Merge branch 'for-4.9/libnvdimm' into libnvdimm-for-next
[cascardo/linux.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/pmem.h>
24 #include <linux/io.h>
25 #include <linux/nd.h>
26 #include <asm/cacheflush.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 unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
40 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
41 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
42
43 /* after three payloads of overflow, it's dead jim */
44 static unsigned int scrub_overflow_abort = 3;
45 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
46 MODULE_PARM_DESC(scrub_overflow_abort,
47                 "Number of times we overflow ARS results before abort");
48
49 static bool disable_vendor_specific;
50 module_param(disable_vendor_specific, bool, S_IRUGO);
51 MODULE_PARM_DESC(disable_vendor_specific,
52                 "Limit commands to the publicly specified set\n");
53
54 LIST_HEAD(acpi_descs);
55 DEFINE_MUTEX(acpi_desc_lock);
56
57 static struct workqueue_struct *nfit_wq;
58
59 struct nfit_table_prev {
60         struct list_head spas;
61         struct list_head memdevs;
62         struct list_head dcrs;
63         struct list_head bdws;
64         struct list_head idts;
65         struct list_head flushes;
66 };
67
68 static u8 nfit_uuid[NFIT_UUID_MAX][16];
69
70 const u8 *to_nfit_uuid(enum nfit_uuids id)
71 {
72         return nfit_uuid[id];
73 }
74 EXPORT_SYMBOL(to_nfit_uuid);
75
76 static struct acpi_nfit_desc *to_acpi_nfit_desc(
77                 struct nvdimm_bus_descriptor *nd_desc)
78 {
79         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
80 }
81
82 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
83 {
84         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
85
86         /*
87          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
88          * acpi_device.
89          */
90         if (!nd_desc->provider_name
91                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
92                 return NULL;
93
94         return to_acpi_device(acpi_desc->dev);
95 }
96
97 static int xlat_status(void *buf, unsigned int cmd, u32 status)
98 {
99         struct nd_cmd_clear_error *clear_err;
100         struct nd_cmd_ars_status *ars_status;
101         u16 flags;
102
103         switch (cmd) {
104         case ND_CMD_ARS_CAP:
105                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
106                         return -ENOTTY;
107
108                 /* Command failed */
109                 if (status & 0xffff)
110                         return -EIO;
111
112                 /* No supported scan types for this range */
113                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
114                 if ((status >> 16 & flags) == 0)
115                         return -ENOTTY;
116                 break;
117         case ND_CMD_ARS_START:
118                 /* ARS is in progress */
119                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
120                         return -EBUSY;
121
122                 /* Command failed */
123                 if (status & 0xffff)
124                         return -EIO;
125                 break;
126         case ND_CMD_ARS_STATUS:
127                 ars_status = buf;
128                 /* Command failed */
129                 if (status & 0xffff)
130                         return -EIO;
131                 /* Check extended status (Upper two bytes) */
132                 if (status == NFIT_ARS_STATUS_DONE)
133                         return 0;
134
135                 /* ARS is in progress */
136                 if (status == NFIT_ARS_STATUS_BUSY)
137                         return -EBUSY;
138
139                 /* No ARS performed for the current boot */
140                 if (status == NFIT_ARS_STATUS_NONE)
141                         return -EAGAIN;
142
143                 /*
144                  * ARS interrupted, either we overflowed or some other
145                  * agent wants the scan to stop.  If we didn't overflow
146                  * then just continue with the returned results.
147                  */
148                 if (status == NFIT_ARS_STATUS_INTR) {
149                         if (ars_status->flags & NFIT_ARS_F_OVERFLOW)
150                                 return -ENOSPC;
151                         return 0;
152                 }
153
154                 /* Unknown status */
155                 if (status >> 16)
156                         return -EIO;
157                 break;
158         case ND_CMD_CLEAR_ERROR:
159                 clear_err = buf;
160                 if (status & 0xffff)
161                         return -EIO;
162                 if (!clear_err->cleared)
163                         return -EIO;
164                 if (clear_err->length > clear_err->cleared)
165                         return clear_err->cleared;
166                 break;
167         default:
168                 break;
169         }
170
171         /* all other non-zero status results in an error */
172         if (status)
173                 return -EIO;
174         return 0;
175 }
176
177 static int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc,
178                 struct nvdimm *nvdimm, unsigned int cmd, void *buf,
179                 unsigned int buf_len, int *cmd_rc)
180 {
181         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
182         union acpi_object in_obj, in_buf, *out_obj;
183         const struct nd_cmd_desc *desc = NULL;
184         struct device *dev = acpi_desc->dev;
185         struct nd_cmd_pkg *call_pkg = NULL;
186         const char *cmd_name, *dimm_name;
187         unsigned long cmd_mask, dsm_mask;
188         u32 offset, fw_status = 0;
189         acpi_handle handle;
190         unsigned int func;
191         const u8 *uuid;
192         int rc, i;
193
194         func = cmd;
195         if (cmd == ND_CMD_CALL) {
196                 call_pkg = buf;
197                 func = call_pkg->nd_command;
198         }
199
200         if (nvdimm) {
201                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
202                 struct acpi_device *adev = nfit_mem->adev;
203
204                 if (!adev)
205                         return -ENOTTY;
206                 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
207                         return -ENOTTY;
208
209                 dimm_name = nvdimm_name(nvdimm);
210                 cmd_name = nvdimm_cmd_name(cmd);
211                 cmd_mask = nvdimm_cmd_mask(nvdimm);
212                 dsm_mask = nfit_mem->dsm_mask;
213                 desc = nd_cmd_dimm_desc(cmd);
214                 uuid = to_nfit_uuid(nfit_mem->family);
215                 handle = adev->handle;
216         } else {
217                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
218
219                 cmd_name = nvdimm_bus_cmd_name(cmd);
220                 cmd_mask = nd_desc->cmd_mask;
221                 dsm_mask = cmd_mask;
222                 desc = nd_cmd_bus_desc(cmd);
223                 uuid = to_nfit_uuid(NFIT_DEV_BUS);
224                 handle = adev->handle;
225                 dimm_name = "bus";
226         }
227
228         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
229                 return -ENOTTY;
230
231         if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
232                 return -ENOTTY;
233
234         in_obj.type = ACPI_TYPE_PACKAGE;
235         in_obj.package.count = 1;
236         in_obj.package.elements = &in_buf;
237         in_buf.type = ACPI_TYPE_BUFFER;
238         in_buf.buffer.pointer = buf;
239         in_buf.buffer.length = 0;
240
241         /* libnvdimm has already validated the input envelope */
242         for (i = 0; i < desc->in_num; i++)
243                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
244                                 i, buf);
245
246         if (call_pkg) {
247                 /* skip over package wrapper */
248                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
249                 in_buf.buffer.length = call_pkg->nd_size_in;
250         }
251
252         if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
253                 dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
254                                 __func__, dimm_name, cmd, func,
255                                 in_buf.buffer.length);
256                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
257                         in_buf.buffer.pointer,
258                         min_t(u32, 256, in_buf.buffer.length), true);
259         }
260
261         out_obj = acpi_evaluate_dsm(handle, uuid, 1, func, &in_obj);
262         if (!out_obj) {
263                 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
264                                 cmd_name);
265                 return -EINVAL;
266         }
267
268         if (call_pkg) {
269                 call_pkg->nd_fw_size = out_obj->buffer.length;
270                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
271                         out_obj->buffer.pointer,
272                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
273
274                 ACPI_FREE(out_obj);
275                 /*
276                  * Need to support FW function w/o known size in advance.
277                  * Caller can determine required size based upon nd_fw_size.
278                  * If we return an error (like elsewhere) then caller wouldn't
279                  * be able to rely upon data returned to make calculation.
280                  */
281                 return 0;
282         }
283
284         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
285                 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
286                                 __func__, dimm_name, cmd_name, out_obj->type);
287                 rc = -EINVAL;
288                 goto out;
289         }
290
291         if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
292                 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__,
293                                 dimm_name, cmd_name, out_obj->buffer.length);
294                 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
295                                 4, out_obj->buffer.pointer, min_t(u32, 128,
296                                         out_obj->buffer.length), true);
297         }
298
299         for (i = 0, offset = 0; i < desc->out_num; i++) {
300                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
301                                 (u32 *) out_obj->buffer.pointer);
302
303                 if (offset + out_size > out_obj->buffer.length) {
304                         dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
305                                         __func__, dimm_name, cmd_name, i);
306                         break;
307                 }
308
309                 if (in_buf.buffer.length + offset + out_size > buf_len) {
310                         dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
311                                         __func__, dimm_name, cmd_name, i);
312                         rc = -ENXIO;
313                         goto out;
314                 }
315                 memcpy(buf + in_buf.buffer.length + offset,
316                                 out_obj->buffer.pointer + offset, out_size);
317                 offset += out_size;
318         }
319
320         /*
321          * Set fw_status for all the commands with a known format to be
322          * later interpreted by xlat_status().
323          */
324         if (i >= 1 && ((cmd >= ND_CMD_ARS_CAP && cmd <= ND_CMD_CLEAR_ERROR)
325                         || (cmd >= ND_CMD_SMART && cmd <= ND_CMD_VENDOR)))
326                 fw_status = *(u32 *) out_obj->buffer.pointer;
327
328         if (offset + in_buf.buffer.length < buf_len) {
329                 if (i >= 1) {
330                         /*
331                          * status valid, return the number of bytes left
332                          * unfilled in the output buffer
333                          */
334                         rc = buf_len - offset - in_buf.buffer.length;
335                         if (cmd_rc)
336                                 *cmd_rc = xlat_status(buf, cmd, fw_status);
337                 } else {
338                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
339                                         __func__, dimm_name, cmd_name, buf_len,
340                                         offset);
341                         rc = -ENXIO;
342                 }
343         } else {
344                 rc = 0;
345                 if (cmd_rc)
346                         *cmd_rc = xlat_status(buf, cmd, fw_status);
347         }
348
349  out:
350         ACPI_FREE(out_obj);
351
352         return rc;
353 }
354
355 static const char *spa_type_name(u16 type)
356 {
357         static const char *to_name[] = {
358                 [NFIT_SPA_VOLATILE] = "volatile",
359                 [NFIT_SPA_PM] = "pmem",
360                 [NFIT_SPA_DCR] = "dimm-control-region",
361                 [NFIT_SPA_BDW] = "block-data-window",
362                 [NFIT_SPA_VDISK] = "volatile-disk",
363                 [NFIT_SPA_VCD] = "volatile-cd",
364                 [NFIT_SPA_PDISK] = "persistent-disk",
365                 [NFIT_SPA_PCD] = "persistent-cd",
366
367         };
368
369         if (type > NFIT_SPA_PCD)
370                 return "unknown";
371
372         return to_name[type];
373 }
374
375 int nfit_spa_type(struct acpi_nfit_system_address *spa)
376 {
377         int i;
378
379         for (i = 0; i < NFIT_UUID_MAX; i++)
380                 if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
381                         return i;
382         return -1;
383 }
384
385 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
386                 struct nfit_table_prev *prev,
387                 struct acpi_nfit_system_address *spa)
388 {
389         struct device *dev = acpi_desc->dev;
390         struct nfit_spa *nfit_spa;
391
392         if (spa->header.length != sizeof(*spa))
393                 return false;
394
395         list_for_each_entry(nfit_spa, &prev->spas, list) {
396                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
397                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
398                         return true;
399                 }
400         }
401
402         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
403                         GFP_KERNEL);
404         if (!nfit_spa)
405                 return false;
406         INIT_LIST_HEAD(&nfit_spa->list);
407         memcpy(nfit_spa->spa, spa, sizeof(*spa));
408         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
409         dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
410                         spa->range_index,
411                         spa_type_name(nfit_spa_type(spa)));
412         return true;
413 }
414
415 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
416                 struct nfit_table_prev *prev,
417                 struct acpi_nfit_memory_map *memdev)
418 {
419         struct device *dev = acpi_desc->dev;
420         struct nfit_memdev *nfit_memdev;
421
422         if (memdev->header.length != sizeof(*memdev))
423                 return false;
424
425         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
426                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
427                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
428                         return true;
429                 }
430
431         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
432                         GFP_KERNEL);
433         if (!nfit_memdev)
434                 return false;
435         INIT_LIST_HEAD(&nfit_memdev->list);
436         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
437         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
438         dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
439                         __func__, memdev->device_handle, memdev->range_index,
440                         memdev->region_index);
441         return true;
442 }
443
444 /*
445  * An implementation may provide a truncated control region if no block windows
446  * are defined.
447  */
448 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
449 {
450         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
451                                 window_size))
452                 return 0;
453         if (dcr->windows)
454                 return sizeof(*dcr);
455         return offsetof(struct acpi_nfit_control_region, window_size);
456 }
457
458 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
459                 struct nfit_table_prev *prev,
460                 struct acpi_nfit_control_region *dcr)
461 {
462         struct device *dev = acpi_desc->dev;
463         struct nfit_dcr *nfit_dcr;
464
465         if (!sizeof_dcr(dcr))
466                 return false;
467
468         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
469                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
470                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
471                         return true;
472                 }
473
474         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
475                         GFP_KERNEL);
476         if (!nfit_dcr)
477                 return false;
478         INIT_LIST_HEAD(&nfit_dcr->list);
479         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
480         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
481         dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
482                         dcr->region_index, dcr->windows);
483         return true;
484 }
485
486 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
487                 struct nfit_table_prev *prev,
488                 struct acpi_nfit_data_region *bdw)
489 {
490         struct device *dev = acpi_desc->dev;
491         struct nfit_bdw *nfit_bdw;
492
493         if (bdw->header.length != sizeof(*bdw))
494                 return false;
495         list_for_each_entry(nfit_bdw, &prev->bdws, list)
496                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
497                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
498                         return true;
499                 }
500
501         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
502                         GFP_KERNEL);
503         if (!nfit_bdw)
504                 return false;
505         INIT_LIST_HEAD(&nfit_bdw->list);
506         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
507         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
508         dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
509                         bdw->region_index, bdw->windows);
510         return true;
511 }
512
513 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
514 {
515         if (idt->header.length < sizeof(*idt))
516                 return 0;
517         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
518 }
519
520 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
521                 struct nfit_table_prev *prev,
522                 struct acpi_nfit_interleave *idt)
523 {
524         struct device *dev = acpi_desc->dev;
525         struct nfit_idt *nfit_idt;
526
527         if (!sizeof_idt(idt))
528                 return false;
529
530         list_for_each_entry(nfit_idt, &prev->idts, list) {
531                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
532                         continue;
533
534                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
535                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
536                         return true;
537                 }
538         }
539
540         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
541                         GFP_KERNEL);
542         if (!nfit_idt)
543                 return false;
544         INIT_LIST_HEAD(&nfit_idt->list);
545         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
546         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
547         dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
548                         idt->interleave_index, idt->line_count);
549         return true;
550 }
551
552 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
553 {
554         if (flush->header.length < sizeof(*flush))
555                 return 0;
556         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
557 }
558
559 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
560                 struct nfit_table_prev *prev,
561                 struct acpi_nfit_flush_address *flush)
562 {
563         struct device *dev = acpi_desc->dev;
564         struct nfit_flush *nfit_flush;
565
566         if (!sizeof_flush(flush))
567                 return false;
568
569         list_for_each_entry(nfit_flush, &prev->flushes, list) {
570                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
571                         continue;
572
573                 if (memcmp(nfit_flush->flush, flush,
574                                         sizeof_flush(flush)) == 0) {
575                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
576                         return true;
577                 }
578         }
579
580         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
581                         + sizeof_flush(flush), GFP_KERNEL);
582         if (!nfit_flush)
583                 return false;
584         INIT_LIST_HEAD(&nfit_flush->list);
585         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
586         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
587         dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
588                         flush->device_handle, flush->hint_count);
589         return true;
590 }
591
592 static void *add_table(struct acpi_nfit_desc *acpi_desc,
593                 struct nfit_table_prev *prev, void *table, const void *end)
594 {
595         struct device *dev = acpi_desc->dev;
596         struct acpi_nfit_header *hdr;
597         void *err = ERR_PTR(-ENOMEM);
598
599         if (table >= end)
600                 return NULL;
601
602         hdr = table;
603         if (!hdr->length) {
604                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
605                         hdr->type);
606                 return NULL;
607         }
608
609         switch (hdr->type) {
610         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
611                 if (!add_spa(acpi_desc, prev, table))
612                         return err;
613                 break;
614         case ACPI_NFIT_TYPE_MEMORY_MAP:
615                 if (!add_memdev(acpi_desc, prev, table))
616                         return err;
617                 break;
618         case ACPI_NFIT_TYPE_CONTROL_REGION:
619                 if (!add_dcr(acpi_desc, prev, table))
620                         return err;
621                 break;
622         case ACPI_NFIT_TYPE_DATA_REGION:
623                 if (!add_bdw(acpi_desc, prev, table))
624                         return err;
625                 break;
626         case ACPI_NFIT_TYPE_INTERLEAVE:
627                 if (!add_idt(acpi_desc, prev, table))
628                         return err;
629                 break;
630         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
631                 if (!add_flush(acpi_desc, prev, table))
632                         return err;
633                 break;
634         case ACPI_NFIT_TYPE_SMBIOS:
635                 dev_dbg(dev, "%s: smbios\n", __func__);
636                 break;
637         default:
638                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
639                 break;
640         }
641
642         return table + hdr->length;
643 }
644
645 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
646                 struct nfit_mem *nfit_mem)
647 {
648         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
649         u16 dcr = nfit_mem->dcr->region_index;
650         struct nfit_spa *nfit_spa;
651
652         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
653                 u16 range_index = nfit_spa->spa->range_index;
654                 int type = nfit_spa_type(nfit_spa->spa);
655                 struct nfit_memdev *nfit_memdev;
656
657                 if (type != NFIT_SPA_BDW)
658                         continue;
659
660                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
661                         if (nfit_memdev->memdev->range_index != range_index)
662                                 continue;
663                         if (nfit_memdev->memdev->device_handle != device_handle)
664                                 continue;
665                         if (nfit_memdev->memdev->region_index != dcr)
666                                 continue;
667
668                         nfit_mem->spa_bdw = nfit_spa->spa;
669                         return;
670                 }
671         }
672
673         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
674                         nfit_mem->spa_dcr->range_index);
675         nfit_mem->bdw = NULL;
676 }
677
678 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
679                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
680 {
681         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
682         struct nfit_memdev *nfit_memdev;
683         struct nfit_bdw *nfit_bdw;
684         struct nfit_idt *nfit_idt;
685         u16 idt_idx, range_index;
686
687         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
688                 if (nfit_bdw->bdw->region_index != dcr)
689                         continue;
690                 nfit_mem->bdw = nfit_bdw->bdw;
691                 break;
692         }
693
694         if (!nfit_mem->bdw)
695                 return;
696
697         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
698
699         if (!nfit_mem->spa_bdw)
700                 return;
701
702         range_index = nfit_mem->spa_bdw->range_index;
703         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
704                 if (nfit_memdev->memdev->range_index != range_index ||
705                                 nfit_memdev->memdev->region_index != dcr)
706                         continue;
707                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
708                 idt_idx = nfit_memdev->memdev->interleave_index;
709                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
710                         if (nfit_idt->idt->interleave_index != idt_idx)
711                                 continue;
712                         nfit_mem->idt_bdw = nfit_idt->idt;
713                         break;
714                 }
715                 break;
716         }
717 }
718
719 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
720                 struct acpi_nfit_system_address *spa)
721 {
722         struct nfit_mem *nfit_mem, *found;
723         struct nfit_memdev *nfit_memdev;
724         int type = nfit_spa_type(spa);
725
726         switch (type) {
727         case NFIT_SPA_DCR:
728         case NFIT_SPA_PM:
729                 break;
730         default:
731                 return 0;
732         }
733
734         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
735                 struct nfit_flush *nfit_flush;
736                 struct nfit_dcr *nfit_dcr;
737                 u32 device_handle;
738                 u16 dcr;
739
740                 if (nfit_memdev->memdev->range_index != spa->range_index)
741                         continue;
742                 found = NULL;
743                 dcr = nfit_memdev->memdev->region_index;
744                 device_handle = nfit_memdev->memdev->device_handle;
745                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
746                         if (__to_nfit_memdev(nfit_mem)->device_handle
747                                         == device_handle) {
748                                 found = nfit_mem;
749                                 break;
750                         }
751
752                 if (found)
753                         nfit_mem = found;
754                 else {
755                         nfit_mem = devm_kzalloc(acpi_desc->dev,
756                                         sizeof(*nfit_mem), GFP_KERNEL);
757                         if (!nfit_mem)
758                                 return -ENOMEM;
759                         INIT_LIST_HEAD(&nfit_mem->list);
760                         nfit_mem->acpi_desc = acpi_desc;
761                         list_add(&nfit_mem->list, &acpi_desc->dimms);
762                 }
763
764                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
765                         if (nfit_dcr->dcr->region_index != dcr)
766                                 continue;
767                         /*
768                          * Record the control region for the dimm.  For
769                          * the ACPI 6.1 case, where there are separate
770                          * control regions for the pmem vs blk
771                          * interfaces, be sure to record the extended
772                          * blk details.
773                          */
774                         if (!nfit_mem->dcr)
775                                 nfit_mem->dcr = nfit_dcr->dcr;
776                         else if (nfit_mem->dcr->windows == 0
777                                         && nfit_dcr->dcr->windows)
778                                 nfit_mem->dcr = nfit_dcr->dcr;
779                         break;
780                 }
781
782                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
783                         struct acpi_nfit_flush_address *flush;
784                         u16 i;
785
786                         if (nfit_flush->flush->device_handle != device_handle)
787                                 continue;
788                         nfit_mem->nfit_flush = nfit_flush;
789                         flush = nfit_flush->flush;
790                         nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
791                                         flush->hint_count
792                                         * sizeof(struct resource), GFP_KERNEL);
793                         if (!nfit_mem->flush_wpq)
794                                 return -ENOMEM;
795                         for (i = 0; i < flush->hint_count; i++) {
796                                 struct resource *res = &nfit_mem->flush_wpq[i];
797
798                                 res->start = flush->hint_address[i];
799                                 res->end = res->start + 8 - 1;
800                         }
801                         break;
802                 }
803
804                 if (dcr && !nfit_mem->dcr) {
805                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
806                                         spa->range_index, dcr);
807                         return -ENODEV;
808                 }
809
810                 if (type == NFIT_SPA_DCR) {
811                         struct nfit_idt *nfit_idt;
812                         u16 idt_idx;
813
814                         /* multiple dimms may share a SPA when interleaved */
815                         nfit_mem->spa_dcr = spa;
816                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
817                         idt_idx = nfit_memdev->memdev->interleave_index;
818                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
819                                 if (nfit_idt->idt->interleave_index != idt_idx)
820                                         continue;
821                                 nfit_mem->idt_dcr = nfit_idt->idt;
822                                 break;
823                         }
824                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
825                 } else {
826                         /*
827                          * A single dimm may belong to multiple SPA-PM
828                          * ranges, record at least one in addition to
829                          * any SPA-DCR range.
830                          */
831                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
832                 }
833         }
834
835         return 0;
836 }
837
838 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
839 {
840         struct nfit_mem *a = container_of(_a, typeof(*a), list);
841         struct nfit_mem *b = container_of(_b, typeof(*b), list);
842         u32 handleA, handleB;
843
844         handleA = __to_nfit_memdev(a)->device_handle;
845         handleB = __to_nfit_memdev(b)->device_handle;
846         if (handleA < handleB)
847                 return -1;
848         else if (handleA > handleB)
849                 return 1;
850         return 0;
851 }
852
853 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
854 {
855         struct nfit_spa *nfit_spa;
856
857         /*
858          * For each SPA-DCR or SPA-PMEM address range find its
859          * corresponding MEMDEV(s).  From each MEMDEV find the
860          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
861          * try to find a SPA-BDW and a corresponding BDW that references
862          * the DCR.  Throw it all into an nfit_mem object.  Note, that
863          * BDWs are optional.
864          */
865         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
866                 int rc;
867
868                 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
869                 if (rc)
870                         return rc;
871         }
872
873         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
874
875         return 0;
876 }
877
878 static ssize_t revision_show(struct device *dev,
879                 struct device_attribute *attr, char *buf)
880 {
881         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
882         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
883         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
884
885         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
886 }
887 static DEVICE_ATTR_RO(revision);
888
889 static ssize_t hw_error_scrub_show(struct device *dev,
890                 struct device_attribute *attr, char *buf)
891 {
892         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
893         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
894         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
895
896         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
897 }
898
899 /*
900  * The 'hw_error_scrub' attribute can have the following values written to it:
901  * '0': Switch to the default mode where an exception will only insert
902  *      the address of the memory error into the poison and badblocks lists.
903  * '1': Enable a full scrub to happen if an exception for a memory error is
904  *      received.
905  */
906 static ssize_t hw_error_scrub_store(struct device *dev,
907                 struct device_attribute *attr, const char *buf, size_t size)
908 {
909         struct nvdimm_bus_descriptor *nd_desc;
910         ssize_t rc;
911         long val;
912
913         rc = kstrtol(buf, 0, &val);
914         if (rc)
915                 return rc;
916
917         device_lock(dev);
918         nd_desc = dev_get_drvdata(dev);
919         if (nd_desc) {
920                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
921
922                 switch (val) {
923                 case HW_ERROR_SCRUB_ON:
924                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
925                         break;
926                 case HW_ERROR_SCRUB_OFF:
927                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
928                         break;
929                 default:
930                         rc = -EINVAL;
931                         break;
932                 }
933         }
934         device_unlock(dev);
935         if (rc)
936                 return rc;
937         return size;
938 }
939 static DEVICE_ATTR_RW(hw_error_scrub);
940
941 /*
942  * This shows the number of full Address Range Scrubs that have been
943  * completed since driver load time. Userspace can wait on this using
944  * select/poll etc. A '+' at the end indicates an ARS is in progress
945  */
946 static ssize_t scrub_show(struct device *dev,
947                 struct device_attribute *attr, char *buf)
948 {
949         struct nvdimm_bus_descriptor *nd_desc;
950         ssize_t rc = -ENXIO;
951
952         device_lock(dev);
953         nd_desc = dev_get_drvdata(dev);
954         if (nd_desc) {
955                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
956
957                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
958                                 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
959         }
960         device_unlock(dev);
961         return rc;
962 }
963
964 static ssize_t scrub_store(struct device *dev,
965                 struct device_attribute *attr, const char *buf, size_t size)
966 {
967         struct nvdimm_bus_descriptor *nd_desc;
968         ssize_t rc;
969         long val;
970
971         rc = kstrtol(buf, 0, &val);
972         if (rc)
973                 return rc;
974         if (val != 1)
975                 return -EINVAL;
976
977         device_lock(dev);
978         nd_desc = dev_get_drvdata(dev);
979         if (nd_desc) {
980                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
981
982                 rc = acpi_nfit_ars_rescan(acpi_desc);
983         }
984         device_unlock(dev);
985         if (rc)
986                 return rc;
987         return size;
988 }
989 static DEVICE_ATTR_RW(scrub);
990
991 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
992 {
993         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
994         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
995                 | 1 << ND_CMD_ARS_STATUS;
996
997         return (nd_desc->cmd_mask & mask) == mask;
998 }
999
1000 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1001 {
1002         struct device *dev = container_of(kobj, struct device, kobj);
1003         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1004
1005         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1006                 return 0;
1007         return a->mode;
1008 }
1009
1010 static struct attribute *acpi_nfit_attributes[] = {
1011         &dev_attr_revision.attr,
1012         &dev_attr_scrub.attr,
1013         &dev_attr_hw_error_scrub.attr,
1014         NULL,
1015 };
1016
1017 static struct attribute_group acpi_nfit_attribute_group = {
1018         .name = "nfit",
1019         .attrs = acpi_nfit_attributes,
1020         .is_visible = nfit_visible,
1021 };
1022
1023 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1024         &nvdimm_bus_attribute_group,
1025         &acpi_nfit_attribute_group,
1026         NULL,
1027 };
1028
1029 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1030 {
1031         struct nvdimm *nvdimm = to_nvdimm(dev);
1032         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1033
1034         return __to_nfit_memdev(nfit_mem);
1035 }
1036
1037 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1038 {
1039         struct nvdimm *nvdimm = to_nvdimm(dev);
1040         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1041
1042         return nfit_mem->dcr;
1043 }
1044
1045 static ssize_t handle_show(struct device *dev,
1046                 struct device_attribute *attr, char *buf)
1047 {
1048         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1049
1050         return sprintf(buf, "%#x\n", memdev->device_handle);
1051 }
1052 static DEVICE_ATTR_RO(handle);
1053
1054 static ssize_t phys_id_show(struct device *dev,
1055                 struct device_attribute *attr, char *buf)
1056 {
1057         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1058
1059         return sprintf(buf, "%#x\n", memdev->physical_id);
1060 }
1061 static DEVICE_ATTR_RO(phys_id);
1062
1063 static ssize_t vendor_show(struct device *dev,
1064                 struct device_attribute *attr, char *buf)
1065 {
1066         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1067
1068         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1069 }
1070 static DEVICE_ATTR_RO(vendor);
1071
1072 static ssize_t rev_id_show(struct device *dev,
1073                 struct device_attribute *attr, char *buf)
1074 {
1075         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1076
1077         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1078 }
1079 static DEVICE_ATTR_RO(rev_id);
1080
1081 static ssize_t device_show(struct device *dev,
1082                 struct device_attribute *attr, char *buf)
1083 {
1084         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1085
1086         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1087 }
1088 static DEVICE_ATTR_RO(device);
1089
1090 static ssize_t subsystem_vendor_show(struct device *dev,
1091                 struct device_attribute *attr, char *buf)
1092 {
1093         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1094
1095         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1096 }
1097 static DEVICE_ATTR_RO(subsystem_vendor);
1098
1099 static ssize_t subsystem_rev_id_show(struct device *dev,
1100                 struct device_attribute *attr, char *buf)
1101 {
1102         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1103
1104         return sprintf(buf, "0x%04x\n",
1105                         be16_to_cpu(dcr->subsystem_revision_id));
1106 }
1107 static DEVICE_ATTR_RO(subsystem_rev_id);
1108
1109 static ssize_t subsystem_device_show(struct device *dev,
1110                 struct device_attribute *attr, char *buf)
1111 {
1112         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1113
1114         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1115 }
1116 static DEVICE_ATTR_RO(subsystem_device);
1117
1118 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1119 {
1120         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1121         int formats = 0;
1122
1123         if (nfit_mem->memdev_pmem)
1124                 formats++;
1125         if (nfit_mem->memdev_bdw)
1126                 formats++;
1127         return formats;
1128 }
1129
1130 static ssize_t format_show(struct device *dev,
1131                 struct device_attribute *attr, char *buf)
1132 {
1133         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1134
1135         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1136 }
1137 static DEVICE_ATTR_RO(format);
1138
1139 static ssize_t format1_show(struct device *dev,
1140                 struct device_attribute *attr, char *buf)
1141 {
1142         u32 handle;
1143         ssize_t rc = -ENXIO;
1144         struct nfit_mem *nfit_mem;
1145         struct nfit_memdev *nfit_memdev;
1146         struct acpi_nfit_desc *acpi_desc;
1147         struct nvdimm *nvdimm = to_nvdimm(dev);
1148         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1149
1150         nfit_mem = nvdimm_provider_data(nvdimm);
1151         acpi_desc = nfit_mem->acpi_desc;
1152         handle = to_nfit_memdev(dev)->device_handle;
1153
1154         /* assumes DIMMs have at most 2 published interface codes */
1155         mutex_lock(&acpi_desc->init_mutex);
1156         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1157                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1158                 struct nfit_dcr *nfit_dcr;
1159
1160                 if (memdev->device_handle != handle)
1161                         continue;
1162
1163                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1164                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1165                                 continue;
1166                         if (nfit_dcr->dcr->code == dcr->code)
1167                                 continue;
1168                         rc = sprintf(buf, "0x%04x\n",
1169                                         le16_to_cpu(nfit_dcr->dcr->code));
1170                         break;
1171                 }
1172                 if (rc != ENXIO)
1173                         break;
1174         }
1175         mutex_unlock(&acpi_desc->init_mutex);
1176         return rc;
1177 }
1178 static DEVICE_ATTR_RO(format1);
1179
1180 static ssize_t formats_show(struct device *dev,
1181                 struct device_attribute *attr, char *buf)
1182 {
1183         struct nvdimm *nvdimm = to_nvdimm(dev);
1184
1185         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1186 }
1187 static DEVICE_ATTR_RO(formats);
1188
1189 static ssize_t serial_show(struct device *dev,
1190                 struct device_attribute *attr, char *buf)
1191 {
1192         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1193
1194         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1195 }
1196 static DEVICE_ATTR_RO(serial);
1197
1198 static ssize_t family_show(struct device *dev,
1199                 struct device_attribute *attr, char *buf)
1200 {
1201         struct nvdimm *nvdimm = to_nvdimm(dev);
1202         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1203
1204         if (nfit_mem->family < 0)
1205                 return -ENXIO;
1206         return sprintf(buf, "%d\n", nfit_mem->family);
1207 }
1208 static DEVICE_ATTR_RO(family);
1209
1210 static ssize_t dsm_mask_show(struct device *dev,
1211                 struct device_attribute *attr, char *buf)
1212 {
1213         struct nvdimm *nvdimm = to_nvdimm(dev);
1214         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1215
1216         if (nfit_mem->family < 0)
1217                 return -ENXIO;
1218         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1219 }
1220 static DEVICE_ATTR_RO(dsm_mask);
1221
1222 static ssize_t flags_show(struct device *dev,
1223                 struct device_attribute *attr, char *buf)
1224 {
1225         u16 flags = to_nfit_memdev(dev)->flags;
1226
1227         return sprintf(buf, "%s%s%s%s%s\n",
1228                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1229                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1230                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1231                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1232                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
1233 }
1234 static DEVICE_ATTR_RO(flags);
1235
1236 static ssize_t id_show(struct device *dev,
1237                 struct device_attribute *attr, char *buf)
1238 {
1239         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1240
1241         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1242                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1243                                 be16_to_cpu(dcr->vendor_id),
1244                                 dcr->manufacturing_location,
1245                                 be16_to_cpu(dcr->manufacturing_date),
1246                                 be32_to_cpu(dcr->serial_number));
1247         else
1248                 return sprintf(buf, "%04x-%08x\n",
1249                                 be16_to_cpu(dcr->vendor_id),
1250                                 be32_to_cpu(dcr->serial_number));
1251 }
1252 static DEVICE_ATTR_RO(id);
1253
1254 static struct attribute *acpi_nfit_dimm_attributes[] = {
1255         &dev_attr_handle.attr,
1256         &dev_attr_phys_id.attr,
1257         &dev_attr_vendor.attr,
1258         &dev_attr_device.attr,
1259         &dev_attr_rev_id.attr,
1260         &dev_attr_subsystem_vendor.attr,
1261         &dev_attr_subsystem_device.attr,
1262         &dev_attr_subsystem_rev_id.attr,
1263         &dev_attr_format.attr,
1264         &dev_attr_formats.attr,
1265         &dev_attr_format1.attr,
1266         &dev_attr_serial.attr,
1267         &dev_attr_flags.attr,
1268         &dev_attr_id.attr,
1269         &dev_attr_family.attr,
1270         &dev_attr_dsm_mask.attr,
1271         NULL,
1272 };
1273
1274 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1275                 struct attribute *a, int n)
1276 {
1277         struct device *dev = container_of(kobj, struct device, kobj);
1278         struct nvdimm *nvdimm = to_nvdimm(dev);
1279
1280         if (!to_nfit_dcr(dev))
1281                 return 0;
1282         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1283                 return 0;
1284         return a->mode;
1285 }
1286
1287 static struct attribute_group acpi_nfit_dimm_attribute_group = {
1288         .name = "nfit",
1289         .attrs = acpi_nfit_dimm_attributes,
1290         .is_visible = acpi_nfit_dimm_attr_visible,
1291 };
1292
1293 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1294         &nvdimm_attribute_group,
1295         &nd_device_attribute_group,
1296         &acpi_nfit_dimm_attribute_group,
1297         NULL,
1298 };
1299
1300 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1301                 u32 device_handle)
1302 {
1303         struct nfit_mem *nfit_mem;
1304
1305         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1306                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1307                         return nfit_mem->nvdimm;
1308
1309         return NULL;
1310 }
1311
1312 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1313 {
1314         struct nfit_mem *nfit_mem;
1315         struct acpi_nfit_desc *acpi_desc;
1316
1317         dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1318                         event);
1319
1320         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1321                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1322                                 event);
1323                 return;
1324         }
1325
1326         acpi_desc = dev_get_drvdata(dev->parent);
1327         if (!acpi_desc)
1328                 return;
1329
1330         /*
1331          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1332          * is still valid.
1333          */
1334         nfit_mem = dev_get_drvdata(dev);
1335         if (nfit_mem && nfit_mem->flags_attr)
1336                 sysfs_notify_dirent(nfit_mem->flags_attr);
1337 }
1338 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1339
1340 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1341 {
1342         struct acpi_device *adev = data;
1343         struct device *dev = &adev->dev;
1344
1345         device_lock(dev->parent);
1346         __acpi_nvdimm_notify(dev, event);
1347         device_unlock(dev->parent);
1348 }
1349
1350 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1351                 struct nfit_mem *nfit_mem, u32 device_handle)
1352 {
1353         struct acpi_device *adev, *adev_dimm;
1354         struct device *dev = acpi_desc->dev;
1355         unsigned long dsm_mask;
1356         const u8 *uuid;
1357         int i;
1358
1359         /* nfit test assumes 1:1 relationship between commands and dsms */
1360         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1361         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1362         adev = to_acpi_dev(acpi_desc);
1363         if (!adev)
1364                 return 0;
1365
1366         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1367         nfit_mem->adev = adev_dimm;
1368         if (!adev_dimm) {
1369                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1370                                 device_handle);
1371                 return force_enable_dimms ? 0 : -ENODEV;
1372         }
1373
1374         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1375                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1376                 dev_err(dev, "%s: notification registration failed\n",
1377                                 dev_name(&adev_dimm->dev));
1378                 return -ENXIO;
1379         }
1380
1381         /*
1382          * Until standardization materializes we need to consider 4
1383          * different command sets.  Note, that checking for function0 (bit0)
1384          * tells us if any commands are reachable through this uuid.
1385          */
1386         for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1387                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1388                         break;
1389
1390         /* limit the supported commands to those that are publicly documented */
1391         nfit_mem->family = i;
1392         if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1393                 dsm_mask = 0x3fe;
1394                 if (disable_vendor_specific)
1395                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1396         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1397                 dsm_mask = 0x1c3c76;
1398         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1399                 dsm_mask = 0x1fe;
1400                 if (disable_vendor_specific)
1401                         dsm_mask &= ~(1 << 8);
1402         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1403                 dsm_mask = 0xffffffff;
1404         } else {
1405                 dev_dbg(dev, "unknown dimm command family\n");
1406                 nfit_mem->family = -1;
1407                 /* DSMs are optional, continue loading the driver... */
1408                 return 0;
1409         }
1410
1411         uuid = to_nfit_uuid(nfit_mem->family);
1412         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1413                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1414                         set_bit(i, &nfit_mem->dsm_mask);
1415
1416         return 0;
1417 }
1418
1419 static void shutdown_dimm_notify(void *data)
1420 {
1421         struct acpi_nfit_desc *acpi_desc = data;
1422         struct nfit_mem *nfit_mem;
1423
1424         mutex_lock(&acpi_desc->init_mutex);
1425         /*
1426          * Clear out the nfit_mem->flags_attr and shut down dimm event
1427          * notifications.
1428          */
1429         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1430                 struct acpi_device *adev_dimm = nfit_mem->adev;
1431
1432                 if (nfit_mem->flags_attr) {
1433                         sysfs_put(nfit_mem->flags_attr);
1434                         nfit_mem->flags_attr = NULL;
1435                 }
1436                 if (adev_dimm)
1437                         acpi_remove_notify_handler(adev_dimm->handle,
1438                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1439         }
1440         mutex_unlock(&acpi_desc->init_mutex);
1441 }
1442
1443 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1444 {
1445         struct nfit_mem *nfit_mem;
1446         int dimm_count = 0, rc;
1447         struct nvdimm *nvdimm;
1448
1449         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1450                 struct acpi_nfit_flush_address *flush;
1451                 unsigned long flags = 0, cmd_mask;
1452                 u32 device_handle;
1453                 u16 mem_flags;
1454
1455                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1456                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1457                 if (nvdimm) {
1458                         dimm_count++;
1459                         continue;
1460                 }
1461
1462                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1463                         flags |= NDD_ALIASING;
1464
1465                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1466                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1467                         flags |= NDD_UNARMED;
1468
1469                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1470                 if (rc)
1471                         continue;
1472
1473                 /*
1474                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1475                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1476                  * userspace interface.
1477                  */
1478                 cmd_mask = 1UL << ND_CMD_CALL;
1479                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1480                         cmd_mask |= nfit_mem->dsm_mask;
1481
1482                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1483                         : NULL;
1484                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1485                                 acpi_nfit_dimm_attribute_groups,
1486                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1487                                 nfit_mem->flush_wpq);
1488                 if (!nvdimm)
1489                         return -ENOMEM;
1490
1491                 nfit_mem->nvdimm = nvdimm;
1492                 dimm_count++;
1493
1494                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1495                         continue;
1496
1497                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1498                                 nvdimm_name(nvdimm),
1499                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1500                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1501                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1502                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1503
1504         }
1505
1506         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1507         if (rc)
1508                 return rc;
1509
1510         /*
1511          * Now that dimms are successfully registered, and async registration
1512          * is flushed, attempt to enable event notification.
1513          */
1514         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1515                 struct kernfs_node *nfit_kernfs;
1516
1517                 nvdimm = nfit_mem->nvdimm;
1518                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1519                 if (nfit_kernfs)
1520                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1521                                         "flags");
1522                 sysfs_put(nfit_kernfs);
1523                 if (!nfit_mem->flags_attr)
1524                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1525                                         nvdimm_name(nvdimm));
1526         }
1527
1528         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1529                         acpi_desc);
1530 }
1531
1532 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1533 {
1534         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1535         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1536         struct acpi_device *adev;
1537         int i;
1538
1539         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1540         adev = to_acpi_dev(acpi_desc);
1541         if (!adev)
1542                 return;
1543
1544         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1545                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1546                         set_bit(i, &nd_desc->cmd_mask);
1547 }
1548
1549 static ssize_t range_index_show(struct device *dev,
1550                 struct device_attribute *attr, char *buf)
1551 {
1552         struct nd_region *nd_region = to_nd_region(dev);
1553         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1554
1555         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1556 }
1557 static DEVICE_ATTR_RO(range_index);
1558
1559 static struct attribute *acpi_nfit_region_attributes[] = {
1560         &dev_attr_range_index.attr,
1561         NULL,
1562 };
1563
1564 static struct attribute_group acpi_nfit_region_attribute_group = {
1565         .name = "nfit",
1566         .attrs = acpi_nfit_region_attributes,
1567 };
1568
1569 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1570         &nd_region_attribute_group,
1571         &nd_mapping_attribute_group,
1572         &nd_device_attribute_group,
1573         &nd_numa_attribute_group,
1574         &acpi_nfit_region_attribute_group,
1575         NULL,
1576 };
1577
1578 /* enough info to uniquely specify an interleave set */
1579 struct nfit_set_info {
1580         struct nfit_set_info_map {
1581                 u64 region_offset;
1582                 u32 serial_number;
1583                 u32 pad;
1584         } mapping[0];
1585 };
1586
1587 static size_t sizeof_nfit_set_info(int num_mappings)
1588 {
1589         return sizeof(struct nfit_set_info)
1590                 + num_mappings * sizeof(struct nfit_set_info_map);
1591 }
1592
1593 static int cmp_map(const void *m0, const void *m1)
1594 {
1595         const struct nfit_set_info_map *map0 = m0;
1596         const struct nfit_set_info_map *map1 = m1;
1597
1598         return memcmp(&map0->region_offset, &map1->region_offset,
1599                         sizeof(u64));
1600 }
1601
1602 /* Retrieve the nth entry referencing this spa */
1603 static struct acpi_nfit_memory_map *memdev_from_spa(
1604                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1605 {
1606         struct nfit_memdev *nfit_memdev;
1607
1608         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1609                 if (nfit_memdev->memdev->range_index == range_index)
1610                         if (n-- == 0)
1611                                 return nfit_memdev->memdev;
1612         return NULL;
1613 }
1614
1615 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1616                 struct nd_region_desc *ndr_desc,
1617                 struct acpi_nfit_system_address *spa)
1618 {
1619         int i, spa_type = nfit_spa_type(spa);
1620         struct device *dev = acpi_desc->dev;
1621         struct nd_interleave_set *nd_set;
1622         u16 nr = ndr_desc->num_mappings;
1623         struct nfit_set_info *info;
1624
1625         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1626                 /* pass */;
1627         else
1628                 return 0;
1629
1630         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1631         if (!nd_set)
1632                 return -ENOMEM;
1633
1634         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1635         if (!info)
1636                 return -ENOMEM;
1637         for (i = 0; i < nr; i++) {
1638                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1639                 struct nfit_set_info_map *map = &info->mapping[i];
1640                 struct nvdimm *nvdimm = mapping->nvdimm;
1641                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1642                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1643                                 spa->range_index, i);
1644
1645                 if (!memdev || !nfit_mem->dcr) {
1646                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1647                         return -ENODEV;
1648                 }
1649
1650                 map->region_offset = memdev->region_offset;
1651                 map->serial_number = nfit_mem->dcr->serial_number;
1652         }
1653
1654         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1655                         cmp_map, NULL);
1656         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1657         ndr_desc->nd_set = nd_set;
1658         devm_kfree(dev, info);
1659
1660         return 0;
1661 }
1662
1663 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1664 {
1665         struct acpi_nfit_interleave *idt = mmio->idt;
1666         u32 sub_line_offset, line_index, line_offset;
1667         u64 line_no, table_skip_count, table_offset;
1668
1669         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1670         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1671         line_offset = idt->line_offset[line_index]
1672                 * mmio->line_size;
1673         table_offset = table_skip_count * mmio->table_size;
1674
1675         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1676 }
1677
1678 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1679 {
1680         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1681         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1682         const u32 STATUS_MASK = 0x80000037;
1683
1684         if (mmio->num_lines)
1685                 offset = to_interleave_offset(offset, mmio);
1686
1687         return readl(mmio->addr.base + offset) & STATUS_MASK;
1688 }
1689
1690 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1691                 resource_size_t dpa, unsigned int len, unsigned int write)
1692 {
1693         u64 cmd, offset;
1694         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1695
1696         enum {
1697                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1698                 BCW_LEN_SHIFT = 48,
1699                 BCW_LEN_MASK = (1ULL << 8) - 1,
1700                 BCW_CMD_SHIFT = 56,
1701         };
1702
1703         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1704         len = len >> L1_CACHE_SHIFT;
1705         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1706         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1707
1708         offset = nfit_blk->cmd_offset + mmio->size * bw;
1709         if (mmio->num_lines)
1710                 offset = to_interleave_offset(offset, mmio);
1711
1712         writeq(cmd, mmio->addr.base + offset);
1713         nvdimm_flush(nfit_blk->nd_region);
1714
1715         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1716                 readq(mmio->addr.base + offset);
1717 }
1718
1719 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1720                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1721                 unsigned int lane)
1722 {
1723         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1724         unsigned int copied = 0;
1725         u64 base_offset;
1726         int rc;
1727
1728         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1729                 + lane * mmio->size;
1730         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1731         while (len) {
1732                 unsigned int c;
1733                 u64 offset;
1734
1735                 if (mmio->num_lines) {
1736                         u32 line_offset;
1737
1738                         offset = to_interleave_offset(base_offset + copied,
1739                                         mmio);
1740                         div_u64_rem(offset, mmio->line_size, &line_offset);
1741                         c = min_t(size_t, len, mmio->line_size - line_offset);
1742                 } else {
1743                         offset = base_offset + nfit_blk->bdw_offset;
1744                         c = len;
1745                 }
1746
1747                 if (rw)
1748                         memcpy_to_pmem(mmio->addr.aperture + offset,
1749                                         iobuf + copied, c);
1750                 else {
1751                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1752                                 mmio_flush_range((void __force *)
1753                                         mmio->addr.aperture + offset, c);
1754
1755                         memcpy_from_pmem(iobuf + copied,
1756                                         mmio->addr.aperture + offset, c);
1757                 }
1758
1759                 copied += c;
1760                 len -= c;
1761         }
1762
1763         if (rw)
1764                 nvdimm_flush(nfit_blk->nd_region);
1765
1766         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1767         return rc;
1768 }
1769
1770 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1771                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1772 {
1773         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1774         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1775         struct nd_region *nd_region = nfit_blk->nd_region;
1776         unsigned int lane, copied = 0;
1777         int rc = 0;
1778
1779         lane = nd_region_acquire_lane(nd_region);
1780         while (len) {
1781                 u64 c = min(len, mmio->size);
1782
1783                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1784                                 iobuf + copied, c, rw, lane);
1785                 if (rc)
1786                         break;
1787
1788                 copied += c;
1789                 len -= c;
1790         }
1791         nd_region_release_lane(nd_region, lane);
1792
1793         return rc;
1794 }
1795
1796 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1797                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1798 {
1799         if (idt) {
1800                 mmio->num_lines = idt->line_count;
1801                 mmio->line_size = idt->line_size;
1802                 if (interleave_ways == 0)
1803                         return -ENXIO;
1804                 mmio->table_size = mmio->num_lines * interleave_ways
1805                         * mmio->line_size;
1806         }
1807
1808         return 0;
1809 }
1810
1811 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1812                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1813 {
1814         struct nd_cmd_dimm_flags flags;
1815         int rc;
1816
1817         memset(&flags, 0, sizeof(flags));
1818         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1819                         sizeof(flags), NULL);
1820
1821         if (rc >= 0 && flags.status == 0)
1822                 nfit_blk->dimm_flags = flags.flags;
1823         else if (rc == -ENOTTY) {
1824                 /* fall back to a conservative default */
1825                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1826                 rc = 0;
1827         } else
1828                 rc = -ENXIO;
1829
1830         return rc;
1831 }
1832
1833 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1834                 struct device *dev)
1835 {
1836         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1837         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1838         struct nfit_blk_mmio *mmio;
1839         struct nfit_blk *nfit_blk;
1840         struct nfit_mem *nfit_mem;
1841         struct nvdimm *nvdimm;
1842         int rc;
1843
1844         nvdimm = nd_blk_region_to_dimm(ndbr);
1845         nfit_mem = nvdimm_provider_data(nvdimm);
1846         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1847                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1848                                 nfit_mem ? "" : " nfit_mem",
1849                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1850                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1851                 return -ENXIO;
1852         }
1853
1854         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1855         if (!nfit_blk)
1856                 return -ENOMEM;
1857         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1858         nfit_blk->nd_region = to_nd_region(dev);
1859
1860         /* map block aperture memory */
1861         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1862         mmio = &nfit_blk->mmio[BDW];
1863         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
1864                         nfit_mem->spa_bdw->length, ARCH_MEMREMAP_PMEM);
1865         if (!mmio->addr.base) {
1866                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1867                                 nvdimm_name(nvdimm));
1868                 return -ENOMEM;
1869         }
1870         mmio->size = nfit_mem->bdw->size;
1871         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1872         mmio->idt = nfit_mem->idt_bdw;
1873         mmio->spa = nfit_mem->spa_bdw;
1874         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1875                         nfit_mem->memdev_bdw->interleave_ways);
1876         if (rc) {
1877                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1878                                 __func__, nvdimm_name(nvdimm));
1879                 return rc;
1880         }
1881
1882         /* map block control memory */
1883         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1884         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1885         mmio = &nfit_blk->mmio[DCR];
1886         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
1887                         nfit_mem->spa_dcr->length);
1888         if (!mmio->addr.base) {
1889                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1890                                 nvdimm_name(nvdimm));
1891                 return -ENOMEM;
1892         }
1893         mmio->size = nfit_mem->dcr->window_size;
1894         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1895         mmio->idt = nfit_mem->idt_dcr;
1896         mmio->spa = nfit_mem->spa_dcr;
1897         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1898                         nfit_mem->memdev_dcr->interleave_ways);
1899         if (rc) {
1900                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1901                                 __func__, nvdimm_name(nvdimm));
1902                 return rc;
1903         }
1904
1905         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1906         if (rc < 0) {
1907                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1908                                 __func__, nvdimm_name(nvdimm));
1909                 return rc;
1910         }
1911
1912         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
1913                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1914
1915         if (mmio->line_size == 0)
1916                 return 0;
1917
1918         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1919                         + 8 > mmio->line_size) {
1920                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1921                 return -ENXIO;
1922         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1923                         + 8 > mmio->line_size) {
1924                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1925                 return -ENXIO;
1926         }
1927
1928         return 0;
1929 }
1930
1931 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1932                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1933 {
1934         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1935         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1936         int cmd_rc, rc;
1937
1938         cmd->address = spa->address;
1939         cmd->length = spa->length;
1940         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1941                         sizeof(*cmd), &cmd_rc);
1942         if (rc < 0)
1943                 return rc;
1944         return cmd_rc;
1945 }
1946
1947 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1948 {
1949         int rc;
1950         int cmd_rc;
1951         struct nd_cmd_ars_start ars_start;
1952         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1953         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1954
1955         memset(&ars_start, 0, sizeof(ars_start));
1956         ars_start.address = spa->address;
1957         ars_start.length = spa->length;
1958         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1959                 ars_start.type = ND_ARS_PERSISTENT;
1960         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1961                 ars_start.type = ND_ARS_VOLATILE;
1962         else
1963                 return -ENOTTY;
1964
1965         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1966                         sizeof(ars_start), &cmd_rc);
1967
1968         if (rc < 0)
1969                 return rc;
1970         return cmd_rc;
1971 }
1972
1973 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1974 {
1975         int rc, cmd_rc;
1976         struct nd_cmd_ars_start ars_start;
1977         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1978         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1979
1980         memset(&ars_start, 0, sizeof(ars_start));
1981         ars_start.address = ars_status->restart_address;
1982         ars_start.length = ars_status->restart_length;
1983         ars_start.type = ars_status->type;
1984         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1985                         sizeof(ars_start), &cmd_rc);
1986         if (rc < 0)
1987                 return rc;
1988         return cmd_rc;
1989 }
1990
1991 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1992 {
1993         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1994         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1995         int rc, cmd_rc;
1996
1997         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1998                         acpi_desc->ars_status_size, &cmd_rc);
1999         if (rc < 0)
2000                 return rc;
2001         return cmd_rc;
2002 }
2003
2004 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
2005                 struct nd_cmd_ars_status *ars_status)
2006 {
2007         int rc;
2008         u32 i;
2009
2010         for (i = 0; i < ars_status->num_records; i++) {
2011                 rc = nvdimm_bus_add_poison(nvdimm_bus,
2012                                 ars_status->records[i].err_address,
2013                                 ars_status->records[i].length);
2014                 if (rc)
2015                         return rc;
2016         }
2017
2018         return 0;
2019 }
2020
2021 static void acpi_nfit_remove_resource(void *data)
2022 {
2023         struct resource *res = data;
2024
2025         remove_resource(res);
2026 }
2027
2028 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2029                 struct nd_region_desc *ndr_desc)
2030 {
2031         struct resource *res, *nd_res = ndr_desc->res;
2032         int is_pmem, ret;
2033
2034         /* No operation if the region is already registered as PMEM */
2035         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2036                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2037         if (is_pmem == REGION_INTERSECTS)
2038                 return 0;
2039
2040         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2041         if (!res)
2042                 return -ENOMEM;
2043
2044         res->name = "Persistent Memory";
2045         res->start = nd_res->start;
2046         res->end = nd_res->end;
2047         res->flags = IORESOURCE_MEM;
2048         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2049
2050         ret = insert_resource(&iomem_resource, res);
2051         if (ret)
2052                 return ret;
2053
2054         ret = devm_add_action_or_reset(acpi_desc->dev,
2055                                         acpi_nfit_remove_resource,
2056                                         res);
2057         if (ret)
2058                 return ret;
2059
2060         return 0;
2061 }
2062
2063 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2064                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2065                 struct acpi_nfit_memory_map *memdev,
2066                 struct nfit_spa *nfit_spa)
2067 {
2068         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2069                         memdev->device_handle);
2070         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2071         struct nd_blk_region_desc *ndbr_desc;
2072         struct nfit_mem *nfit_mem;
2073         int blk_valid = 0;
2074
2075         if (!nvdimm) {
2076                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2077                                 spa->range_index, memdev->device_handle);
2078                 return -ENODEV;
2079         }
2080
2081         mapping->nvdimm = nvdimm;
2082         switch (nfit_spa_type(spa)) {
2083         case NFIT_SPA_PM:
2084         case NFIT_SPA_VOLATILE:
2085                 mapping->start = memdev->address;
2086                 mapping->size = memdev->region_size;
2087                 break;
2088         case NFIT_SPA_DCR:
2089                 nfit_mem = nvdimm_provider_data(nvdimm);
2090                 if (!nfit_mem || !nfit_mem->bdw) {
2091                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2092                                         spa->range_index, nvdimm_name(nvdimm));
2093                 } else {
2094                         mapping->size = nfit_mem->bdw->capacity;
2095                         mapping->start = nfit_mem->bdw->start_address;
2096                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
2097                         blk_valid = 1;
2098                 }
2099
2100                 ndr_desc->mapping = mapping;
2101                 ndr_desc->num_mappings = blk_valid;
2102                 ndbr_desc = to_blk_region_desc(ndr_desc);
2103                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2104                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2105                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2106                                 ndr_desc);
2107                 if (!nfit_spa->nd_region)
2108                         return -ENOMEM;
2109                 break;
2110         }
2111
2112         return 0;
2113 }
2114
2115 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2116 {
2117         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2118                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2119                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2120                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2121 }
2122
2123 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2124                 struct nfit_spa *nfit_spa)
2125 {
2126         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2127         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2128         struct nd_blk_region_desc ndbr_desc;
2129         struct nd_region_desc *ndr_desc;
2130         struct nfit_memdev *nfit_memdev;
2131         struct nvdimm_bus *nvdimm_bus;
2132         struct resource res;
2133         int count = 0, rc;
2134
2135         if (nfit_spa->nd_region)
2136                 return 0;
2137
2138         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2139                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2140                                 __func__);
2141                 return 0;
2142         }
2143
2144         memset(&res, 0, sizeof(res));
2145         memset(&mappings, 0, sizeof(mappings));
2146         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2147         res.start = spa->address;
2148         res.end = res.start + spa->length - 1;
2149         ndr_desc = &ndbr_desc.ndr_desc;
2150         ndr_desc->res = &res;
2151         ndr_desc->provider_data = nfit_spa;
2152         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2153         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2154                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2155                                                 spa->proximity_domain);
2156         else
2157                 ndr_desc->numa_node = NUMA_NO_NODE;
2158
2159         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2160                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2161                 struct nd_mapping_desc *mapping;
2162
2163                 if (memdev->range_index != spa->range_index)
2164                         continue;
2165                 if (count >= ND_MAX_MAPPINGS) {
2166                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2167                                         spa->range_index, ND_MAX_MAPPINGS);
2168                         return -ENXIO;
2169                 }
2170                 mapping = &mappings[count++];
2171                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2172                                 memdev, nfit_spa);
2173                 if (rc)
2174                         goto out;
2175         }
2176
2177         ndr_desc->mapping = mappings;
2178         ndr_desc->num_mappings = count;
2179         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2180         if (rc)
2181                 goto out;
2182
2183         nvdimm_bus = acpi_desc->nvdimm_bus;
2184         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2185                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2186                 if (rc) {
2187                         dev_warn(acpi_desc->dev,
2188                                 "failed to insert pmem resource to iomem: %d\n",
2189                                 rc);
2190                         goto out;
2191                 }
2192
2193                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2194                                 ndr_desc);
2195                 if (!nfit_spa->nd_region)
2196                         rc = -ENOMEM;
2197         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
2198                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2199                                 ndr_desc);
2200                 if (!nfit_spa->nd_region)
2201                         rc = -ENOMEM;
2202         } else if (nfit_spa_is_virtual(spa)) {
2203                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2204                                 ndr_desc);
2205                 if (!nfit_spa->nd_region)
2206                         rc = -ENOMEM;
2207         }
2208
2209  out:
2210         if (rc)
2211                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2212                                 nfit_spa->spa->range_index);
2213         return rc;
2214 }
2215
2216 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2217                 u32 max_ars)
2218 {
2219         struct device *dev = acpi_desc->dev;
2220         struct nd_cmd_ars_status *ars_status;
2221
2222         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2223                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2224                 return 0;
2225         }
2226
2227         if (acpi_desc->ars_status)
2228                 devm_kfree(dev, acpi_desc->ars_status);
2229         acpi_desc->ars_status = NULL;
2230         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2231         if (!ars_status)
2232                 return -ENOMEM;
2233         acpi_desc->ars_status = ars_status;
2234         acpi_desc->ars_status_size = max_ars;
2235         return 0;
2236 }
2237
2238 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2239                 struct nfit_spa *nfit_spa)
2240 {
2241         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2242         int rc;
2243
2244         if (!nfit_spa->max_ars) {
2245                 struct nd_cmd_ars_cap ars_cap;
2246
2247                 memset(&ars_cap, 0, sizeof(ars_cap));
2248                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2249                 if (rc < 0)
2250                         return rc;
2251                 nfit_spa->max_ars = ars_cap.max_ars_out;
2252                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2253                 /* check that the supported scrub types match the spa type */
2254                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2255                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2256                         return -ENOTTY;
2257                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2258                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2259                         return -ENOTTY;
2260         }
2261
2262         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2263                 return -ENOMEM;
2264
2265         rc = ars_get_status(acpi_desc);
2266         if (rc < 0 && rc != -ENOSPC)
2267                 return rc;
2268
2269         if (ars_status_process_records(acpi_desc->nvdimm_bus,
2270                                 acpi_desc->ars_status))
2271                 return -ENOMEM;
2272
2273         return 0;
2274 }
2275
2276 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2277                 struct nfit_spa *nfit_spa)
2278 {
2279         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2280         unsigned int overflow_retry = scrub_overflow_abort;
2281         u64 init_ars_start = 0, init_ars_len = 0;
2282         struct device *dev = acpi_desc->dev;
2283         unsigned int tmo = scrub_timeout;
2284         int rc;
2285
2286         if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2287                 return;
2288
2289         rc = ars_start(acpi_desc, nfit_spa);
2290         /*
2291          * If we timed out the initial scan we'll still be busy here,
2292          * and will wait another timeout before giving up permanently.
2293          */
2294         if (rc < 0 && rc != -EBUSY)
2295                 return;
2296
2297         do {
2298                 u64 ars_start, ars_len;
2299
2300                 if (acpi_desc->cancel)
2301                         break;
2302                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2303                 if (rc == -ENOTTY)
2304                         break;
2305                 if (rc == -EBUSY && !tmo) {
2306                         dev_warn(dev, "range %d ars timeout, aborting\n",
2307                                         spa->range_index);
2308                         break;
2309                 }
2310
2311                 if (rc == -EBUSY) {
2312                         /*
2313                          * Note, entries may be appended to the list
2314                          * while the lock is dropped, but the workqueue
2315                          * being active prevents entries being deleted /
2316                          * freed.
2317                          */
2318                         mutex_unlock(&acpi_desc->init_mutex);
2319                         ssleep(1);
2320                         tmo--;
2321                         mutex_lock(&acpi_desc->init_mutex);
2322                         continue;
2323                 }
2324
2325                 /* we got some results, but there are more pending... */
2326                 if (rc == -ENOSPC && overflow_retry--) {
2327                         if (!init_ars_len) {
2328                                 init_ars_len = acpi_desc->ars_status->length;
2329                                 init_ars_start = acpi_desc->ars_status->address;
2330                         }
2331                         rc = ars_continue(acpi_desc);
2332                 }
2333
2334                 if (rc < 0) {
2335                         dev_warn(dev, "range %d ars continuation failed\n",
2336                                         spa->range_index);
2337                         break;
2338                 }
2339
2340                 if (init_ars_len) {
2341                         ars_start = init_ars_start;
2342                         ars_len = init_ars_len;
2343                 } else {
2344                         ars_start = acpi_desc->ars_status->address;
2345                         ars_len = acpi_desc->ars_status->length;
2346                 }
2347                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2348                                 spa->range_index, ars_start, ars_len);
2349                 /* notify the region about new poison entries */
2350                 nvdimm_region_notify(nfit_spa->nd_region,
2351                                 NVDIMM_REVALIDATE_POISON);
2352                 break;
2353         } while (1);
2354 }
2355
2356 static void acpi_nfit_scrub(struct work_struct *work)
2357 {
2358         struct device *dev;
2359         u64 init_scrub_length = 0;
2360         struct nfit_spa *nfit_spa;
2361         u64 init_scrub_address = 0;
2362         bool init_ars_done = false;
2363         struct acpi_nfit_desc *acpi_desc;
2364         unsigned int tmo = scrub_timeout;
2365         unsigned int overflow_retry = scrub_overflow_abort;
2366
2367         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2368         dev = acpi_desc->dev;
2369
2370         /*
2371          * We scrub in 2 phases.  The first phase waits for any platform
2372          * firmware initiated scrubs to complete and then we go search for the
2373          * affected spa regions to mark them scanned.  In the second phase we
2374          * initiate a directed scrub for every range that was not scrubbed in
2375          * phase 1. If we're called for a 'rescan', we harmlessly pass through
2376          * the first phase, but really only care about running phase 2, where
2377          * regions can be notified of new poison.
2378          */
2379
2380         /* process platform firmware initiated scrubs */
2381  retry:
2382         mutex_lock(&acpi_desc->init_mutex);
2383         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2384                 struct nd_cmd_ars_status *ars_status;
2385                 struct acpi_nfit_system_address *spa;
2386                 u64 ars_start, ars_len;
2387                 int rc;
2388
2389                 if (acpi_desc->cancel)
2390                         break;
2391
2392                 if (nfit_spa->nd_region)
2393                         continue;
2394
2395                 if (init_ars_done) {
2396                         /*
2397                          * No need to re-query, we're now just
2398                          * reconciling all the ranges covered by the
2399                          * initial scrub
2400                          */
2401                         rc = 0;
2402                 } else
2403                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2404
2405                 if (rc == -ENOTTY) {
2406                         /* no ars capability, just register spa and move on */
2407                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2408                         continue;
2409                 }
2410
2411                 if (rc == -EBUSY && !tmo) {
2412                         /* fallthrough to directed scrub in phase 2 */
2413                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2414                         break;
2415                 } else if (rc == -EBUSY) {
2416                         mutex_unlock(&acpi_desc->init_mutex);
2417                         ssleep(1);
2418                         tmo--;
2419                         goto retry;
2420                 }
2421
2422                 /* we got some results, but there are more pending... */
2423                 if (rc == -ENOSPC && overflow_retry--) {
2424                         ars_status = acpi_desc->ars_status;
2425                         /*
2426                          * Record the original scrub range, so that we
2427                          * can recall all the ranges impacted by the
2428                          * initial scrub.
2429                          */
2430                         if (!init_scrub_length) {
2431                                 init_scrub_length = ars_status->length;
2432                                 init_scrub_address = ars_status->address;
2433                         }
2434                         rc = ars_continue(acpi_desc);
2435                         if (rc == 0) {
2436                                 mutex_unlock(&acpi_desc->init_mutex);
2437                                 goto retry;
2438                         }
2439                 }
2440
2441                 if (rc < 0) {
2442                         /*
2443                          * Initial scrub failed, we'll give it one more
2444                          * try below...
2445                          */
2446                         break;
2447                 }
2448
2449                 /* We got some final results, record completed ranges */
2450                 ars_status = acpi_desc->ars_status;
2451                 if (init_scrub_length) {
2452                         ars_start = init_scrub_address;
2453                         ars_len = ars_start + init_scrub_length;
2454                 } else {
2455                         ars_start = ars_status->address;
2456                         ars_len = ars_status->length;
2457                 }
2458                 spa = nfit_spa->spa;
2459
2460                 if (!init_ars_done) {
2461                         init_ars_done = true;
2462                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2463                                         ars_start, ars_len);
2464                 }
2465                 if (ars_start <= spa->address && ars_start + ars_len
2466                                 >= spa->address + spa->length)
2467                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2468         }
2469
2470         /*
2471          * For all the ranges not covered by an initial scrub we still
2472          * want to see if there are errors, but it's ok to discover them
2473          * asynchronously.
2474          */
2475         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2476                 /*
2477                  * Flag all the ranges that still need scrubbing, but
2478                  * register them now to make data available.
2479                  */
2480                 if (!nfit_spa->nd_region) {
2481                         nfit_spa->ars_required = 1;
2482                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2483                 }
2484         }
2485
2486         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2487                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2488         acpi_desc->scrub_count++;
2489         if (acpi_desc->scrub_count_state)
2490                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2491         mutex_unlock(&acpi_desc->init_mutex);
2492 }
2493
2494 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2495 {
2496         struct nfit_spa *nfit_spa;
2497         int rc;
2498
2499         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2500                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2501                         /* BLK regions don't need to wait for ars results */
2502                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2503                         if (rc)
2504                                 return rc;
2505                 }
2506
2507         queue_work(nfit_wq, &acpi_desc->work);
2508         return 0;
2509 }
2510
2511 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2512                 struct nfit_table_prev *prev)
2513 {
2514         struct device *dev = acpi_desc->dev;
2515
2516         if (!list_empty(&prev->spas) ||
2517                         !list_empty(&prev->memdevs) ||
2518                         !list_empty(&prev->dcrs) ||
2519                         !list_empty(&prev->bdws) ||
2520                         !list_empty(&prev->idts) ||
2521                         !list_empty(&prev->flushes)) {
2522                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2523                 return -ENXIO;
2524         }
2525         return 0;
2526 }
2527
2528 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
2529 {
2530         struct device *dev = acpi_desc->dev;
2531         struct kernfs_node *nfit;
2532         struct device *bus_dev;
2533
2534         if (!ars_supported(acpi_desc->nvdimm_bus))
2535                 return 0;
2536
2537         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2538         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
2539         if (!nfit) {
2540                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
2541                 return -ENODEV;
2542         }
2543         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
2544         sysfs_put(nfit);
2545         if (!acpi_desc->scrub_count_state) {
2546                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
2547                 return -ENODEV;
2548         }
2549
2550         return 0;
2551 }
2552
2553 static void acpi_nfit_destruct(void *data)
2554 {
2555         struct acpi_nfit_desc *acpi_desc = data;
2556         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2557
2558         /*
2559          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
2560          * race teardown
2561          */
2562         mutex_lock(&acpi_desc_lock);
2563         acpi_desc->cancel = 1;
2564         /*
2565          * Bounce the nvdimm bus lock to make sure any in-flight
2566          * acpi_nfit_ars_rescan() submissions have had a chance to
2567          * either submit or see ->cancel set.
2568          */
2569         device_lock(bus_dev);
2570         device_unlock(bus_dev);
2571
2572         flush_workqueue(nfit_wq);
2573         if (acpi_desc->scrub_count_state)
2574                 sysfs_put(acpi_desc->scrub_count_state);
2575         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2576         acpi_desc->nvdimm_bus = NULL;
2577         list_del(&acpi_desc->list);
2578         mutex_unlock(&acpi_desc_lock);
2579 }
2580
2581 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
2582 {
2583         struct device *dev = acpi_desc->dev;
2584         struct nfit_table_prev prev;
2585         const void *end;
2586         int rc;
2587
2588         if (!acpi_desc->nvdimm_bus) {
2589                 acpi_nfit_init_dsms(acpi_desc);
2590
2591                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
2592                                 &acpi_desc->nd_desc);
2593                 if (!acpi_desc->nvdimm_bus)
2594                         return -ENOMEM;
2595
2596                 rc = devm_add_action_or_reset(dev, acpi_nfit_destruct,
2597                                 acpi_desc);
2598                 if (rc)
2599                         return rc;
2600
2601                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
2602                 if (rc)
2603                         return rc;
2604
2605                 /* register this acpi_desc for mce notifications */
2606                 mutex_lock(&acpi_desc_lock);
2607                 list_add_tail(&acpi_desc->list, &acpi_descs);
2608                 mutex_unlock(&acpi_desc_lock);
2609         }
2610
2611         mutex_lock(&acpi_desc->init_mutex);
2612
2613         INIT_LIST_HEAD(&prev.spas);
2614         INIT_LIST_HEAD(&prev.memdevs);
2615         INIT_LIST_HEAD(&prev.dcrs);
2616         INIT_LIST_HEAD(&prev.bdws);
2617         INIT_LIST_HEAD(&prev.idts);
2618         INIT_LIST_HEAD(&prev.flushes);
2619
2620         list_cut_position(&prev.spas, &acpi_desc->spas,
2621                                 acpi_desc->spas.prev);
2622         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2623                                 acpi_desc->memdevs.prev);
2624         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2625                                 acpi_desc->dcrs.prev);
2626         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2627                                 acpi_desc->bdws.prev);
2628         list_cut_position(&prev.idts, &acpi_desc->idts,
2629                                 acpi_desc->idts.prev);
2630         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2631                                 acpi_desc->flushes.prev);
2632
2633         end = data + sz;
2634         while (!IS_ERR_OR_NULL(data))
2635                 data = add_table(acpi_desc, &prev, data, end);
2636
2637         if (IS_ERR(data)) {
2638                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2639                                 PTR_ERR(data));
2640                 rc = PTR_ERR(data);
2641                 goto out_unlock;
2642         }
2643
2644         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2645         if (rc)
2646                 goto out_unlock;
2647
2648         rc = nfit_mem_init(acpi_desc);
2649         if (rc)
2650                 goto out_unlock;
2651
2652         rc = acpi_nfit_register_dimms(acpi_desc);
2653         if (rc)
2654                 goto out_unlock;
2655
2656         rc = acpi_nfit_register_regions(acpi_desc);
2657
2658  out_unlock:
2659         mutex_unlock(&acpi_desc->init_mutex);
2660         return rc;
2661 }
2662 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2663
2664 struct acpi_nfit_flush_work {
2665         struct work_struct work;
2666         struct completion cmp;
2667 };
2668
2669 static void flush_probe(struct work_struct *work)
2670 {
2671         struct acpi_nfit_flush_work *flush;
2672
2673         flush = container_of(work, typeof(*flush), work);
2674         complete(&flush->cmp);
2675 }
2676
2677 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2678 {
2679         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2680         struct device *dev = acpi_desc->dev;
2681         struct acpi_nfit_flush_work flush;
2682
2683         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2684         device_lock(dev);
2685         device_unlock(dev);
2686
2687         /*
2688          * Scrub work could take 10s of seconds, userspace may give up so we
2689          * need to be interruptible while waiting.
2690          */
2691         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2692         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2693         queue_work(nfit_wq, &flush.work);
2694         return wait_for_completion_interruptible(&flush.cmp);
2695 }
2696
2697 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2698                 struct nvdimm *nvdimm, unsigned int cmd)
2699 {
2700         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2701
2702         if (nvdimm)
2703                 return 0;
2704         if (cmd != ND_CMD_ARS_START)
2705                 return 0;
2706
2707         /*
2708          * The kernel and userspace may race to initiate a scrub, but
2709          * the scrub thread is prepared to lose that initial race.  It
2710          * just needs guarantees that any ars it initiates are not
2711          * interrupted by any intervening start reqeusts from userspace.
2712          */
2713         if (work_busy(&acpi_desc->work))
2714                 return -EBUSY;
2715
2716         return 0;
2717 }
2718
2719 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc)
2720 {
2721         struct device *dev = acpi_desc->dev;
2722         struct nfit_spa *nfit_spa;
2723
2724         if (work_busy(&acpi_desc->work))
2725                 return -EBUSY;
2726
2727         if (acpi_desc->cancel)
2728                 return 0;
2729
2730         mutex_lock(&acpi_desc->init_mutex);
2731         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2732                 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2733
2734                 if (nfit_spa_type(spa) != NFIT_SPA_PM)
2735                         continue;
2736
2737                 nfit_spa->ars_required = 1;
2738         }
2739         queue_work(nfit_wq, &acpi_desc->work);
2740         dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
2741         mutex_unlock(&acpi_desc->init_mutex);
2742
2743         return 0;
2744 }
2745
2746 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2747 {
2748         struct nvdimm_bus_descriptor *nd_desc;
2749
2750         dev_set_drvdata(dev, acpi_desc);
2751         acpi_desc->dev = dev;
2752         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2753         nd_desc = &acpi_desc->nd_desc;
2754         nd_desc->provider_name = "ACPI.NFIT";
2755         nd_desc->module = THIS_MODULE;
2756         nd_desc->ndctl = acpi_nfit_ctl;
2757         nd_desc->flush_probe = acpi_nfit_flush_probe;
2758         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2759         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2760
2761         INIT_LIST_HEAD(&acpi_desc->spas);
2762         INIT_LIST_HEAD(&acpi_desc->dcrs);
2763         INIT_LIST_HEAD(&acpi_desc->bdws);
2764         INIT_LIST_HEAD(&acpi_desc->idts);
2765         INIT_LIST_HEAD(&acpi_desc->flushes);
2766         INIT_LIST_HEAD(&acpi_desc->memdevs);
2767         INIT_LIST_HEAD(&acpi_desc->dimms);
2768         INIT_LIST_HEAD(&acpi_desc->list);
2769         mutex_init(&acpi_desc->init_mutex);
2770         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2771 }
2772 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2773
2774 static int acpi_nfit_add(struct acpi_device *adev)
2775 {
2776         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2777         struct acpi_nfit_desc *acpi_desc;
2778         struct device *dev = &adev->dev;
2779         struct acpi_table_header *tbl;
2780         acpi_status status = AE_OK;
2781         acpi_size sz;
2782         int rc = 0;
2783
2784         status = acpi_get_table_with_size(ACPI_SIG_NFIT, 0, &tbl, &sz);
2785         if (ACPI_FAILURE(status)) {
2786                 /* This is ok, we could have an nvdimm hotplugged later */
2787                 dev_dbg(dev, "failed to find NFIT at startup\n");
2788                 return 0;
2789         }
2790
2791         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2792         if (!acpi_desc)
2793                 return -ENOMEM;
2794         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2795
2796         /* Save the acpi header for exporting the revision via sysfs */
2797         acpi_desc->acpi_header = *tbl;
2798
2799         /* Evaluate _FIT and override with that if present */
2800         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2801         if (ACPI_SUCCESS(status) && buf.length > 0) {
2802                 union acpi_object *obj = buf.pointer;
2803
2804                 if (obj->type == ACPI_TYPE_BUFFER)
2805                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
2806                                         obj->buffer.length);
2807                 else
2808                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2809                                  __func__, (int) obj->type);
2810                 kfree(buf.pointer);
2811         } else
2812                 /* skip over the lead-in header table */
2813                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
2814                                 + sizeof(struct acpi_table_nfit),
2815                                 sz - sizeof(struct acpi_table_nfit));
2816         return rc;
2817 }
2818
2819 static int acpi_nfit_remove(struct acpi_device *adev)
2820 {
2821         /* see acpi_nfit_destruct */
2822         return 0;
2823 }
2824
2825 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
2826 {
2827         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
2828         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2829         union acpi_object *obj;
2830         acpi_status status;
2831         int ret;
2832
2833         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2834
2835         if (event != NFIT_NOTIFY_UPDATE)
2836                 return;
2837
2838         if (!dev->driver) {
2839                 /* dev->driver may be null if we're being removed */
2840                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2841                 return;
2842         }
2843
2844         if (!acpi_desc) {
2845                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2846                 if (!acpi_desc)
2847                         return;
2848                 acpi_nfit_desc_init(acpi_desc, dev);
2849         } else {
2850                 /*
2851                  * Finish previous registration before considering new
2852                  * regions.
2853                  */
2854                 flush_workqueue(nfit_wq);
2855         }
2856
2857         /* Evaluate _FIT */
2858         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
2859         if (ACPI_FAILURE(status)) {
2860                 dev_err(dev, "failed to evaluate _FIT\n");
2861                 return;
2862         }
2863
2864         obj = buf.pointer;
2865         if (obj->type == ACPI_TYPE_BUFFER) {
2866                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
2867                                 obj->buffer.length);
2868                 if (ret)
2869                         dev_err(dev, "failed to merge updated NFIT\n");
2870         } else
2871                 dev_err(dev, "Invalid _FIT\n");
2872         kfree(buf.pointer);
2873 }
2874 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
2875
2876 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2877 {
2878         device_lock(&adev->dev);
2879         __acpi_nfit_notify(&adev->dev, adev->handle, event);
2880         device_unlock(&adev->dev);
2881 }
2882
2883 static const struct acpi_device_id acpi_nfit_ids[] = {
2884         { "ACPI0012", 0 },
2885         { "", 0 },
2886 };
2887 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2888
2889 static struct acpi_driver acpi_nfit_driver = {
2890         .name = KBUILD_MODNAME,
2891         .ids = acpi_nfit_ids,
2892         .ops = {
2893                 .add = acpi_nfit_add,
2894                 .remove = acpi_nfit_remove,
2895                 .notify = acpi_nfit_notify,
2896         },
2897 };
2898
2899 static __init int nfit_init(void)
2900 {
2901         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2902         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2903         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2904         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2905         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2906         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2907         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2908
2909         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2910         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2911         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2912         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2913         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2914         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2915         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2916         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2917         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2918         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2919         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2920         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2921         acpi_str_to_uuid(UUID_NFIT_DIMM_N_MSFT, nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
2922
2923         nfit_wq = create_singlethread_workqueue("nfit");
2924         if (!nfit_wq)
2925                 return -ENOMEM;
2926
2927         nfit_mce_register();
2928
2929         return acpi_bus_register_driver(&acpi_nfit_driver);
2930 }
2931
2932 static __exit void nfit_exit(void)
2933 {
2934         nfit_mce_unregister();
2935         acpi_bus_unregister_driver(&acpi_nfit_driver);
2936         destroy_workqueue(nfit_wq);
2937         WARN_ON(!list_empty(&acpi_descs));
2938 }
2939
2940 module_init(nfit_init);
2941 module_exit(nfit_exit);
2942 MODULE_LICENSE("GPL v2");
2943 MODULE_AUTHOR("Intel Corporation");