Merge branch 'for-4.16/dax' 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 void *add_table(struct acpi_nfit_desc *acpi_desc,
842                 struct nfit_table_prev *prev, void *table, const void *end)
843 {
844         struct device *dev = acpi_desc->dev;
845         struct acpi_nfit_header *hdr;
846         void *err = ERR_PTR(-ENOMEM);
847
848         if (table >= end)
849                 return NULL;
850
851         hdr = table;
852         if (!hdr->length) {
853                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
854                         hdr->type);
855                 return NULL;
856         }
857
858         switch (hdr->type) {
859         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
860                 if (!add_spa(acpi_desc, prev, table))
861                         return err;
862                 break;
863         case ACPI_NFIT_TYPE_MEMORY_MAP:
864                 if (!add_memdev(acpi_desc, prev, table))
865                         return err;
866                 break;
867         case ACPI_NFIT_TYPE_CONTROL_REGION:
868                 if (!add_dcr(acpi_desc, prev, table))
869                         return err;
870                 break;
871         case ACPI_NFIT_TYPE_DATA_REGION:
872                 if (!add_bdw(acpi_desc, prev, table))
873                         return err;
874                 break;
875         case ACPI_NFIT_TYPE_INTERLEAVE:
876                 if (!add_idt(acpi_desc, prev, table))
877                         return err;
878                 break;
879         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
880                 if (!add_flush(acpi_desc, prev, table))
881                         return err;
882                 break;
883         case ACPI_NFIT_TYPE_SMBIOS:
884                 dev_dbg(dev, "%s: smbios\n", __func__);
885                 break;
886         default:
887                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
888                 break;
889         }
890
891         return table + hdr->length;
892 }
893
894 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
895                 struct nfit_mem *nfit_mem)
896 {
897         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
898         u16 dcr = nfit_mem->dcr->region_index;
899         struct nfit_spa *nfit_spa;
900
901         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
902                 u16 range_index = nfit_spa->spa->range_index;
903                 int type = nfit_spa_type(nfit_spa->spa);
904                 struct nfit_memdev *nfit_memdev;
905
906                 if (type != NFIT_SPA_BDW)
907                         continue;
908
909                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
910                         if (nfit_memdev->memdev->range_index != range_index)
911                                 continue;
912                         if (nfit_memdev->memdev->device_handle != device_handle)
913                                 continue;
914                         if (nfit_memdev->memdev->region_index != dcr)
915                                 continue;
916
917                         nfit_mem->spa_bdw = nfit_spa->spa;
918                         return;
919                 }
920         }
921
922         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
923                         nfit_mem->spa_dcr->range_index);
924         nfit_mem->bdw = NULL;
925 }
926
927 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
928                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
929 {
930         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
931         struct nfit_memdev *nfit_memdev;
932         struct nfit_bdw *nfit_bdw;
933         struct nfit_idt *nfit_idt;
934         u16 idt_idx, range_index;
935
936         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
937                 if (nfit_bdw->bdw->region_index != dcr)
938                         continue;
939                 nfit_mem->bdw = nfit_bdw->bdw;
940                 break;
941         }
942
943         if (!nfit_mem->bdw)
944                 return;
945
946         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
947
948         if (!nfit_mem->spa_bdw)
949                 return;
950
951         range_index = nfit_mem->spa_bdw->range_index;
952         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
953                 if (nfit_memdev->memdev->range_index != range_index ||
954                                 nfit_memdev->memdev->region_index != dcr)
955                         continue;
956                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
957                 idt_idx = nfit_memdev->memdev->interleave_index;
958                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
959                         if (nfit_idt->idt->interleave_index != idt_idx)
960                                 continue;
961                         nfit_mem->idt_bdw = nfit_idt->idt;
962                         break;
963                 }
964                 break;
965         }
966 }
967
968 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
969                 struct acpi_nfit_system_address *spa)
970 {
971         struct nfit_mem *nfit_mem, *found;
972         struct nfit_memdev *nfit_memdev;
973         int type = spa ? nfit_spa_type(spa) : 0;
974
975         switch (type) {
976         case NFIT_SPA_DCR:
977         case NFIT_SPA_PM:
978                 break;
979         default:
980                 if (spa)
981                         return 0;
982         }
983
984         /*
985          * This loop runs in two modes, when a dimm is mapped the loop
986          * adds memdev associations to an existing dimm, or creates a
987          * dimm. In the unmapped dimm case this loop sweeps for memdev
988          * instances with an invalid / zero range_index and adds those
989          * dimms without spa associations.
990          */
991         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
992                 struct nfit_flush *nfit_flush;
993                 struct nfit_dcr *nfit_dcr;
994                 u32 device_handle;
995                 u16 dcr;
996
997                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
998                         continue;
999                 if (!spa && nfit_memdev->memdev->range_index)
1000                         continue;
1001                 found = NULL;
1002                 dcr = nfit_memdev->memdev->region_index;
1003                 device_handle = nfit_memdev->memdev->device_handle;
1004                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1005                         if (__to_nfit_memdev(nfit_mem)->device_handle
1006                                         == device_handle) {
1007                                 found = nfit_mem;
1008                                 break;
1009                         }
1010
1011                 if (found)
1012                         nfit_mem = found;
1013                 else {
1014                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1015                                         sizeof(*nfit_mem), GFP_KERNEL);
1016                         if (!nfit_mem)
1017                                 return -ENOMEM;
1018                         INIT_LIST_HEAD(&nfit_mem->list);
1019                         nfit_mem->acpi_desc = acpi_desc;
1020                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1021                 }
1022
1023                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1024                         if (nfit_dcr->dcr->region_index != dcr)
1025                                 continue;
1026                         /*
1027                          * Record the control region for the dimm.  For
1028                          * the ACPI 6.1 case, where there are separate
1029                          * control regions for the pmem vs blk
1030                          * interfaces, be sure to record the extended
1031                          * blk details.
1032                          */
1033                         if (!nfit_mem->dcr)
1034                                 nfit_mem->dcr = nfit_dcr->dcr;
1035                         else if (nfit_mem->dcr->windows == 0
1036                                         && nfit_dcr->dcr->windows)
1037                                 nfit_mem->dcr = nfit_dcr->dcr;
1038                         break;
1039                 }
1040
1041                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1042                         struct acpi_nfit_flush_address *flush;
1043                         u16 i;
1044
1045                         if (nfit_flush->flush->device_handle != device_handle)
1046                                 continue;
1047                         nfit_mem->nfit_flush = nfit_flush;
1048                         flush = nfit_flush->flush;
1049                         nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
1050                                         flush->hint_count
1051                                         * sizeof(struct resource), GFP_KERNEL);
1052                         if (!nfit_mem->flush_wpq)
1053                                 return -ENOMEM;
1054                         for (i = 0; i < flush->hint_count; i++) {
1055                                 struct resource *res = &nfit_mem->flush_wpq[i];
1056
1057                                 res->start = flush->hint_address[i];
1058                                 res->end = res->start + 8 - 1;
1059                         }
1060                         break;
1061                 }
1062
1063                 if (dcr && !nfit_mem->dcr) {
1064                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1065                                         spa->range_index, dcr);
1066                         return -ENODEV;
1067                 }
1068
1069                 if (type == NFIT_SPA_DCR) {
1070                         struct nfit_idt *nfit_idt;
1071                         u16 idt_idx;
1072
1073                         /* multiple dimms may share a SPA when interleaved */
1074                         nfit_mem->spa_dcr = spa;
1075                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1076                         idt_idx = nfit_memdev->memdev->interleave_index;
1077                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1078                                 if (nfit_idt->idt->interleave_index != idt_idx)
1079                                         continue;
1080                                 nfit_mem->idt_dcr = nfit_idt->idt;
1081                                 break;
1082                         }
1083                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1084                 } else if (type == NFIT_SPA_PM) {
1085                         /*
1086                          * A single dimm may belong to multiple SPA-PM
1087                          * ranges, record at least one in addition to
1088                          * any SPA-DCR range.
1089                          */
1090                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1091                 } else
1092                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1093         }
1094
1095         return 0;
1096 }
1097
1098 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1099 {
1100         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1101         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1102         u32 handleA, handleB;
1103
1104         handleA = __to_nfit_memdev(a)->device_handle;
1105         handleB = __to_nfit_memdev(b)->device_handle;
1106         if (handleA < handleB)
1107                 return -1;
1108         else if (handleA > handleB)
1109                 return 1;
1110         return 0;
1111 }
1112
1113 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1114 {
1115         struct nfit_spa *nfit_spa;
1116         int rc;
1117
1118
1119         /*
1120          * For each SPA-DCR or SPA-PMEM address range find its
1121          * corresponding MEMDEV(s).  From each MEMDEV find the
1122          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1123          * try to find a SPA-BDW and a corresponding BDW that references
1124          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1125          * BDWs are optional.
1126          */
1127         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1128                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1129                 if (rc)
1130                         return rc;
1131         }
1132
1133         /*
1134          * If a DIMM has failed to be mapped into SPA there will be no
1135          * SPA entries above. Find and register all the unmapped DIMMs
1136          * for reporting and recovery purposes.
1137          */
1138         rc = __nfit_mem_init(acpi_desc, NULL);
1139         if (rc)
1140                 return rc;
1141
1142         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1143
1144         return 0;
1145 }
1146
1147 static ssize_t bus_dsm_mask_show(struct device *dev,
1148                 struct device_attribute *attr, char *buf)
1149 {
1150         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1151         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1152
1153         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1154 }
1155 static struct device_attribute dev_attr_bus_dsm_mask =
1156                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1157
1158 static ssize_t revision_show(struct device *dev,
1159                 struct device_attribute *attr, char *buf)
1160 {
1161         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1162         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1163         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1164
1165         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1166 }
1167 static DEVICE_ATTR_RO(revision);
1168
1169 static ssize_t hw_error_scrub_show(struct device *dev,
1170                 struct device_attribute *attr, char *buf)
1171 {
1172         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1173         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1174         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1175
1176         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1177 }
1178
1179 /*
1180  * The 'hw_error_scrub' attribute can have the following values written to it:
1181  * '0': Switch to the default mode where an exception will only insert
1182  *      the address of the memory error into the poison and badblocks lists.
1183  * '1': Enable a full scrub to happen if an exception for a memory error is
1184  *      received.
1185  */
1186 static ssize_t hw_error_scrub_store(struct device *dev,
1187                 struct device_attribute *attr, const char *buf, size_t size)
1188 {
1189         struct nvdimm_bus_descriptor *nd_desc;
1190         ssize_t rc;
1191         long val;
1192
1193         rc = kstrtol(buf, 0, &val);
1194         if (rc)
1195                 return rc;
1196
1197         device_lock(dev);
1198         nd_desc = dev_get_drvdata(dev);
1199         if (nd_desc) {
1200                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1201
1202                 switch (val) {
1203                 case HW_ERROR_SCRUB_ON:
1204                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1205                         break;
1206                 case HW_ERROR_SCRUB_OFF:
1207                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1208                         break;
1209                 default:
1210                         rc = -EINVAL;
1211                         break;
1212                 }
1213         }
1214         device_unlock(dev);
1215         if (rc)
1216                 return rc;
1217         return size;
1218 }
1219 static DEVICE_ATTR_RW(hw_error_scrub);
1220
1221 /*
1222  * This shows the number of full Address Range Scrubs that have been
1223  * completed since driver load time. Userspace can wait on this using
1224  * select/poll etc. A '+' at the end indicates an ARS is in progress
1225  */
1226 static ssize_t scrub_show(struct device *dev,
1227                 struct device_attribute *attr, char *buf)
1228 {
1229         struct nvdimm_bus_descriptor *nd_desc;
1230         ssize_t rc = -ENXIO;
1231
1232         device_lock(dev);
1233         nd_desc = dev_get_drvdata(dev);
1234         if (nd_desc) {
1235                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1236
1237                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1238                                 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
1239         }
1240         device_unlock(dev);
1241         return rc;
1242 }
1243
1244 static ssize_t scrub_store(struct device *dev,
1245                 struct device_attribute *attr, const char *buf, size_t size)
1246 {
1247         struct nvdimm_bus_descriptor *nd_desc;
1248         ssize_t rc;
1249         long val;
1250
1251         rc = kstrtol(buf, 0, &val);
1252         if (rc)
1253                 return rc;
1254         if (val != 1)
1255                 return -EINVAL;
1256
1257         device_lock(dev);
1258         nd_desc = dev_get_drvdata(dev);
1259         if (nd_desc) {
1260                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1261
1262                 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1263         }
1264         device_unlock(dev);
1265         if (rc)
1266                 return rc;
1267         return size;
1268 }
1269 static DEVICE_ATTR_RW(scrub);
1270
1271 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1272 {
1273         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1274         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1275                 | 1 << ND_CMD_ARS_STATUS;
1276
1277         return (nd_desc->cmd_mask & mask) == mask;
1278 }
1279
1280 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1281 {
1282         struct device *dev = container_of(kobj, struct device, kobj);
1283         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1284
1285         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1286                 return 0;
1287         return a->mode;
1288 }
1289
1290 static struct attribute *acpi_nfit_attributes[] = {
1291         &dev_attr_revision.attr,
1292         &dev_attr_scrub.attr,
1293         &dev_attr_hw_error_scrub.attr,
1294         &dev_attr_bus_dsm_mask.attr,
1295         NULL,
1296 };
1297
1298 static const struct attribute_group acpi_nfit_attribute_group = {
1299         .name = "nfit",
1300         .attrs = acpi_nfit_attributes,
1301         .is_visible = nfit_visible,
1302 };
1303
1304 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1305         &nvdimm_bus_attribute_group,
1306         &acpi_nfit_attribute_group,
1307         NULL,
1308 };
1309
1310 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1311 {
1312         struct nvdimm *nvdimm = to_nvdimm(dev);
1313         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1314
1315         return __to_nfit_memdev(nfit_mem);
1316 }
1317
1318 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1319 {
1320         struct nvdimm *nvdimm = to_nvdimm(dev);
1321         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1322
1323         return nfit_mem->dcr;
1324 }
1325
1326 static ssize_t handle_show(struct device *dev,
1327                 struct device_attribute *attr, char *buf)
1328 {
1329         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1330
1331         return sprintf(buf, "%#x\n", memdev->device_handle);
1332 }
1333 static DEVICE_ATTR_RO(handle);
1334
1335 static ssize_t phys_id_show(struct device *dev,
1336                 struct device_attribute *attr, char *buf)
1337 {
1338         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1339
1340         return sprintf(buf, "%#x\n", memdev->physical_id);
1341 }
1342 static DEVICE_ATTR_RO(phys_id);
1343
1344 static ssize_t vendor_show(struct device *dev,
1345                 struct device_attribute *attr, char *buf)
1346 {
1347         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1348
1349         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1350 }
1351 static DEVICE_ATTR_RO(vendor);
1352
1353 static ssize_t rev_id_show(struct device *dev,
1354                 struct device_attribute *attr, char *buf)
1355 {
1356         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1357
1358         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1359 }
1360 static DEVICE_ATTR_RO(rev_id);
1361
1362 static ssize_t device_show(struct device *dev,
1363                 struct device_attribute *attr, char *buf)
1364 {
1365         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1366
1367         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1368 }
1369 static DEVICE_ATTR_RO(device);
1370
1371 static ssize_t subsystem_vendor_show(struct device *dev,
1372                 struct device_attribute *attr, char *buf)
1373 {
1374         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1375
1376         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1377 }
1378 static DEVICE_ATTR_RO(subsystem_vendor);
1379
1380 static ssize_t subsystem_rev_id_show(struct device *dev,
1381                 struct device_attribute *attr, char *buf)
1382 {
1383         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1384
1385         return sprintf(buf, "0x%04x\n",
1386                         be16_to_cpu(dcr->subsystem_revision_id));
1387 }
1388 static DEVICE_ATTR_RO(subsystem_rev_id);
1389
1390 static ssize_t subsystem_device_show(struct device *dev,
1391                 struct device_attribute *attr, char *buf)
1392 {
1393         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1394
1395         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1396 }
1397 static DEVICE_ATTR_RO(subsystem_device);
1398
1399 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1400 {
1401         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1402         int formats = 0;
1403
1404         if (nfit_mem->memdev_pmem)
1405                 formats++;
1406         if (nfit_mem->memdev_bdw)
1407                 formats++;
1408         return formats;
1409 }
1410
1411 static ssize_t format_show(struct device *dev,
1412                 struct device_attribute *attr, char *buf)
1413 {
1414         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1415
1416         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1417 }
1418 static DEVICE_ATTR_RO(format);
1419
1420 static ssize_t format1_show(struct device *dev,
1421                 struct device_attribute *attr, char *buf)
1422 {
1423         u32 handle;
1424         ssize_t rc = -ENXIO;
1425         struct nfit_mem *nfit_mem;
1426         struct nfit_memdev *nfit_memdev;
1427         struct acpi_nfit_desc *acpi_desc;
1428         struct nvdimm *nvdimm = to_nvdimm(dev);
1429         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1430
1431         nfit_mem = nvdimm_provider_data(nvdimm);
1432         acpi_desc = nfit_mem->acpi_desc;
1433         handle = to_nfit_memdev(dev)->device_handle;
1434
1435         /* assumes DIMMs have at most 2 published interface codes */
1436         mutex_lock(&acpi_desc->init_mutex);
1437         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1438                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1439                 struct nfit_dcr *nfit_dcr;
1440
1441                 if (memdev->device_handle != handle)
1442                         continue;
1443
1444                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1445                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1446                                 continue;
1447                         if (nfit_dcr->dcr->code == dcr->code)
1448                                 continue;
1449                         rc = sprintf(buf, "0x%04x\n",
1450                                         le16_to_cpu(nfit_dcr->dcr->code));
1451                         break;
1452                 }
1453                 if (rc != ENXIO)
1454                         break;
1455         }
1456         mutex_unlock(&acpi_desc->init_mutex);
1457         return rc;
1458 }
1459 static DEVICE_ATTR_RO(format1);
1460
1461 static ssize_t formats_show(struct device *dev,
1462                 struct device_attribute *attr, char *buf)
1463 {
1464         struct nvdimm *nvdimm = to_nvdimm(dev);
1465
1466         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1467 }
1468 static DEVICE_ATTR_RO(formats);
1469
1470 static ssize_t serial_show(struct device *dev,
1471                 struct device_attribute *attr, char *buf)
1472 {
1473         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1474
1475         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1476 }
1477 static DEVICE_ATTR_RO(serial);
1478
1479 static ssize_t family_show(struct device *dev,
1480                 struct device_attribute *attr, char *buf)
1481 {
1482         struct nvdimm *nvdimm = to_nvdimm(dev);
1483         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1484
1485         if (nfit_mem->family < 0)
1486                 return -ENXIO;
1487         return sprintf(buf, "%d\n", nfit_mem->family);
1488 }
1489 static DEVICE_ATTR_RO(family);
1490
1491 static ssize_t dsm_mask_show(struct device *dev,
1492                 struct device_attribute *attr, char *buf)
1493 {
1494         struct nvdimm *nvdimm = to_nvdimm(dev);
1495         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1496
1497         if (nfit_mem->family < 0)
1498                 return -ENXIO;
1499         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1500 }
1501 static DEVICE_ATTR_RO(dsm_mask);
1502
1503 static ssize_t flags_show(struct device *dev,
1504                 struct device_attribute *attr, char *buf)
1505 {
1506         u16 flags = to_nfit_memdev(dev)->flags;
1507
1508         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1509                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1510                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1511                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1512                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1513                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1514                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1515                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1516 }
1517 static DEVICE_ATTR_RO(flags);
1518
1519 static ssize_t id_show(struct device *dev,
1520                 struct device_attribute *attr, char *buf)
1521 {
1522         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1523
1524         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1525                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1526                                 be16_to_cpu(dcr->vendor_id),
1527                                 dcr->manufacturing_location,
1528                                 be16_to_cpu(dcr->manufacturing_date),
1529                                 be32_to_cpu(dcr->serial_number));
1530         else
1531                 return sprintf(buf, "%04x-%08x\n",
1532                                 be16_to_cpu(dcr->vendor_id),
1533                                 be32_to_cpu(dcr->serial_number));
1534 }
1535 static DEVICE_ATTR_RO(id);
1536
1537 static struct attribute *acpi_nfit_dimm_attributes[] = {
1538         &dev_attr_handle.attr,
1539         &dev_attr_phys_id.attr,
1540         &dev_attr_vendor.attr,
1541         &dev_attr_device.attr,
1542         &dev_attr_rev_id.attr,
1543         &dev_attr_subsystem_vendor.attr,
1544         &dev_attr_subsystem_device.attr,
1545         &dev_attr_subsystem_rev_id.attr,
1546         &dev_attr_format.attr,
1547         &dev_attr_formats.attr,
1548         &dev_attr_format1.attr,
1549         &dev_attr_serial.attr,
1550         &dev_attr_flags.attr,
1551         &dev_attr_id.attr,
1552         &dev_attr_family.attr,
1553         &dev_attr_dsm_mask.attr,
1554         NULL,
1555 };
1556
1557 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1558                 struct attribute *a, int n)
1559 {
1560         struct device *dev = container_of(kobj, struct device, kobj);
1561         struct nvdimm *nvdimm = to_nvdimm(dev);
1562
1563         if (!to_nfit_dcr(dev)) {
1564                 /* Without a dcr only the memdev attributes can be surfaced */
1565                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1566                                 || a == &dev_attr_flags.attr
1567                                 || a == &dev_attr_family.attr
1568                                 || a == &dev_attr_dsm_mask.attr)
1569                         return a->mode;
1570                 return 0;
1571         }
1572
1573         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1574                 return 0;
1575         return a->mode;
1576 }
1577
1578 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1579         .name = "nfit",
1580         .attrs = acpi_nfit_dimm_attributes,
1581         .is_visible = acpi_nfit_dimm_attr_visible,
1582 };
1583
1584 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1585         &nvdimm_attribute_group,
1586         &nd_device_attribute_group,
1587         &acpi_nfit_dimm_attribute_group,
1588         NULL,
1589 };
1590
1591 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1592                 u32 device_handle)
1593 {
1594         struct nfit_mem *nfit_mem;
1595
1596         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1597                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1598                         return nfit_mem->nvdimm;
1599
1600         return NULL;
1601 }
1602
1603 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1604 {
1605         struct nfit_mem *nfit_mem;
1606         struct acpi_nfit_desc *acpi_desc;
1607
1608         dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1609                         event);
1610
1611         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1612                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1613                                 event);
1614                 return;
1615         }
1616
1617         acpi_desc = dev_get_drvdata(dev->parent);
1618         if (!acpi_desc)
1619                 return;
1620
1621         /*
1622          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1623          * is still valid.
1624          */
1625         nfit_mem = dev_get_drvdata(dev);
1626         if (nfit_mem && nfit_mem->flags_attr)
1627                 sysfs_notify_dirent(nfit_mem->flags_attr);
1628 }
1629 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1630
1631 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1632 {
1633         struct acpi_device *adev = data;
1634         struct device *dev = &adev->dev;
1635
1636         device_lock(dev->parent);
1637         __acpi_nvdimm_notify(dev, event);
1638         device_unlock(dev->parent);
1639 }
1640
1641 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1642                 struct nfit_mem *nfit_mem, u32 device_handle)
1643 {
1644         struct acpi_device *adev, *adev_dimm;
1645         struct device *dev = acpi_desc->dev;
1646         union acpi_object *obj;
1647         unsigned long dsm_mask;
1648         const guid_t *guid;
1649         int i;
1650         int family = -1;
1651
1652         /* nfit test assumes 1:1 relationship between commands and dsms */
1653         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1654         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1655         adev = to_acpi_dev(acpi_desc);
1656         if (!adev)
1657                 return 0;
1658
1659         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1660         nfit_mem->adev = adev_dimm;
1661         if (!adev_dimm) {
1662                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1663                                 device_handle);
1664                 return force_enable_dimms ? 0 : -ENODEV;
1665         }
1666
1667         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1668                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1669                 dev_err(dev, "%s: notification registration failed\n",
1670                                 dev_name(&adev_dimm->dev));
1671                 return -ENXIO;
1672         }
1673         /*
1674          * Record nfit_mem for the notification path to track back to
1675          * the nfit sysfs attributes for this dimm device object.
1676          */
1677         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1678
1679         /*
1680          * Until standardization materializes we need to consider 4
1681          * different command sets.  Note, that checking for function0 (bit0)
1682          * tells us if any commands are reachable through this GUID.
1683          */
1684         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1685                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1686                         if (family < 0 || i == default_dsm_family)
1687                                 family = i;
1688
1689         /* limit the supported commands to those that are publicly documented */
1690         nfit_mem->family = family;
1691         if (override_dsm_mask && !disable_vendor_specific)
1692                 dsm_mask = override_dsm_mask;
1693         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1694                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1695                 if (disable_vendor_specific)
1696                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1697         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1698                 dsm_mask = 0x1c3c76;
1699         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1700                 dsm_mask = 0x1fe;
1701                 if (disable_vendor_specific)
1702                         dsm_mask &= ~(1 << 8);
1703         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1704                 dsm_mask = 0xffffffff;
1705         } else {
1706                 dev_dbg(dev, "unknown dimm command family\n");
1707                 nfit_mem->family = -1;
1708                 /* DSMs are optional, continue loading the driver... */
1709                 return 0;
1710         }
1711
1712         guid = to_nfit_uuid(nfit_mem->family);
1713         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1714                 if (acpi_check_dsm(adev_dimm->handle, guid,
1715                                         nfit_dsm_revid(nfit_mem->family, i),
1716                                         1ULL << i))
1717                         set_bit(i, &nfit_mem->dsm_mask);
1718
1719         obj = acpi_label_info(adev_dimm->handle);
1720         if (obj) {
1721                 ACPI_FREE(obj);
1722                 nfit_mem->has_lsi = 1;
1723                 dev_dbg(dev, "%s: has _LSI\n", dev_name(&adev_dimm->dev));
1724         }
1725
1726         obj = acpi_label_read(adev_dimm->handle, 0, 0);
1727         if (obj) {
1728                 ACPI_FREE(obj);
1729                 nfit_mem->has_lsr = 1;
1730                 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1731         }
1732
1733         obj = acpi_label_write(adev_dimm->handle, 0, 0, NULL);
1734         if (obj) {
1735                 ACPI_FREE(obj);
1736                 nfit_mem->has_lsw = 1;
1737                 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1738         }
1739
1740         return 0;
1741 }
1742
1743 static void shutdown_dimm_notify(void *data)
1744 {
1745         struct acpi_nfit_desc *acpi_desc = data;
1746         struct nfit_mem *nfit_mem;
1747
1748         mutex_lock(&acpi_desc->init_mutex);
1749         /*
1750          * Clear out the nfit_mem->flags_attr and shut down dimm event
1751          * notifications.
1752          */
1753         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1754                 struct acpi_device *adev_dimm = nfit_mem->adev;
1755
1756                 if (nfit_mem->flags_attr) {
1757                         sysfs_put(nfit_mem->flags_attr);
1758                         nfit_mem->flags_attr = NULL;
1759                 }
1760                 if (adev_dimm) {
1761                         acpi_remove_notify_handler(adev_dimm->handle,
1762                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1763                         dev_set_drvdata(&adev_dimm->dev, NULL);
1764                 }
1765         }
1766         mutex_unlock(&acpi_desc->init_mutex);
1767 }
1768
1769 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1770 {
1771         struct nfit_mem *nfit_mem;
1772         int dimm_count = 0, rc;
1773         struct nvdimm *nvdimm;
1774
1775         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1776                 struct acpi_nfit_flush_address *flush;
1777                 unsigned long flags = 0, cmd_mask;
1778                 struct nfit_memdev *nfit_memdev;
1779                 u32 device_handle;
1780                 u16 mem_flags;
1781
1782                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1783                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1784                 if (nvdimm) {
1785                         dimm_count++;
1786                         continue;
1787                 }
1788
1789                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1790                         set_bit(NDD_ALIASING, &flags);
1791
1792                 /* collate flags across all memdevs for this dimm */
1793                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1794                         struct acpi_nfit_memory_map *dimm_memdev;
1795
1796                         dimm_memdev = __to_nfit_memdev(nfit_mem);
1797                         if (dimm_memdev->device_handle
1798                                         != nfit_memdev->memdev->device_handle)
1799                                 continue;
1800                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
1801                 }
1802
1803                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1804                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1805                         set_bit(NDD_UNARMED, &flags);
1806
1807                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1808                 if (rc)
1809                         continue;
1810
1811                 /*
1812                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1813                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1814                  * userspace interface.
1815                  */
1816                 cmd_mask = 1UL << ND_CMD_CALL;
1817                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1818                         /*
1819                          * These commands have a 1:1 correspondence
1820                          * between DSM payload and libnvdimm ioctl
1821                          * payload format.
1822                          */
1823                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1824                 }
1825
1826                 if (nfit_mem->has_lsi)
1827                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1828                 if (nfit_mem->has_lsr)
1829                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1830                 if (nfit_mem->has_lsw)
1831                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1832
1833                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1834                         : NULL;
1835                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1836                                 acpi_nfit_dimm_attribute_groups,
1837                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1838                                 nfit_mem->flush_wpq);
1839                 if (!nvdimm)
1840                         return -ENOMEM;
1841
1842                 nfit_mem->nvdimm = nvdimm;
1843                 dimm_count++;
1844
1845                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1846                         continue;
1847
1848                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1849                                 nvdimm_name(nvdimm),
1850                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1851                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1852                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1853                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1854                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1855
1856         }
1857
1858         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1859         if (rc)
1860                 return rc;
1861
1862         /*
1863          * Now that dimms are successfully registered, and async registration
1864          * is flushed, attempt to enable event notification.
1865          */
1866         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1867                 struct kernfs_node *nfit_kernfs;
1868
1869                 nvdimm = nfit_mem->nvdimm;
1870                 if (!nvdimm)
1871                         continue;
1872
1873                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1874                 if (nfit_kernfs)
1875                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1876                                         "flags");
1877                 sysfs_put(nfit_kernfs);
1878                 if (!nfit_mem->flags_attr)
1879                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1880                                         nvdimm_name(nvdimm));
1881         }
1882
1883         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1884                         acpi_desc);
1885 }
1886
1887 /*
1888  * These constants are private because there are no kernel consumers of
1889  * these commands.
1890  */
1891 enum nfit_aux_cmds {
1892         NFIT_CMD_TRANSLATE_SPA = 5,
1893         NFIT_CMD_ARS_INJECT_SET = 7,
1894         NFIT_CMD_ARS_INJECT_CLEAR = 8,
1895         NFIT_CMD_ARS_INJECT_GET = 9,
1896 };
1897
1898 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1899 {
1900         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1901         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1902         struct acpi_device *adev;
1903         unsigned long dsm_mask;
1904         int i;
1905
1906         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1907         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1908         adev = to_acpi_dev(acpi_desc);
1909         if (!adev)
1910                 return;
1911
1912         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1913                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1914                         set_bit(i, &nd_desc->cmd_mask);
1915         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1916
1917         dsm_mask =
1918                 (1 << ND_CMD_ARS_CAP) |
1919                 (1 << ND_CMD_ARS_START) |
1920                 (1 << ND_CMD_ARS_STATUS) |
1921                 (1 << ND_CMD_CLEAR_ERROR) |
1922                 (1 << NFIT_CMD_TRANSLATE_SPA) |
1923                 (1 << NFIT_CMD_ARS_INJECT_SET) |
1924                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1925                 (1 << NFIT_CMD_ARS_INJECT_GET);
1926         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1927                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1928                         set_bit(i, &nd_desc->bus_dsm_mask);
1929 }
1930
1931 static ssize_t range_index_show(struct device *dev,
1932                 struct device_attribute *attr, char *buf)
1933 {
1934         struct nd_region *nd_region = to_nd_region(dev);
1935         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1936
1937         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1938 }
1939 static DEVICE_ATTR_RO(range_index);
1940
1941 static ssize_t ecc_unit_size_show(struct device *dev,
1942                 struct device_attribute *attr, char *buf)
1943 {
1944         struct nd_region *nd_region = to_nd_region(dev);
1945         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1946
1947         return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1948 }
1949 static DEVICE_ATTR_RO(ecc_unit_size);
1950
1951 static struct attribute *acpi_nfit_region_attributes[] = {
1952         &dev_attr_range_index.attr,
1953         &dev_attr_ecc_unit_size.attr,
1954         NULL,
1955 };
1956
1957 static const struct attribute_group acpi_nfit_region_attribute_group = {
1958         .name = "nfit",
1959         .attrs = acpi_nfit_region_attributes,
1960 };
1961
1962 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1963         &nd_region_attribute_group,
1964         &nd_mapping_attribute_group,
1965         &nd_device_attribute_group,
1966         &nd_numa_attribute_group,
1967         &acpi_nfit_region_attribute_group,
1968         NULL,
1969 };
1970
1971 /* enough info to uniquely specify an interleave set */
1972 struct nfit_set_info {
1973         struct nfit_set_info_map {
1974                 u64 region_offset;
1975                 u32 serial_number;
1976                 u32 pad;
1977         } mapping[0];
1978 };
1979
1980 struct nfit_set_info2 {
1981         struct nfit_set_info_map2 {
1982                 u64 region_offset;
1983                 u32 serial_number;
1984                 u16 vendor_id;
1985                 u16 manufacturing_date;
1986                 u8  manufacturing_location;
1987                 u8  reserved[31];
1988         } mapping[0];
1989 };
1990
1991 static size_t sizeof_nfit_set_info(int num_mappings)
1992 {
1993         return sizeof(struct nfit_set_info)
1994                 + num_mappings * sizeof(struct nfit_set_info_map);
1995 }
1996
1997 static size_t sizeof_nfit_set_info2(int num_mappings)
1998 {
1999         return sizeof(struct nfit_set_info2)
2000                 + num_mappings * sizeof(struct nfit_set_info_map2);
2001 }
2002
2003 static int cmp_map_compat(const void *m0, const void *m1)
2004 {
2005         const struct nfit_set_info_map *map0 = m0;
2006         const struct nfit_set_info_map *map1 = m1;
2007
2008         return memcmp(&map0->region_offset, &map1->region_offset,
2009                         sizeof(u64));
2010 }
2011
2012 static int cmp_map(const void *m0, const void *m1)
2013 {
2014         const struct nfit_set_info_map *map0 = m0;
2015         const struct nfit_set_info_map *map1 = m1;
2016
2017         if (map0->region_offset < map1->region_offset)
2018                 return -1;
2019         else if (map0->region_offset > map1->region_offset)
2020                 return 1;
2021         return 0;
2022 }
2023
2024 static int cmp_map2(const void *m0, const void *m1)
2025 {
2026         const struct nfit_set_info_map2 *map0 = m0;
2027         const struct nfit_set_info_map2 *map1 = m1;
2028
2029         if (map0->region_offset < map1->region_offset)
2030                 return -1;
2031         else if (map0->region_offset > map1->region_offset)
2032                 return 1;
2033         return 0;
2034 }
2035
2036 /* Retrieve the nth entry referencing this spa */
2037 static struct acpi_nfit_memory_map *memdev_from_spa(
2038                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2039 {
2040         struct nfit_memdev *nfit_memdev;
2041
2042         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2043                 if (nfit_memdev->memdev->range_index == range_index)
2044                         if (n-- == 0)
2045                                 return nfit_memdev->memdev;
2046         return NULL;
2047 }
2048
2049 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2050                 struct nd_region_desc *ndr_desc,
2051                 struct acpi_nfit_system_address *spa)
2052 {
2053         struct device *dev = acpi_desc->dev;
2054         struct nd_interleave_set *nd_set;
2055         u16 nr = ndr_desc->num_mappings;
2056         struct nfit_set_info2 *info2;
2057         struct nfit_set_info *info;
2058         int i;
2059
2060         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2061         if (!nd_set)
2062                 return -ENOMEM;
2063         ndr_desc->nd_set = nd_set;
2064         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2065
2066         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2067         if (!info)
2068                 return -ENOMEM;
2069
2070         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2071         if (!info2)
2072                 return -ENOMEM;
2073
2074         for (i = 0; i < nr; i++) {
2075                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2076                 struct nfit_set_info_map *map = &info->mapping[i];
2077                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2078                 struct nvdimm *nvdimm = mapping->nvdimm;
2079                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2080                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2081                                 spa->range_index, i);
2082                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2083
2084                 if (!memdev || !nfit_mem->dcr) {
2085                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2086                         return -ENODEV;
2087                 }
2088
2089                 map->region_offset = memdev->region_offset;
2090                 map->serial_number = dcr->serial_number;
2091
2092                 map2->region_offset = memdev->region_offset;
2093                 map2->serial_number = dcr->serial_number;
2094                 map2->vendor_id = dcr->vendor_id;
2095                 map2->manufacturing_date = dcr->manufacturing_date;
2096                 map2->manufacturing_location = dcr->manufacturing_location;
2097         }
2098
2099         /* v1.1 namespaces */
2100         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2101                         cmp_map, NULL);
2102         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2103
2104         /* v1.2 namespaces */
2105         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2106                         cmp_map2, NULL);
2107         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2108
2109         /* support v1.1 namespaces created with the wrong sort order */
2110         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2111                         cmp_map_compat, NULL);
2112         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2113
2114         /* record the result of the sort for the mapping position */
2115         for (i = 0; i < nr; i++) {
2116                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2117                 int j;
2118
2119                 for (j = 0; j < nr; j++) {
2120                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2121                         struct nvdimm *nvdimm = mapping->nvdimm;
2122                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2123                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2124
2125                         if (map2->serial_number == dcr->serial_number &&
2126                             map2->vendor_id == dcr->vendor_id &&
2127                             map2->manufacturing_date == dcr->manufacturing_date &&
2128                             map2->manufacturing_location
2129                                     == dcr->manufacturing_location) {
2130                                 mapping->position = i;
2131                                 break;
2132                         }
2133                 }
2134         }
2135
2136         ndr_desc->nd_set = nd_set;
2137         devm_kfree(dev, info);
2138         devm_kfree(dev, info2);
2139
2140         return 0;
2141 }
2142
2143 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2144 {
2145         struct acpi_nfit_interleave *idt = mmio->idt;
2146         u32 sub_line_offset, line_index, line_offset;
2147         u64 line_no, table_skip_count, table_offset;
2148
2149         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2150         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2151         line_offset = idt->line_offset[line_index]
2152                 * mmio->line_size;
2153         table_offset = table_skip_count * mmio->table_size;
2154
2155         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2156 }
2157
2158 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2159 {
2160         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2161         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2162         const u32 STATUS_MASK = 0x80000037;
2163
2164         if (mmio->num_lines)
2165                 offset = to_interleave_offset(offset, mmio);
2166
2167         return readl(mmio->addr.base + offset) & STATUS_MASK;
2168 }
2169
2170 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2171                 resource_size_t dpa, unsigned int len, unsigned int write)
2172 {
2173         u64 cmd, offset;
2174         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2175
2176         enum {
2177                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2178                 BCW_LEN_SHIFT = 48,
2179                 BCW_LEN_MASK = (1ULL << 8) - 1,
2180                 BCW_CMD_SHIFT = 56,
2181         };
2182
2183         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2184         len = len >> L1_CACHE_SHIFT;
2185         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2186         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2187
2188         offset = nfit_blk->cmd_offset + mmio->size * bw;
2189         if (mmio->num_lines)
2190                 offset = to_interleave_offset(offset, mmio);
2191
2192         writeq(cmd, mmio->addr.base + offset);
2193         nvdimm_flush(nfit_blk->nd_region);
2194
2195         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2196                 readq(mmio->addr.base + offset);
2197 }
2198
2199 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2200                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2201                 unsigned int lane)
2202 {
2203         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2204         unsigned int copied = 0;
2205         u64 base_offset;
2206         int rc;
2207
2208         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2209                 + lane * mmio->size;
2210         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2211         while (len) {
2212                 unsigned int c;
2213                 u64 offset;
2214
2215                 if (mmio->num_lines) {
2216                         u32 line_offset;
2217
2218                         offset = to_interleave_offset(base_offset + copied,
2219                                         mmio);
2220                         div_u64_rem(offset, mmio->line_size, &line_offset);
2221                         c = min_t(size_t, len, mmio->line_size - line_offset);
2222                 } else {
2223                         offset = base_offset + nfit_blk->bdw_offset;
2224                         c = len;
2225                 }
2226
2227                 if (rw)
2228                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2229                 else {
2230                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2231                                 arch_invalidate_pmem((void __force *)
2232                                         mmio->addr.aperture + offset, c);
2233
2234                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2235                 }
2236
2237                 copied += c;
2238                 len -= c;
2239         }
2240
2241         if (rw)
2242                 nvdimm_flush(nfit_blk->nd_region);
2243
2244         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2245         return rc;
2246 }
2247
2248 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2249                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2250 {
2251         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2252         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2253         struct nd_region *nd_region = nfit_blk->nd_region;
2254         unsigned int lane, copied = 0;
2255         int rc = 0;
2256
2257         lane = nd_region_acquire_lane(nd_region);
2258         while (len) {
2259                 u64 c = min(len, mmio->size);
2260
2261                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2262                                 iobuf + copied, c, rw, lane);
2263                 if (rc)
2264                         break;
2265
2266                 copied += c;
2267                 len -= c;
2268         }
2269         nd_region_release_lane(nd_region, lane);
2270
2271         return rc;
2272 }
2273
2274 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2275                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2276 {
2277         if (idt) {
2278                 mmio->num_lines = idt->line_count;
2279                 mmio->line_size = idt->line_size;
2280                 if (interleave_ways == 0)
2281                         return -ENXIO;
2282                 mmio->table_size = mmio->num_lines * interleave_ways
2283                         * mmio->line_size;
2284         }
2285
2286         return 0;
2287 }
2288
2289 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2290                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2291 {
2292         struct nd_cmd_dimm_flags flags;
2293         int rc;
2294
2295         memset(&flags, 0, sizeof(flags));
2296         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2297                         sizeof(flags), NULL);
2298
2299         if (rc >= 0 && flags.status == 0)
2300                 nfit_blk->dimm_flags = flags.flags;
2301         else if (rc == -ENOTTY) {
2302                 /* fall back to a conservative default */
2303                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2304                 rc = 0;
2305         } else
2306                 rc = -ENXIO;
2307
2308         return rc;
2309 }
2310
2311 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2312                 struct device *dev)
2313 {
2314         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2315         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2316         struct nfit_blk_mmio *mmio;
2317         struct nfit_blk *nfit_blk;
2318         struct nfit_mem *nfit_mem;
2319         struct nvdimm *nvdimm;
2320         int rc;
2321
2322         nvdimm = nd_blk_region_to_dimm(ndbr);
2323         nfit_mem = nvdimm_provider_data(nvdimm);
2324         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2325                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2326                                 nfit_mem ? "" : " nfit_mem",
2327                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2328                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2329                 return -ENXIO;
2330         }
2331
2332         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2333         if (!nfit_blk)
2334                 return -ENOMEM;
2335         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2336         nfit_blk->nd_region = to_nd_region(dev);
2337
2338         /* map block aperture memory */
2339         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2340         mmio = &nfit_blk->mmio[BDW];
2341         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2342                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2343         if (!mmio->addr.base) {
2344                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2345                                 nvdimm_name(nvdimm));
2346                 return -ENOMEM;
2347         }
2348         mmio->size = nfit_mem->bdw->size;
2349         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2350         mmio->idt = nfit_mem->idt_bdw;
2351         mmio->spa = nfit_mem->spa_bdw;
2352         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2353                         nfit_mem->memdev_bdw->interleave_ways);
2354         if (rc) {
2355                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2356                                 __func__, nvdimm_name(nvdimm));
2357                 return rc;
2358         }
2359
2360         /* map block control memory */
2361         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2362         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2363         mmio = &nfit_blk->mmio[DCR];
2364         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2365                         nfit_mem->spa_dcr->length);
2366         if (!mmio->addr.base) {
2367                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2368                                 nvdimm_name(nvdimm));
2369                 return -ENOMEM;
2370         }
2371         mmio->size = nfit_mem->dcr->window_size;
2372         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2373         mmio->idt = nfit_mem->idt_dcr;
2374         mmio->spa = nfit_mem->spa_dcr;
2375         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2376                         nfit_mem->memdev_dcr->interleave_ways);
2377         if (rc) {
2378                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2379                                 __func__, nvdimm_name(nvdimm));
2380                 return rc;
2381         }
2382
2383         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2384         if (rc < 0) {
2385                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2386                                 __func__, nvdimm_name(nvdimm));
2387                 return rc;
2388         }
2389
2390         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2391                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2392
2393         if (mmio->line_size == 0)
2394                 return 0;
2395
2396         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2397                         + 8 > mmio->line_size) {
2398                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2399                 return -ENXIO;
2400         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2401                         + 8 > mmio->line_size) {
2402                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2403                 return -ENXIO;
2404         }
2405
2406         return 0;
2407 }
2408
2409 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2410                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2411 {
2412         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2413         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2414         int cmd_rc, rc;
2415
2416         cmd->address = spa->address;
2417         cmd->length = spa->length;
2418         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2419                         sizeof(*cmd), &cmd_rc);
2420         if (rc < 0)
2421                 return rc;
2422         return cmd_rc;
2423 }
2424
2425 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2426 {
2427         int rc;
2428         int cmd_rc;
2429         struct nd_cmd_ars_start ars_start;
2430         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2431         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2432
2433         memset(&ars_start, 0, sizeof(ars_start));
2434         ars_start.address = spa->address;
2435         ars_start.length = spa->length;
2436         ars_start.flags = acpi_desc->ars_start_flags;
2437         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2438                 ars_start.type = ND_ARS_PERSISTENT;
2439         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2440                 ars_start.type = ND_ARS_VOLATILE;
2441         else
2442                 return -ENOTTY;
2443
2444         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2445                         sizeof(ars_start), &cmd_rc);
2446
2447         if (rc < 0)
2448                 return rc;
2449         return cmd_rc;
2450 }
2451
2452 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2453 {
2454         int rc, cmd_rc;
2455         struct nd_cmd_ars_start ars_start;
2456         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2457         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2458
2459         memset(&ars_start, 0, sizeof(ars_start));
2460         ars_start.address = ars_status->restart_address;
2461         ars_start.length = ars_status->restart_length;
2462         ars_start.type = ars_status->type;
2463         ars_start.flags = acpi_desc->ars_start_flags;
2464         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2465                         sizeof(ars_start), &cmd_rc);
2466         if (rc < 0)
2467                 return rc;
2468         return cmd_rc;
2469 }
2470
2471 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2472 {
2473         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2474         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2475         int rc, cmd_rc;
2476
2477         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2478                         acpi_desc->ars_status_size, &cmd_rc);
2479         if (rc < 0)
2480                 return rc;
2481         return cmd_rc;
2482 }
2483
2484 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2485                 struct nd_cmd_ars_status *ars_status)
2486 {
2487         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2488         int rc;
2489         u32 i;
2490
2491         /*
2492          * First record starts at 44 byte offset from the start of the
2493          * payload.
2494          */
2495         if (ars_status->out_length < 44)
2496                 return 0;
2497         for (i = 0; i < ars_status->num_records; i++) {
2498                 /* only process full records */
2499                 if (ars_status->out_length
2500                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2501                         break;
2502                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2503                                 ars_status->records[i].err_address,
2504                                 ars_status->records[i].length);
2505                 if (rc)
2506                         return rc;
2507         }
2508         if (i < ars_status->num_records)
2509                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2510
2511         return 0;
2512 }
2513
2514 static void acpi_nfit_remove_resource(void *data)
2515 {
2516         struct resource *res = data;
2517
2518         remove_resource(res);
2519 }
2520
2521 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2522                 struct nd_region_desc *ndr_desc)
2523 {
2524         struct resource *res, *nd_res = ndr_desc->res;
2525         int is_pmem, ret;
2526
2527         /* No operation if the region is already registered as PMEM */
2528         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2529                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2530         if (is_pmem == REGION_INTERSECTS)
2531                 return 0;
2532
2533         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2534         if (!res)
2535                 return -ENOMEM;
2536
2537         res->name = "Persistent Memory";
2538         res->start = nd_res->start;
2539         res->end = nd_res->end;
2540         res->flags = IORESOURCE_MEM;
2541         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2542
2543         ret = insert_resource(&iomem_resource, res);
2544         if (ret)
2545                 return ret;
2546
2547         ret = devm_add_action_or_reset(acpi_desc->dev,
2548                                         acpi_nfit_remove_resource,
2549                                         res);
2550         if (ret)
2551                 return ret;
2552
2553         return 0;
2554 }
2555
2556 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2557                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2558                 struct acpi_nfit_memory_map *memdev,
2559                 struct nfit_spa *nfit_spa)
2560 {
2561         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2562                         memdev->device_handle);
2563         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2564         struct nd_blk_region_desc *ndbr_desc;
2565         struct nfit_mem *nfit_mem;
2566         int blk_valid = 0, rc;
2567
2568         if (!nvdimm) {
2569                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2570                                 spa->range_index, memdev->device_handle);
2571                 return -ENODEV;
2572         }
2573
2574         mapping->nvdimm = nvdimm;
2575         switch (nfit_spa_type(spa)) {
2576         case NFIT_SPA_PM:
2577         case NFIT_SPA_VOLATILE:
2578                 mapping->start = memdev->address;
2579                 mapping->size = memdev->region_size;
2580                 break;
2581         case NFIT_SPA_DCR:
2582                 nfit_mem = nvdimm_provider_data(nvdimm);
2583                 if (!nfit_mem || !nfit_mem->bdw) {
2584                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2585                                         spa->range_index, nvdimm_name(nvdimm));
2586                 } else {
2587                         mapping->size = nfit_mem->bdw->capacity;
2588                         mapping->start = nfit_mem->bdw->start_address;
2589                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
2590                         blk_valid = 1;
2591                 }
2592
2593                 ndr_desc->mapping = mapping;
2594                 ndr_desc->num_mappings = blk_valid;
2595                 ndbr_desc = to_blk_region_desc(ndr_desc);
2596                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2597                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2598                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2599                 if (rc)
2600                         return rc;
2601                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2602                                 ndr_desc);
2603                 if (!nfit_spa->nd_region)
2604                         return -ENOMEM;
2605                 break;
2606         }
2607
2608         return 0;
2609 }
2610
2611 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2612 {
2613         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2614                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2615                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2616                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2617 }
2618
2619 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2620 {
2621         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2622                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2623                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2624 }
2625
2626 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2627                 struct nfit_spa *nfit_spa)
2628 {
2629         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2630         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2631         struct nd_blk_region_desc ndbr_desc;
2632         struct nd_region_desc *ndr_desc;
2633         struct nfit_memdev *nfit_memdev;
2634         struct nvdimm_bus *nvdimm_bus;
2635         struct resource res;
2636         int count = 0, rc;
2637
2638         if (nfit_spa->nd_region)
2639                 return 0;
2640
2641         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2642                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2643                                 __func__);
2644                 return 0;
2645         }
2646
2647         memset(&res, 0, sizeof(res));
2648         memset(&mappings, 0, sizeof(mappings));
2649         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2650         res.start = spa->address;
2651         res.end = res.start + spa->length - 1;
2652         ndr_desc = &ndbr_desc.ndr_desc;
2653         ndr_desc->res = &res;
2654         ndr_desc->provider_data = nfit_spa;
2655         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2656         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2657                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2658                                                 spa->proximity_domain);
2659         else
2660                 ndr_desc->numa_node = NUMA_NO_NODE;
2661
2662         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2663                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2664                 struct nd_mapping_desc *mapping;
2665
2666                 if (memdev->range_index != spa->range_index)
2667                         continue;
2668                 if (count >= ND_MAX_MAPPINGS) {
2669                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2670                                         spa->range_index, ND_MAX_MAPPINGS);
2671                         return -ENXIO;
2672                 }
2673                 mapping = &mappings[count++];
2674                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2675                                 memdev, nfit_spa);
2676                 if (rc)
2677                         goto out;
2678         }
2679
2680         ndr_desc->mapping = mappings;
2681         ndr_desc->num_mappings = count;
2682         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2683         if (rc)
2684                 goto out;
2685
2686         nvdimm_bus = acpi_desc->nvdimm_bus;
2687         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2688                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2689                 if (rc) {
2690                         dev_warn(acpi_desc->dev,
2691                                 "failed to insert pmem resource to iomem: %d\n",
2692                                 rc);
2693                         goto out;
2694                 }
2695
2696                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2697                                 ndr_desc);
2698                 if (!nfit_spa->nd_region)
2699                         rc = -ENOMEM;
2700         } else if (nfit_spa_is_volatile(spa)) {
2701                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2702                                 ndr_desc);
2703                 if (!nfit_spa->nd_region)
2704                         rc = -ENOMEM;
2705         } else if (nfit_spa_is_virtual(spa)) {
2706                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2707                                 ndr_desc);
2708                 if (!nfit_spa->nd_region)
2709                         rc = -ENOMEM;
2710         }
2711
2712  out:
2713         if (rc)
2714                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2715                                 nfit_spa->spa->range_index);
2716         return rc;
2717 }
2718
2719 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2720                 u32 max_ars)
2721 {
2722         struct device *dev = acpi_desc->dev;
2723         struct nd_cmd_ars_status *ars_status;
2724
2725         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2726                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2727                 return 0;
2728         }
2729
2730         if (acpi_desc->ars_status)
2731                 devm_kfree(dev, acpi_desc->ars_status);
2732         acpi_desc->ars_status = NULL;
2733         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2734         if (!ars_status)
2735                 return -ENOMEM;
2736         acpi_desc->ars_status = ars_status;
2737         acpi_desc->ars_status_size = max_ars;
2738         return 0;
2739 }
2740
2741 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2742                 struct nfit_spa *nfit_spa)
2743 {
2744         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2745         int rc;
2746
2747         if (!nfit_spa->max_ars) {
2748                 struct nd_cmd_ars_cap ars_cap;
2749
2750                 memset(&ars_cap, 0, sizeof(ars_cap));
2751                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2752                 if (rc < 0)
2753                         return rc;
2754                 nfit_spa->max_ars = ars_cap.max_ars_out;
2755                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2756                 /* check that the supported scrub types match the spa type */
2757                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2758                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2759                         return -ENOTTY;
2760                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2761                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2762                         return -ENOTTY;
2763         }
2764
2765         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2766                 return -ENOMEM;
2767
2768         rc = ars_get_status(acpi_desc);
2769         if (rc < 0 && rc != -ENOSPC)
2770                 return rc;
2771
2772         if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2773                 return -ENOMEM;
2774
2775         return 0;
2776 }
2777
2778 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2779                 struct nfit_spa *nfit_spa)
2780 {
2781         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2782         unsigned int overflow_retry = scrub_overflow_abort;
2783         u64 init_ars_start = 0, init_ars_len = 0;
2784         struct device *dev = acpi_desc->dev;
2785         unsigned int tmo = scrub_timeout;
2786         int rc;
2787
2788         if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2789                 return;
2790
2791         rc = ars_start(acpi_desc, nfit_spa);
2792         /*
2793          * If we timed out the initial scan we'll still be busy here,
2794          * and will wait another timeout before giving up permanently.
2795          */
2796         if (rc < 0 && rc != -EBUSY)
2797                 return;
2798
2799         do {
2800                 u64 ars_start, ars_len;
2801
2802                 if (acpi_desc->cancel)
2803                         break;
2804                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2805                 if (rc == -ENOTTY)
2806                         break;
2807                 if (rc == -EBUSY && !tmo) {
2808                         dev_warn(dev, "range %d ars timeout, aborting\n",
2809                                         spa->range_index);
2810                         break;
2811                 }
2812
2813                 if (rc == -EBUSY) {
2814                         /*
2815                          * Note, entries may be appended to the list
2816                          * while the lock is dropped, but the workqueue
2817                          * being active prevents entries being deleted /
2818                          * freed.
2819                          */
2820                         mutex_unlock(&acpi_desc->init_mutex);
2821                         ssleep(1);
2822                         tmo--;
2823                         mutex_lock(&acpi_desc->init_mutex);
2824                         continue;
2825                 }
2826
2827                 /* we got some results, but there are more pending... */
2828                 if (rc == -ENOSPC && overflow_retry--) {
2829                         if (!init_ars_len) {
2830                                 init_ars_len = acpi_desc->ars_status->length;
2831                                 init_ars_start = acpi_desc->ars_status->address;
2832                         }
2833                         rc = ars_continue(acpi_desc);
2834                 }
2835
2836                 if (rc < 0) {
2837                         dev_warn(dev, "range %d ars continuation failed\n",
2838                                         spa->range_index);
2839                         break;
2840                 }
2841
2842                 if (init_ars_len) {
2843                         ars_start = init_ars_start;
2844                         ars_len = init_ars_len;
2845                 } else {
2846                         ars_start = acpi_desc->ars_status->address;
2847                         ars_len = acpi_desc->ars_status->length;
2848                 }
2849                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2850                                 spa->range_index, ars_start, ars_len);
2851                 /* notify the region about new poison entries */
2852                 nvdimm_region_notify(nfit_spa->nd_region,
2853                                 NVDIMM_REVALIDATE_POISON);
2854                 break;
2855         } while (1);
2856 }
2857
2858 static void acpi_nfit_scrub(struct work_struct *work)
2859 {
2860         struct device *dev;
2861         u64 init_scrub_length = 0;
2862         struct nfit_spa *nfit_spa;
2863         u64 init_scrub_address = 0;
2864         bool init_ars_done = false;
2865         struct acpi_nfit_desc *acpi_desc;
2866         unsigned int tmo = scrub_timeout;
2867         unsigned int overflow_retry = scrub_overflow_abort;
2868
2869         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2870         dev = acpi_desc->dev;
2871
2872         /*
2873          * We scrub in 2 phases.  The first phase waits for any platform
2874          * firmware initiated scrubs to complete and then we go search for the
2875          * affected spa regions to mark them scanned.  In the second phase we
2876          * initiate a directed scrub for every range that was not scrubbed in
2877          * phase 1. If we're called for a 'rescan', we harmlessly pass through
2878          * the first phase, but really only care about running phase 2, where
2879          * regions can be notified of new poison.
2880          */
2881
2882         /* process platform firmware initiated scrubs */
2883  retry:
2884         mutex_lock(&acpi_desc->init_mutex);
2885         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2886                 struct nd_cmd_ars_status *ars_status;
2887                 struct acpi_nfit_system_address *spa;
2888                 u64 ars_start, ars_len;
2889                 int rc;
2890
2891                 if (acpi_desc->cancel)
2892                         break;
2893
2894                 if (nfit_spa->nd_region)
2895                         continue;
2896
2897                 if (init_ars_done) {
2898                         /*
2899                          * No need to re-query, we're now just
2900                          * reconciling all the ranges covered by the
2901                          * initial scrub
2902                          */
2903                         rc = 0;
2904                 } else
2905                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2906
2907                 if (rc == -ENOTTY) {
2908                         /* no ars capability, just register spa and move on */
2909                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2910                         continue;
2911                 }
2912
2913                 if (rc == -EBUSY && !tmo) {
2914                         /* fallthrough to directed scrub in phase 2 */
2915                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2916                         break;
2917                 } else if (rc == -EBUSY) {
2918                         mutex_unlock(&acpi_desc->init_mutex);
2919                         ssleep(1);
2920                         tmo--;
2921                         goto retry;
2922                 }
2923
2924                 /* we got some results, but there are more pending... */
2925                 if (rc == -ENOSPC && overflow_retry--) {
2926                         ars_status = acpi_desc->ars_status;
2927                         /*
2928                          * Record the original scrub range, so that we
2929                          * can recall all the ranges impacted by the
2930                          * initial scrub.
2931                          */
2932                         if (!init_scrub_length) {
2933                                 init_scrub_length = ars_status->length;
2934                                 init_scrub_address = ars_status->address;
2935                         }
2936                         rc = ars_continue(acpi_desc);
2937                         if (rc == 0) {
2938                                 mutex_unlock(&acpi_desc->init_mutex);
2939                                 goto retry;
2940                         }
2941                 }
2942
2943                 if (rc < 0) {
2944                         /*
2945                          * Initial scrub failed, we'll give it one more
2946                          * try below...
2947                          */
2948                         break;
2949                 }
2950
2951                 /* We got some final results, record completed ranges */
2952                 ars_status = acpi_desc->ars_status;
2953                 if (init_scrub_length) {
2954                         ars_start = init_scrub_address;
2955                         ars_len = ars_start + init_scrub_length;
2956                 } else {
2957                         ars_start = ars_status->address;
2958                         ars_len = ars_status->length;
2959                 }
2960                 spa = nfit_spa->spa;
2961
2962                 if (!init_ars_done) {
2963                         init_ars_done = true;
2964                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2965                                         ars_start, ars_len);
2966                 }
2967                 if (ars_start <= spa->address && ars_start + ars_len
2968                                 >= spa->address + spa->length)
2969                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2970         }
2971
2972         /*
2973          * For all the ranges not covered by an initial scrub we still
2974          * want to see if there are errors, but it's ok to discover them
2975          * asynchronously.
2976          */
2977         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2978                 /*
2979                  * Flag all the ranges that still need scrubbing, but
2980                  * register them now to make data available.
2981                  */
2982                 if (!nfit_spa->nd_region) {
2983                         nfit_spa->ars_required = 1;
2984                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2985                 }
2986         }
2987         acpi_desc->init_complete = 1;
2988
2989         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2990                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2991         acpi_desc->scrub_count++;
2992         acpi_desc->ars_start_flags = 0;
2993         if (acpi_desc->scrub_count_state)
2994                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2995         mutex_unlock(&acpi_desc->init_mutex);
2996 }
2997
2998 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2999 {
3000         struct nfit_spa *nfit_spa;
3001         int rc;
3002
3003         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3004                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
3005                         /* BLK regions don't need to wait for ars results */
3006                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3007                         if (rc)
3008                                 return rc;
3009                 }
3010
3011         acpi_desc->ars_start_flags = 0;
3012         if (!acpi_desc->cancel)
3013                 queue_work(nfit_wq, &acpi_desc->work);
3014         return 0;
3015 }
3016
3017 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3018                 struct nfit_table_prev *prev)
3019 {
3020         struct device *dev = acpi_desc->dev;
3021
3022         if (!list_empty(&prev->spas) ||
3023                         !list_empty(&prev->memdevs) ||
3024                         !list_empty(&prev->dcrs) ||
3025                         !list_empty(&prev->bdws) ||
3026                         !list_empty(&prev->idts) ||
3027                         !list_empty(&prev->flushes)) {
3028                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3029                 return -ENXIO;
3030         }
3031         return 0;
3032 }
3033
3034 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3035 {
3036         struct device *dev = acpi_desc->dev;
3037         struct kernfs_node *nfit;
3038         struct device *bus_dev;
3039
3040         if (!ars_supported(acpi_desc->nvdimm_bus))
3041                 return 0;
3042
3043         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3044         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3045         if (!nfit) {
3046                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3047                 return -ENODEV;
3048         }
3049         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3050         sysfs_put(nfit);
3051         if (!acpi_desc->scrub_count_state) {
3052                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3053                 return -ENODEV;
3054         }
3055
3056         return 0;
3057 }
3058
3059 static void acpi_nfit_unregister(void *data)
3060 {
3061         struct acpi_nfit_desc *acpi_desc = data;
3062
3063         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3064 }
3065
3066 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3067 {
3068         struct device *dev = acpi_desc->dev;
3069         struct nfit_table_prev prev;
3070         const void *end;
3071         int rc;
3072
3073         if (!acpi_desc->nvdimm_bus) {
3074                 acpi_nfit_init_dsms(acpi_desc);
3075
3076                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3077                                 &acpi_desc->nd_desc);
3078                 if (!acpi_desc->nvdimm_bus)
3079                         return -ENOMEM;
3080
3081                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3082                                 acpi_desc);
3083                 if (rc)
3084                         return rc;
3085
3086                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3087                 if (rc)
3088                         return rc;
3089
3090                 /* register this acpi_desc for mce notifications */
3091                 mutex_lock(&acpi_desc_lock);
3092                 list_add_tail(&acpi_desc->list, &acpi_descs);
3093                 mutex_unlock(&acpi_desc_lock);
3094         }
3095
3096         mutex_lock(&acpi_desc->init_mutex);
3097
3098         INIT_LIST_HEAD(&prev.spas);
3099         INIT_LIST_HEAD(&prev.memdevs);
3100         INIT_LIST_HEAD(&prev.dcrs);
3101         INIT_LIST_HEAD(&prev.bdws);
3102         INIT_LIST_HEAD(&prev.idts);
3103         INIT_LIST_HEAD(&prev.flushes);
3104
3105         list_cut_position(&prev.spas, &acpi_desc->spas,
3106                                 acpi_desc->spas.prev);
3107         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3108                                 acpi_desc->memdevs.prev);
3109         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3110                                 acpi_desc->dcrs.prev);
3111         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3112                                 acpi_desc->bdws.prev);
3113         list_cut_position(&prev.idts, &acpi_desc->idts,
3114                                 acpi_desc->idts.prev);
3115         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3116                                 acpi_desc->flushes.prev);
3117
3118         end = data + sz;
3119         while (!IS_ERR_OR_NULL(data))
3120                 data = add_table(acpi_desc, &prev, data, end);
3121
3122         if (IS_ERR(data)) {
3123                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
3124                                 PTR_ERR(data));
3125                 rc = PTR_ERR(data);
3126                 goto out_unlock;
3127         }
3128
3129         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3130         if (rc)
3131                 goto out_unlock;
3132
3133         rc = nfit_mem_init(acpi_desc);
3134         if (rc)
3135                 goto out_unlock;
3136
3137         rc = acpi_nfit_register_dimms(acpi_desc);
3138         if (rc)
3139                 goto out_unlock;
3140
3141         rc = acpi_nfit_register_regions(acpi_desc);
3142
3143  out_unlock:
3144         mutex_unlock(&acpi_desc->init_mutex);
3145         return rc;
3146 }
3147 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3148
3149 struct acpi_nfit_flush_work {
3150         struct work_struct work;
3151         struct completion cmp;
3152 };
3153
3154 static void flush_probe(struct work_struct *work)
3155 {
3156         struct acpi_nfit_flush_work *flush;
3157
3158         flush = container_of(work, typeof(*flush), work);
3159         complete(&flush->cmp);
3160 }
3161
3162 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3163 {
3164         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3165         struct device *dev = acpi_desc->dev;
3166         struct acpi_nfit_flush_work flush;
3167         int rc;
3168
3169         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3170         device_lock(dev);
3171         device_unlock(dev);
3172
3173         /* bounce the init_mutex to make init_complete valid */
3174         mutex_lock(&acpi_desc->init_mutex);
3175         if (acpi_desc->cancel || acpi_desc->init_complete) {
3176                 mutex_unlock(&acpi_desc->init_mutex);
3177                 return 0;
3178         }
3179
3180         /*
3181          * Scrub work could take 10s of seconds, userspace may give up so we
3182          * need to be interruptible while waiting.
3183          */
3184         INIT_WORK_ONSTACK(&flush.work, flush_probe);
3185         init_completion(&flush.cmp);
3186         queue_work(nfit_wq, &flush.work);
3187         mutex_unlock(&acpi_desc->init_mutex);
3188
3189         rc = wait_for_completion_interruptible(&flush.cmp);
3190         cancel_work_sync(&flush.work);
3191         return rc;
3192 }
3193
3194 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3195                 struct nvdimm *nvdimm, unsigned int cmd)
3196 {
3197         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3198
3199         if (nvdimm)
3200                 return 0;
3201         if (cmd != ND_CMD_ARS_START)
3202                 return 0;
3203
3204         /*
3205          * The kernel and userspace may race to initiate a scrub, but
3206          * the scrub thread is prepared to lose that initial race.  It
3207          * just needs guarantees that any ars it initiates are not
3208          * interrupted by any intervening start reqeusts from userspace.
3209          */
3210         if (work_busy(&acpi_desc->work))
3211                 return -EBUSY;
3212
3213         return 0;
3214 }
3215
3216 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
3217 {
3218         struct device *dev = acpi_desc->dev;
3219         struct nfit_spa *nfit_spa;
3220
3221         if (work_busy(&acpi_desc->work))
3222                 return -EBUSY;
3223
3224         mutex_lock(&acpi_desc->init_mutex);
3225         if (acpi_desc->cancel) {
3226                 mutex_unlock(&acpi_desc->init_mutex);
3227                 return 0;
3228         }
3229
3230         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3231                 struct acpi_nfit_system_address *spa = nfit_spa->spa;
3232
3233                 if (nfit_spa_type(spa) != NFIT_SPA_PM)
3234                         continue;
3235
3236                 nfit_spa->ars_required = 1;
3237         }
3238         acpi_desc->ars_start_flags = flags;
3239         queue_work(nfit_wq, &acpi_desc->work);
3240         dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
3241         mutex_unlock(&acpi_desc->init_mutex);
3242
3243         return 0;
3244 }
3245
3246 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3247 {
3248         struct nvdimm_bus_descriptor *nd_desc;
3249
3250         dev_set_drvdata(dev, acpi_desc);
3251         acpi_desc->dev = dev;
3252         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3253         nd_desc = &acpi_desc->nd_desc;
3254         nd_desc->provider_name = "ACPI.NFIT";
3255         nd_desc->module = THIS_MODULE;
3256         nd_desc->ndctl = acpi_nfit_ctl;
3257         nd_desc->flush_probe = acpi_nfit_flush_probe;
3258         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3259         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3260
3261         INIT_LIST_HEAD(&acpi_desc->spas);
3262         INIT_LIST_HEAD(&acpi_desc->dcrs);
3263         INIT_LIST_HEAD(&acpi_desc->bdws);
3264         INIT_LIST_HEAD(&acpi_desc->idts);
3265         INIT_LIST_HEAD(&acpi_desc->flushes);
3266         INIT_LIST_HEAD(&acpi_desc->memdevs);
3267         INIT_LIST_HEAD(&acpi_desc->dimms);
3268         INIT_LIST_HEAD(&acpi_desc->list);
3269         mutex_init(&acpi_desc->init_mutex);
3270         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3271 }
3272 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3273
3274 static void acpi_nfit_put_table(void *table)
3275 {
3276         acpi_put_table(table);
3277 }
3278
3279 void acpi_nfit_shutdown(void *data)
3280 {
3281         struct acpi_nfit_desc *acpi_desc = data;
3282         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3283
3284         /*
3285          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3286          * race teardown
3287          */
3288         mutex_lock(&acpi_desc_lock);
3289         list_del(&acpi_desc->list);
3290         mutex_unlock(&acpi_desc_lock);
3291
3292         mutex_lock(&acpi_desc->init_mutex);
3293         acpi_desc->cancel = 1;
3294         mutex_unlock(&acpi_desc->init_mutex);
3295
3296         /*
3297          * Bounce the nvdimm bus lock to make sure any in-flight
3298          * acpi_nfit_ars_rescan() submissions have had a chance to
3299          * either submit or see ->cancel set.
3300          */
3301         device_lock(bus_dev);
3302         device_unlock(bus_dev);
3303
3304         flush_workqueue(nfit_wq);
3305 }
3306 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3307
3308 static int acpi_nfit_add(struct acpi_device *adev)
3309 {
3310         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3311         struct acpi_nfit_desc *acpi_desc;
3312         struct device *dev = &adev->dev;
3313         struct acpi_table_header *tbl;
3314         acpi_status status = AE_OK;
3315         acpi_size sz;
3316         int rc = 0;
3317
3318         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3319         if (ACPI_FAILURE(status)) {
3320                 /* This is ok, we could have an nvdimm hotplugged later */
3321                 dev_dbg(dev, "failed to find NFIT at startup\n");
3322                 return 0;
3323         }
3324
3325         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3326         if (rc)
3327                 return rc;
3328         sz = tbl->length;
3329
3330         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3331         if (!acpi_desc)
3332                 return -ENOMEM;
3333         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3334
3335         /* Save the acpi header for exporting the revision via sysfs */
3336         acpi_desc->acpi_header = *tbl;
3337
3338         /* Evaluate _FIT and override with that if present */
3339         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3340         if (ACPI_SUCCESS(status) && buf.length > 0) {
3341                 union acpi_object *obj = buf.pointer;
3342
3343                 if (obj->type == ACPI_TYPE_BUFFER)
3344                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3345                                         obj->buffer.length);
3346                 else
3347                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
3348                                  __func__, (int) obj->type);
3349                 kfree(buf.pointer);
3350         } else
3351                 /* skip over the lead-in header table */
3352                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3353                                 + sizeof(struct acpi_table_nfit),
3354                                 sz - sizeof(struct acpi_table_nfit));
3355
3356         if (rc)
3357                 return rc;
3358         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3359 }
3360
3361 static int acpi_nfit_remove(struct acpi_device *adev)
3362 {
3363         /* see acpi_nfit_unregister */
3364         return 0;
3365 }
3366
3367 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3368 {
3369         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3370         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3371         union acpi_object *obj;
3372         acpi_status status;
3373         int ret;
3374
3375         if (!dev->driver) {
3376                 /* dev->driver may be null if we're being removed */
3377                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
3378                 return;
3379         }
3380
3381         if (!acpi_desc) {
3382                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3383                 if (!acpi_desc)
3384                         return;
3385                 acpi_nfit_desc_init(acpi_desc, dev);
3386         } else {
3387                 /*
3388                  * Finish previous registration before considering new
3389                  * regions.
3390                  */
3391                 flush_workqueue(nfit_wq);
3392         }
3393
3394         /* Evaluate _FIT */
3395         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3396         if (ACPI_FAILURE(status)) {
3397                 dev_err(dev, "failed to evaluate _FIT\n");
3398                 return;
3399         }
3400
3401         obj = buf.pointer;
3402         if (obj->type == ACPI_TYPE_BUFFER) {
3403                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3404                                 obj->buffer.length);
3405                 if (ret)
3406                         dev_err(dev, "failed to merge updated NFIT\n");
3407         } else
3408                 dev_err(dev, "Invalid _FIT\n");
3409         kfree(buf.pointer);
3410 }
3411
3412 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3413 {
3414         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3415         u8 flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3416                         0 : ND_ARS_RETURN_PREV_DATA;
3417
3418         acpi_nfit_ars_rescan(acpi_desc, flags);
3419 }
3420
3421 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3422 {
3423         dev_dbg(dev, "%s: event: 0x%x\n", __func__, event);
3424
3425         switch (event) {
3426         case NFIT_NOTIFY_UPDATE:
3427                 return acpi_nfit_update_notify(dev, handle);
3428         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3429                 return acpi_nfit_uc_error_notify(dev, handle);
3430         default:
3431                 return;
3432         }
3433 }
3434 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3435
3436 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3437 {
3438         device_lock(&adev->dev);
3439         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3440         device_unlock(&adev->dev);
3441 }
3442
3443 static const struct acpi_device_id acpi_nfit_ids[] = {
3444         { "ACPI0012", 0 },
3445         { "", 0 },
3446 };
3447 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3448
3449 static struct acpi_driver acpi_nfit_driver = {
3450         .name = KBUILD_MODNAME,
3451         .ids = acpi_nfit_ids,
3452         .ops = {
3453                 .add = acpi_nfit_add,
3454                 .remove = acpi_nfit_remove,
3455                 .notify = acpi_nfit_notify,
3456         },
3457 };
3458
3459 static __init int nfit_init(void)
3460 {
3461         int ret;
3462
3463         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3464         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3465         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3466         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3467         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3468         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3469         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3470
3471         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3472         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3473         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3474         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3475         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3476         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3477         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3478         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3479         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3480         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3481         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3482         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3483         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3484
3485         nfit_wq = create_singlethread_workqueue("nfit");
3486         if (!nfit_wq)
3487                 return -ENOMEM;
3488
3489         nfit_mce_register();
3490         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3491         if (ret) {
3492                 nfit_mce_unregister();
3493                 destroy_workqueue(nfit_wq);
3494         }
3495
3496         return ret;
3497
3498 }
3499
3500 static __exit void nfit_exit(void)
3501 {
3502         nfit_mce_unregister();
3503         acpi_bus_unregister_driver(&acpi_nfit_driver);
3504         destroy_workqueue(nfit_wq);
3505         WARN_ON(!list_empty(&acpi_descs));
3506 }
3507
3508 module_init(nfit_init);
3509 module_exit(nfit_exit);
3510 MODULE_LICENSE("GPL v2");
3511 MODULE_AUTHOR("Intel Corporation");