Merge branch 'for-4.16/nfit' into libnvdimm-for-next
[sfrench/cifs-2.6.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                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1254                                 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
1255         }
1256         device_unlock(dev);
1257         return rc;
1258 }
1259
1260 static ssize_t scrub_store(struct device *dev,
1261                 struct device_attribute *attr, const char *buf, size_t size)
1262 {
1263         struct nvdimm_bus_descriptor *nd_desc;
1264         ssize_t rc;
1265         long val;
1266
1267         rc = kstrtol(buf, 0, &val);
1268         if (rc)
1269                 return rc;
1270         if (val != 1)
1271                 return -EINVAL;
1272
1273         device_lock(dev);
1274         nd_desc = dev_get_drvdata(dev);
1275         if (nd_desc) {
1276                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1277
1278                 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1279         }
1280         device_unlock(dev);
1281         if (rc)
1282                 return rc;
1283         return size;
1284 }
1285 static DEVICE_ATTR_RW(scrub);
1286
1287 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1288 {
1289         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1290         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1291                 | 1 << ND_CMD_ARS_STATUS;
1292
1293         return (nd_desc->cmd_mask & mask) == mask;
1294 }
1295
1296 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1297 {
1298         struct device *dev = container_of(kobj, struct device, kobj);
1299         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1300
1301         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1302                 return 0;
1303         return a->mode;
1304 }
1305
1306 static struct attribute *acpi_nfit_attributes[] = {
1307         &dev_attr_revision.attr,
1308         &dev_attr_scrub.attr,
1309         &dev_attr_hw_error_scrub.attr,
1310         &dev_attr_bus_dsm_mask.attr,
1311         NULL,
1312 };
1313
1314 static const struct attribute_group acpi_nfit_attribute_group = {
1315         .name = "nfit",
1316         .attrs = acpi_nfit_attributes,
1317         .is_visible = nfit_visible,
1318 };
1319
1320 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1321         &nvdimm_bus_attribute_group,
1322         &acpi_nfit_attribute_group,
1323         NULL,
1324 };
1325
1326 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1327 {
1328         struct nvdimm *nvdimm = to_nvdimm(dev);
1329         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1330
1331         return __to_nfit_memdev(nfit_mem);
1332 }
1333
1334 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1335 {
1336         struct nvdimm *nvdimm = to_nvdimm(dev);
1337         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1338
1339         return nfit_mem->dcr;
1340 }
1341
1342 static ssize_t handle_show(struct device *dev,
1343                 struct device_attribute *attr, char *buf)
1344 {
1345         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1346
1347         return sprintf(buf, "%#x\n", memdev->device_handle);
1348 }
1349 static DEVICE_ATTR_RO(handle);
1350
1351 static ssize_t phys_id_show(struct device *dev,
1352                 struct device_attribute *attr, char *buf)
1353 {
1354         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1355
1356         return sprintf(buf, "%#x\n", memdev->physical_id);
1357 }
1358 static DEVICE_ATTR_RO(phys_id);
1359
1360 static ssize_t vendor_show(struct device *dev,
1361                 struct device_attribute *attr, char *buf)
1362 {
1363         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1364
1365         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1366 }
1367 static DEVICE_ATTR_RO(vendor);
1368
1369 static ssize_t rev_id_show(struct device *dev,
1370                 struct device_attribute *attr, char *buf)
1371 {
1372         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1373
1374         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1375 }
1376 static DEVICE_ATTR_RO(rev_id);
1377
1378 static ssize_t device_show(struct device *dev,
1379                 struct device_attribute *attr, char *buf)
1380 {
1381         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1382
1383         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1384 }
1385 static DEVICE_ATTR_RO(device);
1386
1387 static ssize_t subsystem_vendor_show(struct device *dev,
1388                 struct device_attribute *attr, char *buf)
1389 {
1390         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1391
1392         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1393 }
1394 static DEVICE_ATTR_RO(subsystem_vendor);
1395
1396 static ssize_t subsystem_rev_id_show(struct device *dev,
1397                 struct device_attribute *attr, char *buf)
1398 {
1399         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1400
1401         return sprintf(buf, "0x%04x\n",
1402                         be16_to_cpu(dcr->subsystem_revision_id));
1403 }
1404 static DEVICE_ATTR_RO(subsystem_rev_id);
1405
1406 static ssize_t subsystem_device_show(struct device *dev,
1407                 struct device_attribute *attr, char *buf)
1408 {
1409         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1410
1411         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1412 }
1413 static DEVICE_ATTR_RO(subsystem_device);
1414
1415 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1416 {
1417         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1418         int formats = 0;
1419
1420         if (nfit_mem->memdev_pmem)
1421                 formats++;
1422         if (nfit_mem->memdev_bdw)
1423                 formats++;
1424         return formats;
1425 }
1426
1427 static ssize_t format_show(struct device *dev,
1428                 struct device_attribute *attr, char *buf)
1429 {
1430         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1431
1432         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1433 }
1434 static DEVICE_ATTR_RO(format);
1435
1436 static ssize_t format1_show(struct device *dev,
1437                 struct device_attribute *attr, char *buf)
1438 {
1439         u32 handle;
1440         ssize_t rc = -ENXIO;
1441         struct nfit_mem *nfit_mem;
1442         struct nfit_memdev *nfit_memdev;
1443         struct acpi_nfit_desc *acpi_desc;
1444         struct nvdimm *nvdimm = to_nvdimm(dev);
1445         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1446
1447         nfit_mem = nvdimm_provider_data(nvdimm);
1448         acpi_desc = nfit_mem->acpi_desc;
1449         handle = to_nfit_memdev(dev)->device_handle;
1450
1451         /* assumes DIMMs have at most 2 published interface codes */
1452         mutex_lock(&acpi_desc->init_mutex);
1453         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1454                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1455                 struct nfit_dcr *nfit_dcr;
1456
1457                 if (memdev->device_handle != handle)
1458                         continue;
1459
1460                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1461                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1462                                 continue;
1463                         if (nfit_dcr->dcr->code == dcr->code)
1464                                 continue;
1465                         rc = sprintf(buf, "0x%04x\n",
1466                                         le16_to_cpu(nfit_dcr->dcr->code));
1467                         break;
1468                 }
1469                 if (rc != ENXIO)
1470                         break;
1471         }
1472         mutex_unlock(&acpi_desc->init_mutex);
1473         return rc;
1474 }
1475 static DEVICE_ATTR_RO(format1);
1476
1477 static ssize_t formats_show(struct device *dev,
1478                 struct device_attribute *attr, char *buf)
1479 {
1480         struct nvdimm *nvdimm = to_nvdimm(dev);
1481
1482         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1483 }
1484 static DEVICE_ATTR_RO(formats);
1485
1486 static ssize_t serial_show(struct device *dev,
1487                 struct device_attribute *attr, char *buf)
1488 {
1489         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1490
1491         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1492 }
1493 static DEVICE_ATTR_RO(serial);
1494
1495 static ssize_t family_show(struct device *dev,
1496                 struct device_attribute *attr, char *buf)
1497 {
1498         struct nvdimm *nvdimm = to_nvdimm(dev);
1499         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1500
1501         if (nfit_mem->family < 0)
1502                 return -ENXIO;
1503         return sprintf(buf, "%d\n", nfit_mem->family);
1504 }
1505 static DEVICE_ATTR_RO(family);
1506
1507 static ssize_t dsm_mask_show(struct device *dev,
1508                 struct device_attribute *attr, char *buf)
1509 {
1510         struct nvdimm *nvdimm = to_nvdimm(dev);
1511         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1512
1513         if (nfit_mem->family < 0)
1514                 return -ENXIO;
1515         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1516 }
1517 static DEVICE_ATTR_RO(dsm_mask);
1518
1519 static ssize_t flags_show(struct device *dev,
1520                 struct device_attribute *attr, char *buf)
1521 {
1522         u16 flags = to_nfit_memdev(dev)->flags;
1523
1524         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1525                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1526                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1527                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1528                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1529                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1530                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1531                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1532 }
1533 static DEVICE_ATTR_RO(flags);
1534
1535 static ssize_t id_show(struct device *dev,
1536                 struct device_attribute *attr, char *buf)
1537 {
1538         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1539
1540         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1541                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1542                                 be16_to_cpu(dcr->vendor_id),
1543                                 dcr->manufacturing_location,
1544                                 be16_to_cpu(dcr->manufacturing_date),
1545                                 be32_to_cpu(dcr->serial_number));
1546         else
1547                 return sprintf(buf, "%04x-%08x\n",
1548                                 be16_to_cpu(dcr->vendor_id),
1549                                 be32_to_cpu(dcr->serial_number));
1550 }
1551 static DEVICE_ATTR_RO(id);
1552
1553 static struct attribute *acpi_nfit_dimm_attributes[] = {
1554         &dev_attr_handle.attr,
1555         &dev_attr_phys_id.attr,
1556         &dev_attr_vendor.attr,
1557         &dev_attr_device.attr,
1558         &dev_attr_rev_id.attr,
1559         &dev_attr_subsystem_vendor.attr,
1560         &dev_attr_subsystem_device.attr,
1561         &dev_attr_subsystem_rev_id.attr,
1562         &dev_attr_format.attr,
1563         &dev_attr_formats.attr,
1564         &dev_attr_format1.attr,
1565         &dev_attr_serial.attr,
1566         &dev_attr_flags.attr,
1567         &dev_attr_id.attr,
1568         &dev_attr_family.attr,
1569         &dev_attr_dsm_mask.attr,
1570         NULL,
1571 };
1572
1573 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1574                 struct attribute *a, int n)
1575 {
1576         struct device *dev = container_of(kobj, struct device, kobj);
1577         struct nvdimm *nvdimm = to_nvdimm(dev);
1578
1579         if (!to_nfit_dcr(dev)) {
1580                 /* Without a dcr only the memdev attributes can be surfaced */
1581                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1582                                 || a == &dev_attr_flags.attr
1583                                 || a == &dev_attr_family.attr
1584                                 || a == &dev_attr_dsm_mask.attr)
1585                         return a->mode;
1586                 return 0;
1587         }
1588
1589         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1590                 return 0;
1591         return a->mode;
1592 }
1593
1594 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1595         .name = "nfit",
1596         .attrs = acpi_nfit_dimm_attributes,
1597         .is_visible = acpi_nfit_dimm_attr_visible,
1598 };
1599
1600 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1601         &nvdimm_attribute_group,
1602         &nd_device_attribute_group,
1603         &acpi_nfit_dimm_attribute_group,
1604         NULL,
1605 };
1606
1607 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1608                 u32 device_handle)
1609 {
1610         struct nfit_mem *nfit_mem;
1611
1612         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1613                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1614                         return nfit_mem->nvdimm;
1615
1616         return NULL;
1617 }
1618
1619 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1620 {
1621         struct nfit_mem *nfit_mem;
1622         struct acpi_nfit_desc *acpi_desc;
1623
1624         dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1625                         event);
1626
1627         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1628                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1629                                 event);
1630                 return;
1631         }
1632
1633         acpi_desc = dev_get_drvdata(dev->parent);
1634         if (!acpi_desc)
1635                 return;
1636
1637         /*
1638          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1639          * is still valid.
1640          */
1641         nfit_mem = dev_get_drvdata(dev);
1642         if (nfit_mem && nfit_mem->flags_attr)
1643                 sysfs_notify_dirent(nfit_mem->flags_attr);
1644 }
1645 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1646
1647 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1648 {
1649         struct acpi_device *adev = data;
1650         struct device *dev = &adev->dev;
1651
1652         device_lock(dev->parent);
1653         __acpi_nvdimm_notify(dev, event);
1654         device_unlock(dev->parent);
1655 }
1656
1657 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1658                 struct nfit_mem *nfit_mem, u32 device_handle)
1659 {
1660         struct acpi_device *adev, *adev_dimm;
1661         struct device *dev = acpi_desc->dev;
1662         union acpi_object *obj;
1663         unsigned long dsm_mask;
1664         const guid_t *guid;
1665         int i;
1666         int family = -1;
1667
1668         /* nfit test assumes 1:1 relationship between commands and dsms */
1669         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1670         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1671         adev = to_acpi_dev(acpi_desc);
1672         if (!adev)
1673                 return 0;
1674
1675         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1676         nfit_mem->adev = adev_dimm;
1677         if (!adev_dimm) {
1678                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1679                                 device_handle);
1680                 return force_enable_dimms ? 0 : -ENODEV;
1681         }
1682
1683         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1684                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1685                 dev_err(dev, "%s: notification registration failed\n",
1686                                 dev_name(&adev_dimm->dev));
1687                 return -ENXIO;
1688         }
1689         /*
1690          * Record nfit_mem for the notification path to track back to
1691          * the nfit sysfs attributes for this dimm device object.
1692          */
1693         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1694
1695         /*
1696          * Until standardization materializes we need to consider 4
1697          * different command sets.  Note, that checking for function0 (bit0)
1698          * tells us if any commands are reachable through this GUID.
1699          */
1700         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1701                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1702                         if (family < 0 || i == default_dsm_family)
1703                                 family = i;
1704
1705         /* limit the supported commands to those that are publicly documented */
1706         nfit_mem->family = family;
1707         if (override_dsm_mask && !disable_vendor_specific)
1708                 dsm_mask = override_dsm_mask;
1709         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1710                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1711                 if (disable_vendor_specific)
1712                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1713         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1714                 dsm_mask = 0x1c3c76;
1715         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1716                 dsm_mask = 0x1fe;
1717                 if (disable_vendor_specific)
1718                         dsm_mask &= ~(1 << 8);
1719         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1720                 dsm_mask = 0xffffffff;
1721         } else {
1722                 dev_dbg(dev, "unknown dimm command family\n");
1723                 nfit_mem->family = -1;
1724                 /* DSMs are optional, continue loading the driver... */
1725                 return 0;
1726         }
1727
1728         guid = to_nfit_uuid(nfit_mem->family);
1729         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1730                 if (acpi_check_dsm(adev_dimm->handle, guid,
1731                                         nfit_dsm_revid(nfit_mem->family, i),
1732                                         1ULL << i))
1733                         set_bit(i, &nfit_mem->dsm_mask);
1734
1735         obj = acpi_label_info(adev_dimm->handle);
1736         if (obj) {
1737                 ACPI_FREE(obj);
1738                 nfit_mem->has_lsi = 1;
1739                 dev_dbg(dev, "%s: has _LSI\n", dev_name(&adev_dimm->dev));
1740         }
1741
1742         obj = acpi_label_read(adev_dimm->handle, 0, 0);
1743         if (obj) {
1744                 ACPI_FREE(obj);
1745                 nfit_mem->has_lsr = 1;
1746                 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1747         }
1748
1749         obj = acpi_label_write(adev_dimm->handle, 0, 0, NULL);
1750         if (obj) {
1751                 ACPI_FREE(obj);
1752                 nfit_mem->has_lsw = 1;
1753                 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1754         }
1755
1756         return 0;
1757 }
1758
1759 static void shutdown_dimm_notify(void *data)
1760 {
1761         struct acpi_nfit_desc *acpi_desc = data;
1762         struct nfit_mem *nfit_mem;
1763
1764         mutex_lock(&acpi_desc->init_mutex);
1765         /*
1766          * Clear out the nfit_mem->flags_attr and shut down dimm event
1767          * notifications.
1768          */
1769         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1770                 struct acpi_device *adev_dimm = nfit_mem->adev;
1771
1772                 if (nfit_mem->flags_attr) {
1773                         sysfs_put(nfit_mem->flags_attr);
1774                         nfit_mem->flags_attr = NULL;
1775                 }
1776                 if (adev_dimm) {
1777                         acpi_remove_notify_handler(adev_dimm->handle,
1778                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1779                         dev_set_drvdata(&adev_dimm->dev, NULL);
1780                 }
1781         }
1782         mutex_unlock(&acpi_desc->init_mutex);
1783 }
1784
1785 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1786 {
1787         struct nfit_mem *nfit_mem;
1788         int dimm_count = 0, rc;
1789         struct nvdimm *nvdimm;
1790
1791         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1792                 struct acpi_nfit_flush_address *flush;
1793                 unsigned long flags = 0, cmd_mask;
1794                 struct nfit_memdev *nfit_memdev;
1795                 u32 device_handle;
1796                 u16 mem_flags;
1797
1798                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1799                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1800                 if (nvdimm) {
1801                         dimm_count++;
1802                         continue;
1803                 }
1804
1805                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1806                         set_bit(NDD_ALIASING, &flags);
1807
1808                 /* collate flags across all memdevs for this dimm */
1809                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1810                         struct acpi_nfit_memory_map *dimm_memdev;
1811
1812                         dimm_memdev = __to_nfit_memdev(nfit_mem);
1813                         if (dimm_memdev->device_handle
1814                                         != nfit_memdev->memdev->device_handle)
1815                                 continue;
1816                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
1817                 }
1818
1819                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1820                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1821                         set_bit(NDD_UNARMED, &flags);
1822
1823                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1824                 if (rc)
1825                         continue;
1826
1827                 /*
1828                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1829                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1830                  * userspace interface.
1831                  */
1832                 cmd_mask = 1UL << ND_CMD_CALL;
1833                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1834                         /*
1835                          * These commands have a 1:1 correspondence
1836                          * between DSM payload and libnvdimm ioctl
1837                          * payload format.
1838                          */
1839                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1840                 }
1841
1842                 if (nfit_mem->has_lsi)
1843                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1844                 if (nfit_mem->has_lsr)
1845                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1846                 if (nfit_mem->has_lsw)
1847                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1848
1849                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1850                         : NULL;
1851                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1852                                 acpi_nfit_dimm_attribute_groups,
1853                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1854                                 nfit_mem->flush_wpq);
1855                 if (!nvdimm)
1856                         return -ENOMEM;
1857
1858                 nfit_mem->nvdimm = nvdimm;
1859                 dimm_count++;
1860
1861                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1862                         continue;
1863
1864                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1865                                 nvdimm_name(nvdimm),
1866                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1867                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1868                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1869                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1870                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1871
1872         }
1873
1874         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1875         if (rc)
1876                 return rc;
1877
1878         /*
1879          * Now that dimms are successfully registered, and async registration
1880          * is flushed, attempt to enable event notification.
1881          */
1882         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1883                 struct kernfs_node *nfit_kernfs;
1884
1885                 nvdimm = nfit_mem->nvdimm;
1886                 if (!nvdimm)
1887                         continue;
1888
1889                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1890                 if (nfit_kernfs)
1891                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1892                                         "flags");
1893                 sysfs_put(nfit_kernfs);
1894                 if (!nfit_mem->flags_attr)
1895                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1896                                         nvdimm_name(nvdimm));
1897         }
1898
1899         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1900                         acpi_desc);
1901 }
1902
1903 /*
1904  * These constants are private because there are no kernel consumers of
1905  * these commands.
1906  */
1907 enum nfit_aux_cmds {
1908         NFIT_CMD_TRANSLATE_SPA = 5,
1909         NFIT_CMD_ARS_INJECT_SET = 7,
1910         NFIT_CMD_ARS_INJECT_CLEAR = 8,
1911         NFIT_CMD_ARS_INJECT_GET = 9,
1912 };
1913
1914 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1915 {
1916         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1917         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1918         struct acpi_device *adev;
1919         unsigned long dsm_mask;
1920         int i;
1921
1922         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1923         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1924         adev = to_acpi_dev(acpi_desc);
1925         if (!adev)
1926                 return;
1927
1928         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1929                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1930                         set_bit(i, &nd_desc->cmd_mask);
1931         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1932
1933         dsm_mask =
1934                 (1 << ND_CMD_ARS_CAP) |
1935                 (1 << ND_CMD_ARS_START) |
1936                 (1 << ND_CMD_ARS_STATUS) |
1937                 (1 << ND_CMD_CLEAR_ERROR) |
1938                 (1 << NFIT_CMD_TRANSLATE_SPA) |
1939                 (1 << NFIT_CMD_ARS_INJECT_SET) |
1940                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1941                 (1 << NFIT_CMD_ARS_INJECT_GET);
1942         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1943                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1944                         set_bit(i, &nd_desc->bus_dsm_mask);
1945 }
1946
1947 static ssize_t range_index_show(struct device *dev,
1948                 struct device_attribute *attr, char *buf)
1949 {
1950         struct nd_region *nd_region = to_nd_region(dev);
1951         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1952
1953         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1954 }
1955 static DEVICE_ATTR_RO(range_index);
1956
1957 static ssize_t ecc_unit_size_show(struct device *dev,
1958                 struct device_attribute *attr, char *buf)
1959 {
1960         struct nd_region *nd_region = to_nd_region(dev);
1961         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1962
1963         return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1964 }
1965 static DEVICE_ATTR_RO(ecc_unit_size);
1966
1967 static struct attribute *acpi_nfit_region_attributes[] = {
1968         &dev_attr_range_index.attr,
1969         &dev_attr_ecc_unit_size.attr,
1970         NULL,
1971 };
1972
1973 static const struct attribute_group acpi_nfit_region_attribute_group = {
1974         .name = "nfit",
1975         .attrs = acpi_nfit_region_attributes,
1976 };
1977
1978 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1979         &nd_region_attribute_group,
1980         &nd_mapping_attribute_group,
1981         &nd_device_attribute_group,
1982         &nd_numa_attribute_group,
1983         &acpi_nfit_region_attribute_group,
1984         NULL,
1985 };
1986
1987 /* enough info to uniquely specify an interleave set */
1988 struct nfit_set_info {
1989         struct nfit_set_info_map {
1990                 u64 region_offset;
1991                 u32 serial_number;
1992                 u32 pad;
1993         } mapping[0];
1994 };
1995
1996 struct nfit_set_info2 {
1997         struct nfit_set_info_map2 {
1998                 u64 region_offset;
1999                 u32 serial_number;
2000                 u16 vendor_id;
2001                 u16 manufacturing_date;
2002                 u8  manufacturing_location;
2003                 u8  reserved[31];
2004         } mapping[0];
2005 };
2006
2007 static size_t sizeof_nfit_set_info(int num_mappings)
2008 {
2009         return sizeof(struct nfit_set_info)
2010                 + num_mappings * sizeof(struct nfit_set_info_map);
2011 }
2012
2013 static size_t sizeof_nfit_set_info2(int num_mappings)
2014 {
2015         return sizeof(struct nfit_set_info2)
2016                 + num_mappings * sizeof(struct nfit_set_info_map2);
2017 }
2018
2019 static int cmp_map_compat(const void *m0, const void *m1)
2020 {
2021         const struct nfit_set_info_map *map0 = m0;
2022         const struct nfit_set_info_map *map1 = m1;
2023
2024         return memcmp(&map0->region_offset, &map1->region_offset,
2025                         sizeof(u64));
2026 }
2027
2028 static int cmp_map(const void *m0, const void *m1)
2029 {
2030         const struct nfit_set_info_map *map0 = m0;
2031         const struct nfit_set_info_map *map1 = m1;
2032
2033         if (map0->region_offset < map1->region_offset)
2034                 return -1;
2035         else if (map0->region_offset > map1->region_offset)
2036                 return 1;
2037         return 0;
2038 }
2039
2040 static int cmp_map2(const void *m0, const void *m1)
2041 {
2042         const struct nfit_set_info_map2 *map0 = m0;
2043         const struct nfit_set_info_map2 *map1 = m1;
2044
2045         if (map0->region_offset < map1->region_offset)
2046                 return -1;
2047         else if (map0->region_offset > map1->region_offset)
2048                 return 1;
2049         return 0;
2050 }
2051
2052 /* Retrieve the nth entry referencing this spa */
2053 static struct acpi_nfit_memory_map *memdev_from_spa(
2054                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2055 {
2056         struct nfit_memdev *nfit_memdev;
2057
2058         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2059                 if (nfit_memdev->memdev->range_index == range_index)
2060                         if (n-- == 0)
2061                                 return nfit_memdev->memdev;
2062         return NULL;
2063 }
2064
2065 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2066                 struct nd_region_desc *ndr_desc,
2067                 struct acpi_nfit_system_address *spa)
2068 {
2069         struct device *dev = acpi_desc->dev;
2070         struct nd_interleave_set *nd_set;
2071         u16 nr = ndr_desc->num_mappings;
2072         struct nfit_set_info2 *info2;
2073         struct nfit_set_info *info;
2074         int i;
2075
2076         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2077         if (!nd_set)
2078                 return -ENOMEM;
2079         ndr_desc->nd_set = nd_set;
2080         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2081
2082         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2083         if (!info)
2084                 return -ENOMEM;
2085
2086         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2087         if (!info2)
2088                 return -ENOMEM;
2089
2090         for (i = 0; i < nr; i++) {
2091                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2092                 struct nfit_set_info_map *map = &info->mapping[i];
2093                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2094                 struct nvdimm *nvdimm = mapping->nvdimm;
2095                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2096                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2097                                 spa->range_index, i);
2098                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2099
2100                 if (!memdev || !nfit_mem->dcr) {
2101                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2102                         return -ENODEV;
2103                 }
2104
2105                 map->region_offset = memdev->region_offset;
2106                 map->serial_number = dcr->serial_number;
2107
2108                 map2->region_offset = memdev->region_offset;
2109                 map2->serial_number = dcr->serial_number;
2110                 map2->vendor_id = dcr->vendor_id;
2111                 map2->manufacturing_date = dcr->manufacturing_date;
2112                 map2->manufacturing_location = dcr->manufacturing_location;
2113         }
2114
2115         /* v1.1 namespaces */
2116         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2117                         cmp_map, NULL);
2118         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2119
2120         /* v1.2 namespaces */
2121         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2122                         cmp_map2, NULL);
2123         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2124
2125         /* support v1.1 namespaces created with the wrong sort order */
2126         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2127                         cmp_map_compat, NULL);
2128         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2129
2130         /* record the result of the sort for the mapping position */
2131         for (i = 0; i < nr; i++) {
2132                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2133                 int j;
2134
2135                 for (j = 0; j < nr; j++) {
2136                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2137                         struct nvdimm *nvdimm = mapping->nvdimm;
2138                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2139                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2140
2141                         if (map2->serial_number == dcr->serial_number &&
2142                             map2->vendor_id == dcr->vendor_id &&
2143                             map2->manufacturing_date == dcr->manufacturing_date &&
2144                             map2->manufacturing_location
2145                                     == dcr->manufacturing_location) {
2146                                 mapping->position = i;
2147                                 break;
2148                         }
2149                 }
2150         }
2151
2152         ndr_desc->nd_set = nd_set;
2153         devm_kfree(dev, info);
2154         devm_kfree(dev, info2);
2155
2156         return 0;
2157 }
2158
2159 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2160 {
2161         struct acpi_nfit_interleave *idt = mmio->idt;
2162         u32 sub_line_offset, line_index, line_offset;
2163         u64 line_no, table_skip_count, table_offset;
2164
2165         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2166         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2167         line_offset = idt->line_offset[line_index]
2168                 * mmio->line_size;
2169         table_offset = table_skip_count * mmio->table_size;
2170
2171         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2172 }
2173
2174 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2175 {
2176         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2177         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2178         const u32 STATUS_MASK = 0x80000037;
2179
2180         if (mmio->num_lines)
2181                 offset = to_interleave_offset(offset, mmio);
2182
2183         return readl(mmio->addr.base + offset) & STATUS_MASK;
2184 }
2185
2186 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2187                 resource_size_t dpa, unsigned int len, unsigned int write)
2188 {
2189         u64 cmd, offset;
2190         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2191
2192         enum {
2193                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2194                 BCW_LEN_SHIFT = 48,
2195                 BCW_LEN_MASK = (1ULL << 8) - 1,
2196                 BCW_CMD_SHIFT = 56,
2197         };
2198
2199         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2200         len = len >> L1_CACHE_SHIFT;
2201         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2202         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2203
2204         offset = nfit_blk->cmd_offset + mmio->size * bw;
2205         if (mmio->num_lines)
2206                 offset = to_interleave_offset(offset, mmio);
2207
2208         writeq(cmd, mmio->addr.base + offset);
2209         nvdimm_flush(nfit_blk->nd_region);
2210
2211         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2212                 readq(mmio->addr.base + offset);
2213 }
2214
2215 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2216                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2217                 unsigned int lane)
2218 {
2219         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2220         unsigned int copied = 0;
2221         u64 base_offset;
2222         int rc;
2223
2224         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2225                 + lane * mmio->size;
2226         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2227         while (len) {
2228                 unsigned int c;
2229                 u64 offset;
2230
2231                 if (mmio->num_lines) {
2232                         u32 line_offset;
2233
2234                         offset = to_interleave_offset(base_offset + copied,
2235                                         mmio);
2236                         div_u64_rem(offset, mmio->line_size, &line_offset);
2237                         c = min_t(size_t, len, mmio->line_size - line_offset);
2238                 } else {
2239                         offset = base_offset + nfit_blk->bdw_offset;
2240                         c = len;
2241                 }
2242
2243                 if (rw)
2244                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2245                 else {
2246                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2247                                 arch_invalidate_pmem((void __force *)
2248                                         mmio->addr.aperture + offset, c);
2249
2250                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2251                 }
2252
2253                 copied += c;
2254                 len -= c;
2255         }
2256
2257         if (rw)
2258                 nvdimm_flush(nfit_blk->nd_region);
2259
2260         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2261         return rc;
2262 }
2263
2264 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2265                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2266 {
2267         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2268         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2269         struct nd_region *nd_region = nfit_blk->nd_region;
2270         unsigned int lane, copied = 0;
2271         int rc = 0;
2272
2273         lane = nd_region_acquire_lane(nd_region);
2274         while (len) {
2275                 u64 c = min(len, mmio->size);
2276
2277                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2278                                 iobuf + copied, c, rw, lane);
2279                 if (rc)
2280                         break;
2281
2282                 copied += c;
2283                 len -= c;
2284         }
2285         nd_region_release_lane(nd_region, lane);
2286
2287         return rc;
2288 }
2289
2290 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2291                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2292 {
2293         if (idt) {
2294                 mmio->num_lines = idt->line_count;
2295                 mmio->line_size = idt->line_size;
2296                 if (interleave_ways == 0)
2297                         return -ENXIO;
2298                 mmio->table_size = mmio->num_lines * interleave_ways
2299                         * mmio->line_size;
2300         }
2301
2302         return 0;
2303 }
2304
2305 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2306                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2307 {
2308         struct nd_cmd_dimm_flags flags;
2309         int rc;
2310
2311         memset(&flags, 0, sizeof(flags));
2312         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2313                         sizeof(flags), NULL);
2314
2315         if (rc >= 0 && flags.status == 0)
2316                 nfit_blk->dimm_flags = flags.flags;
2317         else if (rc == -ENOTTY) {
2318                 /* fall back to a conservative default */
2319                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2320                 rc = 0;
2321         } else
2322                 rc = -ENXIO;
2323
2324         return rc;
2325 }
2326
2327 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2328                 struct device *dev)
2329 {
2330         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2331         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2332         struct nfit_blk_mmio *mmio;
2333         struct nfit_blk *nfit_blk;
2334         struct nfit_mem *nfit_mem;
2335         struct nvdimm *nvdimm;
2336         int rc;
2337
2338         nvdimm = nd_blk_region_to_dimm(ndbr);
2339         nfit_mem = nvdimm_provider_data(nvdimm);
2340         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2341                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2342                                 nfit_mem ? "" : " nfit_mem",
2343                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2344                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2345                 return -ENXIO;
2346         }
2347
2348         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2349         if (!nfit_blk)
2350                 return -ENOMEM;
2351         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2352         nfit_blk->nd_region = to_nd_region(dev);
2353
2354         /* map block aperture memory */
2355         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2356         mmio = &nfit_blk->mmio[BDW];
2357         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2358                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2359         if (!mmio->addr.base) {
2360                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2361                                 nvdimm_name(nvdimm));
2362                 return -ENOMEM;
2363         }
2364         mmio->size = nfit_mem->bdw->size;
2365         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2366         mmio->idt = nfit_mem->idt_bdw;
2367         mmio->spa = nfit_mem->spa_bdw;
2368         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2369                         nfit_mem->memdev_bdw->interleave_ways);
2370         if (rc) {
2371                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2372                                 __func__, nvdimm_name(nvdimm));
2373                 return rc;
2374         }
2375
2376         /* map block control memory */
2377         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2378         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2379         mmio = &nfit_blk->mmio[DCR];
2380         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2381                         nfit_mem->spa_dcr->length);
2382         if (!mmio->addr.base) {
2383                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2384                                 nvdimm_name(nvdimm));
2385                 return -ENOMEM;
2386         }
2387         mmio->size = nfit_mem->dcr->window_size;
2388         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2389         mmio->idt = nfit_mem->idt_dcr;
2390         mmio->spa = nfit_mem->spa_dcr;
2391         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2392                         nfit_mem->memdev_dcr->interleave_ways);
2393         if (rc) {
2394                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2395                                 __func__, nvdimm_name(nvdimm));
2396                 return rc;
2397         }
2398
2399         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2400         if (rc < 0) {
2401                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2402                                 __func__, nvdimm_name(nvdimm));
2403                 return rc;
2404         }
2405
2406         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2407                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2408
2409         if (mmio->line_size == 0)
2410                 return 0;
2411
2412         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2413                         + 8 > mmio->line_size) {
2414                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2415                 return -ENXIO;
2416         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2417                         + 8 > mmio->line_size) {
2418                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2419                 return -ENXIO;
2420         }
2421
2422         return 0;
2423 }
2424
2425 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2426                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2427 {
2428         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2429         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2430         int cmd_rc, rc;
2431
2432         cmd->address = spa->address;
2433         cmd->length = spa->length;
2434         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2435                         sizeof(*cmd), &cmd_rc);
2436         if (rc < 0)
2437                 return rc;
2438         return cmd_rc;
2439 }
2440
2441 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2442 {
2443         int rc;
2444         int cmd_rc;
2445         struct nd_cmd_ars_start ars_start;
2446         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2447         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2448
2449         memset(&ars_start, 0, sizeof(ars_start));
2450         ars_start.address = spa->address;
2451         ars_start.length = spa->length;
2452         ars_start.flags = acpi_desc->ars_start_flags;
2453         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2454                 ars_start.type = ND_ARS_PERSISTENT;
2455         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2456                 ars_start.type = ND_ARS_VOLATILE;
2457         else
2458                 return -ENOTTY;
2459
2460         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2461                         sizeof(ars_start), &cmd_rc);
2462
2463         if (rc < 0)
2464                 return rc;
2465         return cmd_rc;
2466 }
2467
2468 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2469 {
2470         int rc, cmd_rc;
2471         struct nd_cmd_ars_start ars_start;
2472         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2473         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2474
2475         memset(&ars_start, 0, sizeof(ars_start));
2476         ars_start.address = ars_status->restart_address;
2477         ars_start.length = ars_status->restart_length;
2478         ars_start.type = ars_status->type;
2479         ars_start.flags = acpi_desc->ars_start_flags;
2480         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2481                         sizeof(ars_start), &cmd_rc);
2482         if (rc < 0)
2483                 return rc;
2484         return cmd_rc;
2485 }
2486
2487 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2488 {
2489         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2490         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2491         int rc, cmd_rc;
2492
2493         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2494                         acpi_desc->ars_status_size, &cmd_rc);
2495         if (rc < 0)
2496                 return rc;
2497         return cmd_rc;
2498 }
2499
2500 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2501                 struct nd_cmd_ars_status *ars_status)
2502 {
2503         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2504         int rc;
2505         u32 i;
2506
2507         /*
2508          * First record starts at 44 byte offset from the start of the
2509          * payload.
2510          */
2511         if (ars_status->out_length < 44)
2512                 return 0;
2513         for (i = 0; i < ars_status->num_records; i++) {
2514                 /* only process full records */
2515                 if (ars_status->out_length
2516                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2517                         break;
2518                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2519                                 ars_status->records[i].err_address,
2520                                 ars_status->records[i].length);
2521                 if (rc)
2522                         return rc;
2523         }
2524         if (i < ars_status->num_records)
2525                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2526
2527         return 0;
2528 }
2529
2530 static void acpi_nfit_remove_resource(void *data)
2531 {
2532         struct resource *res = data;
2533
2534         remove_resource(res);
2535 }
2536
2537 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2538                 struct nd_region_desc *ndr_desc)
2539 {
2540         struct resource *res, *nd_res = ndr_desc->res;
2541         int is_pmem, ret;
2542
2543         /* No operation if the region is already registered as PMEM */
2544         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2545                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2546         if (is_pmem == REGION_INTERSECTS)
2547                 return 0;
2548
2549         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2550         if (!res)
2551                 return -ENOMEM;
2552
2553         res->name = "Persistent Memory";
2554         res->start = nd_res->start;
2555         res->end = nd_res->end;
2556         res->flags = IORESOURCE_MEM;
2557         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2558
2559         ret = insert_resource(&iomem_resource, res);
2560         if (ret)
2561                 return ret;
2562
2563         ret = devm_add_action_or_reset(acpi_desc->dev,
2564                                         acpi_nfit_remove_resource,
2565                                         res);
2566         if (ret)
2567                 return ret;
2568
2569         return 0;
2570 }
2571
2572 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2573                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2574                 struct acpi_nfit_memory_map *memdev,
2575                 struct nfit_spa *nfit_spa)
2576 {
2577         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2578                         memdev->device_handle);
2579         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2580         struct nd_blk_region_desc *ndbr_desc;
2581         struct nfit_mem *nfit_mem;
2582         int blk_valid = 0, rc;
2583
2584         if (!nvdimm) {
2585                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2586                                 spa->range_index, memdev->device_handle);
2587                 return -ENODEV;
2588         }
2589
2590         mapping->nvdimm = nvdimm;
2591         switch (nfit_spa_type(spa)) {
2592         case NFIT_SPA_PM:
2593         case NFIT_SPA_VOLATILE:
2594                 mapping->start = memdev->address;
2595                 mapping->size = memdev->region_size;
2596                 break;
2597         case NFIT_SPA_DCR:
2598                 nfit_mem = nvdimm_provider_data(nvdimm);
2599                 if (!nfit_mem || !nfit_mem->bdw) {
2600                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2601                                         spa->range_index, nvdimm_name(nvdimm));
2602                 } else {
2603                         mapping->size = nfit_mem->bdw->capacity;
2604                         mapping->start = nfit_mem->bdw->start_address;
2605                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
2606                         blk_valid = 1;
2607                 }
2608
2609                 ndr_desc->mapping = mapping;
2610                 ndr_desc->num_mappings = blk_valid;
2611                 ndbr_desc = to_blk_region_desc(ndr_desc);
2612                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2613                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2614                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2615                 if (rc)
2616                         return rc;
2617                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2618                                 ndr_desc);
2619                 if (!nfit_spa->nd_region)
2620                         return -ENOMEM;
2621                 break;
2622         }
2623
2624         return 0;
2625 }
2626
2627 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2628 {
2629         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2630                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2631                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2632                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2633 }
2634
2635 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2636 {
2637         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2638                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2639                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2640 }
2641
2642 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2643                 struct nfit_spa *nfit_spa)
2644 {
2645         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2646         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2647         struct nd_blk_region_desc ndbr_desc;
2648         struct nd_region_desc *ndr_desc;
2649         struct nfit_memdev *nfit_memdev;
2650         struct nvdimm_bus *nvdimm_bus;
2651         struct resource res;
2652         int count = 0, rc;
2653
2654         if (nfit_spa->nd_region)
2655                 return 0;
2656
2657         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2658                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2659                                 __func__);
2660                 return 0;
2661         }
2662
2663         memset(&res, 0, sizeof(res));
2664         memset(&mappings, 0, sizeof(mappings));
2665         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2666         res.start = spa->address;
2667         res.end = res.start + spa->length - 1;
2668         ndr_desc = &ndbr_desc.ndr_desc;
2669         ndr_desc->res = &res;
2670         ndr_desc->provider_data = nfit_spa;
2671         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2672         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2673                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2674                                                 spa->proximity_domain);
2675         else
2676                 ndr_desc->numa_node = NUMA_NO_NODE;
2677
2678         if(acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2679                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2680
2681         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2682                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2683
2684         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2685                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2686                 struct nd_mapping_desc *mapping;
2687
2688                 if (memdev->range_index != spa->range_index)
2689                         continue;
2690                 if (count >= ND_MAX_MAPPINGS) {
2691                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2692                                         spa->range_index, ND_MAX_MAPPINGS);
2693                         return -ENXIO;
2694                 }
2695                 mapping = &mappings[count++];
2696                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2697                                 memdev, nfit_spa);
2698                 if (rc)
2699                         goto out;
2700         }
2701
2702         ndr_desc->mapping = mappings;
2703         ndr_desc->num_mappings = count;
2704         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2705         if (rc)
2706                 goto out;
2707
2708         nvdimm_bus = acpi_desc->nvdimm_bus;
2709         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2710                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2711                 if (rc) {
2712                         dev_warn(acpi_desc->dev,
2713                                 "failed to insert pmem resource to iomem: %d\n",
2714                                 rc);
2715                         goto out;
2716                 }
2717
2718                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2719                                 ndr_desc);
2720                 if (!nfit_spa->nd_region)
2721                         rc = -ENOMEM;
2722         } else if (nfit_spa_is_volatile(spa)) {
2723                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2724                                 ndr_desc);
2725                 if (!nfit_spa->nd_region)
2726                         rc = -ENOMEM;
2727         } else if (nfit_spa_is_virtual(spa)) {
2728                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2729                                 ndr_desc);
2730                 if (!nfit_spa->nd_region)
2731                         rc = -ENOMEM;
2732         }
2733
2734  out:
2735         if (rc)
2736                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2737                                 nfit_spa->spa->range_index);
2738         return rc;
2739 }
2740
2741 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2742                 u32 max_ars)
2743 {
2744         struct device *dev = acpi_desc->dev;
2745         struct nd_cmd_ars_status *ars_status;
2746
2747         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2748                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2749                 return 0;
2750         }
2751
2752         if (acpi_desc->ars_status)
2753                 devm_kfree(dev, acpi_desc->ars_status);
2754         acpi_desc->ars_status = NULL;
2755         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2756         if (!ars_status)
2757                 return -ENOMEM;
2758         acpi_desc->ars_status = ars_status;
2759         acpi_desc->ars_status_size = max_ars;
2760         return 0;
2761 }
2762
2763 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2764                 struct nfit_spa *nfit_spa)
2765 {
2766         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2767         int rc;
2768
2769         if (!nfit_spa->max_ars) {
2770                 struct nd_cmd_ars_cap ars_cap;
2771
2772                 memset(&ars_cap, 0, sizeof(ars_cap));
2773                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2774                 if (rc < 0)
2775                         return rc;
2776                 nfit_spa->max_ars = ars_cap.max_ars_out;
2777                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2778                 /* check that the supported scrub types match the spa type */
2779                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2780                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2781                         return -ENOTTY;
2782                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2783                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2784                         return -ENOTTY;
2785         }
2786
2787         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2788                 return -ENOMEM;
2789
2790         rc = ars_get_status(acpi_desc);
2791         if (rc < 0 && rc != -ENOSPC)
2792                 return rc;
2793
2794         if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2795                 return -ENOMEM;
2796
2797         return 0;
2798 }
2799
2800 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2801                 struct nfit_spa *nfit_spa)
2802 {
2803         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2804         unsigned int overflow_retry = scrub_overflow_abort;
2805         u64 init_ars_start = 0, init_ars_len = 0;
2806         struct device *dev = acpi_desc->dev;
2807         unsigned int tmo = scrub_timeout;
2808         int rc;
2809
2810         if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2811                 return;
2812
2813         rc = ars_start(acpi_desc, nfit_spa);
2814         /*
2815          * If we timed out the initial scan we'll still be busy here,
2816          * and will wait another timeout before giving up permanently.
2817          */
2818         if (rc < 0 && rc != -EBUSY)
2819                 return;
2820
2821         do {
2822                 u64 ars_start, ars_len;
2823
2824                 if (acpi_desc->cancel)
2825                         break;
2826                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2827                 if (rc == -ENOTTY)
2828                         break;
2829                 if (rc == -EBUSY && !tmo) {
2830                         dev_warn(dev, "range %d ars timeout, aborting\n",
2831                                         spa->range_index);
2832                         break;
2833                 }
2834
2835                 if (rc == -EBUSY) {
2836                         /*
2837                          * Note, entries may be appended to the list
2838                          * while the lock is dropped, but the workqueue
2839                          * being active prevents entries being deleted /
2840                          * freed.
2841                          */
2842                         mutex_unlock(&acpi_desc->init_mutex);
2843                         ssleep(1);
2844                         tmo--;
2845                         mutex_lock(&acpi_desc->init_mutex);
2846                         continue;
2847                 }
2848
2849                 /* we got some results, but there are more pending... */
2850                 if (rc == -ENOSPC && overflow_retry--) {
2851                         if (!init_ars_len) {
2852                                 init_ars_len = acpi_desc->ars_status->length;
2853                                 init_ars_start = acpi_desc->ars_status->address;
2854                         }
2855                         rc = ars_continue(acpi_desc);
2856                 }
2857
2858                 if (rc < 0) {
2859                         dev_warn(dev, "range %d ars continuation failed\n",
2860                                         spa->range_index);
2861                         break;
2862                 }
2863
2864                 if (init_ars_len) {
2865                         ars_start = init_ars_start;
2866                         ars_len = init_ars_len;
2867                 } else {
2868                         ars_start = acpi_desc->ars_status->address;
2869                         ars_len = acpi_desc->ars_status->length;
2870                 }
2871                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2872                                 spa->range_index, ars_start, ars_len);
2873                 /* notify the region about new poison entries */
2874                 nvdimm_region_notify(nfit_spa->nd_region,
2875                                 NVDIMM_REVALIDATE_POISON);
2876                 break;
2877         } while (1);
2878 }
2879
2880 static void acpi_nfit_scrub(struct work_struct *work)
2881 {
2882         struct device *dev;
2883         u64 init_scrub_length = 0;
2884         struct nfit_spa *nfit_spa;
2885         u64 init_scrub_address = 0;
2886         bool init_ars_done = false;
2887         struct acpi_nfit_desc *acpi_desc;
2888         unsigned int tmo = scrub_timeout;
2889         unsigned int overflow_retry = scrub_overflow_abort;
2890
2891         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2892         dev = acpi_desc->dev;
2893
2894         /*
2895          * We scrub in 2 phases.  The first phase waits for any platform
2896          * firmware initiated scrubs to complete and then we go search for the
2897          * affected spa regions to mark them scanned.  In the second phase we
2898          * initiate a directed scrub for every range that was not scrubbed in
2899          * phase 1. If we're called for a 'rescan', we harmlessly pass through
2900          * the first phase, but really only care about running phase 2, where
2901          * regions can be notified of new poison.
2902          */
2903
2904         /* process platform firmware initiated scrubs */
2905  retry:
2906         mutex_lock(&acpi_desc->init_mutex);
2907         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2908                 struct nd_cmd_ars_status *ars_status;
2909                 struct acpi_nfit_system_address *spa;
2910                 u64 ars_start, ars_len;
2911                 int rc;
2912
2913                 if (acpi_desc->cancel)
2914                         break;
2915
2916                 if (nfit_spa->nd_region)
2917                         continue;
2918
2919                 if (init_ars_done) {
2920                         /*
2921                          * No need to re-query, we're now just
2922                          * reconciling all the ranges covered by the
2923                          * initial scrub
2924                          */
2925                         rc = 0;
2926                 } else
2927                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2928
2929                 if (rc == -ENOTTY) {
2930                         /* no ars capability, just register spa and move on */
2931                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2932                         continue;
2933                 }
2934
2935                 if (rc == -EBUSY && !tmo) {
2936                         /* fallthrough to directed scrub in phase 2 */
2937                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2938                         break;
2939                 } else if (rc == -EBUSY) {
2940                         mutex_unlock(&acpi_desc->init_mutex);
2941                         ssleep(1);
2942                         tmo--;
2943                         goto retry;
2944                 }
2945
2946                 /* we got some results, but there are more pending... */
2947                 if (rc == -ENOSPC && overflow_retry--) {
2948                         ars_status = acpi_desc->ars_status;
2949                         /*
2950                          * Record the original scrub range, so that we
2951                          * can recall all the ranges impacted by the
2952                          * initial scrub.
2953                          */
2954                         if (!init_scrub_length) {
2955                                 init_scrub_length = ars_status->length;
2956                                 init_scrub_address = ars_status->address;
2957                         }
2958                         rc = ars_continue(acpi_desc);
2959                         if (rc == 0) {
2960                                 mutex_unlock(&acpi_desc->init_mutex);
2961                                 goto retry;
2962                         }
2963                 }
2964
2965                 if (rc < 0) {
2966                         /*
2967                          * Initial scrub failed, we'll give it one more
2968                          * try below...
2969                          */
2970                         break;
2971                 }
2972
2973                 /* We got some final results, record completed ranges */
2974                 ars_status = acpi_desc->ars_status;
2975                 if (init_scrub_length) {
2976                         ars_start = init_scrub_address;
2977                         ars_len = ars_start + init_scrub_length;
2978                 } else {
2979                         ars_start = ars_status->address;
2980                         ars_len = ars_status->length;
2981                 }
2982                 spa = nfit_spa->spa;
2983
2984                 if (!init_ars_done) {
2985                         init_ars_done = true;
2986                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2987                                         ars_start, ars_len);
2988                 }
2989                 if (ars_start <= spa->address && ars_start + ars_len
2990                                 >= spa->address + spa->length)
2991                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2992         }
2993
2994         /*
2995          * For all the ranges not covered by an initial scrub we still
2996          * want to see if there are errors, but it's ok to discover them
2997          * asynchronously.
2998          */
2999         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3000                 /*
3001                  * Flag all the ranges that still need scrubbing, but
3002                  * register them now to make data available.
3003                  */
3004                 if (!nfit_spa->nd_region) {
3005                         nfit_spa->ars_required = 1;
3006                         acpi_nfit_register_region(acpi_desc, nfit_spa);
3007                 }
3008         }
3009         acpi_desc->init_complete = 1;
3010
3011         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3012                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
3013         acpi_desc->scrub_count++;
3014         acpi_desc->ars_start_flags = 0;
3015         if (acpi_desc->scrub_count_state)
3016                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3017         mutex_unlock(&acpi_desc->init_mutex);
3018 }
3019
3020 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3021 {
3022         struct nfit_spa *nfit_spa;
3023         int rc;
3024
3025         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3026                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
3027                         /* BLK regions don't need to wait for ars results */
3028                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3029                         if (rc)
3030                                 return rc;
3031                 }
3032
3033         acpi_desc->ars_start_flags = 0;
3034         if (!acpi_desc->cancel)
3035                 queue_work(nfit_wq, &acpi_desc->work);
3036         return 0;
3037 }
3038
3039 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3040                 struct nfit_table_prev *prev)
3041 {
3042         struct device *dev = acpi_desc->dev;
3043
3044         if (!list_empty(&prev->spas) ||
3045                         !list_empty(&prev->memdevs) ||
3046                         !list_empty(&prev->dcrs) ||
3047                         !list_empty(&prev->bdws) ||
3048                         !list_empty(&prev->idts) ||
3049                         !list_empty(&prev->flushes)) {
3050                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3051                 return -ENXIO;
3052         }
3053         return 0;
3054 }
3055
3056 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3057 {
3058         struct device *dev = acpi_desc->dev;
3059         struct kernfs_node *nfit;
3060         struct device *bus_dev;
3061
3062         if (!ars_supported(acpi_desc->nvdimm_bus))
3063                 return 0;
3064
3065         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3066         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3067         if (!nfit) {
3068                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3069                 return -ENODEV;
3070         }
3071         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3072         sysfs_put(nfit);
3073         if (!acpi_desc->scrub_count_state) {
3074                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3075                 return -ENODEV;
3076         }
3077
3078         return 0;
3079 }
3080
3081 static void acpi_nfit_unregister(void *data)
3082 {
3083         struct acpi_nfit_desc *acpi_desc = data;
3084
3085         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3086 }
3087
3088 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3089 {
3090         struct device *dev = acpi_desc->dev;
3091         struct nfit_table_prev prev;
3092         const void *end;
3093         int rc;
3094
3095         if (!acpi_desc->nvdimm_bus) {
3096                 acpi_nfit_init_dsms(acpi_desc);
3097
3098                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3099                                 &acpi_desc->nd_desc);
3100                 if (!acpi_desc->nvdimm_bus)
3101                         return -ENOMEM;
3102
3103                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3104                                 acpi_desc);
3105                 if (rc)
3106                         return rc;
3107
3108                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3109                 if (rc)
3110                         return rc;
3111
3112                 /* register this acpi_desc for mce notifications */
3113                 mutex_lock(&acpi_desc_lock);
3114                 list_add_tail(&acpi_desc->list, &acpi_descs);
3115                 mutex_unlock(&acpi_desc_lock);
3116         }
3117
3118         mutex_lock(&acpi_desc->init_mutex);
3119
3120         INIT_LIST_HEAD(&prev.spas);
3121         INIT_LIST_HEAD(&prev.memdevs);
3122         INIT_LIST_HEAD(&prev.dcrs);
3123         INIT_LIST_HEAD(&prev.bdws);
3124         INIT_LIST_HEAD(&prev.idts);
3125         INIT_LIST_HEAD(&prev.flushes);
3126
3127         list_cut_position(&prev.spas, &acpi_desc->spas,
3128                                 acpi_desc->spas.prev);
3129         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3130                                 acpi_desc->memdevs.prev);
3131         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3132                                 acpi_desc->dcrs.prev);
3133         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3134                                 acpi_desc->bdws.prev);
3135         list_cut_position(&prev.idts, &acpi_desc->idts,
3136                                 acpi_desc->idts.prev);
3137         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3138                                 acpi_desc->flushes.prev);
3139
3140         end = data + sz;
3141         while (!IS_ERR_OR_NULL(data))
3142                 data = add_table(acpi_desc, &prev, data, end);
3143
3144         if (IS_ERR(data)) {
3145                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
3146                                 PTR_ERR(data));
3147                 rc = PTR_ERR(data);
3148                 goto out_unlock;
3149         }
3150