netfilter: nf_tables: move dumper state allocation into ->start
[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 <acpi/nfit.h>
27 #include "nfit.h"
28
29 /*
30  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
31  * irrelevant.
32  */
33 #include <linux/io-64-nonatomic-hi-lo.h>
34
35 static bool force_enable_dimms;
36 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
37 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
38
39 static bool disable_vendor_specific;
40 module_param(disable_vendor_specific, bool, S_IRUGO);
41 MODULE_PARM_DESC(disable_vendor_specific,
42                 "Limit commands to the publicly specified set");
43
44 static unsigned long override_dsm_mask;
45 module_param(override_dsm_mask, ulong, S_IRUGO);
46 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
47
48 static int default_dsm_family = -1;
49 module_param(default_dsm_family, int, S_IRUGO);
50 MODULE_PARM_DESC(default_dsm_family,
51                 "Try this DSM type first when identifying NVDIMM family");
52
53 static bool no_init_ars;
54 module_param(no_init_ars, bool, 0644);
55 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
56
57 LIST_HEAD(acpi_descs);
58 DEFINE_MUTEX(acpi_desc_lock);
59
60 static struct workqueue_struct *nfit_wq;
61
62 struct nfit_table_prev {
63         struct list_head spas;
64         struct list_head memdevs;
65         struct list_head dcrs;
66         struct list_head bdws;
67         struct list_head idts;
68         struct list_head flushes;
69 };
70
71 static guid_t nfit_uuid[NFIT_UUID_MAX];
72
73 const guid_t *to_nfit_uuid(enum nfit_uuids id)
74 {
75         return &nfit_uuid[id];
76 }
77 EXPORT_SYMBOL(to_nfit_uuid);
78
79 static struct acpi_nfit_desc *to_acpi_nfit_desc(
80                 struct nvdimm_bus_descriptor *nd_desc)
81 {
82         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
83 }
84
85 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
86 {
87         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
88
89         /*
90          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
91          * acpi_device.
92          */
93         if (!nd_desc->provider_name
94                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
95                 return NULL;
96
97         return to_acpi_device(acpi_desc->dev);
98 }
99
100 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
101 {
102         struct nd_cmd_clear_error *clear_err;
103         struct nd_cmd_ars_status *ars_status;
104         u16 flags;
105
106         switch (cmd) {
107         case ND_CMD_ARS_CAP:
108                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
109                         return -ENOTTY;
110
111                 /* Command failed */
112                 if (status & 0xffff)
113                         return -EIO;
114
115                 /* No supported scan types for this range */
116                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
117                 if ((status >> 16 & flags) == 0)
118                         return -ENOTTY;
119                 return 0;
120         case ND_CMD_ARS_START:
121                 /* ARS is in progress */
122                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
123                         return -EBUSY;
124
125                 /* Command failed */
126                 if (status & 0xffff)
127                         return -EIO;
128                 return 0;
129         case ND_CMD_ARS_STATUS:
130                 ars_status = buf;
131                 /* Command failed */
132                 if (status & 0xffff)
133                         return -EIO;
134                 /* Check extended status (Upper two bytes) */
135                 if (status == NFIT_ARS_STATUS_DONE)
136                         return 0;
137
138                 /* ARS is in progress */
139                 if (status == NFIT_ARS_STATUS_BUSY)
140                         return -EBUSY;
141
142                 /* No ARS performed for the current boot */
143                 if (status == NFIT_ARS_STATUS_NONE)
144                         return -EAGAIN;
145
146                 /*
147                  * ARS interrupted, either we overflowed or some other
148                  * agent wants the scan to stop.  If we didn't overflow
149                  * then just continue with the returned results.
150                  */
151                 if (status == NFIT_ARS_STATUS_INTR) {
152                         if (ars_status->out_length >= 40 && (ars_status->flags
153                                                 & NFIT_ARS_F_OVERFLOW))
154                                 return -ENOSPC;
155                         return 0;
156                 }
157
158                 /* Unknown status */
159                 if (status >> 16)
160                         return -EIO;
161                 return 0;
162         case ND_CMD_CLEAR_ERROR:
163                 clear_err = buf;
164                 if (status & 0xffff)
165                         return -EIO;
166                 if (!clear_err->cleared)
167                         return -EIO;
168                 if (clear_err->length > clear_err->cleared)
169                         return clear_err->cleared;
170                 return 0;
171         default:
172                 break;
173         }
174
175         /* all other non-zero status results in an error */
176         if (status)
177                 return -EIO;
178         return 0;
179 }
180
181 #define ACPI_LABELS_LOCKED 3
182
183 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
184                 u32 status)
185 {
186         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
187
188         switch (cmd) {
189         case ND_CMD_GET_CONFIG_SIZE:
190                 /*
191                  * In the _LSI, _LSR, _LSW case the locked status is
192                  * communicated via the read/write commands
193                  */
194                 if (nfit_mem->has_lsr)
195                         break;
196
197                 if (status >> 16 & ND_CONFIG_LOCKED)
198                         return -EACCES;
199                 break;
200         case ND_CMD_GET_CONFIG_DATA:
201                 if (nfit_mem->has_lsr && status == ACPI_LABELS_LOCKED)
202                         return -EACCES;
203                 break;
204         case ND_CMD_SET_CONFIG_DATA:
205                 if (nfit_mem->has_lsw && status == ACPI_LABELS_LOCKED)
206                         return -EACCES;
207                 break;
208         default:
209                 break;
210         }
211
212         /* all other non-zero status results in an error */
213         if (status)
214                 return -EIO;
215         return 0;
216 }
217
218 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
219                 u32 status)
220 {
221         if (!nvdimm)
222                 return xlat_bus_status(buf, cmd, status);
223         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
224 }
225
226 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
227 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
228 {
229         int i;
230         void *dst;
231         size_t size = 0;
232         union acpi_object *buf = NULL;
233
234         if (pkg->type != ACPI_TYPE_PACKAGE) {
235                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
236                                 pkg->type);
237                 goto err;
238         }
239
240         for (i = 0; i < pkg->package.count; i++) {
241                 union acpi_object *obj = &pkg->package.elements[i];
242
243                 if (obj->type == ACPI_TYPE_INTEGER)
244                         size += 4;
245                 else if (obj->type == ACPI_TYPE_BUFFER)
246                         size += obj->buffer.length;
247                 else {
248                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
249                                         obj->type);
250                         goto err;
251                 }
252         }
253
254         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
255         if (!buf)
256                 goto err;
257
258         dst = buf + 1;
259         buf->type = ACPI_TYPE_BUFFER;
260         buf->buffer.length = size;
261         buf->buffer.pointer = dst;
262         for (i = 0; i < pkg->package.count; i++) {
263                 union acpi_object *obj = &pkg->package.elements[i];
264
265                 if (obj->type == ACPI_TYPE_INTEGER) {
266                         memcpy(dst, &obj->integer.value, 4);
267                         dst += 4;
268                 } else if (obj->type == ACPI_TYPE_BUFFER) {
269                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
270                         dst += obj->buffer.length;
271                 }
272         }
273 err:
274         ACPI_FREE(pkg);
275         return buf;
276 }
277
278 static union acpi_object *int_to_buf(union acpi_object *integer)
279 {
280         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
281         void *dst = NULL;
282
283         if (!buf)
284                 goto err;
285
286         if (integer->type != ACPI_TYPE_INTEGER) {
287                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
288                                 integer->type);
289                 goto err;
290         }
291
292         dst = buf + 1;
293         buf->type = ACPI_TYPE_BUFFER;
294         buf->buffer.length = 4;
295         buf->buffer.pointer = dst;
296         memcpy(dst, &integer->integer.value, 4);
297 err:
298         ACPI_FREE(integer);
299         return buf;
300 }
301
302 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
303                 u32 len, void *data)
304 {
305         acpi_status rc;
306         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
307         struct acpi_object_list input = {
308                 .count = 3,
309                 .pointer = (union acpi_object []) {
310                         [0] = {
311                                 .integer.type = ACPI_TYPE_INTEGER,
312                                 .integer.value = offset,
313                         },
314                         [1] = {
315                                 .integer.type = ACPI_TYPE_INTEGER,
316                                 .integer.value = len,
317                         },
318                         [2] = {
319                                 .buffer.type = ACPI_TYPE_BUFFER,
320                                 .buffer.pointer = data,
321                                 .buffer.length = len,
322                         },
323                 },
324         };
325
326         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
327         if (ACPI_FAILURE(rc))
328                 return NULL;
329         return int_to_buf(buf.pointer);
330 }
331
332 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
333                 u32 len)
334 {
335         acpi_status rc;
336         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
337         struct acpi_object_list input = {
338                 .count = 2,
339                 .pointer = (union acpi_object []) {
340                         [0] = {
341                                 .integer.type = ACPI_TYPE_INTEGER,
342                                 .integer.value = offset,
343                         },
344                         [1] = {
345                                 .integer.type = ACPI_TYPE_INTEGER,
346                                 .integer.value = len,
347                         },
348                 },
349         };
350
351         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
352         if (ACPI_FAILURE(rc))
353                 return NULL;
354         return pkg_to_buf(buf.pointer);
355 }
356
357 static union acpi_object *acpi_label_info(acpi_handle handle)
358 {
359         acpi_status rc;
360         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
361
362         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
363         if (ACPI_FAILURE(rc))
364                 return NULL;
365         return pkg_to_buf(buf.pointer);
366 }
367
368 static u8 nfit_dsm_revid(unsigned family, unsigned func)
369 {
370         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
371                 [NVDIMM_FAMILY_INTEL] = {
372                         [NVDIMM_INTEL_GET_MODES] = 2,
373                         [NVDIMM_INTEL_GET_FWINFO] = 2,
374                         [NVDIMM_INTEL_START_FWUPDATE] = 2,
375                         [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
376                         [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
377                         [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
378                         [NVDIMM_INTEL_SET_THRESHOLD] = 2,
379                         [NVDIMM_INTEL_INJECT_ERROR] = 2,
380                 },
381         };
382         u8 id;
383
384         if (family > NVDIMM_FAMILY_MAX)
385                 return 0;
386         if (func > 31)
387                 return 0;
388         id = revid_table[family][func];
389         if (id == 0)
390                 return 1; /* default */
391         return id;
392 }
393
394 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
395                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
396 {
397         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
398         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
399         union acpi_object in_obj, in_buf, *out_obj;
400         const struct nd_cmd_desc *desc = NULL;
401         struct device *dev = acpi_desc->dev;
402         struct nd_cmd_pkg *call_pkg = NULL;
403         const char *cmd_name, *dimm_name;
404         unsigned long cmd_mask, dsm_mask;
405         u32 offset, fw_status = 0;
406         acpi_handle handle;
407         unsigned int func;
408         const guid_t *guid;
409         int rc, i;
410
411         func = cmd;
412         if (cmd == ND_CMD_CALL) {
413                 call_pkg = buf;
414                 func = call_pkg->nd_command;
415
416                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
417                         if (call_pkg->nd_reserved2[i])
418                                 return -EINVAL;
419         }
420
421         if (nvdimm) {
422                 struct acpi_device *adev = nfit_mem->adev;
423
424                 if (!adev)
425                         return -ENOTTY;
426                 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
427                         return -ENOTTY;
428
429                 dimm_name = nvdimm_name(nvdimm);
430                 cmd_name = nvdimm_cmd_name(cmd);
431                 cmd_mask = nvdimm_cmd_mask(nvdimm);
432                 dsm_mask = nfit_mem->dsm_mask;
433                 desc = nd_cmd_dimm_desc(cmd);
434                 guid = to_nfit_uuid(nfit_mem->family);
435                 handle = adev->handle;
436         } else {
437                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
438
439                 cmd_name = nvdimm_bus_cmd_name(cmd);
440                 cmd_mask = nd_desc->cmd_mask;
441                 dsm_mask = cmd_mask;
442                 if (cmd == ND_CMD_CALL)
443                         dsm_mask = nd_desc->bus_dsm_mask;
444                 desc = nd_cmd_bus_desc(cmd);
445                 guid = to_nfit_uuid(NFIT_DEV_BUS);
446                 handle = adev->handle;
447                 dimm_name = "bus";
448         }
449
450         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
451                 return -ENOTTY;
452
453         if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
454                 return -ENOTTY;
455
456         in_obj.type = ACPI_TYPE_PACKAGE;
457         in_obj.package.count = 1;
458         in_obj.package.elements = &in_buf;
459         in_buf.type = ACPI_TYPE_BUFFER;
460         in_buf.buffer.pointer = buf;
461         in_buf.buffer.length = 0;
462
463         /* libnvdimm has already validated the input envelope */
464         for (i = 0; i < desc->in_num; i++)
465                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
466                                 i, buf);
467
468         if (call_pkg) {
469                 /* skip over package wrapper */
470                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
471                 in_buf.buffer.length = call_pkg->nd_size_in;
472         }
473
474         dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
475                 dimm_name, cmd, func, in_buf.buffer.length);
476         print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
477                         in_buf.buffer.pointer,
478                         min_t(u32, 256, in_buf.buffer.length), true);
479
480         /* call the BIOS, prefer the named methods over _DSM if available */
481         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE && nfit_mem->has_lsr)
482                 out_obj = acpi_label_info(handle);
483         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && nfit_mem->has_lsr) {
484                 struct nd_cmd_get_config_data_hdr *p = buf;
485
486                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
487         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
488                         && nfit_mem->has_lsw) {
489                 struct nd_cmd_set_config_hdr *p = buf;
490
491                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
492                                 p->in_buf);
493         } else {
494                 u8 revid;
495
496                 if (nvdimm)
497                         revid = nfit_dsm_revid(nfit_mem->family, func);
498                 else
499                         revid = 1;
500                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
501         }
502
503         if (!out_obj) {
504                 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
505                 return -EINVAL;
506         }
507
508         if (call_pkg) {
509                 call_pkg->nd_fw_size = out_obj->buffer.length;
510                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
511                         out_obj->buffer.pointer,
512                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
513
514                 ACPI_FREE(out_obj);
515                 /*
516                  * Need to support FW function w/o known size in advance.
517                  * Caller can determine required size based upon nd_fw_size.
518                  * If we return an error (like elsewhere) then caller wouldn't
519                  * be able to rely upon data returned to make calculation.
520                  */
521                 return 0;
522         }
523
524         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
525                 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
526                                 dimm_name, cmd_name, out_obj->type);
527                 rc = -EINVAL;
528                 goto out;
529         }
530
531         dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
532                         cmd_name, out_obj->buffer.length);
533         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
534                         out_obj->buffer.pointer,
535                         min_t(u32, 128, out_obj->buffer.length), true);
536
537         for (i = 0, offset = 0; i < desc->out_num; i++) {
538                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
539                                 (u32 *) out_obj->buffer.pointer,
540                                 out_obj->buffer.length - offset);
541
542                 if (offset + out_size > out_obj->buffer.length) {
543                         dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
544                                         dimm_name, cmd_name, i);
545                         break;
546                 }
547
548                 if (in_buf.buffer.length + offset + out_size > buf_len) {
549                         dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
550                                         dimm_name, cmd_name, i);
551                         rc = -ENXIO;
552                         goto out;
553                 }
554                 memcpy(buf + in_buf.buffer.length + offset,
555                                 out_obj->buffer.pointer + offset, out_size);
556                 offset += out_size;
557         }
558
559         /*
560          * Set fw_status for all the commands with a known format to be
561          * later interpreted by xlat_status().
562          */
563         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
564                                         && cmd <= ND_CMD_CLEAR_ERROR)
565                                 || (nvdimm && cmd >= ND_CMD_SMART
566                                         && cmd <= ND_CMD_VENDOR)))
567                 fw_status = *(u32 *) out_obj->buffer.pointer;
568
569         if (offset + in_buf.buffer.length < buf_len) {
570                 if (i >= 1) {
571                         /*
572                          * status valid, return the number of bytes left
573                          * unfilled in the output buffer
574                          */
575                         rc = buf_len - offset - in_buf.buffer.length;
576                         if (cmd_rc)
577                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
578                                                 fw_status);
579                 } else {
580                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
581                                         __func__, dimm_name, cmd_name, buf_len,
582                                         offset);
583                         rc = -ENXIO;
584                 }
585         } else {
586                 rc = 0;
587                 if (cmd_rc)
588                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
589         }
590
591  out:
592         ACPI_FREE(out_obj);
593
594         return rc;
595 }
596 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
597
598 static const char *spa_type_name(u16 type)
599 {
600         static const char *to_name[] = {
601                 [NFIT_SPA_VOLATILE] = "volatile",
602                 [NFIT_SPA_PM] = "pmem",
603                 [NFIT_SPA_DCR] = "dimm-control-region",
604                 [NFIT_SPA_BDW] = "block-data-window",
605                 [NFIT_SPA_VDISK] = "volatile-disk",
606                 [NFIT_SPA_VCD] = "volatile-cd",
607                 [NFIT_SPA_PDISK] = "persistent-disk",
608                 [NFIT_SPA_PCD] = "persistent-cd",
609
610         };
611
612         if (type > NFIT_SPA_PCD)
613                 return "unknown";
614
615         return to_name[type];
616 }
617
618 int nfit_spa_type(struct acpi_nfit_system_address *spa)
619 {
620         int i;
621
622         for (i = 0; i < NFIT_UUID_MAX; i++)
623                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
624                         return i;
625         return -1;
626 }
627
628 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
629                 struct nfit_table_prev *prev,
630                 struct acpi_nfit_system_address *spa)
631 {
632         struct device *dev = acpi_desc->dev;
633         struct nfit_spa *nfit_spa;
634
635         if (spa->header.length != sizeof(*spa))
636                 return false;
637
638         list_for_each_entry(nfit_spa, &prev->spas, list) {
639                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
640                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
641                         return true;
642                 }
643         }
644
645         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
646                         GFP_KERNEL);
647         if (!nfit_spa)
648                 return false;
649         INIT_LIST_HEAD(&nfit_spa->list);
650         memcpy(nfit_spa->spa, spa, sizeof(*spa));
651         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
652         dev_dbg(dev, "spa index: %d type: %s\n",
653                         spa->range_index,
654                         spa_type_name(nfit_spa_type(spa)));
655         return true;
656 }
657
658 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
659                 struct nfit_table_prev *prev,
660                 struct acpi_nfit_memory_map *memdev)
661 {
662         struct device *dev = acpi_desc->dev;
663         struct nfit_memdev *nfit_memdev;
664
665         if (memdev->header.length != sizeof(*memdev))
666                 return false;
667
668         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
669                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
670                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
671                         return true;
672                 }
673
674         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
675                         GFP_KERNEL);
676         if (!nfit_memdev)
677                 return false;
678         INIT_LIST_HEAD(&nfit_memdev->list);
679         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
680         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
681         dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
682                         memdev->device_handle, memdev->range_index,
683                         memdev->region_index, memdev->flags);
684         return true;
685 }
686
687 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
688 {
689         struct acpi_nfit_memory_map *memdev;
690         struct acpi_nfit_desc *acpi_desc;
691         struct nfit_mem *nfit_mem;
692
693         mutex_lock(&acpi_desc_lock);
694         list_for_each_entry(acpi_desc, &acpi_descs, list) {
695                 mutex_lock(&acpi_desc->init_mutex);
696                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
697                         memdev = __to_nfit_memdev(nfit_mem);
698                         if (memdev->device_handle == device_handle) {
699                                 mutex_unlock(&acpi_desc->init_mutex);
700                                 mutex_unlock(&acpi_desc_lock);
701                                 *flags = memdev->flags;
702                                 return memdev->physical_id;
703                         }
704                 }
705                 mutex_unlock(&acpi_desc->init_mutex);
706         }
707         mutex_unlock(&acpi_desc_lock);
708
709         return -ENODEV;
710 }
711 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
712
713 /*
714  * An implementation may provide a truncated control region if no block windows
715  * are defined.
716  */
717 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
718 {
719         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
720                                 window_size))
721                 return 0;
722         if (dcr->windows)
723                 return sizeof(*dcr);
724         return offsetof(struct acpi_nfit_control_region, window_size);
725 }
726
727 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
728                 struct nfit_table_prev *prev,
729                 struct acpi_nfit_control_region *dcr)
730 {
731         struct device *dev = acpi_desc->dev;
732         struct nfit_dcr *nfit_dcr;
733
734         if (!sizeof_dcr(dcr))
735                 return false;
736
737         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
738                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
739                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
740                         return true;
741                 }
742
743         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
744                         GFP_KERNEL);
745         if (!nfit_dcr)
746                 return false;
747         INIT_LIST_HEAD(&nfit_dcr->list);
748         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
749         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
750         dev_dbg(dev, "dcr index: %d windows: %d\n",
751                         dcr->region_index, dcr->windows);
752         return true;
753 }
754
755 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
756                 struct nfit_table_prev *prev,
757                 struct acpi_nfit_data_region *bdw)
758 {
759         struct device *dev = acpi_desc->dev;
760         struct nfit_bdw *nfit_bdw;
761
762         if (bdw->header.length != sizeof(*bdw))
763                 return false;
764         list_for_each_entry(nfit_bdw, &prev->bdws, list)
765                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
766                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
767                         return true;
768                 }
769
770         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
771                         GFP_KERNEL);
772         if (!nfit_bdw)
773                 return false;
774         INIT_LIST_HEAD(&nfit_bdw->list);
775         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
776         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
777         dev_dbg(dev, "bdw dcr: %d windows: %d\n",
778                         bdw->region_index, bdw->windows);
779         return true;
780 }
781
782 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
783 {
784         if (idt->header.length < sizeof(*idt))
785                 return 0;
786         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
787 }
788
789 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
790                 struct nfit_table_prev *prev,
791                 struct acpi_nfit_interleave *idt)
792 {
793         struct device *dev = acpi_desc->dev;
794         struct nfit_idt *nfit_idt;
795
796         if (!sizeof_idt(idt))
797                 return false;
798
799         list_for_each_entry(nfit_idt, &prev->idts, list) {
800                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
801                         continue;
802
803                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
804                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
805                         return true;
806                 }
807         }
808
809         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
810                         GFP_KERNEL);
811         if (!nfit_idt)
812                 return false;
813         INIT_LIST_HEAD(&nfit_idt->list);
814         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
815         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
816         dev_dbg(dev, "idt index: %d num_lines: %d\n",
817                         idt->interleave_index, idt->line_count);
818         return true;
819 }
820
821 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
822 {
823         if (flush->header.length < sizeof(*flush))
824                 return 0;
825         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
826 }
827
828 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
829                 struct nfit_table_prev *prev,
830                 struct acpi_nfit_flush_address *flush)
831 {
832         struct device *dev = acpi_desc->dev;
833         struct nfit_flush *nfit_flush;
834
835         if (!sizeof_flush(flush))
836                 return false;
837
838         list_for_each_entry(nfit_flush, &prev->flushes, list) {
839                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
840                         continue;
841
842                 if (memcmp(nfit_flush->flush, flush,
843                                         sizeof_flush(flush)) == 0) {
844                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
845                         return true;
846                 }
847         }
848
849         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
850                         + sizeof_flush(flush), GFP_KERNEL);
851         if (!nfit_flush)
852                 return false;
853         INIT_LIST_HEAD(&nfit_flush->list);
854         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
855         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
856         dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
857                         flush->device_handle, flush->hint_count);
858         return true;
859 }
860
861 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
862                 struct acpi_nfit_capabilities *pcap)
863 {
864         struct device *dev = acpi_desc->dev;
865         u32 mask;
866
867         mask = (1 << (pcap->highest_capability + 1)) - 1;
868         acpi_desc->platform_cap = pcap->capabilities & mask;
869         dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
870         return true;
871 }
872
873 static void *add_table(struct acpi_nfit_desc *acpi_desc,
874                 struct nfit_table_prev *prev, void *table, const void *end)
875 {
876         struct device *dev = acpi_desc->dev;
877         struct acpi_nfit_header *hdr;
878         void *err = ERR_PTR(-ENOMEM);
879
880         if (table >= end)
881                 return NULL;
882
883         hdr = table;
884         if (!hdr->length) {
885                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
886                         hdr->type);
887                 return NULL;
888         }
889
890         switch (hdr->type) {
891         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
892                 if (!add_spa(acpi_desc, prev, table))
893                         return err;
894                 break;
895         case ACPI_NFIT_TYPE_MEMORY_MAP:
896                 if (!add_memdev(acpi_desc, prev, table))
897                         return err;
898                 break;
899         case ACPI_NFIT_TYPE_CONTROL_REGION:
900                 if (!add_dcr(acpi_desc, prev, table))
901                         return err;
902                 break;
903         case ACPI_NFIT_TYPE_DATA_REGION:
904                 if (!add_bdw(acpi_desc, prev, table))
905                         return err;
906                 break;
907         case ACPI_NFIT_TYPE_INTERLEAVE:
908                 if (!add_idt(acpi_desc, prev, table))
909                         return err;
910                 break;
911         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
912                 if (!add_flush(acpi_desc, prev, table))
913                         return err;
914                 break;
915         case ACPI_NFIT_TYPE_SMBIOS:
916                 dev_dbg(dev, "smbios\n");
917                 break;
918         case ACPI_NFIT_TYPE_CAPABILITIES:
919                 if (!add_platform_cap(acpi_desc, table))
920                         return err;
921                 break;
922         default:
923                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
924                 break;
925         }
926
927         return table + hdr->length;
928 }
929
930 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
931                 struct nfit_mem *nfit_mem)
932 {
933         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
934         u16 dcr = nfit_mem->dcr->region_index;
935         struct nfit_spa *nfit_spa;
936
937         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
938                 u16 range_index = nfit_spa->spa->range_index;
939                 int type = nfit_spa_type(nfit_spa->spa);
940                 struct nfit_memdev *nfit_memdev;
941
942                 if (type != NFIT_SPA_BDW)
943                         continue;
944
945                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
946                         if (nfit_memdev->memdev->range_index != range_index)
947                                 continue;
948                         if (nfit_memdev->memdev->device_handle != device_handle)
949                                 continue;
950                         if (nfit_memdev->memdev->region_index != dcr)
951                                 continue;
952
953                         nfit_mem->spa_bdw = nfit_spa->spa;
954                         return;
955                 }
956         }
957
958         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
959                         nfit_mem->spa_dcr->range_index);
960         nfit_mem->bdw = NULL;
961 }
962
963 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
964                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
965 {
966         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
967         struct nfit_memdev *nfit_memdev;
968         struct nfit_bdw *nfit_bdw;
969         struct nfit_idt *nfit_idt;
970         u16 idt_idx, range_index;
971
972         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
973                 if (nfit_bdw->bdw->region_index != dcr)
974                         continue;
975                 nfit_mem->bdw = nfit_bdw->bdw;
976                 break;
977         }
978
979         if (!nfit_mem->bdw)
980                 return;
981
982         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
983
984         if (!nfit_mem->spa_bdw)
985                 return;
986
987         range_index = nfit_mem->spa_bdw->range_index;
988         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
989                 if (nfit_memdev->memdev->range_index != range_index ||
990                                 nfit_memdev->memdev->region_index != dcr)
991                         continue;
992                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
993                 idt_idx = nfit_memdev->memdev->interleave_index;
994                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
995                         if (nfit_idt->idt->interleave_index != idt_idx)
996                                 continue;
997                         nfit_mem->idt_bdw = nfit_idt->idt;
998                         break;
999                 }
1000                 break;
1001         }
1002 }
1003
1004 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1005                 struct acpi_nfit_system_address *spa)
1006 {
1007         struct nfit_mem *nfit_mem, *found;
1008         struct nfit_memdev *nfit_memdev;
1009         int type = spa ? nfit_spa_type(spa) : 0;
1010
1011         switch (type) {
1012         case NFIT_SPA_DCR:
1013         case NFIT_SPA_PM:
1014                 break;
1015         default:
1016                 if (spa)
1017                         return 0;
1018         }
1019
1020         /*
1021          * This loop runs in two modes, when a dimm is mapped the loop
1022          * adds memdev associations to an existing dimm, or creates a
1023          * dimm. In the unmapped dimm case this loop sweeps for memdev
1024          * instances with an invalid / zero range_index and adds those
1025          * dimms without spa associations.
1026          */
1027         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1028                 struct nfit_flush *nfit_flush;
1029                 struct nfit_dcr *nfit_dcr;
1030                 u32 device_handle;
1031                 u16 dcr;
1032
1033                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1034                         continue;
1035                 if (!spa && nfit_memdev->memdev->range_index)
1036                         continue;
1037                 found = NULL;
1038                 dcr = nfit_memdev->memdev->region_index;
1039                 device_handle = nfit_memdev->memdev->device_handle;
1040                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1041                         if (__to_nfit_memdev(nfit_mem)->device_handle
1042                                         == device_handle) {
1043                                 found = nfit_mem;
1044                                 break;
1045                         }
1046
1047                 if (found)
1048                         nfit_mem = found;
1049                 else {
1050                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1051                                         sizeof(*nfit_mem), GFP_KERNEL);
1052                         if (!nfit_mem)
1053                                 return -ENOMEM;
1054                         INIT_LIST_HEAD(&nfit_mem->list);
1055                         nfit_mem->acpi_desc = acpi_desc;
1056                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1057                 }
1058
1059                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1060                         if (nfit_dcr->dcr->region_index != dcr)
1061                                 continue;
1062                         /*
1063                          * Record the control region for the dimm.  For
1064                          * the ACPI 6.1 case, where there are separate
1065                          * control regions for the pmem vs blk
1066                          * interfaces, be sure to record the extended
1067                          * blk details.
1068                          */
1069                         if (!nfit_mem->dcr)
1070                                 nfit_mem->dcr = nfit_dcr->dcr;
1071                         else if (nfit_mem->dcr->windows == 0
1072                                         && nfit_dcr->dcr->windows)
1073                                 nfit_mem->dcr = nfit_dcr->dcr;
1074                         break;
1075                 }
1076
1077                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1078                         struct acpi_nfit_flush_address *flush;
1079                         u16 i;
1080
1081                         if (nfit_flush->flush->device_handle != device_handle)
1082                                 continue;
1083                         nfit_mem->nfit_flush = nfit_flush;
1084                         flush = nfit_flush->flush;
1085                         nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1086                                         flush->hint_count,
1087                                         sizeof(struct resource),
1088                                         GFP_KERNEL);
1089                         if (!nfit_mem->flush_wpq)
1090                                 return -ENOMEM;
1091                         for (i = 0; i < flush->hint_count; i++) {
1092                                 struct resource *res = &nfit_mem->flush_wpq[i];
1093
1094                                 res->start = flush->hint_address[i];
1095                                 res->end = res->start + 8 - 1;
1096                         }
1097                         break;
1098                 }
1099
1100                 if (dcr && !nfit_mem->dcr) {
1101                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1102                                         spa->range_index, dcr);
1103                         return -ENODEV;
1104                 }
1105
1106                 if (type == NFIT_SPA_DCR) {
1107                         struct nfit_idt *nfit_idt;
1108                         u16 idt_idx;
1109
1110                         /* multiple dimms may share a SPA when interleaved */
1111                         nfit_mem->spa_dcr = spa;
1112                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1113                         idt_idx = nfit_memdev->memdev->interleave_index;
1114                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1115                                 if (nfit_idt->idt->interleave_index != idt_idx)
1116                                         continue;
1117                                 nfit_mem->idt_dcr = nfit_idt->idt;
1118                                 break;
1119                         }
1120                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1121                 } else if (type == NFIT_SPA_PM) {
1122                         /*
1123                          * A single dimm may belong to multiple SPA-PM
1124                          * ranges, record at least one in addition to
1125                          * any SPA-DCR range.
1126                          */
1127                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1128                 } else
1129                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1130         }
1131
1132         return 0;
1133 }
1134
1135 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1136 {
1137         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1138         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1139         u32 handleA, handleB;
1140
1141         handleA = __to_nfit_memdev(a)->device_handle;
1142         handleB = __to_nfit_memdev(b)->device_handle;
1143         if (handleA < handleB)
1144                 return -1;
1145         else if (handleA > handleB)
1146                 return 1;
1147         return 0;
1148 }
1149
1150 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1151 {
1152         struct nfit_spa *nfit_spa;
1153         int rc;
1154
1155
1156         /*
1157          * For each SPA-DCR or SPA-PMEM address range find its
1158          * corresponding MEMDEV(s).  From each MEMDEV find the
1159          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1160          * try to find a SPA-BDW and a corresponding BDW that references
1161          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1162          * BDWs are optional.
1163          */
1164         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1165                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1166                 if (rc)
1167                         return rc;
1168         }
1169
1170         /*
1171          * If a DIMM has failed to be mapped into SPA there will be no
1172          * SPA entries above. Find and register all the unmapped DIMMs
1173          * for reporting and recovery purposes.
1174          */
1175         rc = __nfit_mem_init(acpi_desc, NULL);
1176         if (rc)
1177                 return rc;
1178
1179         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1180
1181         return 0;
1182 }
1183
1184 static ssize_t bus_dsm_mask_show(struct device *dev,
1185                 struct device_attribute *attr, char *buf)
1186 {
1187         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1188         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1189
1190         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1191 }
1192 static struct device_attribute dev_attr_bus_dsm_mask =
1193                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1194
1195 static ssize_t revision_show(struct device *dev,
1196                 struct device_attribute *attr, char *buf)
1197 {
1198         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1199         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1200         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1201
1202         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1203 }
1204 static DEVICE_ATTR_RO(revision);
1205
1206 static ssize_t hw_error_scrub_show(struct device *dev,
1207                 struct device_attribute *attr, char *buf)
1208 {
1209         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1210         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1211         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1212
1213         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1214 }
1215
1216 /*
1217  * The 'hw_error_scrub' attribute can have the following values written to it:
1218  * '0': Switch to the default mode where an exception will only insert
1219  *      the address of the memory error into the poison and badblocks lists.
1220  * '1': Enable a full scrub to happen if an exception for a memory error is
1221  *      received.
1222  */
1223 static ssize_t hw_error_scrub_store(struct device *dev,
1224                 struct device_attribute *attr, const char *buf, size_t size)
1225 {
1226         struct nvdimm_bus_descriptor *nd_desc;
1227         ssize_t rc;
1228         long val;
1229
1230         rc = kstrtol(buf, 0, &val);
1231         if (rc)
1232                 return rc;
1233
1234         device_lock(dev);
1235         nd_desc = dev_get_drvdata(dev);
1236         if (nd_desc) {
1237                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1238
1239                 switch (val) {
1240                 case HW_ERROR_SCRUB_ON:
1241                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1242                         break;
1243                 case HW_ERROR_SCRUB_OFF:
1244                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1245                         break;
1246                 default:
1247                         rc = -EINVAL;
1248                         break;
1249                 }
1250         }
1251         device_unlock(dev);
1252         if (rc)
1253                 return rc;
1254         return size;
1255 }
1256 static DEVICE_ATTR_RW(hw_error_scrub);
1257
1258 /*
1259  * This shows the number of full Address Range Scrubs that have been
1260  * completed since driver load time. Userspace can wait on this using
1261  * select/poll etc. A '+' at the end indicates an ARS is in progress
1262  */
1263 static ssize_t scrub_show(struct device *dev,
1264                 struct device_attribute *attr, char *buf)
1265 {
1266         struct nvdimm_bus_descriptor *nd_desc;
1267         ssize_t rc = -ENXIO;
1268
1269         device_lock(dev);
1270         nd_desc = dev_get_drvdata(dev);
1271         if (nd_desc) {
1272                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1273
1274                 mutex_lock(&acpi_desc->init_mutex);
1275                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1276                                 work_busy(&acpi_desc->dwork.work)
1277                                 && !acpi_desc->cancel ? "+\n" : "\n");
1278                 mutex_unlock(&acpi_desc->init_mutex);
1279         }
1280         device_unlock(dev);
1281         return rc;
1282 }
1283
1284 static ssize_t scrub_store(struct device *dev,
1285                 struct device_attribute *attr, const char *buf, size_t size)
1286 {
1287         struct nvdimm_bus_descriptor *nd_desc;
1288         ssize_t rc;
1289         long val;
1290
1291         rc = kstrtol(buf, 0, &val);
1292         if (rc)
1293                 return rc;
1294         if (val != 1)
1295                 return -EINVAL;
1296
1297         device_lock(dev);
1298         nd_desc = dev_get_drvdata(dev);
1299         if (nd_desc) {
1300                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1301
1302                 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1303         }
1304         device_unlock(dev);
1305         if (rc)
1306                 return rc;
1307         return size;
1308 }
1309 static DEVICE_ATTR_RW(scrub);
1310
1311 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1312 {
1313         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1314         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1315                 | 1 << ND_CMD_ARS_STATUS;
1316
1317         return (nd_desc->cmd_mask & mask) == mask;
1318 }
1319
1320 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1321 {
1322         struct device *dev = container_of(kobj, struct device, kobj);
1323         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1324
1325         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1326                 return 0;
1327         return a->mode;
1328 }
1329
1330 static struct attribute *acpi_nfit_attributes[] = {
1331         &dev_attr_revision.attr,
1332         &dev_attr_scrub.attr,
1333         &dev_attr_hw_error_scrub.attr,
1334         &dev_attr_bus_dsm_mask.attr,
1335         NULL,
1336 };
1337
1338 static const struct attribute_group acpi_nfit_attribute_group = {
1339         .name = "nfit",
1340         .attrs = acpi_nfit_attributes,
1341         .is_visible = nfit_visible,
1342 };
1343
1344 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1345         &nvdimm_bus_attribute_group,
1346         &acpi_nfit_attribute_group,
1347         NULL,
1348 };
1349
1350 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1351 {
1352         struct nvdimm *nvdimm = to_nvdimm(dev);
1353         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1354
1355         return __to_nfit_memdev(nfit_mem);
1356 }
1357
1358 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1359 {
1360         struct nvdimm *nvdimm = to_nvdimm(dev);
1361         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1362
1363         return nfit_mem->dcr;
1364 }
1365
1366 static ssize_t handle_show(struct device *dev,
1367                 struct device_attribute *attr, char *buf)
1368 {
1369         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1370
1371         return sprintf(buf, "%#x\n", memdev->device_handle);
1372 }
1373 static DEVICE_ATTR_RO(handle);
1374
1375 static ssize_t phys_id_show(struct device *dev,
1376                 struct device_attribute *attr, char *buf)
1377 {
1378         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1379
1380         return sprintf(buf, "%#x\n", memdev->physical_id);
1381 }
1382 static DEVICE_ATTR_RO(phys_id);
1383
1384 static ssize_t vendor_show(struct device *dev,
1385                 struct device_attribute *attr, char *buf)
1386 {
1387         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1388
1389         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1390 }
1391 static DEVICE_ATTR_RO(vendor);
1392
1393 static ssize_t rev_id_show(struct device *dev,
1394                 struct device_attribute *attr, char *buf)
1395 {
1396         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1397
1398         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1399 }
1400 static DEVICE_ATTR_RO(rev_id);
1401
1402 static ssize_t device_show(struct device *dev,
1403                 struct device_attribute *attr, char *buf)
1404 {
1405         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1406
1407         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1408 }
1409 static DEVICE_ATTR_RO(device);
1410
1411 static ssize_t subsystem_vendor_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", be16_to_cpu(dcr->subsystem_vendor_id));
1417 }
1418 static DEVICE_ATTR_RO(subsystem_vendor);
1419
1420 static ssize_t subsystem_rev_id_show(struct device *dev,
1421                 struct device_attribute *attr, char *buf)
1422 {
1423         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1424
1425         return sprintf(buf, "0x%04x\n",
1426                         be16_to_cpu(dcr->subsystem_revision_id));
1427 }
1428 static DEVICE_ATTR_RO(subsystem_rev_id);
1429
1430 static ssize_t subsystem_device_show(struct device *dev,
1431                 struct device_attribute *attr, char *buf)
1432 {
1433         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1434
1435         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1436 }
1437 static DEVICE_ATTR_RO(subsystem_device);
1438
1439 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1440 {
1441         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1442         int formats = 0;
1443
1444         if (nfit_mem->memdev_pmem)
1445                 formats++;
1446         if (nfit_mem->memdev_bdw)
1447                 formats++;
1448         return formats;
1449 }
1450
1451 static ssize_t format_show(struct device *dev,
1452                 struct device_attribute *attr, char *buf)
1453 {
1454         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1455
1456         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1457 }
1458 static DEVICE_ATTR_RO(format);
1459
1460 static ssize_t format1_show(struct device *dev,
1461                 struct device_attribute *attr, char *buf)
1462 {
1463         u32 handle;
1464         ssize_t rc = -ENXIO;
1465         struct nfit_mem *nfit_mem;
1466         struct nfit_memdev *nfit_memdev;
1467         struct acpi_nfit_desc *acpi_desc;
1468         struct nvdimm *nvdimm = to_nvdimm(dev);
1469         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1470
1471         nfit_mem = nvdimm_provider_data(nvdimm);
1472         acpi_desc = nfit_mem->acpi_desc;
1473         handle = to_nfit_memdev(dev)->device_handle;
1474
1475         /* assumes DIMMs have at most 2 published interface codes */
1476         mutex_lock(&acpi_desc->init_mutex);
1477         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1478                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1479                 struct nfit_dcr *nfit_dcr;
1480
1481                 if (memdev->device_handle != handle)
1482                         continue;
1483
1484                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1485                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1486                                 continue;
1487                         if (nfit_dcr->dcr->code == dcr->code)
1488                                 continue;
1489                         rc = sprintf(buf, "0x%04x\n",
1490                                         le16_to_cpu(nfit_dcr->dcr->code));
1491                         break;
1492                 }
1493                 if (rc != ENXIO)
1494                         break;
1495         }
1496         mutex_unlock(&acpi_desc->init_mutex);
1497         return rc;
1498 }
1499 static DEVICE_ATTR_RO(format1);
1500
1501 static ssize_t formats_show(struct device *dev,
1502                 struct device_attribute *attr, char *buf)
1503 {
1504         struct nvdimm *nvdimm = to_nvdimm(dev);
1505
1506         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1507 }
1508 static DEVICE_ATTR_RO(formats);
1509
1510 static ssize_t serial_show(struct device *dev,
1511                 struct device_attribute *attr, char *buf)
1512 {
1513         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1514
1515         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1516 }
1517 static DEVICE_ATTR_RO(serial);
1518
1519 static ssize_t family_show(struct device *dev,
1520                 struct device_attribute *attr, char *buf)
1521 {
1522         struct nvdimm *nvdimm = to_nvdimm(dev);
1523         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1524
1525         if (nfit_mem->family < 0)
1526                 return -ENXIO;
1527         return sprintf(buf, "%d\n", nfit_mem->family);
1528 }
1529 static DEVICE_ATTR_RO(family);
1530
1531 static ssize_t dsm_mask_show(struct device *dev,
1532                 struct device_attribute *attr, char *buf)
1533 {
1534         struct nvdimm *nvdimm = to_nvdimm(dev);
1535         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1536
1537         if (nfit_mem->family < 0)
1538                 return -ENXIO;
1539         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1540 }
1541 static DEVICE_ATTR_RO(dsm_mask);
1542
1543 static ssize_t flags_show(struct device *dev,
1544                 struct device_attribute *attr, char *buf)
1545 {
1546         u16 flags = to_nfit_memdev(dev)->flags;
1547
1548         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1549                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1550                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1551                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1552                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1553                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1554                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1555                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1556 }
1557 static DEVICE_ATTR_RO(flags);
1558
1559 static ssize_t id_show(struct device *dev,
1560                 struct device_attribute *attr, char *buf)
1561 {
1562         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1563
1564         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1565                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1566                                 be16_to_cpu(dcr->vendor_id),
1567                                 dcr->manufacturing_location,
1568                                 be16_to_cpu(dcr->manufacturing_date),
1569                                 be32_to_cpu(dcr->serial_number));
1570         else
1571                 return sprintf(buf, "%04x-%08x\n",
1572                                 be16_to_cpu(dcr->vendor_id),
1573                                 be32_to_cpu(dcr->serial_number));
1574 }
1575 static DEVICE_ATTR_RO(id);
1576
1577 static struct attribute *acpi_nfit_dimm_attributes[] = {
1578         &dev_attr_handle.attr,
1579         &dev_attr_phys_id.attr,
1580         &dev_attr_vendor.attr,
1581         &dev_attr_device.attr,
1582         &dev_attr_rev_id.attr,
1583         &dev_attr_subsystem_vendor.attr,
1584         &dev_attr_subsystem_device.attr,
1585         &dev_attr_subsystem_rev_id.attr,
1586         &dev_attr_format.attr,
1587         &dev_attr_formats.attr,
1588         &dev_attr_format1.attr,
1589         &dev_attr_serial.attr,
1590         &dev_attr_flags.attr,
1591         &dev_attr_id.attr,
1592         &dev_attr_family.attr,
1593         &dev_attr_dsm_mask.attr,
1594         NULL,
1595 };
1596
1597 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1598                 struct attribute *a, int n)
1599 {
1600         struct device *dev = container_of(kobj, struct device, kobj);
1601         struct nvdimm *nvdimm = to_nvdimm(dev);
1602
1603         if (!to_nfit_dcr(dev)) {
1604                 /* Without a dcr only the memdev attributes can be surfaced */
1605                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1606                                 || a == &dev_attr_flags.attr
1607                                 || a == &dev_attr_family.attr
1608                                 || a == &dev_attr_dsm_mask.attr)
1609                         return a->mode;
1610                 return 0;
1611         }
1612
1613         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1614                 return 0;
1615         return a->mode;
1616 }
1617
1618 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1619         .name = "nfit",
1620         .attrs = acpi_nfit_dimm_attributes,
1621         .is_visible = acpi_nfit_dimm_attr_visible,
1622 };
1623
1624 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1625         &nvdimm_attribute_group,
1626         &nd_device_attribute_group,
1627         &acpi_nfit_dimm_attribute_group,
1628         NULL,
1629 };
1630
1631 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1632                 u32 device_handle)
1633 {
1634         struct nfit_mem *nfit_mem;
1635
1636         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1637                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1638                         return nfit_mem->nvdimm;
1639
1640         return NULL;
1641 }
1642
1643 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1644 {
1645         struct nfit_mem *nfit_mem;
1646         struct acpi_nfit_desc *acpi_desc;
1647
1648         dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1649                         event);
1650
1651         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1652                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1653                                 event);
1654                 return;
1655         }
1656
1657         acpi_desc = dev_get_drvdata(dev->parent);
1658         if (!acpi_desc)
1659                 return;
1660
1661         /*
1662          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1663          * is still valid.
1664          */
1665         nfit_mem = dev_get_drvdata(dev);
1666         if (nfit_mem && nfit_mem->flags_attr)
1667                 sysfs_notify_dirent(nfit_mem->flags_attr);
1668 }
1669 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1670
1671 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1672 {
1673         struct acpi_device *adev = data;
1674         struct device *dev = &adev->dev;
1675
1676         device_lock(dev->parent);
1677         __acpi_nvdimm_notify(dev, event);
1678         device_unlock(dev->parent);
1679 }
1680
1681 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1682 {
1683         acpi_handle handle;
1684         acpi_status status;
1685
1686         status = acpi_get_handle(adev->handle, method, &handle);
1687
1688         if (ACPI_SUCCESS(status))
1689                 return true;
1690         return false;
1691 }
1692
1693 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1694                 struct nfit_mem *nfit_mem, u32 device_handle)
1695 {
1696         struct acpi_device *adev, *adev_dimm;
1697         struct device *dev = acpi_desc->dev;
1698         unsigned long dsm_mask;
1699         const guid_t *guid;
1700         int i;
1701         int family = -1;
1702
1703         /* nfit test assumes 1:1 relationship between commands and dsms */
1704         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1705         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1706         adev = to_acpi_dev(acpi_desc);
1707         if (!adev)
1708                 return 0;
1709
1710         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1711         nfit_mem->adev = adev_dimm;
1712         if (!adev_dimm) {
1713                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1714                                 device_handle);
1715                 return force_enable_dimms ? 0 : -ENODEV;
1716         }
1717
1718         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1719                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1720                 dev_err(dev, "%s: notification registration failed\n",
1721                                 dev_name(&adev_dimm->dev));
1722                 return -ENXIO;
1723         }
1724         /*
1725          * Record nfit_mem for the notification path to track back to
1726          * the nfit sysfs attributes for this dimm device object.
1727          */
1728         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1729
1730         /*
1731          * Until standardization materializes we need to consider 4
1732          * different command sets.  Note, that checking for function0 (bit0)
1733          * tells us if any commands are reachable through this GUID.
1734          */
1735         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1736                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1737                         if (family < 0 || i == default_dsm_family)
1738                                 family = i;
1739
1740         /* limit the supported commands to those that are publicly documented */
1741         nfit_mem->family = family;
1742         if (override_dsm_mask && !disable_vendor_specific)
1743                 dsm_mask = override_dsm_mask;
1744         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1745                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1746                 if (disable_vendor_specific)
1747                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1748         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1749                 dsm_mask = 0x1c3c76;
1750         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1751                 dsm_mask = 0x1fe;
1752                 if (disable_vendor_specific)
1753                         dsm_mask &= ~(1 << 8);
1754         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1755                 dsm_mask = 0xffffffff;
1756         } else {
1757                 dev_dbg(dev, "unknown dimm command family\n");
1758                 nfit_mem->family = -1;
1759                 /* DSMs are optional, continue loading the driver... */
1760                 return 0;
1761         }
1762
1763         guid = to_nfit_uuid(nfit_mem->family);
1764         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1765                 if (acpi_check_dsm(adev_dimm->handle, guid,
1766                                         nfit_dsm_revid(nfit_mem->family, i),
1767                                         1ULL << i))
1768                         set_bit(i, &nfit_mem->dsm_mask);
1769
1770         if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1771                         && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1772                 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1773                 nfit_mem->has_lsr = true;
1774         }
1775
1776         if (nfit_mem->has_lsr && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1777                 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1778                 nfit_mem->has_lsw = true;
1779         }
1780
1781         return 0;
1782 }
1783
1784 static void shutdown_dimm_notify(void *data)
1785 {
1786         struct acpi_nfit_desc *acpi_desc = data;
1787         struct nfit_mem *nfit_mem;
1788
1789         mutex_lock(&acpi_desc->init_mutex);
1790         /*
1791          * Clear out the nfit_mem->flags_attr and shut down dimm event
1792          * notifications.
1793          */
1794         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1795                 struct acpi_device *adev_dimm = nfit_mem->adev;
1796
1797                 if (nfit_mem->flags_attr) {
1798                         sysfs_put(nfit_mem->flags_attr);
1799                         nfit_mem->flags_attr = NULL;
1800                 }
1801                 if (adev_dimm) {
1802                         acpi_remove_notify_handler(adev_dimm->handle,
1803                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1804                         dev_set_drvdata(&adev_dimm->dev, NULL);
1805                 }
1806         }
1807         mutex_unlock(&acpi_desc->init_mutex);
1808 }
1809
1810 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1811 {
1812         struct nfit_mem *nfit_mem;
1813         int dimm_count = 0, rc;
1814         struct nvdimm *nvdimm;
1815
1816         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1817                 struct acpi_nfit_flush_address *flush;
1818                 unsigned long flags = 0, cmd_mask;
1819                 struct nfit_memdev *nfit_memdev;
1820                 u32 device_handle;
1821                 u16 mem_flags;
1822
1823                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1824                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1825                 if (nvdimm) {
1826                         dimm_count++;
1827                         continue;
1828                 }
1829
1830                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1831                         set_bit(NDD_ALIASING, &flags);
1832
1833                 /* collate flags across all memdevs for this dimm */
1834                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1835                         struct acpi_nfit_memory_map *dimm_memdev;
1836
1837                         dimm_memdev = __to_nfit_memdev(nfit_mem);
1838                         if (dimm_memdev->device_handle
1839                                         != nfit_memdev->memdev->device_handle)
1840                                 continue;
1841                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
1842                 }
1843
1844                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1845                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1846                         set_bit(NDD_UNARMED, &flags);
1847
1848                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1849                 if (rc)
1850                         continue;
1851
1852                 /*
1853                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1854                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1855                  * userspace interface.
1856                  */
1857                 cmd_mask = 1UL << ND_CMD_CALL;
1858                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1859                         /*
1860                          * These commands have a 1:1 correspondence
1861                          * between DSM payload and libnvdimm ioctl
1862                          * payload format.
1863                          */
1864                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1865                 }
1866
1867                 if (nfit_mem->has_lsr) {
1868                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1869                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1870                 }
1871                 if (nfit_mem->has_lsw)
1872                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1873
1874                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1875                         : NULL;
1876                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1877                                 acpi_nfit_dimm_attribute_groups,
1878                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1879                                 nfit_mem->flush_wpq);
1880                 if (!nvdimm)
1881                         return -ENOMEM;
1882
1883                 nfit_mem->nvdimm = nvdimm;
1884                 dimm_count++;
1885
1886                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1887                         continue;
1888
1889                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1890                                 nvdimm_name(nvdimm),
1891                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1892                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1893                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1894                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1895                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1896
1897         }
1898
1899         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1900         if (rc)
1901                 return rc;
1902
1903         /*
1904          * Now that dimms are successfully registered, and async registration
1905          * is flushed, attempt to enable event notification.
1906          */
1907         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1908                 struct kernfs_node *nfit_kernfs;
1909
1910                 nvdimm = nfit_mem->nvdimm;
1911                 if (!nvdimm)
1912                         continue;
1913
1914                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1915                 if (nfit_kernfs)
1916                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1917                                         "flags");
1918                 sysfs_put(nfit_kernfs);
1919                 if (!nfit_mem->flags_attr)
1920                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1921                                         nvdimm_name(nvdimm));
1922         }
1923
1924         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1925                         acpi_desc);
1926 }
1927
1928 /*
1929  * These constants are private because there are no kernel consumers of
1930  * these commands.
1931  */
1932 enum nfit_aux_cmds {
1933         NFIT_CMD_TRANSLATE_SPA = 5,
1934         NFIT_CMD_ARS_INJECT_SET = 7,
1935         NFIT_CMD_ARS_INJECT_CLEAR = 8,
1936         NFIT_CMD_ARS_INJECT_GET = 9,
1937 };
1938
1939 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1940 {
1941         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1942         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1943         struct acpi_device *adev;
1944         unsigned long dsm_mask;
1945         int i;
1946
1947         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1948         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1949         adev = to_acpi_dev(acpi_desc);
1950         if (!adev)
1951                 return;
1952
1953         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1954                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1955                         set_bit(i, &nd_desc->cmd_mask);
1956         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1957
1958         dsm_mask =
1959                 (1 << ND_CMD_ARS_CAP) |
1960                 (1 << ND_CMD_ARS_START) |
1961                 (1 << ND_CMD_ARS_STATUS) |
1962                 (1 << ND_CMD_CLEAR_ERROR) |
1963                 (1 << NFIT_CMD_TRANSLATE_SPA) |
1964                 (1 << NFIT_CMD_ARS_INJECT_SET) |
1965                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1966                 (1 << NFIT_CMD_ARS_INJECT_GET);
1967         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1968                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1969                         set_bit(i, &nd_desc->bus_dsm_mask);
1970 }
1971
1972 static ssize_t range_index_show(struct device *dev,
1973                 struct device_attribute *attr, char *buf)
1974 {
1975         struct nd_region *nd_region = to_nd_region(dev);
1976         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1977
1978         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1979 }
1980 static DEVICE_ATTR_RO(range_index);
1981
1982 static struct attribute *acpi_nfit_region_attributes[] = {
1983         &dev_attr_range_index.attr,
1984         NULL,
1985 };
1986
1987 static const struct attribute_group acpi_nfit_region_attribute_group = {
1988         .name = "nfit",
1989         .attrs = acpi_nfit_region_attributes,
1990 };
1991
1992 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1993         &nd_region_attribute_group,
1994         &nd_mapping_attribute_group,
1995         &nd_device_attribute_group,
1996         &nd_numa_attribute_group,
1997         &acpi_nfit_region_attribute_group,
1998         NULL,
1999 };
2000
2001 /* enough info to uniquely specify an interleave set */
2002 struct nfit_set_info {
2003         struct nfit_set_info_map {
2004                 u64 region_offset;
2005                 u32 serial_number;
2006                 u32 pad;
2007         } mapping[0];
2008 };
2009
2010 struct nfit_set_info2 {
2011         struct nfit_set_info_map2 {
2012                 u64 region_offset;
2013                 u32 serial_number;
2014                 u16 vendor_id;
2015                 u16 manufacturing_date;
2016                 u8  manufacturing_location;
2017                 u8  reserved[31];
2018         } mapping[0];
2019 };
2020
2021 static size_t sizeof_nfit_set_info(int num_mappings)
2022 {
2023         return sizeof(struct nfit_set_info)
2024                 + num_mappings * sizeof(struct nfit_set_info_map);
2025 }
2026
2027 static size_t sizeof_nfit_set_info2(int num_mappings)
2028 {
2029         return sizeof(struct nfit_set_info2)
2030                 + num_mappings * sizeof(struct nfit_set_info_map2);
2031 }
2032
2033 static int cmp_map_compat(const void *m0, const void *m1)
2034 {
2035         const struct nfit_set_info_map *map0 = m0;
2036         const struct nfit_set_info_map *map1 = m1;
2037
2038         return memcmp(&map0->region_offset, &map1->region_offset,
2039                         sizeof(u64));
2040 }
2041
2042 static int cmp_map(const void *m0, const void *m1)
2043 {
2044         const struct nfit_set_info_map *map0 = m0;
2045         const struct nfit_set_info_map *map1 = m1;
2046
2047         if (map0->region_offset < map1->region_offset)
2048                 return -1;
2049         else if (map0->region_offset > map1->region_offset)
2050                 return 1;
2051         return 0;
2052 }
2053
2054 static int cmp_map2(const void *m0, const void *m1)
2055 {
2056         const struct nfit_set_info_map2 *map0 = m0;
2057         const struct nfit_set_info_map2 *map1 = m1;
2058
2059         if (map0->region_offset < map1->region_offset)
2060                 return -1;
2061         else if (map0->region_offset > map1->region_offset)
2062                 return 1;
2063         return 0;
2064 }
2065
2066 /* Retrieve the nth entry referencing this spa */
2067 static struct acpi_nfit_memory_map *memdev_from_spa(
2068                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2069 {
2070         struct nfit_memdev *nfit_memdev;
2071
2072         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2073                 if (nfit_memdev->memdev->range_index == range_index)
2074                         if (n-- == 0)
2075                                 return nfit_memdev->memdev;
2076         return NULL;
2077 }
2078
2079 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2080                 struct nd_region_desc *ndr_desc,
2081                 struct acpi_nfit_system_address *spa)
2082 {
2083         struct device *dev = acpi_desc->dev;
2084         struct nd_interleave_set *nd_set;
2085         u16 nr = ndr_desc->num_mappings;
2086         struct nfit_set_info2 *info2;
2087         struct nfit_set_info *info;
2088         int i;
2089
2090         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2091         if (!nd_set)
2092                 return -ENOMEM;
2093         ndr_desc->nd_set = nd_set;
2094         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2095
2096         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2097         if (!info)
2098                 return -ENOMEM;
2099
2100         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2101         if (!info2)
2102                 return -ENOMEM;
2103
2104         for (i = 0; i < nr; i++) {
2105                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2106                 struct nfit_set_info_map *map = &info->mapping[i];
2107                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2108                 struct nvdimm *nvdimm = mapping->nvdimm;
2109                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2110                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2111                                 spa->range_index, i);
2112                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2113
2114                 if (!memdev || !nfit_mem->dcr) {
2115                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2116                         return -ENODEV;
2117                 }
2118
2119                 map->region_offset = memdev->region_offset;
2120                 map->serial_number = dcr->serial_number;
2121
2122                 map2->region_offset = memdev->region_offset;
2123                 map2->serial_number = dcr->serial_number;
2124                 map2->vendor_id = dcr->vendor_id;
2125                 map2->manufacturing_date = dcr->manufacturing_date;
2126                 map2->manufacturing_location = dcr->manufacturing_location;
2127         }
2128
2129         /* v1.1 namespaces */
2130         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2131                         cmp_map, NULL);
2132         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2133
2134         /* v1.2 namespaces */
2135         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2136                         cmp_map2, NULL);
2137         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2138
2139         /* support v1.1 namespaces created with the wrong sort order */
2140         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2141                         cmp_map_compat, NULL);
2142         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2143
2144         /* record the result of the sort for the mapping position */
2145         for (i = 0; i < nr; i++) {
2146                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2147                 int j;
2148
2149                 for (j = 0; j < nr; j++) {
2150                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2151                         struct nvdimm *nvdimm = mapping->nvdimm;
2152                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2153                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2154
2155                         if (map2->serial_number == dcr->serial_number &&
2156                             map2->vendor_id == dcr->vendor_id &&
2157                             map2->manufacturing_date == dcr->manufacturing_date &&
2158                             map2->manufacturing_location
2159                                     == dcr->manufacturing_location) {
2160                                 mapping->position = i;
2161                                 break;
2162                         }
2163                 }
2164         }
2165
2166         ndr_desc->nd_set = nd_set;
2167         devm_kfree(dev, info);
2168         devm_kfree(dev, info2);
2169
2170         return 0;
2171 }
2172
2173 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2174 {
2175         struct acpi_nfit_interleave *idt = mmio->idt;
2176         u32 sub_line_offset, line_index, line_offset;
2177         u64 line_no, table_skip_count, table_offset;
2178
2179         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2180         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2181         line_offset = idt->line_offset[line_index]
2182                 * mmio->line_size;
2183         table_offset = table_skip_count * mmio->table_size;
2184
2185         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2186 }
2187
2188 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2189 {
2190         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2191         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2192         const u32 STATUS_MASK = 0x80000037;
2193
2194         if (mmio->num_lines)
2195                 offset = to_interleave_offset(offset, mmio);
2196
2197         return readl(mmio->addr.base + offset) & STATUS_MASK;
2198 }
2199
2200 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2201                 resource_size_t dpa, unsigned int len, unsigned int write)
2202 {
2203         u64 cmd, offset;
2204         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2205
2206         enum {
2207                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2208                 BCW_LEN_SHIFT = 48,
2209                 BCW_LEN_MASK = (1ULL << 8) - 1,
2210                 BCW_CMD_SHIFT = 56,
2211         };
2212
2213         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2214         len = len >> L1_CACHE_SHIFT;
2215         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2216         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2217
2218         offset = nfit_blk->cmd_offset + mmio->size * bw;
2219         if (mmio->num_lines)
2220                 offset = to_interleave_offset(offset, mmio);
2221
2222         writeq(cmd, mmio->addr.base + offset);
2223         nvdimm_flush(nfit_blk->nd_region);
2224
2225         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2226                 readq(mmio->addr.base + offset);
2227 }
2228
2229 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2230                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2231                 unsigned int lane)
2232 {
2233         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2234         unsigned int copied = 0;
2235         u64 base_offset;
2236         int rc;
2237
2238         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2239                 + lane * mmio->size;
2240         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2241         while (len) {
2242                 unsigned int c;
2243                 u64 offset;
2244
2245                 if (mmio->num_lines) {
2246                         u32 line_offset;
2247
2248                         offset = to_interleave_offset(base_offset + copied,
2249                                         mmio);
2250                         div_u64_rem(offset, mmio->line_size, &line_offset);
2251                         c = min_t(size_t, len, mmio->line_size - line_offset);
2252                 } else {
2253                         offset = base_offset + nfit_blk->bdw_offset;
2254                         c = len;
2255                 }
2256
2257                 if (rw)
2258                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2259                 else {
2260                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2261                                 arch_invalidate_pmem((void __force *)
2262                                         mmio->addr.aperture + offset, c);
2263
2264                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2265                 }
2266
2267                 copied += c;
2268                 len -= c;
2269         }
2270
2271         if (rw)
2272                 nvdimm_flush(nfit_blk->nd_region);
2273
2274         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2275         return rc;
2276 }
2277
2278 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2279                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2280 {
2281         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2282         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2283         struct nd_region *nd_region = nfit_blk->nd_region;
2284         unsigned int lane, copied = 0;
2285         int rc = 0;
2286
2287         lane = nd_region_acquire_lane(nd_region);
2288         while (len) {
2289                 u64 c = min(len, mmio->size);
2290
2291                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2292                                 iobuf + copied, c, rw, lane);
2293                 if (rc)
2294                         break;
2295
2296                 copied += c;
2297                 len -= c;
2298         }
2299         nd_region_release_lane(nd_region, lane);
2300
2301         return rc;
2302 }
2303
2304 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2305                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2306 {
2307         if (idt) {
2308                 mmio->num_lines = idt->line_count;
2309                 mmio->line_size = idt->line_size;
2310                 if (interleave_ways == 0)
2311                         return -ENXIO;
2312                 mmio->table_size = mmio->num_lines * interleave_ways
2313                         * mmio->line_size;
2314         }
2315
2316         return 0;
2317 }
2318
2319 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2320                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2321 {
2322         struct nd_cmd_dimm_flags flags;
2323         int rc;
2324
2325         memset(&flags, 0, sizeof(flags));
2326         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2327                         sizeof(flags), NULL);
2328
2329         if (rc >= 0 && flags.status == 0)
2330                 nfit_blk->dimm_flags = flags.flags;
2331         else if (rc == -ENOTTY) {
2332                 /* fall back to a conservative default */
2333                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2334                 rc = 0;
2335         } else
2336                 rc = -ENXIO;
2337
2338         return rc;
2339 }
2340
2341 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2342                 struct device *dev)
2343 {
2344         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2345         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2346         struct nfit_blk_mmio *mmio;
2347         struct nfit_blk *nfit_blk;
2348         struct nfit_mem *nfit_mem;
2349         struct nvdimm *nvdimm;
2350         int rc;
2351
2352         nvdimm = nd_blk_region_to_dimm(ndbr);
2353         nfit_mem = nvdimm_provider_data(nvdimm);
2354         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2355                 dev_dbg(dev, "missing%s%s%s\n",
2356                                 nfit_mem ? "" : " nfit_mem",
2357                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2358                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2359                 return -ENXIO;
2360         }
2361
2362         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2363         if (!nfit_blk)
2364                 return -ENOMEM;
2365         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2366         nfit_blk->nd_region = to_nd_region(dev);
2367
2368         /* map block aperture memory */
2369         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2370         mmio = &nfit_blk->mmio[BDW];
2371         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2372                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2373         if (!mmio->addr.base) {
2374                 dev_dbg(dev, "%s failed to map bdw\n",
2375                                 nvdimm_name(nvdimm));
2376                 return -ENOMEM;
2377         }
2378         mmio->size = nfit_mem->bdw->size;
2379         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2380         mmio->idt = nfit_mem->idt_bdw;
2381         mmio->spa = nfit_mem->spa_bdw;
2382         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2383                         nfit_mem->memdev_bdw->interleave_ways);
2384         if (rc) {
2385                 dev_dbg(dev, "%s failed to init bdw interleave\n",
2386                                 nvdimm_name(nvdimm));
2387                 return rc;
2388         }
2389
2390         /* map block control memory */
2391         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2392         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2393         mmio = &nfit_blk->mmio[DCR];
2394         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2395                         nfit_mem->spa_dcr->length);
2396         if (!mmio->addr.base) {
2397                 dev_dbg(dev, "%s failed to map dcr\n",
2398                                 nvdimm_name(nvdimm));
2399                 return -ENOMEM;
2400         }
2401         mmio->size = nfit_mem->dcr->window_size;
2402         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2403         mmio->idt = nfit_mem->idt_dcr;
2404         mmio->spa = nfit_mem->spa_dcr;
2405         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2406                         nfit_mem->memdev_dcr->interleave_ways);
2407         if (rc) {
2408                 dev_dbg(dev, "%s failed to init dcr interleave\n",
2409                                 nvdimm_name(nvdimm));
2410                 return rc;
2411         }
2412
2413         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2414         if (rc < 0) {
2415                 dev_dbg(dev, "%s failed get DIMM flags\n",
2416                                 nvdimm_name(nvdimm));
2417                 return rc;
2418         }
2419
2420         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2421                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2422
2423         if (mmio->line_size == 0)
2424                 return 0;
2425
2426         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2427                         + 8 > mmio->line_size) {
2428                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2429                 return -ENXIO;
2430         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2431                         + 8 > mmio->line_size) {
2432                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2433                 return -ENXIO;
2434         }
2435
2436         return 0;
2437 }
2438
2439 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2440                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2441 {
2442         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2443         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2444         int cmd_rc, rc;
2445
2446         cmd->address = spa->address;
2447         cmd->length = spa->length;
2448         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2449                         sizeof(*cmd), &cmd_rc);
2450         if (rc < 0)
2451                 return rc;
2452         return cmd_rc;
2453 }
2454
2455 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2456 {
2457         int rc;
2458         int cmd_rc;
2459         struct nd_cmd_ars_start ars_start;
2460         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2461         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2462
2463         memset(&ars_start, 0, sizeof(ars_start));
2464         ars_start.address = spa->address;
2465         ars_start.length = spa->length;
2466         if (test_bit(ARS_SHORT, &nfit_spa->ars_state))
2467                 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2468         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2469                 ars_start.type = ND_ARS_PERSISTENT;
2470         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2471                 ars_start.type = ND_ARS_VOLATILE;
2472         else
2473                 return -ENOTTY;
2474
2475         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2476                         sizeof(ars_start), &cmd_rc);
2477
2478         if (rc < 0)
2479                 return rc;
2480         return cmd_rc;
2481 }
2482
2483 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2484 {
2485         int rc, cmd_rc;
2486         struct nd_cmd_ars_start ars_start;
2487         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2488         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2489
2490         memset(&ars_start, 0, sizeof(ars_start));
2491         ars_start.address = ars_status->restart_address;
2492         ars_start.length = ars_status->restart_length;
2493         ars_start.type = ars_status->type;
2494         ars_start.flags = acpi_desc->ars_start_flags;
2495         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2496                         sizeof(ars_start), &cmd_rc);
2497         if (rc < 0)
2498                 return rc;
2499         return cmd_rc;
2500 }
2501
2502 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2503 {
2504         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2505         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2506         int rc, cmd_rc;
2507
2508         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2509                         acpi_desc->max_ars, &cmd_rc);
2510         if (rc < 0)
2511                 return rc;
2512         return cmd_rc;
2513 }
2514
2515 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2516                 struct nfit_spa *nfit_spa)
2517 {
2518         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2519         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2520         struct nd_region *nd_region = nfit_spa->nd_region;
2521         struct device *dev;
2522
2523         if ((ars_status->address >= spa->address && ars_status->address
2524                                 < spa->address + spa->length)
2525                         || (ars_status->address < spa->address)) {
2526                 /*
2527                  * Assume that if a scrub starts at an offset from the
2528                  * start of nfit_spa that we are in the continuation
2529                  * case.
2530                  *
2531                  * Otherwise, if the scrub covers the spa range, mark
2532                  * any pending request complete.
2533                  */
2534                 if (ars_status->address + ars_status->length
2535                                 >= spa->address + spa->length)
2536                                 /* complete */;
2537                 else
2538                         return;
2539         } else
2540                 return;
2541
2542         if (test_bit(ARS_DONE, &nfit_spa->ars_state))
2543                 return;
2544
2545         if (!test_and_clear_bit(ARS_REQ, &nfit_spa->ars_state))
2546                 return;
2547
2548         if (nd_region) {
2549                 dev = nd_region_dev(nd_region);
2550                 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2551         } else
2552                 dev = acpi_desc->dev;
2553
2554         dev_dbg(dev, "ARS: range %d %s complete\n", spa->range_index,
2555                         test_bit(ARS_SHORT, &nfit_spa->ars_state)
2556                         ? "short" : "long");
2557         clear_bit(ARS_SHORT, &nfit_spa->ars_state);
2558         set_bit(ARS_DONE, &nfit_spa->ars_state);
2559 }
2560
2561 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2562 {
2563         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2564         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2565         int rc;
2566         u32 i;
2567
2568         /*
2569          * First record starts at 44 byte offset from the start of the
2570          * payload.
2571          */
2572         if (ars_status->out_length < 44)
2573                 return 0;
2574         for (i = 0; i < ars_status->num_records; i++) {
2575                 /* only process full records */
2576                 if (ars_status->out_length
2577                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2578                         break;
2579                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2580                                 ars_status->records[i].err_address,
2581                                 ars_status->records[i].length);
2582                 if (rc)
2583                         return rc;
2584         }
2585         if (i < ars_status->num_records)
2586                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2587
2588         return 0;
2589 }
2590
2591 static void acpi_nfit_remove_resource(void *data)
2592 {
2593         struct resource *res = data;
2594
2595         remove_resource(res);
2596 }
2597
2598 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2599                 struct nd_region_desc *ndr_desc)
2600 {
2601         struct resource *res, *nd_res = ndr_desc->res;
2602         int is_pmem, ret;
2603
2604         /* No operation if the region is already registered as PMEM */
2605         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2606                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2607         if (is_pmem == REGION_INTERSECTS)
2608                 return 0;
2609
2610         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2611         if (!res)
2612                 return -ENOMEM;
2613
2614         res->name = "Persistent Memory";
2615         res->start = nd_res->start;
2616         res->end = nd_res->end;
2617         res->flags = IORESOURCE_MEM;
2618         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2619
2620         ret = insert_resource(&iomem_resource, res);
2621         if (ret)
2622                 return ret;
2623
2624         ret = devm_add_action_or_reset(acpi_desc->dev,
2625                                         acpi_nfit_remove_resource,
2626                                         res);
2627         if (ret)
2628                 return ret;
2629
2630         return 0;
2631 }
2632
2633 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2634                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2635                 struct acpi_nfit_memory_map *memdev,
2636                 struct nfit_spa *nfit_spa)
2637 {
2638         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2639                         memdev->device_handle);
2640         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2641         struct nd_blk_region_desc *ndbr_desc;
2642         struct nfit_mem *nfit_mem;
2643         int rc;
2644
2645         if (!nvdimm) {
2646                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2647                                 spa->range_index, memdev->device_handle);
2648                 return -ENODEV;
2649         }
2650
2651         mapping->nvdimm = nvdimm;
2652         switch (nfit_spa_type(spa)) {
2653         case NFIT_SPA_PM:
2654         case NFIT_SPA_VOLATILE:
2655                 mapping->start = memdev->address;
2656                 mapping->size = memdev->region_size;
2657                 break;
2658         case NFIT_SPA_DCR:
2659                 nfit_mem = nvdimm_provider_data(nvdimm);
2660                 if (!nfit_mem || !nfit_mem->bdw) {
2661                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2662                                         spa->range_index, nvdimm_name(nvdimm));
2663                         break;
2664                 }
2665
2666                 mapping->size = nfit_mem->bdw->capacity;
2667                 mapping->start = nfit_mem->bdw->start_address;
2668                 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2669                 ndr_desc->mapping = mapping;
2670                 ndr_desc->num_mappings = 1;
2671                 ndbr_desc = to_blk_region_desc(ndr_desc);
2672                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2673                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2674                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2675                 if (rc)
2676                         return rc;
2677                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2678                                 ndr_desc);
2679                 if (!nfit_spa->nd_region)
2680                         return -ENOMEM;
2681                 break;
2682         }
2683
2684         return 0;
2685 }
2686
2687 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2688 {
2689         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2690                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2691                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2692                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2693 }
2694
2695 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2696 {
2697         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2698                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2699                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2700 }
2701
2702 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2703                 struct nfit_spa *nfit_spa)
2704 {
2705         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2706         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2707         struct nd_blk_region_desc ndbr_desc;
2708         struct nd_region_desc *ndr_desc;
2709         struct nfit_memdev *nfit_memdev;
2710         struct nvdimm_bus *nvdimm_bus;
2711         struct resource res;
2712         int count = 0, rc;
2713
2714         if (nfit_spa->nd_region)
2715                 return 0;
2716
2717         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2718                 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2719                 return 0;
2720         }
2721
2722         memset(&res, 0, sizeof(res));
2723         memset(&mappings, 0, sizeof(mappings));
2724         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2725         res.start = spa->address;
2726         res.end = res.start + spa->length - 1;
2727         ndr_desc = &ndbr_desc.ndr_desc;
2728         ndr_desc->res = &res;
2729         ndr_desc->provider_data = nfit_spa;
2730         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2731         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2732                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2733                                                 spa->proximity_domain);
2734         else
2735                 ndr_desc->numa_node = NUMA_NO_NODE;
2736
2737         /*
2738          * Persistence domain bits are hierarchical, if
2739          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2740          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2741          */
2742         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2743                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2744         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2745                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2746
2747         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2748                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2749                 struct nd_mapping_desc *mapping;
2750
2751                 if (memdev->range_index != spa->range_index)
2752                         continue;
2753                 if (count >= ND_MAX_MAPPINGS) {
2754                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2755                                         spa->range_index, ND_MAX_MAPPINGS);
2756                         return -ENXIO;
2757                 }
2758                 mapping = &mappings[count++];
2759                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2760                                 memdev, nfit_spa);
2761                 if (rc)
2762                         goto out;
2763         }
2764
2765         ndr_desc->mapping = mappings;
2766         ndr_desc->num_mappings = count;
2767         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2768         if (rc)
2769                 goto out;
2770
2771         nvdimm_bus = acpi_desc->nvdimm_bus;
2772         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2773                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2774                 if (rc) {
2775                         dev_warn(acpi_desc->dev,
2776                                 "failed to insert pmem resource to iomem: %d\n",
2777                                 rc);
2778                         goto out;
2779                 }
2780
2781                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2782                                 ndr_desc);
2783                 if (!nfit_spa->nd_region)
2784                         rc = -ENOMEM;
2785         } else if (nfit_spa_is_volatile(spa)) {
2786                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2787                                 ndr_desc);
2788                 if (!nfit_spa->nd_region)
2789                         rc = -ENOMEM;
2790         } else if (nfit_spa_is_virtual(spa)) {
2791                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2792                                 ndr_desc);
2793                 if (!nfit_spa->nd_region)
2794                         rc = -ENOMEM;
2795         }
2796
2797  out:
2798         if (rc)
2799                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2800                                 nfit_spa->spa->range_index);
2801         return rc;
2802 }
2803
2804 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
2805 {
2806         struct device *dev = acpi_desc->dev;
2807         struct nd_cmd_ars_status *ars_status;
2808
2809         if (acpi_desc->ars_status) {
2810                 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2811                 return 0;
2812         }
2813
2814         ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
2815         if (!ars_status)
2816                 return -ENOMEM;
2817         acpi_desc->ars_status = ars_status;
2818         return 0;
2819 }
2820
2821 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
2822 {
2823         int rc;
2824
2825         if (ars_status_alloc(acpi_desc))
2826                 return -ENOMEM;
2827
2828         rc = ars_get_status(acpi_desc);
2829
2830         if (rc < 0 && rc != -ENOSPC)
2831                 return rc;
2832
2833         if (ars_status_process_records(acpi_desc))
2834                 return -ENOMEM;
2835
2836         return 0;
2837 }
2838
2839 static int ars_register(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa,
2840                 int *query_rc)
2841 {
2842         int rc = *query_rc;
2843
2844         if (no_init_ars)
2845                 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2846
2847         set_bit(ARS_REQ, &nfit_spa->ars_state);
2848         set_bit(ARS_SHORT, &nfit_spa->ars_state);
2849
2850         switch (rc) {
2851         case 0:
2852         case -EAGAIN:
2853                 rc = ars_start(acpi_desc, nfit_spa);
2854                 if (rc == -EBUSY) {
2855                         *query_rc = rc;
2856                         break;
2857                 } else if (rc == 0) {
2858                         rc = acpi_nfit_query_poison(acpi_desc);
2859                 } else {
2860                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
2861                         break;
2862                 }
2863                 if (rc == -EAGAIN)
2864                         clear_bit(ARS_SHORT, &nfit_spa->ars_state);
2865                 else if (rc == 0)
2866                         ars_complete(acpi_desc, nfit_spa);
2867                 break;
2868         case -EBUSY:
2869         case -ENOSPC:
2870                 break;
2871         default:
2872                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2873                 break;
2874         }
2875
2876         if (test_and_clear_bit(ARS_DONE, &nfit_spa->ars_state))
2877                 set_bit(ARS_REQ, &nfit_spa->ars_state);
2878
2879         return acpi_nfit_register_region(acpi_desc, nfit_spa);
2880 }
2881
2882 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
2883 {
2884         struct nfit_spa *nfit_spa;
2885
2886         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2887                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2888                         continue;
2889                 ars_complete(acpi_desc, nfit_spa);
2890         }
2891 }
2892
2893 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
2894                 int query_rc)
2895 {
2896         unsigned int tmo = acpi_desc->scrub_tmo;
2897         struct device *dev = acpi_desc->dev;
2898         struct nfit_spa *nfit_spa;
2899
2900         if (acpi_desc->cancel)
2901                 return 0;
2902
2903         if (query_rc == -EBUSY) {
2904                 dev_dbg(dev, "ARS: ARS busy\n");
2905                 return min(30U * 60U, tmo * 2);
2906         }
2907         if (query_rc == -ENOSPC) {
2908                 dev_dbg(dev, "ARS: ARS continue\n");
2909                 ars_continue(acpi_desc);
2910                 return 1;
2911         }
2912         if (query_rc && query_rc != -EAGAIN) {
2913                 unsigned long long addr, end;
2914
2915                 addr = acpi_desc->ars_status->address;
2916                 end = addr + acpi_desc->ars_status->length;
2917                 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
2918                                 query_rc);
2919         }
2920
2921         ars_complete_all(acpi_desc);
2922         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2923                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2924                         continue;
2925                 if (test_bit(ARS_REQ, &nfit_spa->ars_state)) {
2926                         int rc = ars_start(acpi_desc, nfit_spa);
2927
2928                         clear_bit(ARS_DONE, &nfit_spa->ars_state);
2929                         dev = nd_region_dev(nfit_spa->nd_region);
2930                         dev_dbg(dev, "ARS: range %d ARS start (%d)\n",
2931                                         nfit_spa->spa->range_index, rc);
2932                         if (rc == 0 || rc == -EBUSY)
2933                                 return 1;
2934                         dev_err(dev, "ARS: range %d ARS failed (%d)\n",
2935                                         nfit_spa->spa->range_index, rc);
2936                         set_bit(ARS_FAILED, &nfit_spa->ars_state);
2937                 }
2938         }
2939         return 0;
2940 }
2941
2942 static void acpi_nfit_scrub(struct work_struct *work)
2943 {
2944         struct acpi_nfit_desc *acpi_desc;
2945         unsigned int tmo;
2946         int query_rc;
2947
2948         acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
2949         mutex_lock(&acpi_desc->init_mutex);
2950         query_rc = acpi_nfit_query_poison(acpi_desc);
2951         tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
2952         if (tmo) {
2953                 queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
2954                 acpi_desc->scrub_tmo = tmo;
2955         } else {
2956                 acpi_desc->scrub_count++;
2957                 if (acpi_desc->scrub_count_state)
2958                         sysfs_notify_dirent(acpi_desc->scrub_count_state);
2959         }
2960         memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2961         mutex_unlock(&acpi_desc->init_mutex);
2962 }
2963
2964 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
2965                 struct nfit_spa *nfit_spa)
2966 {
2967         int type = nfit_spa_type(nfit_spa->spa);
2968         struct nd_cmd_ars_cap ars_cap;
2969         int rc;
2970
2971         memset(&ars_cap, 0, sizeof(ars_cap));
2972         rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2973         if (rc < 0)
2974                 return;
2975         /* check that the supported scrub types match the spa type */
2976         if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
2977                                 & ND_ARS_VOLATILE) == 0)
2978                 return;
2979         if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
2980                                 & ND_ARS_PERSISTENT) == 0)
2981                 return;
2982
2983         nfit_spa->max_ars = ars_cap.max_ars_out;
2984         nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2985         acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
2986         clear_bit(ARS_FAILED, &nfit_spa->ars_state);
2987         set_bit(ARS_REQ, &nfit_spa->ars_state);
2988 }
2989
2990 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2991 {
2992         struct nfit_spa *nfit_spa;
2993         int rc, query_rc;
2994
2995         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2996                 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2997                 switch (nfit_spa_type(nfit_spa->spa)) {
2998                 case NFIT_SPA_VOLATILE:
2999                 case NFIT_SPA_PM:
3000                         acpi_nfit_init_ars(acpi_desc, nfit_spa);
3001                         break;
3002                 }
3003         }
3004
3005         /*
3006          * Reap any results that might be pending before starting new
3007          * short requests.
3008          */
3009         query_rc = acpi_nfit_query_poison(acpi_desc);
3010         if (query_rc == 0)
3011                 ars_complete_all(acpi_desc);
3012
3013         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3014                 switch (nfit_spa_type(nfit_spa->spa)) {
3015                 case NFIT_SPA_VOLATILE:
3016                 case NFIT_SPA_PM:
3017                         /* register regions and kick off initial ARS run */
3018                         rc = ars_register(acpi_desc, nfit_spa, &query_rc);
3019                         if (rc)
3020                                 return rc;
3021                         break;
3022                 case NFIT_SPA_BDW:
3023                         /* nothing to register */
3024                         break;
3025                 case NFIT_SPA_DCR:
3026                 case NFIT_SPA_VDISK:
3027                 case NFIT_SPA_VCD:
3028                 case NFIT_SPA_PDISK:
3029                 case NFIT_SPA_PCD:
3030                         /* register known regions that don't support ARS */
3031                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3032                         if (rc)
3033                                 return rc;
3034                         break;
3035                 default:
3036                         /* don't register unknown regions */
3037                         break;
3038                 }
3039
3040         queue_delayed_work(nfit_wq, &acpi_desc->dwork, 0);
3041         return 0;
3042 }
3043
3044 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3045                 struct nfit_table_prev *prev)
3046 {
3047         struct device *dev = acpi_desc->dev;
3048
3049         if (!list_empty(&prev->spas) ||
3050                         !list_empty(&prev->memdevs) ||
3051                         !list_empty(&prev->dcrs) ||
3052                         !list_empty(&prev->bdws) ||
3053                         !list_empty(&prev->idts) ||
3054                         !list_empty(&prev->flushes)) {
3055                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3056                 return -ENXIO;
3057         }
3058         return 0;
3059 }
3060
3061 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3062 {
3063         struct device *dev = acpi_desc->dev;
3064         struct kernfs_node *nfit;
3065         struct device *bus_dev;
3066
3067         if (!ars_supported(acpi_desc->nvdimm_bus))
3068                 return 0;
3069
3070         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3071         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3072         if (!nfit) {
3073                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3074                 return -ENODEV;
3075         }
3076         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3077         sysfs_put(nfit);
3078         if (!acpi_desc->scrub_count_state) {
3079                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3080                 return -ENODEV;
3081         }
3082
3083         return 0;
3084 }
3085
3086 static void acpi_nfit_unregister(void *data)
3087 {
3088         struct acpi_nfit_desc *acpi_desc = data;
3089
3090         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3091 }
3092
3093 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3094 {
3095         struct device *dev = acpi_desc->dev;
3096         struct nfit_table_prev prev;
3097         const void *end;
3098         int rc;
3099
3100         if (!acpi_desc->nvdimm_bus) {
3101                 acpi_nfit_init_dsms(acpi_desc);
3102
3103                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3104                                 &acpi_desc->nd_desc);
3105                 if (!acpi_desc->nvdimm_bus)
3106                         return -ENOMEM;
3107
3108                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3109                                 acpi_desc);
3110                 if (rc)
3111                         return rc;
3112
3113                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3114                 if (rc)
3115                         return rc;
3116
3117                 /* register this acpi_desc for mce notifications */
3118                 mutex_lock(&acpi_desc_lock);
3119                 list_add_tail(&acpi_desc->list, &acpi_descs);
3120                 mutex_unlock(&acpi_desc_lock);
3121         }
3122
3123         mutex_lock(&acpi_desc->init_mutex);
3124
3125         INIT_LIST_HEAD(&prev.spas);
3126         INIT_LIST_HEAD(&prev.memdevs);
3127         INIT_LIST_HEAD(&prev.dcrs);
3128         INIT_LIST_HEAD(&prev.bdws);
3129         INIT_LIST_HEAD(&prev.idts);
3130         INIT_LIST_HEAD(&prev.flushes);
3131
3132         list_cut_position(&prev.spas, &acpi_desc->spas,
3133                                 acpi_desc->spas.prev);
3134         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3135                                 acpi_desc->memdevs.prev);
3136         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3137                                 acpi_desc->dcrs.prev);
3138         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3139                                 acpi_desc->bdws.prev);
3140         list_cut_position(&prev.idts, &acpi_desc->idts,
3141                                 acpi_desc->idts.prev);
3142         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3143                                 acpi_desc->flushes.prev);
3144
3145         end = data + sz;
3146         while (!IS_ERR_OR_NULL(data))
3147                 data = add_table(acpi_desc, &prev, data, end);
3148
3149         if (IS_ERR(data)) {
3150                 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3151                 rc = PTR_ERR(data);
3152                 goto out_unlock;
3153         }
3154
3155         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3156         if (rc)
3157                 goto out_unlock;
3158
3159         rc = nfit_mem_init(acpi_desc);
3160         if (rc)
3161                 goto out_unlock;
3162
3163         rc = acpi_nfit_register_dimms(acpi_desc);
3164         if (rc)
3165                 goto out_unlock;
3166
3167         rc = acpi_nfit_register_regions(acpi_desc);
3168
3169  out_unlock:
3170         mutex_unlock(&acpi_desc->init_mutex);
3171         return rc;
3172 }
3173 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3174
3175 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3176 {
3177         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3178         struct device *dev = acpi_desc->dev;
3179
3180         /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3181         device_lock(dev);
3182         device_unlock(dev);
3183
3184         /* Bounce the init_mutex to complete initial registration */
3185         mutex_lock(&acpi_desc->init_mutex);
3186         mutex_unlock(&acpi_desc->init_mutex);
3187
3188         return 0;
3189 }
3190
3191 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3192                 struct nvdimm *nvdimm, unsigned int cmd)
3193 {
3194         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3195
3196         if (nvdimm)
3197                 return 0;
3198         if (cmd != ND_CMD_ARS_START)
3199                 return 0;
3200
3201         /*
3202          * The kernel and userspace may race to initiate a scrub, but
3203          * the scrub thread is prepared to lose that initial race.  It
3204          * just needs guarantees that any ars it initiates are not
3205          * interrupted by any intervening start reqeusts from userspace.
3206          */
3207         if (work_busy(&acpi_desc->dwork.work))
3208                 return -EBUSY;
3209
3210         return 0;
3211 }
3212
3213 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, unsigned long flags)
3214 {
3215         struct device *dev = acpi_desc->dev;
3216         int scheduled = 0, busy = 0;
3217         struct nfit_spa *nfit_spa;
3218
3219         mutex_lock(&acpi_desc->init_mutex);
3220         if (acpi_desc->cancel) {
3221                 mutex_unlock(&acpi_desc->init_mutex);
3222                 return 0;
3223         }
3224
3225         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3226                 int type = nfit_spa_type(nfit_spa->spa);
3227
3228                 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3229                         continue;
3230                 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3231                         continue;
3232
3233                 if (test_and_set_bit(ARS_REQ, &nfit_spa->ars_state))
3234                         busy++;
3235                 else {
3236                         if (test_bit(ARS_SHORT, &flags))
3237                                 set_bit(ARS_SHORT, &nfit_spa->ars_state);
3238                         scheduled++;
3239                 }
3240         }
3241         if (scheduled) {
3242                 queue_delayed_work(nfit_wq, &acpi_desc->dwork, 0);
3243                 dev_dbg(dev, "ars_scan triggered\n");
3244         }
3245         mutex_unlock(&acpi_desc->init_mutex);
3246
3247         if (scheduled)
3248                 return 0;
3249         if (busy)
3250                 return -EBUSY;
3251         return -ENOTTY;
3252 }
3253
3254 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3255 {
3256         struct nvdimm_bus_descriptor *nd_desc;
3257
3258         dev_set_drvdata(dev, acpi_desc);
3259         acpi_desc->dev = dev;
3260         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3261         nd_desc = &acpi_desc->nd_desc;
3262         nd_desc->provider_name = "ACPI.NFIT";
3263         nd_desc->module = THIS_MODULE;
3264         nd_desc->ndctl = acpi_nfit_ctl;
3265         nd_desc->flush_probe = acpi_nfit_flush_probe;
3266         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3267         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3268
3269         INIT_LIST_HEAD(&acpi_desc->spas);
3270         INIT_LIST_HEAD(&acpi_desc->dcrs);
3271         INIT_LIST_HEAD(&acpi_desc->bdws);
3272         INIT_LIST_HEAD(&acpi_desc->idts);
3273         INIT_LIST_HEAD(&acpi_desc->flushes);
3274         INIT_LIST_HEAD(&acpi_desc->memdevs);
3275         INIT_LIST_HEAD(&acpi_desc->dimms);
3276         INIT_LIST_HEAD(&acpi_desc->list);
3277         mutex_init(&acpi_desc->init_mutex);
3278         acpi_desc->scrub_tmo = 1;
3279         INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3280 }
3281 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3282
3283 static void acpi_nfit_put_table(void *table)
3284 {
3285         acpi_put_table(table);
3286 }
3287
3288 void acpi_nfit_shutdown(void *data)
3289 {
3290         struct acpi_nfit_desc *acpi_desc = data;
3291         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3292
3293         /*
3294          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3295          * race teardown
3296          */
3297         mutex_lock(&acpi_desc_lock);
3298         list_del(&acpi_desc->list);
3299         mutex_unlock(&acpi_desc_lock);
3300
3301         mutex_lock(&acpi_desc->init_mutex);
3302         acpi_desc->cancel = 1;
3303         cancel_delayed_work_sync(&acpi_desc->dwork);
3304         mutex_unlock(&acpi_desc->init_mutex);
3305
3306         /*
3307          * Bounce the nvdimm bus lock to make sure any in-flight
3308          * acpi_nfit_ars_rescan() submissions have had a chance to
3309          * either submit or see ->cancel set.
3310          */
3311         device_lock(bus_dev);
3312         device_unlock(bus_dev);
3313
3314         flush_workqueue(nfit_wq);
3315 }
3316 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3317
3318 static int acpi_nfit_add(struct acpi_device *adev)
3319 {
3320         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3321         struct acpi_nfit_desc *acpi_desc;
3322         struct device *dev = &adev->dev;
3323         struct acpi_table_header *tbl;
3324         acpi_status status = AE_OK;
3325         acpi_size sz;
3326         int rc = 0;
3327
3328         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3329         if (ACPI_FAILURE(status)) {
3330                 /* This is ok, we could have an nvdimm hotplugged later */
3331                 dev_dbg(dev, "failed to find NFIT at startup\n");
3332                 return 0;
3333         }
3334
3335         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3336         if (rc)
3337                 return rc;
3338         sz = tbl->length;
3339
3340         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3341         if (!acpi_desc)
3342                 return -ENOMEM;
3343         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3344
3345         /* Save the acpi header for exporting the revision via sysfs */
3346         acpi_desc->acpi_header = *tbl;
3347
3348         /* Evaluate _FIT and override with that if present */
3349         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3350         if (ACPI_SUCCESS(status) && buf.length > 0) {
3351                 union acpi_object *obj = buf.pointer;
3352
3353                 if (obj->type == ACPI_TYPE_BUFFER)
3354                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3355                                         obj->buffer.length);
3356                 else
3357                         dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3358                                 (int) obj->type);
3359                 kfree(buf.pointer);
3360         } else
3361                 /* skip over the lead-in header table */
3362                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3363                                 + sizeof(struct acpi_table_nfit),
3364                                 sz - sizeof(struct acpi_table_nfit));
3365
3366         if (rc)
3367                 return rc;
3368         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3369 }
3370
3371 static int acpi_nfit_remove(struct acpi_device *adev)
3372 {
3373         /* see acpi_nfit_unregister */
3374         return 0;
3375 }
3376
3377 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3378 {
3379         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3380         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3381         union acpi_object *obj;
3382         acpi_status status;
3383         int ret;
3384
3385         if (!dev->driver) {
3386                 /* dev->driver may be null if we're being removed */
3387                 dev_dbg(dev, "no driver found for dev\n");
3388                 return;
3389         }
3390
3391         if (!acpi_desc) {
3392                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3393                 if (!acpi_desc)
3394                         return;
3395                 acpi_nfit_desc_init(acpi_desc, dev);
3396         } else {
3397                 /*
3398                  * Finish previous registration before considering new
3399                  * regions.
3400                  */
3401                 flush_workqueue(nfit_wq);
3402         }
3403
3404         /* Evaluate _FIT */
3405         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3406         if (ACPI_FAILURE(status)) {
3407                 dev_err(dev, "failed to evaluate _FIT\n");
3408                 return;
3409         }
3410
3411         obj = buf.pointer;
3412         if (obj->type == ACPI_TYPE_BUFFER) {
3413                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3414                                 obj->buffer.length);
3415                 if (ret)
3416                         dev_err(dev, "failed to merge updated NFIT\n");
3417         } else
3418                 dev_err(dev, "Invalid _FIT\n");
3419         kfree(buf.pointer);
3420 }
3421
3422 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3423 {
3424         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3425         unsigned long flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3426                         0 : 1 << ARS_SHORT;
3427
3428         acpi_nfit_ars_rescan(acpi_desc, flags);
3429 }
3430
3431 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3432 {
3433         dev_dbg(dev, "event: 0x%x\n", event);
3434
3435         switch (event) {
3436         case NFIT_NOTIFY_UPDATE:
3437                 return acpi_nfit_update_notify(dev, handle);
3438         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3439                 return acpi_nfit_uc_error_notify(dev, handle);
3440         default:
3441                 return;
3442         }
3443 }
3444 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3445
3446 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3447 {
3448         device_lock(&adev->dev);
3449         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3450         device_unlock(&adev->dev);
3451 }
3452
3453 static const struct acpi_device_id acpi_nfit_ids[] = {
3454         { "ACPI0012", 0 },
3455         { "", 0 },
3456 };
3457 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3458
3459 static struct acpi_driver acpi_nfit_driver = {
3460         .name = KBUILD_MODNAME,
3461         .ids = acpi_nfit_ids,
3462         .ops = {
3463                 .add = acpi_nfit_add,
3464                 .remove = acpi_nfit_remove,
3465                 .notify = acpi_nfit_notify,
3466         },
3467 };
3468
3469 static __init int nfit_init(void)
3470 {
3471         int ret;
3472
3473         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3474         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3475         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3476         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3477         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3478         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3479         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3480         BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3481
3482         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3483         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3484         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3485         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3486         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3487         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3488         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3489         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3490         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3491         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3492         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3493         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3494         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3495
3496         nfit_wq = create_singlethread_workqueue("nfit");
3497         if (!nfit_wq)
3498                 return -ENOMEM;
3499
3500         nfit_mce_register();
3501         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3502         if (ret) {
3503                 nfit_mce_unregister();
3504                 destroy_workqueue(nfit_wq);
3505         }
3506
3507         return ret;
3508
3509 }
3510
3511 static __exit void nfit_exit(void)
3512 {
3513         nfit_mce_unregister();
3514         acpi_bus_unregister_driver(&acpi_nfit_driver);
3515         destroy_workqueue(nfit_wq);
3516         WARN_ON(!list_empty(&acpi_descs));
3517 }
3518
3519 module_init(nfit_init);
3520 module_exit(nfit_exit);
3521 MODULE_LICENSE("GPL v2");
3522 MODULE_AUTHOR("Intel Corporation");