xref: /openbmc/linux/drivers/acpi/nfit/core.c (revision 8730046c)
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_bus_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 		return 0;
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 		return 0;
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->out_length >= 40 && (ars_status->flags
150 						& NFIT_ARS_F_OVERFLOW))
151 				return -ENOSPC;
152 			return 0;
153 		}
154 
155 		/* Unknown status */
156 		if (status >> 16)
157 			return -EIO;
158 		return 0;
159 	case ND_CMD_CLEAR_ERROR:
160 		clear_err = buf;
161 		if (status & 0xffff)
162 			return -EIO;
163 		if (!clear_err->cleared)
164 			return -EIO;
165 		if (clear_err->length > clear_err->cleared)
166 			return clear_err->cleared;
167 		return 0;
168 	default:
169 		break;
170 	}
171 
172 	/* all other non-zero status results in an error */
173 	if (status)
174 		return -EIO;
175 	return 0;
176 }
177 
178 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
179 		u32 status)
180 {
181 	if (!nvdimm)
182 		return xlat_bus_status(buf, cmd, status);
183 	if (status)
184 		return -EIO;
185 	return 0;
186 }
187 
188 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
189 		unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
190 {
191 	struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
192 	union acpi_object in_obj, in_buf, *out_obj;
193 	const struct nd_cmd_desc *desc = NULL;
194 	struct device *dev = acpi_desc->dev;
195 	struct nd_cmd_pkg *call_pkg = NULL;
196 	const char *cmd_name, *dimm_name;
197 	unsigned long cmd_mask, dsm_mask;
198 	u32 offset, fw_status = 0;
199 	acpi_handle handle;
200 	unsigned int func;
201 	const u8 *uuid;
202 	int rc, i;
203 
204 	func = cmd;
205 	if (cmd == ND_CMD_CALL) {
206 		call_pkg = buf;
207 		func = call_pkg->nd_command;
208 	}
209 
210 	if (nvdimm) {
211 		struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
212 		struct acpi_device *adev = nfit_mem->adev;
213 
214 		if (!adev)
215 			return -ENOTTY;
216 		if (call_pkg && nfit_mem->family != call_pkg->nd_family)
217 			return -ENOTTY;
218 
219 		dimm_name = nvdimm_name(nvdimm);
220 		cmd_name = nvdimm_cmd_name(cmd);
221 		cmd_mask = nvdimm_cmd_mask(nvdimm);
222 		dsm_mask = nfit_mem->dsm_mask;
223 		desc = nd_cmd_dimm_desc(cmd);
224 		uuid = to_nfit_uuid(nfit_mem->family);
225 		handle = adev->handle;
226 	} else {
227 		struct acpi_device *adev = to_acpi_dev(acpi_desc);
228 
229 		cmd_name = nvdimm_bus_cmd_name(cmd);
230 		cmd_mask = nd_desc->cmd_mask;
231 		dsm_mask = cmd_mask;
232 		desc = nd_cmd_bus_desc(cmd);
233 		uuid = to_nfit_uuid(NFIT_DEV_BUS);
234 		handle = adev->handle;
235 		dimm_name = "bus";
236 	}
237 
238 	if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
239 		return -ENOTTY;
240 
241 	if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
242 		return -ENOTTY;
243 
244 	in_obj.type = ACPI_TYPE_PACKAGE;
245 	in_obj.package.count = 1;
246 	in_obj.package.elements = &in_buf;
247 	in_buf.type = ACPI_TYPE_BUFFER;
248 	in_buf.buffer.pointer = buf;
249 	in_buf.buffer.length = 0;
250 
251 	/* libnvdimm has already validated the input envelope */
252 	for (i = 0; i < desc->in_num; i++)
253 		in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
254 				i, buf);
255 
256 	if (call_pkg) {
257 		/* skip over package wrapper */
258 		in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
259 		in_buf.buffer.length = call_pkg->nd_size_in;
260 	}
261 
262 	if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
263 		dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
264 				__func__, dimm_name, cmd, func,
265 				in_buf.buffer.length);
266 		print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
267 			in_buf.buffer.pointer,
268 			min_t(u32, 256, in_buf.buffer.length), true);
269 	}
270 
271 	out_obj = acpi_evaluate_dsm(handle, uuid, 1, func, &in_obj);
272 	if (!out_obj) {
273 		dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
274 				cmd_name);
275 		return -EINVAL;
276 	}
277 
278 	if (call_pkg) {
279 		call_pkg->nd_fw_size = out_obj->buffer.length;
280 		memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
281 			out_obj->buffer.pointer,
282 			min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
283 
284 		ACPI_FREE(out_obj);
285 		/*
286 		 * Need to support FW function w/o known size in advance.
287 		 * Caller can determine required size based upon nd_fw_size.
288 		 * If we return an error (like elsewhere) then caller wouldn't
289 		 * be able to rely upon data returned to make calculation.
290 		 */
291 		return 0;
292 	}
293 
294 	if (out_obj->package.type != ACPI_TYPE_BUFFER) {
295 		dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
296 				__func__, dimm_name, cmd_name, out_obj->type);
297 		rc = -EINVAL;
298 		goto out;
299 	}
300 
301 	if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
302 		dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__,
303 				dimm_name, cmd_name, out_obj->buffer.length);
304 		print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
305 				4, out_obj->buffer.pointer, min_t(u32, 128,
306 					out_obj->buffer.length), true);
307 	}
308 
309 	for (i = 0, offset = 0; i < desc->out_num; i++) {
310 		u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
311 				(u32 *) out_obj->buffer.pointer,
312 				out_obj->buffer.length - offset);
313 
314 		if (offset + out_size > out_obj->buffer.length) {
315 			dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
316 					__func__, dimm_name, cmd_name, i);
317 			break;
318 		}
319 
320 		if (in_buf.buffer.length + offset + out_size > buf_len) {
321 			dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
322 					__func__, dimm_name, cmd_name, i);
323 			rc = -ENXIO;
324 			goto out;
325 		}
326 		memcpy(buf + in_buf.buffer.length + offset,
327 				out_obj->buffer.pointer + offset, out_size);
328 		offset += out_size;
329 	}
330 
331 	/*
332 	 * Set fw_status for all the commands with a known format to be
333 	 * later interpreted by xlat_status().
334 	 */
335 	if (i >= 1 && ((cmd >= ND_CMD_ARS_CAP && cmd <= ND_CMD_CLEAR_ERROR)
336 			|| (cmd >= ND_CMD_SMART && cmd <= ND_CMD_VENDOR)))
337 		fw_status = *(u32 *) out_obj->buffer.pointer;
338 
339 	if (offset + in_buf.buffer.length < buf_len) {
340 		if (i >= 1) {
341 			/*
342 			 * status valid, return the number of bytes left
343 			 * unfilled in the output buffer
344 			 */
345 			rc = buf_len - offset - in_buf.buffer.length;
346 			if (cmd_rc)
347 				*cmd_rc = xlat_status(nvdimm, buf, cmd,
348 						fw_status);
349 		} else {
350 			dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
351 					__func__, dimm_name, cmd_name, buf_len,
352 					offset);
353 			rc = -ENXIO;
354 		}
355 	} else {
356 		rc = 0;
357 		if (cmd_rc)
358 			*cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
359 	}
360 
361  out:
362 	ACPI_FREE(out_obj);
363 
364 	return rc;
365 }
366 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
367 
368 static const char *spa_type_name(u16 type)
369 {
370 	static const char *to_name[] = {
371 		[NFIT_SPA_VOLATILE] = "volatile",
372 		[NFIT_SPA_PM] = "pmem",
373 		[NFIT_SPA_DCR] = "dimm-control-region",
374 		[NFIT_SPA_BDW] = "block-data-window",
375 		[NFIT_SPA_VDISK] = "volatile-disk",
376 		[NFIT_SPA_VCD] = "volatile-cd",
377 		[NFIT_SPA_PDISK] = "persistent-disk",
378 		[NFIT_SPA_PCD] = "persistent-cd",
379 
380 	};
381 
382 	if (type > NFIT_SPA_PCD)
383 		return "unknown";
384 
385 	return to_name[type];
386 }
387 
388 int nfit_spa_type(struct acpi_nfit_system_address *spa)
389 {
390 	int i;
391 
392 	for (i = 0; i < NFIT_UUID_MAX; i++)
393 		if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
394 			return i;
395 	return -1;
396 }
397 
398 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
399 		struct nfit_table_prev *prev,
400 		struct acpi_nfit_system_address *spa)
401 {
402 	struct device *dev = acpi_desc->dev;
403 	struct nfit_spa *nfit_spa;
404 
405 	if (spa->header.length != sizeof(*spa))
406 		return false;
407 
408 	list_for_each_entry(nfit_spa, &prev->spas, list) {
409 		if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
410 			list_move_tail(&nfit_spa->list, &acpi_desc->spas);
411 			return true;
412 		}
413 	}
414 
415 	nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
416 			GFP_KERNEL);
417 	if (!nfit_spa)
418 		return false;
419 	INIT_LIST_HEAD(&nfit_spa->list);
420 	memcpy(nfit_spa->spa, spa, sizeof(*spa));
421 	list_add_tail(&nfit_spa->list, &acpi_desc->spas);
422 	dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
423 			spa->range_index,
424 			spa_type_name(nfit_spa_type(spa)));
425 	return true;
426 }
427 
428 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
429 		struct nfit_table_prev *prev,
430 		struct acpi_nfit_memory_map *memdev)
431 {
432 	struct device *dev = acpi_desc->dev;
433 	struct nfit_memdev *nfit_memdev;
434 
435 	if (memdev->header.length != sizeof(*memdev))
436 		return false;
437 
438 	list_for_each_entry(nfit_memdev, &prev->memdevs, list)
439 		if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
440 			list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
441 			return true;
442 		}
443 
444 	nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
445 			GFP_KERNEL);
446 	if (!nfit_memdev)
447 		return false;
448 	INIT_LIST_HEAD(&nfit_memdev->list);
449 	memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
450 	list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
451 	dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
452 			__func__, memdev->device_handle, memdev->range_index,
453 			memdev->region_index);
454 	return true;
455 }
456 
457 /*
458  * An implementation may provide a truncated control region if no block windows
459  * are defined.
460  */
461 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
462 {
463 	if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
464 				window_size))
465 		return 0;
466 	if (dcr->windows)
467 		return sizeof(*dcr);
468 	return offsetof(struct acpi_nfit_control_region, window_size);
469 }
470 
471 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
472 		struct nfit_table_prev *prev,
473 		struct acpi_nfit_control_region *dcr)
474 {
475 	struct device *dev = acpi_desc->dev;
476 	struct nfit_dcr *nfit_dcr;
477 
478 	if (!sizeof_dcr(dcr))
479 		return false;
480 
481 	list_for_each_entry(nfit_dcr, &prev->dcrs, list)
482 		if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
483 			list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
484 			return true;
485 		}
486 
487 	nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
488 			GFP_KERNEL);
489 	if (!nfit_dcr)
490 		return false;
491 	INIT_LIST_HEAD(&nfit_dcr->list);
492 	memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
493 	list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
494 	dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
495 			dcr->region_index, dcr->windows);
496 	return true;
497 }
498 
499 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
500 		struct nfit_table_prev *prev,
501 		struct acpi_nfit_data_region *bdw)
502 {
503 	struct device *dev = acpi_desc->dev;
504 	struct nfit_bdw *nfit_bdw;
505 
506 	if (bdw->header.length != sizeof(*bdw))
507 		return false;
508 	list_for_each_entry(nfit_bdw, &prev->bdws, list)
509 		if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
510 			list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
511 			return true;
512 		}
513 
514 	nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
515 			GFP_KERNEL);
516 	if (!nfit_bdw)
517 		return false;
518 	INIT_LIST_HEAD(&nfit_bdw->list);
519 	memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
520 	list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
521 	dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
522 			bdw->region_index, bdw->windows);
523 	return true;
524 }
525 
526 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
527 {
528 	if (idt->header.length < sizeof(*idt))
529 		return 0;
530 	return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
531 }
532 
533 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
534 		struct nfit_table_prev *prev,
535 		struct acpi_nfit_interleave *idt)
536 {
537 	struct device *dev = acpi_desc->dev;
538 	struct nfit_idt *nfit_idt;
539 
540 	if (!sizeof_idt(idt))
541 		return false;
542 
543 	list_for_each_entry(nfit_idt, &prev->idts, list) {
544 		if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
545 			continue;
546 
547 		if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
548 			list_move_tail(&nfit_idt->list, &acpi_desc->idts);
549 			return true;
550 		}
551 	}
552 
553 	nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
554 			GFP_KERNEL);
555 	if (!nfit_idt)
556 		return false;
557 	INIT_LIST_HEAD(&nfit_idt->list);
558 	memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
559 	list_add_tail(&nfit_idt->list, &acpi_desc->idts);
560 	dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
561 			idt->interleave_index, idt->line_count);
562 	return true;
563 }
564 
565 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
566 {
567 	if (flush->header.length < sizeof(*flush))
568 		return 0;
569 	return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
570 }
571 
572 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
573 		struct nfit_table_prev *prev,
574 		struct acpi_nfit_flush_address *flush)
575 {
576 	struct device *dev = acpi_desc->dev;
577 	struct nfit_flush *nfit_flush;
578 
579 	if (!sizeof_flush(flush))
580 		return false;
581 
582 	list_for_each_entry(nfit_flush, &prev->flushes, list) {
583 		if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
584 			continue;
585 
586 		if (memcmp(nfit_flush->flush, flush,
587 					sizeof_flush(flush)) == 0) {
588 			list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
589 			return true;
590 		}
591 	}
592 
593 	nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
594 			+ sizeof_flush(flush), GFP_KERNEL);
595 	if (!nfit_flush)
596 		return false;
597 	INIT_LIST_HEAD(&nfit_flush->list);
598 	memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
599 	list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
600 	dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
601 			flush->device_handle, flush->hint_count);
602 	return true;
603 }
604 
605 static void *add_table(struct acpi_nfit_desc *acpi_desc,
606 		struct nfit_table_prev *prev, void *table, const void *end)
607 {
608 	struct device *dev = acpi_desc->dev;
609 	struct acpi_nfit_header *hdr;
610 	void *err = ERR_PTR(-ENOMEM);
611 
612 	if (table >= end)
613 		return NULL;
614 
615 	hdr = table;
616 	if (!hdr->length) {
617 		dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
618 			hdr->type);
619 		return NULL;
620 	}
621 
622 	switch (hdr->type) {
623 	case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
624 		if (!add_spa(acpi_desc, prev, table))
625 			return err;
626 		break;
627 	case ACPI_NFIT_TYPE_MEMORY_MAP:
628 		if (!add_memdev(acpi_desc, prev, table))
629 			return err;
630 		break;
631 	case ACPI_NFIT_TYPE_CONTROL_REGION:
632 		if (!add_dcr(acpi_desc, prev, table))
633 			return err;
634 		break;
635 	case ACPI_NFIT_TYPE_DATA_REGION:
636 		if (!add_bdw(acpi_desc, prev, table))
637 			return err;
638 		break;
639 	case ACPI_NFIT_TYPE_INTERLEAVE:
640 		if (!add_idt(acpi_desc, prev, table))
641 			return err;
642 		break;
643 	case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
644 		if (!add_flush(acpi_desc, prev, table))
645 			return err;
646 		break;
647 	case ACPI_NFIT_TYPE_SMBIOS:
648 		dev_dbg(dev, "%s: smbios\n", __func__);
649 		break;
650 	default:
651 		dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
652 		break;
653 	}
654 
655 	return table + hdr->length;
656 }
657 
658 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
659 		struct nfit_mem *nfit_mem)
660 {
661 	u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
662 	u16 dcr = nfit_mem->dcr->region_index;
663 	struct nfit_spa *nfit_spa;
664 
665 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
666 		u16 range_index = nfit_spa->spa->range_index;
667 		int type = nfit_spa_type(nfit_spa->spa);
668 		struct nfit_memdev *nfit_memdev;
669 
670 		if (type != NFIT_SPA_BDW)
671 			continue;
672 
673 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
674 			if (nfit_memdev->memdev->range_index != range_index)
675 				continue;
676 			if (nfit_memdev->memdev->device_handle != device_handle)
677 				continue;
678 			if (nfit_memdev->memdev->region_index != dcr)
679 				continue;
680 
681 			nfit_mem->spa_bdw = nfit_spa->spa;
682 			return;
683 		}
684 	}
685 
686 	dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
687 			nfit_mem->spa_dcr->range_index);
688 	nfit_mem->bdw = NULL;
689 }
690 
691 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
692 		struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
693 {
694 	u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
695 	struct nfit_memdev *nfit_memdev;
696 	struct nfit_bdw *nfit_bdw;
697 	struct nfit_idt *nfit_idt;
698 	u16 idt_idx, range_index;
699 
700 	list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
701 		if (nfit_bdw->bdw->region_index != dcr)
702 			continue;
703 		nfit_mem->bdw = nfit_bdw->bdw;
704 		break;
705 	}
706 
707 	if (!nfit_mem->bdw)
708 		return;
709 
710 	nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
711 
712 	if (!nfit_mem->spa_bdw)
713 		return;
714 
715 	range_index = nfit_mem->spa_bdw->range_index;
716 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
717 		if (nfit_memdev->memdev->range_index != range_index ||
718 				nfit_memdev->memdev->region_index != dcr)
719 			continue;
720 		nfit_mem->memdev_bdw = nfit_memdev->memdev;
721 		idt_idx = nfit_memdev->memdev->interleave_index;
722 		list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
723 			if (nfit_idt->idt->interleave_index != idt_idx)
724 				continue;
725 			nfit_mem->idt_bdw = nfit_idt->idt;
726 			break;
727 		}
728 		break;
729 	}
730 }
731 
732 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
733 		struct acpi_nfit_system_address *spa)
734 {
735 	struct nfit_mem *nfit_mem, *found;
736 	struct nfit_memdev *nfit_memdev;
737 	int type = nfit_spa_type(spa);
738 
739 	switch (type) {
740 	case NFIT_SPA_DCR:
741 	case NFIT_SPA_PM:
742 		break;
743 	default:
744 		return 0;
745 	}
746 
747 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
748 		struct nfit_flush *nfit_flush;
749 		struct nfit_dcr *nfit_dcr;
750 		u32 device_handle;
751 		u16 dcr;
752 
753 		if (nfit_memdev->memdev->range_index != spa->range_index)
754 			continue;
755 		found = NULL;
756 		dcr = nfit_memdev->memdev->region_index;
757 		device_handle = nfit_memdev->memdev->device_handle;
758 		list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
759 			if (__to_nfit_memdev(nfit_mem)->device_handle
760 					== device_handle) {
761 				found = nfit_mem;
762 				break;
763 			}
764 
765 		if (found)
766 			nfit_mem = found;
767 		else {
768 			nfit_mem = devm_kzalloc(acpi_desc->dev,
769 					sizeof(*nfit_mem), GFP_KERNEL);
770 			if (!nfit_mem)
771 				return -ENOMEM;
772 			INIT_LIST_HEAD(&nfit_mem->list);
773 			nfit_mem->acpi_desc = acpi_desc;
774 			list_add(&nfit_mem->list, &acpi_desc->dimms);
775 		}
776 
777 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
778 			if (nfit_dcr->dcr->region_index != dcr)
779 				continue;
780 			/*
781 			 * Record the control region for the dimm.  For
782 			 * the ACPI 6.1 case, where there are separate
783 			 * control regions for the pmem vs blk
784 			 * interfaces, be sure to record the extended
785 			 * blk details.
786 			 */
787 			if (!nfit_mem->dcr)
788 				nfit_mem->dcr = nfit_dcr->dcr;
789 			else if (nfit_mem->dcr->windows == 0
790 					&& nfit_dcr->dcr->windows)
791 				nfit_mem->dcr = nfit_dcr->dcr;
792 			break;
793 		}
794 
795 		list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
796 			struct acpi_nfit_flush_address *flush;
797 			u16 i;
798 
799 			if (nfit_flush->flush->device_handle != device_handle)
800 				continue;
801 			nfit_mem->nfit_flush = nfit_flush;
802 			flush = nfit_flush->flush;
803 			nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
804 					flush->hint_count
805 					* sizeof(struct resource), GFP_KERNEL);
806 			if (!nfit_mem->flush_wpq)
807 				return -ENOMEM;
808 			for (i = 0; i < flush->hint_count; i++) {
809 				struct resource *res = &nfit_mem->flush_wpq[i];
810 
811 				res->start = flush->hint_address[i];
812 				res->end = res->start + 8 - 1;
813 			}
814 			break;
815 		}
816 
817 		if (dcr && !nfit_mem->dcr) {
818 			dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
819 					spa->range_index, dcr);
820 			return -ENODEV;
821 		}
822 
823 		if (type == NFIT_SPA_DCR) {
824 			struct nfit_idt *nfit_idt;
825 			u16 idt_idx;
826 
827 			/* multiple dimms may share a SPA when interleaved */
828 			nfit_mem->spa_dcr = spa;
829 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
830 			idt_idx = nfit_memdev->memdev->interleave_index;
831 			list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
832 				if (nfit_idt->idt->interleave_index != idt_idx)
833 					continue;
834 				nfit_mem->idt_dcr = nfit_idt->idt;
835 				break;
836 			}
837 			nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
838 		} else {
839 			/*
840 			 * A single dimm may belong to multiple SPA-PM
841 			 * ranges, record at least one in addition to
842 			 * any SPA-DCR range.
843 			 */
844 			nfit_mem->memdev_pmem = nfit_memdev->memdev;
845 		}
846 	}
847 
848 	return 0;
849 }
850 
851 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
852 {
853 	struct nfit_mem *a = container_of(_a, typeof(*a), list);
854 	struct nfit_mem *b = container_of(_b, typeof(*b), list);
855 	u32 handleA, handleB;
856 
857 	handleA = __to_nfit_memdev(a)->device_handle;
858 	handleB = __to_nfit_memdev(b)->device_handle;
859 	if (handleA < handleB)
860 		return -1;
861 	else if (handleA > handleB)
862 		return 1;
863 	return 0;
864 }
865 
866 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
867 {
868 	struct nfit_spa *nfit_spa;
869 
870 	/*
871 	 * For each SPA-DCR or SPA-PMEM address range find its
872 	 * corresponding MEMDEV(s).  From each MEMDEV find the
873 	 * corresponding DCR.  Then, if we're operating on a SPA-DCR,
874 	 * try to find a SPA-BDW and a corresponding BDW that references
875 	 * the DCR.  Throw it all into an nfit_mem object.  Note, that
876 	 * BDWs are optional.
877 	 */
878 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
879 		int rc;
880 
881 		rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
882 		if (rc)
883 			return rc;
884 	}
885 
886 	list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
887 
888 	return 0;
889 }
890 
891 static ssize_t revision_show(struct device *dev,
892 		struct device_attribute *attr, char *buf)
893 {
894 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
895 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
896 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
897 
898 	return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
899 }
900 static DEVICE_ATTR_RO(revision);
901 
902 static ssize_t hw_error_scrub_show(struct device *dev,
903 		struct device_attribute *attr, char *buf)
904 {
905 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
906 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
907 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
908 
909 	return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
910 }
911 
912 /*
913  * The 'hw_error_scrub' attribute can have the following values written to it:
914  * '0': Switch to the default mode where an exception will only insert
915  *      the address of the memory error into the poison and badblocks lists.
916  * '1': Enable a full scrub to happen if an exception for a memory error is
917  *      received.
918  */
919 static ssize_t hw_error_scrub_store(struct device *dev,
920 		struct device_attribute *attr, const char *buf, size_t size)
921 {
922 	struct nvdimm_bus_descriptor *nd_desc;
923 	ssize_t rc;
924 	long val;
925 
926 	rc = kstrtol(buf, 0, &val);
927 	if (rc)
928 		return rc;
929 
930 	device_lock(dev);
931 	nd_desc = dev_get_drvdata(dev);
932 	if (nd_desc) {
933 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
934 
935 		switch (val) {
936 		case HW_ERROR_SCRUB_ON:
937 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
938 			break;
939 		case HW_ERROR_SCRUB_OFF:
940 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
941 			break;
942 		default:
943 			rc = -EINVAL;
944 			break;
945 		}
946 	}
947 	device_unlock(dev);
948 	if (rc)
949 		return rc;
950 	return size;
951 }
952 static DEVICE_ATTR_RW(hw_error_scrub);
953 
954 /*
955  * This shows the number of full Address Range Scrubs that have been
956  * completed since driver load time. Userspace can wait on this using
957  * select/poll etc. A '+' at the end indicates an ARS is in progress
958  */
959 static ssize_t scrub_show(struct device *dev,
960 		struct device_attribute *attr, char *buf)
961 {
962 	struct nvdimm_bus_descriptor *nd_desc;
963 	ssize_t rc = -ENXIO;
964 
965 	device_lock(dev);
966 	nd_desc = dev_get_drvdata(dev);
967 	if (nd_desc) {
968 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
969 
970 		rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
971 				(work_busy(&acpi_desc->work)) ? "+\n" : "\n");
972 	}
973 	device_unlock(dev);
974 	return rc;
975 }
976 
977 static ssize_t scrub_store(struct device *dev,
978 		struct device_attribute *attr, const char *buf, size_t size)
979 {
980 	struct nvdimm_bus_descriptor *nd_desc;
981 	ssize_t rc;
982 	long val;
983 
984 	rc = kstrtol(buf, 0, &val);
985 	if (rc)
986 		return rc;
987 	if (val != 1)
988 		return -EINVAL;
989 
990 	device_lock(dev);
991 	nd_desc = dev_get_drvdata(dev);
992 	if (nd_desc) {
993 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
994 
995 		rc = acpi_nfit_ars_rescan(acpi_desc);
996 	}
997 	device_unlock(dev);
998 	if (rc)
999 		return rc;
1000 	return size;
1001 }
1002 static DEVICE_ATTR_RW(scrub);
1003 
1004 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1005 {
1006 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1007 	const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1008 		| 1 << ND_CMD_ARS_STATUS;
1009 
1010 	return (nd_desc->cmd_mask & mask) == mask;
1011 }
1012 
1013 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1014 {
1015 	struct device *dev = container_of(kobj, struct device, kobj);
1016 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1017 
1018 	if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1019 		return 0;
1020 	return a->mode;
1021 }
1022 
1023 static struct attribute *acpi_nfit_attributes[] = {
1024 	&dev_attr_revision.attr,
1025 	&dev_attr_scrub.attr,
1026 	&dev_attr_hw_error_scrub.attr,
1027 	NULL,
1028 };
1029 
1030 static struct attribute_group acpi_nfit_attribute_group = {
1031 	.name = "nfit",
1032 	.attrs = acpi_nfit_attributes,
1033 	.is_visible = nfit_visible,
1034 };
1035 
1036 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1037 	&nvdimm_bus_attribute_group,
1038 	&acpi_nfit_attribute_group,
1039 	NULL,
1040 };
1041 
1042 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1043 {
1044 	struct nvdimm *nvdimm = to_nvdimm(dev);
1045 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1046 
1047 	return __to_nfit_memdev(nfit_mem);
1048 }
1049 
1050 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1051 {
1052 	struct nvdimm *nvdimm = to_nvdimm(dev);
1053 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1054 
1055 	return nfit_mem->dcr;
1056 }
1057 
1058 static ssize_t handle_show(struct device *dev,
1059 		struct device_attribute *attr, char *buf)
1060 {
1061 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1062 
1063 	return sprintf(buf, "%#x\n", memdev->device_handle);
1064 }
1065 static DEVICE_ATTR_RO(handle);
1066 
1067 static ssize_t phys_id_show(struct device *dev,
1068 		struct device_attribute *attr, char *buf)
1069 {
1070 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1071 
1072 	return sprintf(buf, "%#x\n", memdev->physical_id);
1073 }
1074 static DEVICE_ATTR_RO(phys_id);
1075 
1076 static ssize_t vendor_show(struct device *dev,
1077 		struct device_attribute *attr, char *buf)
1078 {
1079 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1080 
1081 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1082 }
1083 static DEVICE_ATTR_RO(vendor);
1084 
1085 static ssize_t rev_id_show(struct device *dev,
1086 		struct device_attribute *attr, char *buf)
1087 {
1088 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1089 
1090 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1091 }
1092 static DEVICE_ATTR_RO(rev_id);
1093 
1094 static ssize_t device_show(struct device *dev,
1095 		struct device_attribute *attr, char *buf)
1096 {
1097 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1098 
1099 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1100 }
1101 static DEVICE_ATTR_RO(device);
1102 
1103 static ssize_t subsystem_vendor_show(struct device *dev,
1104 		struct device_attribute *attr, char *buf)
1105 {
1106 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1107 
1108 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1109 }
1110 static DEVICE_ATTR_RO(subsystem_vendor);
1111 
1112 static ssize_t subsystem_rev_id_show(struct device *dev,
1113 		struct device_attribute *attr, char *buf)
1114 {
1115 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1116 
1117 	return sprintf(buf, "0x%04x\n",
1118 			be16_to_cpu(dcr->subsystem_revision_id));
1119 }
1120 static DEVICE_ATTR_RO(subsystem_rev_id);
1121 
1122 static ssize_t subsystem_device_show(struct device *dev,
1123 		struct device_attribute *attr, char *buf)
1124 {
1125 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1126 
1127 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1128 }
1129 static DEVICE_ATTR_RO(subsystem_device);
1130 
1131 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1132 {
1133 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1134 	int formats = 0;
1135 
1136 	if (nfit_mem->memdev_pmem)
1137 		formats++;
1138 	if (nfit_mem->memdev_bdw)
1139 		formats++;
1140 	return formats;
1141 }
1142 
1143 static ssize_t format_show(struct device *dev,
1144 		struct device_attribute *attr, char *buf)
1145 {
1146 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1147 
1148 	return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1149 }
1150 static DEVICE_ATTR_RO(format);
1151 
1152 static ssize_t format1_show(struct device *dev,
1153 		struct device_attribute *attr, char *buf)
1154 {
1155 	u32 handle;
1156 	ssize_t rc = -ENXIO;
1157 	struct nfit_mem *nfit_mem;
1158 	struct nfit_memdev *nfit_memdev;
1159 	struct acpi_nfit_desc *acpi_desc;
1160 	struct nvdimm *nvdimm = to_nvdimm(dev);
1161 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1162 
1163 	nfit_mem = nvdimm_provider_data(nvdimm);
1164 	acpi_desc = nfit_mem->acpi_desc;
1165 	handle = to_nfit_memdev(dev)->device_handle;
1166 
1167 	/* assumes DIMMs have at most 2 published interface codes */
1168 	mutex_lock(&acpi_desc->init_mutex);
1169 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1170 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1171 		struct nfit_dcr *nfit_dcr;
1172 
1173 		if (memdev->device_handle != handle)
1174 			continue;
1175 
1176 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1177 			if (nfit_dcr->dcr->region_index != memdev->region_index)
1178 				continue;
1179 			if (nfit_dcr->dcr->code == dcr->code)
1180 				continue;
1181 			rc = sprintf(buf, "0x%04x\n",
1182 					le16_to_cpu(nfit_dcr->dcr->code));
1183 			break;
1184 		}
1185 		if (rc != ENXIO)
1186 			break;
1187 	}
1188 	mutex_unlock(&acpi_desc->init_mutex);
1189 	return rc;
1190 }
1191 static DEVICE_ATTR_RO(format1);
1192 
1193 static ssize_t formats_show(struct device *dev,
1194 		struct device_attribute *attr, char *buf)
1195 {
1196 	struct nvdimm *nvdimm = to_nvdimm(dev);
1197 
1198 	return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1199 }
1200 static DEVICE_ATTR_RO(formats);
1201 
1202 static ssize_t serial_show(struct device *dev,
1203 		struct device_attribute *attr, char *buf)
1204 {
1205 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1206 
1207 	return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1208 }
1209 static DEVICE_ATTR_RO(serial);
1210 
1211 static ssize_t family_show(struct device *dev,
1212 		struct device_attribute *attr, char *buf)
1213 {
1214 	struct nvdimm *nvdimm = to_nvdimm(dev);
1215 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1216 
1217 	if (nfit_mem->family < 0)
1218 		return -ENXIO;
1219 	return sprintf(buf, "%d\n", nfit_mem->family);
1220 }
1221 static DEVICE_ATTR_RO(family);
1222 
1223 static ssize_t dsm_mask_show(struct device *dev,
1224 		struct device_attribute *attr, char *buf)
1225 {
1226 	struct nvdimm *nvdimm = to_nvdimm(dev);
1227 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1228 
1229 	if (nfit_mem->family < 0)
1230 		return -ENXIO;
1231 	return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1232 }
1233 static DEVICE_ATTR_RO(dsm_mask);
1234 
1235 static ssize_t flags_show(struct device *dev,
1236 		struct device_attribute *attr, char *buf)
1237 {
1238 	u16 flags = to_nfit_memdev(dev)->flags;
1239 
1240 	return sprintf(buf, "%s%s%s%s%s\n",
1241 		flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1242 		flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1243 		flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1244 		flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1245 		flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
1246 }
1247 static DEVICE_ATTR_RO(flags);
1248 
1249 static ssize_t id_show(struct device *dev,
1250 		struct device_attribute *attr, char *buf)
1251 {
1252 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1253 
1254 	if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1255 		return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1256 				be16_to_cpu(dcr->vendor_id),
1257 				dcr->manufacturing_location,
1258 				be16_to_cpu(dcr->manufacturing_date),
1259 				be32_to_cpu(dcr->serial_number));
1260 	else
1261 		return sprintf(buf, "%04x-%08x\n",
1262 				be16_to_cpu(dcr->vendor_id),
1263 				be32_to_cpu(dcr->serial_number));
1264 }
1265 static DEVICE_ATTR_RO(id);
1266 
1267 static struct attribute *acpi_nfit_dimm_attributes[] = {
1268 	&dev_attr_handle.attr,
1269 	&dev_attr_phys_id.attr,
1270 	&dev_attr_vendor.attr,
1271 	&dev_attr_device.attr,
1272 	&dev_attr_rev_id.attr,
1273 	&dev_attr_subsystem_vendor.attr,
1274 	&dev_attr_subsystem_device.attr,
1275 	&dev_attr_subsystem_rev_id.attr,
1276 	&dev_attr_format.attr,
1277 	&dev_attr_formats.attr,
1278 	&dev_attr_format1.attr,
1279 	&dev_attr_serial.attr,
1280 	&dev_attr_flags.attr,
1281 	&dev_attr_id.attr,
1282 	&dev_attr_family.attr,
1283 	&dev_attr_dsm_mask.attr,
1284 	NULL,
1285 };
1286 
1287 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1288 		struct attribute *a, int n)
1289 {
1290 	struct device *dev = container_of(kobj, struct device, kobj);
1291 	struct nvdimm *nvdimm = to_nvdimm(dev);
1292 
1293 	if (!to_nfit_dcr(dev))
1294 		return 0;
1295 	if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1296 		return 0;
1297 	return a->mode;
1298 }
1299 
1300 static struct attribute_group acpi_nfit_dimm_attribute_group = {
1301 	.name = "nfit",
1302 	.attrs = acpi_nfit_dimm_attributes,
1303 	.is_visible = acpi_nfit_dimm_attr_visible,
1304 };
1305 
1306 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1307 	&nvdimm_attribute_group,
1308 	&nd_device_attribute_group,
1309 	&acpi_nfit_dimm_attribute_group,
1310 	NULL,
1311 };
1312 
1313 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1314 		u32 device_handle)
1315 {
1316 	struct nfit_mem *nfit_mem;
1317 
1318 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1319 		if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1320 			return nfit_mem->nvdimm;
1321 
1322 	return NULL;
1323 }
1324 
1325 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1326 {
1327 	struct nfit_mem *nfit_mem;
1328 	struct acpi_nfit_desc *acpi_desc;
1329 
1330 	dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1331 			event);
1332 
1333 	if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1334 		dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1335 				event);
1336 		return;
1337 	}
1338 
1339 	acpi_desc = dev_get_drvdata(dev->parent);
1340 	if (!acpi_desc)
1341 		return;
1342 
1343 	/*
1344 	 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1345 	 * is still valid.
1346 	 */
1347 	nfit_mem = dev_get_drvdata(dev);
1348 	if (nfit_mem && nfit_mem->flags_attr)
1349 		sysfs_notify_dirent(nfit_mem->flags_attr);
1350 }
1351 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1352 
1353 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1354 {
1355 	struct acpi_device *adev = data;
1356 	struct device *dev = &adev->dev;
1357 
1358 	device_lock(dev->parent);
1359 	__acpi_nvdimm_notify(dev, event);
1360 	device_unlock(dev->parent);
1361 }
1362 
1363 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1364 		struct nfit_mem *nfit_mem, u32 device_handle)
1365 {
1366 	struct acpi_device *adev, *adev_dimm;
1367 	struct device *dev = acpi_desc->dev;
1368 	unsigned long dsm_mask;
1369 	const u8 *uuid;
1370 	int i;
1371 
1372 	/* nfit test assumes 1:1 relationship between commands and dsms */
1373 	nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1374 	nfit_mem->family = NVDIMM_FAMILY_INTEL;
1375 	adev = to_acpi_dev(acpi_desc);
1376 	if (!adev)
1377 		return 0;
1378 
1379 	adev_dimm = acpi_find_child_device(adev, device_handle, false);
1380 	nfit_mem->adev = adev_dimm;
1381 	if (!adev_dimm) {
1382 		dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1383 				device_handle);
1384 		return force_enable_dimms ? 0 : -ENODEV;
1385 	}
1386 
1387 	if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1388 		ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1389 		dev_err(dev, "%s: notification registration failed\n",
1390 				dev_name(&adev_dimm->dev));
1391 		return -ENXIO;
1392 	}
1393 
1394 	/*
1395 	 * Until standardization materializes we need to consider 4
1396 	 * different command sets.  Note, that checking for function0 (bit0)
1397 	 * tells us if any commands are reachable through this uuid.
1398 	 */
1399 	for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1400 		if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1401 			break;
1402 
1403 	/* limit the supported commands to those that are publicly documented */
1404 	nfit_mem->family = i;
1405 	if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1406 		dsm_mask = 0x3fe;
1407 		if (disable_vendor_specific)
1408 			dsm_mask &= ~(1 << ND_CMD_VENDOR);
1409 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1410 		dsm_mask = 0x1c3c76;
1411 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1412 		dsm_mask = 0x1fe;
1413 		if (disable_vendor_specific)
1414 			dsm_mask &= ~(1 << 8);
1415 	} else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1416 		dsm_mask = 0xffffffff;
1417 	} else {
1418 		dev_dbg(dev, "unknown dimm command family\n");
1419 		nfit_mem->family = -1;
1420 		/* DSMs are optional, continue loading the driver... */
1421 		return 0;
1422 	}
1423 
1424 	uuid = to_nfit_uuid(nfit_mem->family);
1425 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1426 		if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1427 			set_bit(i, &nfit_mem->dsm_mask);
1428 
1429 	return 0;
1430 }
1431 
1432 static void shutdown_dimm_notify(void *data)
1433 {
1434 	struct acpi_nfit_desc *acpi_desc = data;
1435 	struct nfit_mem *nfit_mem;
1436 
1437 	mutex_lock(&acpi_desc->init_mutex);
1438 	/*
1439 	 * Clear out the nfit_mem->flags_attr and shut down dimm event
1440 	 * notifications.
1441 	 */
1442 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1443 		struct acpi_device *adev_dimm = nfit_mem->adev;
1444 
1445 		if (nfit_mem->flags_attr) {
1446 			sysfs_put(nfit_mem->flags_attr);
1447 			nfit_mem->flags_attr = NULL;
1448 		}
1449 		if (adev_dimm)
1450 			acpi_remove_notify_handler(adev_dimm->handle,
1451 					ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1452 	}
1453 	mutex_unlock(&acpi_desc->init_mutex);
1454 }
1455 
1456 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1457 {
1458 	struct nfit_mem *nfit_mem;
1459 	int dimm_count = 0, rc;
1460 	struct nvdimm *nvdimm;
1461 
1462 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1463 		struct acpi_nfit_flush_address *flush;
1464 		unsigned long flags = 0, cmd_mask;
1465 		u32 device_handle;
1466 		u16 mem_flags;
1467 
1468 		device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1469 		nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1470 		if (nvdimm) {
1471 			dimm_count++;
1472 			continue;
1473 		}
1474 
1475 		if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1476 			flags |= NDD_ALIASING;
1477 
1478 		mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1479 		if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1480 			flags |= NDD_UNARMED;
1481 
1482 		rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1483 		if (rc)
1484 			continue;
1485 
1486 		/*
1487 		 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1488 		 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1489 		 * userspace interface.
1490 		 */
1491 		cmd_mask = 1UL << ND_CMD_CALL;
1492 		if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1493 			cmd_mask |= nfit_mem->dsm_mask;
1494 
1495 		flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1496 			: NULL;
1497 		nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1498 				acpi_nfit_dimm_attribute_groups,
1499 				flags, cmd_mask, flush ? flush->hint_count : 0,
1500 				nfit_mem->flush_wpq);
1501 		if (!nvdimm)
1502 			return -ENOMEM;
1503 
1504 		nfit_mem->nvdimm = nvdimm;
1505 		dimm_count++;
1506 
1507 		if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1508 			continue;
1509 
1510 		dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1511 				nvdimm_name(nvdimm),
1512 		  mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1513 		  mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1514 		  mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1515 		  mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1516 
1517 	}
1518 
1519 	rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1520 	if (rc)
1521 		return rc;
1522 
1523 	/*
1524 	 * Now that dimms are successfully registered, and async registration
1525 	 * is flushed, attempt to enable event notification.
1526 	 */
1527 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1528 		struct kernfs_node *nfit_kernfs;
1529 
1530 		nvdimm = nfit_mem->nvdimm;
1531 		nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1532 		if (nfit_kernfs)
1533 			nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1534 					"flags");
1535 		sysfs_put(nfit_kernfs);
1536 		if (!nfit_mem->flags_attr)
1537 			dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1538 					nvdimm_name(nvdimm));
1539 	}
1540 
1541 	return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1542 			acpi_desc);
1543 }
1544 
1545 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1546 {
1547 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1548 	const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1549 	struct acpi_device *adev;
1550 	int i;
1551 
1552 	nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1553 	adev = to_acpi_dev(acpi_desc);
1554 	if (!adev)
1555 		return;
1556 
1557 	for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1558 		if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1559 			set_bit(i, &nd_desc->cmd_mask);
1560 }
1561 
1562 static ssize_t range_index_show(struct device *dev,
1563 		struct device_attribute *attr, char *buf)
1564 {
1565 	struct nd_region *nd_region = to_nd_region(dev);
1566 	struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1567 
1568 	return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1569 }
1570 static DEVICE_ATTR_RO(range_index);
1571 
1572 static struct attribute *acpi_nfit_region_attributes[] = {
1573 	&dev_attr_range_index.attr,
1574 	NULL,
1575 };
1576 
1577 static struct attribute_group acpi_nfit_region_attribute_group = {
1578 	.name = "nfit",
1579 	.attrs = acpi_nfit_region_attributes,
1580 };
1581 
1582 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1583 	&nd_region_attribute_group,
1584 	&nd_mapping_attribute_group,
1585 	&nd_device_attribute_group,
1586 	&nd_numa_attribute_group,
1587 	&acpi_nfit_region_attribute_group,
1588 	NULL,
1589 };
1590 
1591 /* enough info to uniquely specify an interleave set */
1592 struct nfit_set_info {
1593 	struct nfit_set_info_map {
1594 		u64 region_offset;
1595 		u32 serial_number;
1596 		u32 pad;
1597 	} mapping[0];
1598 };
1599 
1600 static size_t sizeof_nfit_set_info(int num_mappings)
1601 {
1602 	return sizeof(struct nfit_set_info)
1603 		+ num_mappings * sizeof(struct nfit_set_info_map);
1604 }
1605 
1606 static int cmp_map(const void *m0, const void *m1)
1607 {
1608 	const struct nfit_set_info_map *map0 = m0;
1609 	const struct nfit_set_info_map *map1 = m1;
1610 
1611 	return memcmp(&map0->region_offset, &map1->region_offset,
1612 			sizeof(u64));
1613 }
1614 
1615 /* Retrieve the nth entry referencing this spa */
1616 static struct acpi_nfit_memory_map *memdev_from_spa(
1617 		struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1618 {
1619 	struct nfit_memdev *nfit_memdev;
1620 
1621 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1622 		if (nfit_memdev->memdev->range_index == range_index)
1623 			if (n-- == 0)
1624 				return nfit_memdev->memdev;
1625 	return NULL;
1626 }
1627 
1628 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1629 		struct nd_region_desc *ndr_desc,
1630 		struct acpi_nfit_system_address *spa)
1631 {
1632 	int i, spa_type = nfit_spa_type(spa);
1633 	struct device *dev = acpi_desc->dev;
1634 	struct nd_interleave_set *nd_set;
1635 	u16 nr = ndr_desc->num_mappings;
1636 	struct nfit_set_info *info;
1637 
1638 	if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1639 		/* pass */;
1640 	else
1641 		return 0;
1642 
1643 	nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1644 	if (!nd_set)
1645 		return -ENOMEM;
1646 
1647 	info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1648 	if (!info)
1649 		return -ENOMEM;
1650 	for (i = 0; i < nr; i++) {
1651 		struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1652 		struct nfit_set_info_map *map = &info->mapping[i];
1653 		struct nvdimm *nvdimm = mapping->nvdimm;
1654 		struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1655 		struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1656 				spa->range_index, i);
1657 
1658 		if (!memdev || !nfit_mem->dcr) {
1659 			dev_err(dev, "%s: failed to find DCR\n", __func__);
1660 			return -ENODEV;
1661 		}
1662 
1663 		map->region_offset = memdev->region_offset;
1664 		map->serial_number = nfit_mem->dcr->serial_number;
1665 	}
1666 
1667 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1668 			cmp_map, NULL);
1669 	nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1670 	ndr_desc->nd_set = nd_set;
1671 	devm_kfree(dev, info);
1672 
1673 	return 0;
1674 }
1675 
1676 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1677 {
1678 	struct acpi_nfit_interleave *idt = mmio->idt;
1679 	u32 sub_line_offset, line_index, line_offset;
1680 	u64 line_no, table_skip_count, table_offset;
1681 
1682 	line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1683 	table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1684 	line_offset = idt->line_offset[line_index]
1685 		* mmio->line_size;
1686 	table_offset = table_skip_count * mmio->table_size;
1687 
1688 	return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1689 }
1690 
1691 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1692 {
1693 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1694 	u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1695 	const u32 STATUS_MASK = 0x80000037;
1696 
1697 	if (mmio->num_lines)
1698 		offset = to_interleave_offset(offset, mmio);
1699 
1700 	return readl(mmio->addr.base + offset) & STATUS_MASK;
1701 }
1702 
1703 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1704 		resource_size_t dpa, unsigned int len, unsigned int write)
1705 {
1706 	u64 cmd, offset;
1707 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1708 
1709 	enum {
1710 		BCW_OFFSET_MASK = (1ULL << 48)-1,
1711 		BCW_LEN_SHIFT = 48,
1712 		BCW_LEN_MASK = (1ULL << 8) - 1,
1713 		BCW_CMD_SHIFT = 56,
1714 	};
1715 
1716 	cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1717 	len = len >> L1_CACHE_SHIFT;
1718 	cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1719 	cmd |= ((u64) write) << BCW_CMD_SHIFT;
1720 
1721 	offset = nfit_blk->cmd_offset + mmio->size * bw;
1722 	if (mmio->num_lines)
1723 		offset = to_interleave_offset(offset, mmio);
1724 
1725 	writeq(cmd, mmio->addr.base + offset);
1726 	nvdimm_flush(nfit_blk->nd_region);
1727 
1728 	if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1729 		readq(mmio->addr.base + offset);
1730 }
1731 
1732 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1733 		resource_size_t dpa, void *iobuf, size_t len, int rw,
1734 		unsigned int lane)
1735 {
1736 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1737 	unsigned int copied = 0;
1738 	u64 base_offset;
1739 	int rc;
1740 
1741 	base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1742 		+ lane * mmio->size;
1743 	write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1744 	while (len) {
1745 		unsigned int c;
1746 		u64 offset;
1747 
1748 		if (mmio->num_lines) {
1749 			u32 line_offset;
1750 
1751 			offset = to_interleave_offset(base_offset + copied,
1752 					mmio);
1753 			div_u64_rem(offset, mmio->line_size, &line_offset);
1754 			c = min_t(size_t, len, mmio->line_size - line_offset);
1755 		} else {
1756 			offset = base_offset + nfit_blk->bdw_offset;
1757 			c = len;
1758 		}
1759 
1760 		if (rw)
1761 			memcpy_to_pmem(mmio->addr.aperture + offset,
1762 					iobuf + copied, c);
1763 		else {
1764 			if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1765 				mmio_flush_range((void __force *)
1766 					mmio->addr.aperture + offset, c);
1767 
1768 			memcpy_from_pmem(iobuf + copied,
1769 					mmio->addr.aperture + offset, c);
1770 		}
1771 
1772 		copied += c;
1773 		len -= c;
1774 	}
1775 
1776 	if (rw)
1777 		nvdimm_flush(nfit_blk->nd_region);
1778 
1779 	rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1780 	return rc;
1781 }
1782 
1783 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1784 		resource_size_t dpa, void *iobuf, u64 len, int rw)
1785 {
1786 	struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1787 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1788 	struct nd_region *nd_region = nfit_blk->nd_region;
1789 	unsigned int lane, copied = 0;
1790 	int rc = 0;
1791 
1792 	lane = nd_region_acquire_lane(nd_region);
1793 	while (len) {
1794 		u64 c = min(len, mmio->size);
1795 
1796 		rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1797 				iobuf + copied, c, rw, lane);
1798 		if (rc)
1799 			break;
1800 
1801 		copied += c;
1802 		len -= c;
1803 	}
1804 	nd_region_release_lane(nd_region, lane);
1805 
1806 	return rc;
1807 }
1808 
1809 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1810 		struct acpi_nfit_interleave *idt, u16 interleave_ways)
1811 {
1812 	if (idt) {
1813 		mmio->num_lines = idt->line_count;
1814 		mmio->line_size = idt->line_size;
1815 		if (interleave_ways == 0)
1816 			return -ENXIO;
1817 		mmio->table_size = mmio->num_lines * interleave_ways
1818 			* mmio->line_size;
1819 	}
1820 
1821 	return 0;
1822 }
1823 
1824 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1825 		struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1826 {
1827 	struct nd_cmd_dimm_flags flags;
1828 	int rc;
1829 
1830 	memset(&flags, 0, sizeof(flags));
1831 	rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1832 			sizeof(flags), NULL);
1833 
1834 	if (rc >= 0 && flags.status == 0)
1835 		nfit_blk->dimm_flags = flags.flags;
1836 	else if (rc == -ENOTTY) {
1837 		/* fall back to a conservative default */
1838 		nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1839 		rc = 0;
1840 	} else
1841 		rc = -ENXIO;
1842 
1843 	return rc;
1844 }
1845 
1846 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1847 		struct device *dev)
1848 {
1849 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1850 	struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1851 	struct nfit_blk_mmio *mmio;
1852 	struct nfit_blk *nfit_blk;
1853 	struct nfit_mem *nfit_mem;
1854 	struct nvdimm *nvdimm;
1855 	int rc;
1856 
1857 	nvdimm = nd_blk_region_to_dimm(ndbr);
1858 	nfit_mem = nvdimm_provider_data(nvdimm);
1859 	if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1860 		dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1861 				nfit_mem ? "" : " nfit_mem",
1862 				(nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1863 				(nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1864 		return -ENXIO;
1865 	}
1866 
1867 	nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1868 	if (!nfit_blk)
1869 		return -ENOMEM;
1870 	nd_blk_region_set_provider_data(ndbr, nfit_blk);
1871 	nfit_blk->nd_region = to_nd_region(dev);
1872 
1873 	/* map block aperture memory */
1874 	nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1875 	mmio = &nfit_blk->mmio[BDW];
1876 	mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
1877                         nfit_mem->spa_bdw->length, ARCH_MEMREMAP_PMEM);
1878 	if (!mmio->addr.base) {
1879 		dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1880 				nvdimm_name(nvdimm));
1881 		return -ENOMEM;
1882 	}
1883 	mmio->size = nfit_mem->bdw->size;
1884 	mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1885 	mmio->idt = nfit_mem->idt_bdw;
1886 	mmio->spa = nfit_mem->spa_bdw;
1887 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1888 			nfit_mem->memdev_bdw->interleave_ways);
1889 	if (rc) {
1890 		dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1891 				__func__, nvdimm_name(nvdimm));
1892 		return rc;
1893 	}
1894 
1895 	/* map block control memory */
1896 	nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1897 	nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1898 	mmio = &nfit_blk->mmio[DCR];
1899 	mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
1900 			nfit_mem->spa_dcr->length);
1901 	if (!mmio->addr.base) {
1902 		dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1903 				nvdimm_name(nvdimm));
1904 		return -ENOMEM;
1905 	}
1906 	mmio->size = nfit_mem->dcr->window_size;
1907 	mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1908 	mmio->idt = nfit_mem->idt_dcr;
1909 	mmio->spa = nfit_mem->spa_dcr;
1910 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1911 			nfit_mem->memdev_dcr->interleave_ways);
1912 	if (rc) {
1913 		dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1914 				__func__, nvdimm_name(nvdimm));
1915 		return rc;
1916 	}
1917 
1918 	rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1919 	if (rc < 0) {
1920 		dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1921 				__func__, nvdimm_name(nvdimm));
1922 		return rc;
1923 	}
1924 
1925 	if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
1926 		dev_warn(dev, "unable to guarantee persistence of writes\n");
1927 
1928 	if (mmio->line_size == 0)
1929 		return 0;
1930 
1931 	if ((u32) nfit_blk->cmd_offset % mmio->line_size
1932 			+ 8 > mmio->line_size) {
1933 		dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1934 		return -ENXIO;
1935 	} else if ((u32) nfit_blk->stat_offset % mmio->line_size
1936 			+ 8 > mmio->line_size) {
1937 		dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1938 		return -ENXIO;
1939 	}
1940 
1941 	return 0;
1942 }
1943 
1944 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1945 		struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1946 {
1947 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1948 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
1949 	int cmd_rc, rc;
1950 
1951 	cmd->address = spa->address;
1952 	cmd->length = spa->length;
1953 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1954 			sizeof(*cmd), &cmd_rc);
1955 	if (rc < 0)
1956 		return rc;
1957 	return cmd_rc;
1958 }
1959 
1960 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1961 {
1962 	int rc;
1963 	int cmd_rc;
1964 	struct nd_cmd_ars_start ars_start;
1965 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
1966 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1967 
1968 	memset(&ars_start, 0, sizeof(ars_start));
1969 	ars_start.address = spa->address;
1970 	ars_start.length = spa->length;
1971 	if (nfit_spa_type(spa) == NFIT_SPA_PM)
1972 		ars_start.type = ND_ARS_PERSISTENT;
1973 	else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1974 		ars_start.type = ND_ARS_VOLATILE;
1975 	else
1976 		return -ENOTTY;
1977 
1978 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1979 			sizeof(ars_start), &cmd_rc);
1980 
1981 	if (rc < 0)
1982 		return rc;
1983 	return cmd_rc;
1984 }
1985 
1986 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1987 {
1988 	int rc, cmd_rc;
1989 	struct nd_cmd_ars_start ars_start;
1990 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1991 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1992 
1993 	memset(&ars_start, 0, sizeof(ars_start));
1994 	ars_start.address = ars_status->restart_address;
1995 	ars_start.length = ars_status->restart_length;
1996 	ars_start.type = ars_status->type;
1997 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1998 			sizeof(ars_start), &cmd_rc);
1999 	if (rc < 0)
2000 		return rc;
2001 	return cmd_rc;
2002 }
2003 
2004 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2005 {
2006 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2007 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2008 	int rc, cmd_rc;
2009 
2010 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2011 			acpi_desc->ars_status_size, &cmd_rc);
2012 	if (rc < 0)
2013 		return rc;
2014 	return cmd_rc;
2015 }
2016 
2017 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2018 		struct nd_cmd_ars_status *ars_status)
2019 {
2020 	struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2021 	int rc;
2022 	u32 i;
2023 
2024 	/*
2025 	 * First record starts at 44 byte offset from the start of the
2026 	 * payload.
2027 	 */
2028 	if (ars_status->out_length < 44)
2029 		return 0;
2030 	for (i = 0; i < ars_status->num_records; i++) {
2031 		/* only process full records */
2032 		if (ars_status->out_length
2033 				< 44 + sizeof(struct nd_ars_record) * (i + 1))
2034 			break;
2035 		rc = nvdimm_bus_add_poison(nvdimm_bus,
2036 				ars_status->records[i].err_address,
2037 				ars_status->records[i].length);
2038 		if (rc)
2039 			return rc;
2040 	}
2041 	if (i < ars_status->num_records)
2042 		dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2043 
2044 	return 0;
2045 }
2046 
2047 static void acpi_nfit_remove_resource(void *data)
2048 {
2049 	struct resource *res = data;
2050 
2051 	remove_resource(res);
2052 }
2053 
2054 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2055 		struct nd_region_desc *ndr_desc)
2056 {
2057 	struct resource *res, *nd_res = ndr_desc->res;
2058 	int is_pmem, ret;
2059 
2060 	/* No operation if the region is already registered as PMEM */
2061 	is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2062 				IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2063 	if (is_pmem == REGION_INTERSECTS)
2064 		return 0;
2065 
2066 	res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2067 	if (!res)
2068 		return -ENOMEM;
2069 
2070 	res->name = "Persistent Memory";
2071 	res->start = nd_res->start;
2072 	res->end = nd_res->end;
2073 	res->flags = IORESOURCE_MEM;
2074 	res->desc = IORES_DESC_PERSISTENT_MEMORY;
2075 
2076 	ret = insert_resource(&iomem_resource, res);
2077 	if (ret)
2078 		return ret;
2079 
2080 	ret = devm_add_action_or_reset(acpi_desc->dev,
2081 					acpi_nfit_remove_resource,
2082 					res);
2083 	if (ret)
2084 		return ret;
2085 
2086 	return 0;
2087 }
2088 
2089 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2090 		struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2091 		struct acpi_nfit_memory_map *memdev,
2092 		struct nfit_spa *nfit_spa)
2093 {
2094 	struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2095 			memdev->device_handle);
2096 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2097 	struct nd_blk_region_desc *ndbr_desc;
2098 	struct nfit_mem *nfit_mem;
2099 	int blk_valid = 0;
2100 
2101 	if (!nvdimm) {
2102 		dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2103 				spa->range_index, memdev->device_handle);
2104 		return -ENODEV;
2105 	}
2106 
2107 	mapping->nvdimm = nvdimm;
2108 	switch (nfit_spa_type(spa)) {
2109 	case NFIT_SPA_PM:
2110 	case NFIT_SPA_VOLATILE:
2111 		mapping->start = memdev->address;
2112 		mapping->size = memdev->region_size;
2113 		break;
2114 	case NFIT_SPA_DCR:
2115 		nfit_mem = nvdimm_provider_data(nvdimm);
2116 		if (!nfit_mem || !nfit_mem->bdw) {
2117 			dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2118 					spa->range_index, nvdimm_name(nvdimm));
2119 		} else {
2120 			mapping->size = nfit_mem->bdw->capacity;
2121 			mapping->start = nfit_mem->bdw->start_address;
2122 			ndr_desc->num_lanes = nfit_mem->bdw->windows;
2123 			blk_valid = 1;
2124 		}
2125 
2126 		ndr_desc->mapping = mapping;
2127 		ndr_desc->num_mappings = blk_valid;
2128 		ndbr_desc = to_blk_region_desc(ndr_desc);
2129 		ndbr_desc->enable = acpi_nfit_blk_region_enable;
2130 		ndbr_desc->do_io = acpi_desc->blk_do_io;
2131 		nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2132 				ndr_desc);
2133 		if (!nfit_spa->nd_region)
2134 			return -ENOMEM;
2135 		break;
2136 	}
2137 
2138 	return 0;
2139 }
2140 
2141 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2142 {
2143 	return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2144 		nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2145 		nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2146 		nfit_spa_type(spa) == NFIT_SPA_PCD);
2147 }
2148 
2149 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2150 		struct nfit_spa *nfit_spa)
2151 {
2152 	static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2153 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2154 	struct nd_blk_region_desc ndbr_desc;
2155 	struct nd_region_desc *ndr_desc;
2156 	struct nfit_memdev *nfit_memdev;
2157 	struct nvdimm_bus *nvdimm_bus;
2158 	struct resource res;
2159 	int count = 0, rc;
2160 
2161 	if (nfit_spa->nd_region)
2162 		return 0;
2163 
2164 	if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2165 		dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2166 				__func__);
2167 		return 0;
2168 	}
2169 
2170 	memset(&res, 0, sizeof(res));
2171 	memset(&mappings, 0, sizeof(mappings));
2172 	memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2173 	res.start = spa->address;
2174 	res.end = res.start + spa->length - 1;
2175 	ndr_desc = &ndbr_desc.ndr_desc;
2176 	ndr_desc->res = &res;
2177 	ndr_desc->provider_data = nfit_spa;
2178 	ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2179 	if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2180 		ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2181 						spa->proximity_domain);
2182 	else
2183 		ndr_desc->numa_node = NUMA_NO_NODE;
2184 
2185 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2186 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2187 		struct nd_mapping_desc *mapping;
2188 
2189 		if (memdev->range_index != spa->range_index)
2190 			continue;
2191 		if (count >= ND_MAX_MAPPINGS) {
2192 			dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2193 					spa->range_index, ND_MAX_MAPPINGS);
2194 			return -ENXIO;
2195 		}
2196 		mapping = &mappings[count++];
2197 		rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2198 				memdev, nfit_spa);
2199 		if (rc)
2200 			goto out;
2201 	}
2202 
2203 	ndr_desc->mapping = mappings;
2204 	ndr_desc->num_mappings = count;
2205 	rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2206 	if (rc)
2207 		goto out;
2208 
2209 	nvdimm_bus = acpi_desc->nvdimm_bus;
2210 	if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2211 		rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2212 		if (rc) {
2213 			dev_warn(acpi_desc->dev,
2214 				"failed to insert pmem resource to iomem: %d\n",
2215 				rc);
2216 			goto out;
2217 		}
2218 
2219 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2220 				ndr_desc);
2221 		if (!nfit_spa->nd_region)
2222 			rc = -ENOMEM;
2223 	} else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
2224 		nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2225 				ndr_desc);
2226 		if (!nfit_spa->nd_region)
2227 			rc = -ENOMEM;
2228 	} else if (nfit_spa_is_virtual(spa)) {
2229 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2230 				ndr_desc);
2231 		if (!nfit_spa->nd_region)
2232 			rc = -ENOMEM;
2233 	}
2234 
2235  out:
2236 	if (rc)
2237 		dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2238 				nfit_spa->spa->range_index);
2239 	return rc;
2240 }
2241 
2242 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2243 		u32 max_ars)
2244 {
2245 	struct device *dev = acpi_desc->dev;
2246 	struct nd_cmd_ars_status *ars_status;
2247 
2248 	if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2249 		memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2250 		return 0;
2251 	}
2252 
2253 	if (acpi_desc->ars_status)
2254 		devm_kfree(dev, acpi_desc->ars_status);
2255 	acpi_desc->ars_status = NULL;
2256 	ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2257 	if (!ars_status)
2258 		return -ENOMEM;
2259 	acpi_desc->ars_status = ars_status;
2260 	acpi_desc->ars_status_size = max_ars;
2261 	return 0;
2262 }
2263 
2264 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2265 		struct nfit_spa *nfit_spa)
2266 {
2267 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2268 	int rc;
2269 
2270 	if (!nfit_spa->max_ars) {
2271 		struct nd_cmd_ars_cap ars_cap;
2272 
2273 		memset(&ars_cap, 0, sizeof(ars_cap));
2274 		rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2275 		if (rc < 0)
2276 			return rc;
2277 		nfit_spa->max_ars = ars_cap.max_ars_out;
2278 		nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2279 		/* check that the supported scrub types match the spa type */
2280 		if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2281 				((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2282 			return -ENOTTY;
2283 		else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2284 				((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2285 			return -ENOTTY;
2286 	}
2287 
2288 	if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2289 		return -ENOMEM;
2290 
2291 	rc = ars_get_status(acpi_desc);
2292 	if (rc < 0 && rc != -ENOSPC)
2293 		return rc;
2294 
2295 	if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2296 		return -ENOMEM;
2297 
2298 	return 0;
2299 }
2300 
2301 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2302 		struct nfit_spa *nfit_spa)
2303 {
2304 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2305 	unsigned int overflow_retry = scrub_overflow_abort;
2306 	u64 init_ars_start = 0, init_ars_len = 0;
2307 	struct device *dev = acpi_desc->dev;
2308 	unsigned int tmo = scrub_timeout;
2309 	int rc;
2310 
2311 	if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2312 		return;
2313 
2314 	rc = ars_start(acpi_desc, nfit_spa);
2315 	/*
2316 	 * If we timed out the initial scan we'll still be busy here,
2317 	 * and will wait another timeout before giving up permanently.
2318 	 */
2319 	if (rc < 0 && rc != -EBUSY)
2320 		return;
2321 
2322 	do {
2323 		u64 ars_start, ars_len;
2324 
2325 		if (acpi_desc->cancel)
2326 			break;
2327 		rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2328 		if (rc == -ENOTTY)
2329 			break;
2330 		if (rc == -EBUSY && !tmo) {
2331 			dev_warn(dev, "range %d ars timeout, aborting\n",
2332 					spa->range_index);
2333 			break;
2334 		}
2335 
2336 		if (rc == -EBUSY) {
2337 			/*
2338 			 * Note, entries may be appended to the list
2339 			 * while the lock is dropped, but the workqueue
2340 			 * being active prevents entries being deleted /
2341 			 * freed.
2342 			 */
2343 			mutex_unlock(&acpi_desc->init_mutex);
2344 			ssleep(1);
2345 			tmo--;
2346 			mutex_lock(&acpi_desc->init_mutex);
2347 			continue;
2348 		}
2349 
2350 		/* we got some results, but there are more pending... */
2351 		if (rc == -ENOSPC && overflow_retry--) {
2352 			if (!init_ars_len) {
2353 				init_ars_len = acpi_desc->ars_status->length;
2354 				init_ars_start = acpi_desc->ars_status->address;
2355 			}
2356 			rc = ars_continue(acpi_desc);
2357 		}
2358 
2359 		if (rc < 0) {
2360 			dev_warn(dev, "range %d ars continuation failed\n",
2361 					spa->range_index);
2362 			break;
2363 		}
2364 
2365 		if (init_ars_len) {
2366 			ars_start = init_ars_start;
2367 			ars_len = init_ars_len;
2368 		} else {
2369 			ars_start = acpi_desc->ars_status->address;
2370 			ars_len = acpi_desc->ars_status->length;
2371 		}
2372 		dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2373 				spa->range_index, ars_start, ars_len);
2374 		/* notify the region about new poison entries */
2375 		nvdimm_region_notify(nfit_spa->nd_region,
2376 				NVDIMM_REVALIDATE_POISON);
2377 		break;
2378 	} while (1);
2379 }
2380 
2381 static void acpi_nfit_scrub(struct work_struct *work)
2382 {
2383 	struct device *dev;
2384 	u64 init_scrub_length = 0;
2385 	struct nfit_spa *nfit_spa;
2386 	u64 init_scrub_address = 0;
2387 	bool init_ars_done = false;
2388 	struct acpi_nfit_desc *acpi_desc;
2389 	unsigned int tmo = scrub_timeout;
2390 	unsigned int overflow_retry = scrub_overflow_abort;
2391 
2392 	acpi_desc = container_of(work, typeof(*acpi_desc), work);
2393 	dev = acpi_desc->dev;
2394 
2395 	/*
2396 	 * We scrub in 2 phases.  The first phase waits for any platform
2397 	 * firmware initiated scrubs to complete and then we go search for the
2398 	 * affected spa regions to mark them scanned.  In the second phase we
2399 	 * initiate a directed scrub for every range that was not scrubbed in
2400 	 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2401 	 * the first phase, but really only care about running phase 2, where
2402 	 * regions can be notified of new poison.
2403 	 */
2404 
2405 	/* process platform firmware initiated scrubs */
2406  retry:
2407 	mutex_lock(&acpi_desc->init_mutex);
2408 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2409 		struct nd_cmd_ars_status *ars_status;
2410 		struct acpi_nfit_system_address *spa;
2411 		u64 ars_start, ars_len;
2412 		int rc;
2413 
2414 		if (acpi_desc->cancel)
2415 			break;
2416 
2417 		if (nfit_spa->nd_region)
2418 			continue;
2419 
2420 		if (init_ars_done) {
2421 			/*
2422 			 * No need to re-query, we're now just
2423 			 * reconciling all the ranges covered by the
2424 			 * initial scrub
2425 			 */
2426 			rc = 0;
2427 		} else
2428 			rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2429 
2430 		if (rc == -ENOTTY) {
2431 			/* no ars capability, just register spa and move on */
2432 			acpi_nfit_register_region(acpi_desc, nfit_spa);
2433 			continue;
2434 		}
2435 
2436 		if (rc == -EBUSY && !tmo) {
2437 			/* fallthrough to directed scrub in phase 2 */
2438 			dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2439 			break;
2440 		} else if (rc == -EBUSY) {
2441 			mutex_unlock(&acpi_desc->init_mutex);
2442 			ssleep(1);
2443 			tmo--;
2444 			goto retry;
2445 		}
2446 
2447 		/* we got some results, but there are more pending... */
2448 		if (rc == -ENOSPC && overflow_retry--) {
2449 			ars_status = acpi_desc->ars_status;
2450 			/*
2451 			 * Record the original scrub range, so that we
2452 			 * can recall all the ranges impacted by the
2453 			 * initial scrub.
2454 			 */
2455 			if (!init_scrub_length) {
2456 				init_scrub_length = ars_status->length;
2457 				init_scrub_address = ars_status->address;
2458 			}
2459 			rc = ars_continue(acpi_desc);
2460 			if (rc == 0) {
2461 				mutex_unlock(&acpi_desc->init_mutex);
2462 				goto retry;
2463 			}
2464 		}
2465 
2466 		if (rc < 0) {
2467 			/*
2468 			 * Initial scrub failed, we'll give it one more
2469 			 * try below...
2470 			 */
2471 			break;
2472 		}
2473 
2474 		/* We got some final results, record completed ranges */
2475 		ars_status = acpi_desc->ars_status;
2476 		if (init_scrub_length) {
2477 			ars_start = init_scrub_address;
2478 			ars_len = ars_start + init_scrub_length;
2479 		} else {
2480 			ars_start = ars_status->address;
2481 			ars_len = ars_status->length;
2482 		}
2483 		spa = nfit_spa->spa;
2484 
2485 		if (!init_ars_done) {
2486 			init_ars_done = true;
2487 			dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2488 					ars_start, ars_len);
2489 		}
2490 		if (ars_start <= spa->address && ars_start + ars_len
2491 				>= spa->address + spa->length)
2492 			acpi_nfit_register_region(acpi_desc, nfit_spa);
2493 	}
2494 
2495 	/*
2496 	 * For all the ranges not covered by an initial scrub we still
2497 	 * want to see if there are errors, but it's ok to discover them
2498 	 * asynchronously.
2499 	 */
2500 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2501 		/*
2502 		 * Flag all the ranges that still need scrubbing, but
2503 		 * register them now to make data available.
2504 		 */
2505 		if (!nfit_spa->nd_region) {
2506 			nfit_spa->ars_required = 1;
2507 			acpi_nfit_register_region(acpi_desc, nfit_spa);
2508 		}
2509 	}
2510 
2511 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2512 		acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2513 	acpi_desc->scrub_count++;
2514 	if (acpi_desc->scrub_count_state)
2515 		sysfs_notify_dirent(acpi_desc->scrub_count_state);
2516 	mutex_unlock(&acpi_desc->init_mutex);
2517 }
2518 
2519 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2520 {
2521 	struct nfit_spa *nfit_spa;
2522 	int rc;
2523 
2524 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2525 		if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2526 			/* BLK regions don't need to wait for ars results */
2527 			rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2528 			if (rc)
2529 				return rc;
2530 		}
2531 
2532 	queue_work(nfit_wq, &acpi_desc->work);
2533 	return 0;
2534 }
2535 
2536 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2537 		struct nfit_table_prev *prev)
2538 {
2539 	struct device *dev = acpi_desc->dev;
2540 
2541 	if (!list_empty(&prev->spas) ||
2542 			!list_empty(&prev->memdevs) ||
2543 			!list_empty(&prev->dcrs) ||
2544 			!list_empty(&prev->bdws) ||
2545 			!list_empty(&prev->idts) ||
2546 			!list_empty(&prev->flushes)) {
2547 		dev_err(dev, "new nfit deletes entries (unsupported)\n");
2548 		return -ENXIO;
2549 	}
2550 	return 0;
2551 }
2552 
2553 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
2554 {
2555 	struct device *dev = acpi_desc->dev;
2556 	struct kernfs_node *nfit;
2557 	struct device *bus_dev;
2558 
2559 	if (!ars_supported(acpi_desc->nvdimm_bus))
2560 		return 0;
2561 
2562 	bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2563 	nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
2564 	if (!nfit) {
2565 		dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
2566 		return -ENODEV;
2567 	}
2568 	acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
2569 	sysfs_put(nfit);
2570 	if (!acpi_desc->scrub_count_state) {
2571 		dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
2572 		return -ENODEV;
2573 	}
2574 
2575 	return 0;
2576 }
2577 
2578 static void acpi_nfit_destruct(void *data)
2579 {
2580 	struct acpi_nfit_desc *acpi_desc = data;
2581 	struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2582 
2583 	/*
2584 	 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
2585 	 * race teardown
2586 	 */
2587 	mutex_lock(&acpi_desc_lock);
2588 	acpi_desc->cancel = 1;
2589 	/*
2590 	 * Bounce the nvdimm bus lock to make sure any in-flight
2591 	 * acpi_nfit_ars_rescan() submissions have had a chance to
2592 	 * either submit or see ->cancel set.
2593 	 */
2594 	device_lock(bus_dev);
2595 	device_unlock(bus_dev);
2596 
2597 	flush_workqueue(nfit_wq);
2598 	if (acpi_desc->scrub_count_state)
2599 		sysfs_put(acpi_desc->scrub_count_state);
2600 	nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2601 	acpi_desc->nvdimm_bus = NULL;
2602 	list_del(&acpi_desc->list);
2603 	mutex_unlock(&acpi_desc_lock);
2604 }
2605 
2606 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
2607 {
2608 	struct device *dev = acpi_desc->dev;
2609 	struct nfit_table_prev prev;
2610 	const void *end;
2611 	int rc;
2612 
2613 	if (!acpi_desc->nvdimm_bus) {
2614 		acpi_nfit_init_dsms(acpi_desc);
2615 
2616 		acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
2617 				&acpi_desc->nd_desc);
2618 		if (!acpi_desc->nvdimm_bus)
2619 			return -ENOMEM;
2620 
2621 		rc = devm_add_action_or_reset(dev, acpi_nfit_destruct,
2622 				acpi_desc);
2623 		if (rc)
2624 			return rc;
2625 
2626 		rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
2627 		if (rc)
2628 			return rc;
2629 
2630 		/* register this acpi_desc for mce notifications */
2631 		mutex_lock(&acpi_desc_lock);
2632 		list_add_tail(&acpi_desc->list, &acpi_descs);
2633 		mutex_unlock(&acpi_desc_lock);
2634 	}
2635 
2636 	mutex_lock(&acpi_desc->init_mutex);
2637 
2638 	INIT_LIST_HEAD(&prev.spas);
2639 	INIT_LIST_HEAD(&prev.memdevs);
2640 	INIT_LIST_HEAD(&prev.dcrs);
2641 	INIT_LIST_HEAD(&prev.bdws);
2642 	INIT_LIST_HEAD(&prev.idts);
2643 	INIT_LIST_HEAD(&prev.flushes);
2644 
2645 	list_cut_position(&prev.spas, &acpi_desc->spas,
2646 				acpi_desc->spas.prev);
2647 	list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2648 				acpi_desc->memdevs.prev);
2649 	list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2650 				acpi_desc->dcrs.prev);
2651 	list_cut_position(&prev.bdws, &acpi_desc->bdws,
2652 				acpi_desc->bdws.prev);
2653 	list_cut_position(&prev.idts, &acpi_desc->idts,
2654 				acpi_desc->idts.prev);
2655 	list_cut_position(&prev.flushes, &acpi_desc->flushes,
2656 				acpi_desc->flushes.prev);
2657 
2658 	end = data + sz;
2659 	while (!IS_ERR_OR_NULL(data))
2660 		data = add_table(acpi_desc, &prev, data, end);
2661 
2662 	if (IS_ERR(data)) {
2663 		dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2664 				PTR_ERR(data));
2665 		rc = PTR_ERR(data);
2666 		goto out_unlock;
2667 	}
2668 
2669 	rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2670 	if (rc)
2671 		goto out_unlock;
2672 
2673 	rc = nfit_mem_init(acpi_desc);
2674 	if (rc)
2675 		goto out_unlock;
2676 
2677 	rc = acpi_nfit_register_dimms(acpi_desc);
2678 	if (rc)
2679 		goto out_unlock;
2680 
2681 	rc = acpi_nfit_register_regions(acpi_desc);
2682 
2683  out_unlock:
2684 	mutex_unlock(&acpi_desc->init_mutex);
2685 	return rc;
2686 }
2687 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2688 
2689 struct acpi_nfit_flush_work {
2690 	struct work_struct work;
2691 	struct completion cmp;
2692 };
2693 
2694 static void flush_probe(struct work_struct *work)
2695 {
2696 	struct acpi_nfit_flush_work *flush;
2697 
2698 	flush = container_of(work, typeof(*flush), work);
2699 	complete(&flush->cmp);
2700 }
2701 
2702 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2703 {
2704 	struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2705 	struct device *dev = acpi_desc->dev;
2706 	struct acpi_nfit_flush_work flush;
2707 
2708 	/* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2709 	device_lock(dev);
2710 	device_unlock(dev);
2711 
2712 	/*
2713 	 * Scrub work could take 10s of seconds, userspace may give up so we
2714 	 * need to be interruptible while waiting.
2715 	 */
2716 	INIT_WORK_ONSTACK(&flush.work, flush_probe);
2717 	COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2718 	queue_work(nfit_wq, &flush.work);
2719 	return wait_for_completion_interruptible(&flush.cmp);
2720 }
2721 
2722 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2723 		struct nvdimm *nvdimm, unsigned int cmd)
2724 {
2725 	struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2726 
2727 	if (nvdimm)
2728 		return 0;
2729 	if (cmd != ND_CMD_ARS_START)
2730 		return 0;
2731 
2732 	/*
2733 	 * The kernel and userspace may race to initiate a scrub, but
2734 	 * the scrub thread is prepared to lose that initial race.  It
2735 	 * just needs guarantees that any ars it initiates are not
2736 	 * interrupted by any intervening start reqeusts from userspace.
2737 	 */
2738 	if (work_busy(&acpi_desc->work))
2739 		return -EBUSY;
2740 
2741 	return 0;
2742 }
2743 
2744 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc)
2745 {
2746 	struct device *dev = acpi_desc->dev;
2747 	struct nfit_spa *nfit_spa;
2748 
2749 	if (work_busy(&acpi_desc->work))
2750 		return -EBUSY;
2751 
2752 	if (acpi_desc->cancel)
2753 		return 0;
2754 
2755 	mutex_lock(&acpi_desc->init_mutex);
2756 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2757 		struct acpi_nfit_system_address *spa = nfit_spa->spa;
2758 
2759 		if (nfit_spa_type(spa) != NFIT_SPA_PM)
2760 			continue;
2761 
2762 		nfit_spa->ars_required = 1;
2763 	}
2764 	queue_work(nfit_wq, &acpi_desc->work);
2765 	dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
2766 	mutex_unlock(&acpi_desc->init_mutex);
2767 
2768 	return 0;
2769 }
2770 
2771 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2772 {
2773 	struct nvdimm_bus_descriptor *nd_desc;
2774 
2775 	dev_set_drvdata(dev, acpi_desc);
2776 	acpi_desc->dev = dev;
2777 	acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2778 	nd_desc = &acpi_desc->nd_desc;
2779 	nd_desc->provider_name = "ACPI.NFIT";
2780 	nd_desc->module = THIS_MODULE;
2781 	nd_desc->ndctl = acpi_nfit_ctl;
2782 	nd_desc->flush_probe = acpi_nfit_flush_probe;
2783 	nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2784 	nd_desc->attr_groups = acpi_nfit_attribute_groups;
2785 
2786 	INIT_LIST_HEAD(&acpi_desc->spas);
2787 	INIT_LIST_HEAD(&acpi_desc->dcrs);
2788 	INIT_LIST_HEAD(&acpi_desc->bdws);
2789 	INIT_LIST_HEAD(&acpi_desc->idts);
2790 	INIT_LIST_HEAD(&acpi_desc->flushes);
2791 	INIT_LIST_HEAD(&acpi_desc->memdevs);
2792 	INIT_LIST_HEAD(&acpi_desc->dimms);
2793 	INIT_LIST_HEAD(&acpi_desc->list);
2794 	mutex_init(&acpi_desc->init_mutex);
2795 	INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2796 }
2797 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2798 
2799 static int acpi_nfit_add(struct acpi_device *adev)
2800 {
2801 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2802 	struct acpi_nfit_desc *acpi_desc;
2803 	struct device *dev = &adev->dev;
2804 	struct acpi_table_header *tbl;
2805 	acpi_status status = AE_OK;
2806 	acpi_size sz;
2807 	int rc = 0;
2808 
2809 	status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
2810 	if (ACPI_FAILURE(status)) {
2811 		/* This is ok, we could have an nvdimm hotplugged later */
2812 		dev_dbg(dev, "failed to find NFIT at startup\n");
2813 		return 0;
2814 	}
2815 	sz = tbl->length;
2816 
2817 	acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2818 	if (!acpi_desc)
2819 		return -ENOMEM;
2820 	acpi_nfit_desc_init(acpi_desc, &adev->dev);
2821 
2822 	/* Save the acpi header for exporting the revision via sysfs */
2823 	acpi_desc->acpi_header = *tbl;
2824 
2825 	/* Evaluate _FIT and override with that if present */
2826 	status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2827 	if (ACPI_SUCCESS(status) && buf.length > 0) {
2828 		union acpi_object *obj = buf.pointer;
2829 
2830 		if (obj->type == ACPI_TYPE_BUFFER)
2831 			rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
2832 					obj->buffer.length);
2833 		else
2834 			dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2835 				 __func__, (int) obj->type);
2836 		kfree(buf.pointer);
2837 	} else
2838 		/* skip over the lead-in header table */
2839 		rc = acpi_nfit_init(acpi_desc, (void *) tbl
2840 				+ sizeof(struct acpi_table_nfit),
2841 				sz - sizeof(struct acpi_table_nfit));
2842 	return rc;
2843 }
2844 
2845 static int acpi_nfit_remove(struct acpi_device *adev)
2846 {
2847 	/* see acpi_nfit_destruct */
2848 	return 0;
2849 }
2850 
2851 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
2852 {
2853 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
2854 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2855 	union acpi_object *obj;
2856 	acpi_status status;
2857 	int ret;
2858 
2859 	dev_dbg(dev, "%s: event: %d\n", __func__, event);
2860 
2861 	if (event != NFIT_NOTIFY_UPDATE)
2862 		return;
2863 
2864 	if (!dev->driver) {
2865 		/* dev->driver may be null if we're being removed */
2866 		dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2867 		return;
2868 	}
2869 
2870 	if (!acpi_desc) {
2871 		acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2872 		if (!acpi_desc)
2873 			return;
2874 		acpi_nfit_desc_init(acpi_desc, dev);
2875 	} else {
2876 		/*
2877 		 * Finish previous registration before considering new
2878 		 * regions.
2879 		 */
2880 		flush_workqueue(nfit_wq);
2881 	}
2882 
2883 	/* Evaluate _FIT */
2884 	status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
2885 	if (ACPI_FAILURE(status)) {
2886 		dev_err(dev, "failed to evaluate _FIT\n");
2887 		return;
2888 	}
2889 
2890 	obj = buf.pointer;
2891 	if (obj->type == ACPI_TYPE_BUFFER) {
2892 		ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
2893 				obj->buffer.length);
2894 		if (ret)
2895 			dev_err(dev, "failed to merge updated NFIT\n");
2896 	} else
2897 		dev_err(dev, "Invalid _FIT\n");
2898 	kfree(buf.pointer);
2899 }
2900 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
2901 
2902 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2903 {
2904 	device_lock(&adev->dev);
2905 	__acpi_nfit_notify(&adev->dev, adev->handle, event);
2906 	device_unlock(&adev->dev);
2907 }
2908 
2909 static const struct acpi_device_id acpi_nfit_ids[] = {
2910 	{ "ACPI0012", 0 },
2911 	{ "", 0 },
2912 };
2913 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2914 
2915 static struct acpi_driver acpi_nfit_driver = {
2916 	.name = KBUILD_MODNAME,
2917 	.ids = acpi_nfit_ids,
2918 	.ops = {
2919 		.add = acpi_nfit_add,
2920 		.remove = acpi_nfit_remove,
2921 		.notify = acpi_nfit_notify,
2922 	},
2923 };
2924 
2925 static __init int nfit_init(void)
2926 {
2927 	BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2928 	BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2929 	BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2930 	BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2931 	BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2932 	BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2933 	BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2934 
2935 	acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2936 	acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2937 	acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2938 	acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2939 	acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2940 	acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2941 	acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2942 	acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2943 	acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2944 	acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2945 	acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2946 	acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2947 	acpi_str_to_uuid(UUID_NFIT_DIMM_N_MSFT, nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
2948 
2949 	nfit_wq = create_singlethread_workqueue("nfit");
2950 	if (!nfit_wq)
2951 		return -ENOMEM;
2952 
2953 	nfit_mce_register();
2954 
2955 	return acpi_bus_register_driver(&acpi_nfit_driver);
2956 }
2957 
2958 static __exit void nfit_exit(void)
2959 {
2960 	nfit_mce_unregister();
2961 	acpi_bus_unregister_driver(&acpi_nfit_driver);
2962 	destroy_workqueue(nfit_wq);
2963 	WARN_ON(!list_empty(&acpi_descs));
2964 }
2965 
2966 module_init(nfit_init);
2967 module_exit(nfit_exit);
2968 MODULE_LICENSE("GPL v2");
2969 MODULE_AUTHOR("Intel Corporation");
2970