_spa *nfit_spa;
3030
3031 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3032 int rc, type = nfit_spa_type(nfit_spa->spa);
3033
3034 /* PMEM and VMEM will be registered by the ARS workqueue */
3035 if (type == NFIT_SPA_PM || type == NFIT_SPA_VOLATILE)
3036 continue;
3037 /* BLK apertures belong to BLK region registration below */
3038 if (type == NFIT_SPA_BDW)
3039 continue;
3040 /* BLK regions don't need to wait for ARS results */
3041 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3042 if (rc)
3043 return rc;
3044 }
3045
3046 acpi_desc->ars_start_flags = 0;
3047 if (!acpi_desc->cancel)
3048 queue_work(nfit_wq, &acpi_desc->work);
3049 return 0;
3050 }
3051
3052 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3053 struct nfit_table_prev *prev)
3054 {
3055 struct device *dev = acpi_desc->dev;
3056
3057 if (!list_empty(&prev->spas) ||
3058 !list_empty(&prev->memdevs) ||
3059 !list_empty(&prev->dcrs) ||
3060 !list_empty(&prev->bdws) ||
3061 !list_empty(&prev->idts) ||
3062 !list_empty(&prev->flushes)) {
3063 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3064 return -ENXIO;
3065 }
3066 return 0;
3067 }
3068
3069 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3070 {
3071 struct device *dev = acpi_desc->dev;
3072 struct kernfs_node *nfit;
3073 struct device *bus_dev;
3074
3075 if (!ars_supported(acpi_desc->nvdimm_bus))
3076 return 0;
3077
3078 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3079 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3080 if (!nfit) {
3081 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3082 return -ENODEV;
3083 }
3084 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3085 sysfs_put(nfit);
3086 if (!acpi_desc->scrub_count_state) {
3087 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3088 return -ENODEV;
3089 }
3090
3091 return 0;
3092 }
3093
3094 static void acpi_nfit_unregister(void *data)
3095 {
3096 struct acpi_nfit_desc *acpi_desc = data;
3097
3098 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3099 }
3100
3101 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3102 {
3103 struct device *dev = acpi_desc->dev;
3104 struct nfit_table_prev prev;
3105 const void *end;
3106 int rc;
3107
3108 if (!acpi_desc->nvdimm_bus) {
3109 acpi_nfit_init_dsms(acpi_desc);
3110
3111 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3112 &acpi_desc->nd_desc);
3113 if (!acpi_desc->nvdimm_bus)
3114 return -ENOMEM;
3115
3116 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3117 acpi_desc);
3118 if (rc)
3119 return rc;
3120
3121 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3122 if (rc)
3123 return rc;
3124
3125 /* register this acpi_desc for mce notifications */
3126 mutex_lock(&acpi_desc_lock);
3127 list_add_tail(&acpi_desc->list, &acpi_descs);
3128 mutex_unlock(&acpi_desc_lock);
3129 }
3130
3131 mutex_lock(&acpi_desc->init_mutex);
3132
3133 INIT_LIST_HEAD(&prev.spas);
3134 INIT_LIST_HEAD(&prev.memdevs);
3135 INIT_LIST_HEAD(&prev.dcrs);
3136 INIT_LIST_HEAD(&prev.bdws);
3137 INIT_LIST_HEAD(&prev.idts);
3138 INIT_LIST_HEAD(&prev.flushes);
3139
3140 list_cut_position(&prev.spas, &acpi_desc->spas,
3141 acpi_desc->spas.prev);
3142 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3143 acpi_desc->memdevs.prev);
3144 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3145 acpi_desc->dcrs.prev);
3146 list_cut_position(&prev.bdws, &acpi_desc->bdws,
3147 acpi_desc->bdws.prev);
3148 list_cut_position(&prev.idts, &acpi_desc->idts,
3149 acpi_desc->idts.prev);
3150 list_cut_position(&prev.flushes, &acpi_desc->flushes,
3151 acpi_desc->flushes.prev);
3152
3153 end = data + sz;
3154 while (!IS_ERR_OR_NULL(data))
3155 data = add_table(acpi_desc, &prev, data, end);
3156
3157 if (IS_ERR(data)) {
3158 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
3159 PTR_ERR(data));
3160 rc = PTR_ERR(data);
3161 goto out_unlock;
3162 }
3163
3164 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3165 if (rc)
3166 goto out_unlock;
3167
3168 rc = nfit_mem_init(acpi_desc);
3169 if (rc)
3170 goto out_unlock;
3171
3172 rc = acpi_nfit_register_dimms(acpi_desc);
3173 if (rc)
3174 goto out_unlock;
3175
3176 rc = acpi_nfit_register_regions(acpi_desc);
3177
3178 out_unlock:
3179 mutex_unlock(&acpi_desc->init_mutex);
3180 return rc;
3181 }
3182 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3183
3184 struct acpi_nfit_flush_work {
3185 struct work_struct work;
3186 struct completion cmp;
3187 };
3188
3189 static void flush_probe(struct work_struct *work)
3190 {
3191 struct acpi_nfit_flush_work *flush;
3192
3193 flush = container_of(work, typeof(*flush), work);
3194 complete(&flush->cmp);
3195 }
3196
3197 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3198 {
3199 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3200 struct device *dev = acpi_desc->dev;
3201 struct acpi_nfit_flush_work flush;
3202 int rc;
3203
3204 /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3205 device_lock(dev);
3206 device_unlock(dev);
3207
3208 /* bounce the init_mutex to make init_complete valid */
3209 mutex_lock(&acpi_desc->init_mutex);
3210 if (acpi_desc->cancel || acpi_desc->init_complete) {
3211 mutex_unlock(&acpi_desc->init_mutex);
3212 return 0;
3213 }
3214
3215 /*
3216 * Scrub work could take 10s of seconds, userspace may give up so we
3217 * need to be interruptible while waiting.
3218 */
3219 INIT_WORK_ONSTACK(&flush.work, flush_probe);
3220 init_completion(&flush.cmp);
3221 queue_work(nfit_wq, &flush.work);
3222 mutex_unlock(&acpi_desc->init_mutex);
3223
3224 rc = wait_for_completion_interruptible(&flush.cmp);
3225 cancel_work_sync(&flush.work);
3226 return rc;
3227 }
3228
3229 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3230 struct nvdimm *nvdimm, unsigned int cmd)
3231 {
3232 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3233
3234 if (nvdimm)
3235 return 0;
3236 if (cmd != ND_CMD_ARS_START)
3237 return 0;
3238
3239 /*
3240 * The kernel and userspace may race to initiate a scrub, but
3241 * the scrub thread is prepared to lose that initial race. It
3242 * just needs guarantees that any ars it initiates are not
3243 * interrupted by any intervening start reqeusts from userspace.
3244 */
3245 if (work_busy(&acpi_desc->work))
3246 return -EBUSY;
3247
3248 return 0;
3249 }
3250
3251 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
3252 {
3253 struct device *dev = acpi_desc->dev;
3254 struct nfit_spa *nfit_spa;
3255
3256 if (work_busy(&acpi_desc->work))
3257 return -EBUSY;
3258
3259 mutex_lock(&acpi_desc->init_mutex);
3260 if (acpi_desc->cancel) {
3261 mutex_unlock(&acpi_desc->init_mutex);
3262 return 0;
3263 }
3264
3265 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3266 struct acpi_nfit_system_address *spa = nfit_spa->spa;
3267
3268 if (nfit_spa_type(spa) != NFIT_SPA_PM)
3269 continue;
3270
3271 nfit_spa->ars_required = 1;
3272 }
3273 acpi_desc->ars_start_flags = flags;
3274 queue_work(nfit_wq, &acpi_desc->work);
3275 dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
3276 mutex_unlock(&acpi_desc->init_mutex);
3277
3278 return 0;
3279 }
3280
3281 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3282 {
3283 struct nvdimm_bus_descriptor *nd_desc;
3284
3285 dev_set_drvdata(dev, acpi_desc);
3286 acpi_desc->dev = dev;
3287 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3288 nd_desc = &acpi_desc->nd_desc;
3289 nd_desc->provider_name = "ACPI.NFIT";
3290 nd_desc->module = THIS_MODULE;
3291 nd_desc->ndctl = acpi_nfit_ctl;
3292 nd_desc->flush_probe = acpi_nfit_flush_probe;
3293 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3294 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3295
3296 INIT_LIST_HEAD(&acpi_desc->spas);
3297 INIT_LIST_HEAD(&acpi_desc->dcrs);
3298 INIT_LIST_HEAD(&acpi_desc->bdws);
3299 INIT_LIST_HEAD(&acpi_desc->idts);
3300 INIT_LIST_HEAD(&acpi_desc->flushes);
3301 INIT_LIST_HEAD(&acpi_desc->memdevs);
3302 INIT_LIST_HEAD(&acpi_desc->dimms);
3303 INIT_LIST_HEAD(&acpi_desc->list);
3304 mutex_init(&acpi_desc->init_mutex);
3305 INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3306 }
3307 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3308
3309 static void acpi_nfit_put_table(void *table)
3310 {
3311 acpi_put_table(table);
3312 }
3313
3314 void acpi_nfit_shutdown(void *data)
3315 {
3316 struct acpi_nfit_desc *acpi_desc = data;
3317 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3318
3319 /*
3320 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3321 * race teardown
3322 */
3323 mutex_lock(&acpi_desc_lock);
3324 list_del(&acpi_desc->list);
3325 mutex_unlock(&acpi_desc_lock);
3326
3327 mutex_lock(&acpi_desc->init_mutex);
3328 acpi_desc->cancel = 1;
3329 mutex_unlock(&acpi_desc->init_mutex);
3330
3331 /*
3332 * Bounce the nvdimm bus lock to make sure any in-flight
3333 * acpi_nfit_ars_rescan() submissions have had a chance to
3334 * either submit or see ->cancel set.
3335 */
3336 device_lock(bus_dev);
3337 device_unlock(bus_dev);
3338
3339 flush_workqueue(nfit_wq);
3340 }
3341 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3342
3343 static int acpi_nfit_add(struct acpi_device *adev)
3344 {
3345 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3346 struct acpi_nfit_desc *acpi_desc;
3347 struct device *dev = &adev->dev;
3348 struct acpi_table_header *tbl;
3349 acpi_status status = AE_OK;
3350 acpi_size sz;
3351 int rc = 0;
3352
3353 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3354 if (ACPI_FAILURE(status)) {
3355 /* This is ok, we could have an nvdimm hotplugged later */
3356 dev_dbg(dev, "failed to find NFIT at startup\n");
3357 return 0;
3358 }
3359
3360 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3361 if (rc)
3362 return rc;
3363 sz = tbl->length;
3364
3365 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3366 if (!acpi_desc)
3367 return -ENOMEM;
3368 acpi_nfit_desc_init(acpi_desc, &adev->dev);
3369
3370 /* Save the acpi header for exporting the revision via sysfs */
3371 acpi_desc->acpi_header = *tbl;
3372
3373 /* Evaluate _FIT and override with that if present */
3374 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3375 if (ACPI_SUCCESS(status) && buf.length > 0) {
3376 union acpi_object *obj = buf.pointer;
3377
3378 if (obj->type == ACPI_TYPE_BUFFER)
3379 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3380 obj->buffer.length);
3381 else
3382 dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
3383 __func__, (int) obj->type);
3384 kfree(buf.pointer);
3385 } else
3386 /* skip over the lead-in header table */
3387 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3388 + sizeof(struct acpi_table_nfit),
3389 sz - sizeof(struct acpi_table_nfit));
3390
3391 if (rc)
3392 return rc;
3393 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3394 }
3395
3396 static int acpi_nfit_remove(struct acpi_device *adev)
3397 {
3398 /* see acpi_nfit_unregister */
3399 return 0;
3400 }
3401
3402 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3403 {
3404 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3405 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3406 union acpi_object *obj;
3407 acpi_status status;
3408 int ret;
3409
3410 if (!dev->driver) {
3411 /* dev->driver may be null if we're being removed */
3412 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
3413 return;
3414 }
3415
3416 if (!acpi_desc) {
3417 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3418 if (!acpi_desc)
3419 return;
3420 acpi_nfit_desc_init(acpi_desc, dev);
3421 } else {
3422 /*
3423 * Finish previous registration before considering new
3424 * regions.
3425 */
3426 flush_workqueue(nfit_wq);
3427 }
3428
3429 /* Evaluate _FIT */
3430 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3431 if (ACPI_FAILURE(status)) {
3432 dev_err(dev, "failed to evaluate _FIT\n");
3433 return;
3434 }
3435
3436 obj = buf.pointer;
3437 if (obj->type == ACPI_TYPE_BUFFER) {
3438 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3439 obj->buffer.length);
3440 if (ret)
3441 dev_err(dev, "failed to merge updated NFIT\n");
3442 } else
3443 dev_err(dev, "Invalid _FIT\n");
3444 kfree(buf.pointer);
3445 }
3446
3447 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3448 {
3449 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3450 u8 flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3451 0 : ND_ARS_RETURN_PREV_DATA;
3452
3453 acpi_nfit_ars_rescan(acpi_desc, flags);
3454 }
3455
3456 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3457 {
3458 dev_dbg(dev, "%s: event: 0x%x\n", __func__, event);
3459
3460 switch (event) {
3461 case NFIT_NOTIFY_UPDATE:
3462 return acpi_nfit_update_notify(dev, handle);
3463 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3464 return acpi_nfit_uc_error_notify(dev, handle);
3465 default:
3466 return;
3467 }
3468 }
3469 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3470
3471 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3472 {
3473 device_lock(&adev->dev);
3474 __acpi_nfit_notify(&adev->dev, adev->handle, event);
3475 device_unlock(&adev->dev);
3476 }
3477
3478 static const struct acpi_device_id acpi_nfit_ids[] = {
3479 { "ACPI0012", 0 },
3480 { "", 0 },
3481 };
3482 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3483
3484 static struct acpi_driver acpi_nfit_driver = {
3485 .name = KBUILD_MODNAME,
3486 .ids = acpi_nfit_ids,
3487 .ops = {
3488 .add = acpi_nfit_add,
3489 .remove = acpi_nfit_remove,
3490 .notify = acpi_nfit_notify,
3491 },
3492 };
3493
3494 static __init int nfit_init(void)
3495 {
3496 int ret;
3497
3498 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3499 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3500 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3501 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3502 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3503 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3504 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3505 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3506
3507 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3508 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3509 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3510 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3511 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3512 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3513 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3514 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3515 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3516 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3517 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3518 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3519 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3520
3521 nfit_wq = create_singlethread_workqueue("nfit");
3522 if (!nfit_wq)
3523 return -ENOMEM;
3524
3525 nfit_mce_register();
3526 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3527 if (ret) {
3528 nfit_mce_unregister();
3529 destroy_workqueue(nfit_wq);
3530 }
3531
3532 return ret;
3533
3534 }
3535
3536 static __exit void nfit_exit(void)
3537 {
3538 nfit_mce_unregister();
3539 acpi_bus_unregister_driver(&acpi_nfit_driver);
3540 destroy_workqueue(nfit_wq);
3541 WARN_ON(!list_empty(&acpi_descs));
3542 }
3543
3544 module_init(nfit_init);
3545 module_exit(nfit_exit);
3546 MODULE_LICENSE("GPL v2");
3547 MODULE_AUTHOR("Intel Corporation");
ubuntu-bionic-kernel mirror