xref: /openbmc/linux/drivers/acpi/nfit/core.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
1 /*
2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/list_sort.h>
14 #include <linux/libnvdimm.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/ndctl.h>
18 #include <linux/sysfs.h>
19 #include <linux/delay.h>
20 #include <linux/list.h>
21 #include <linux/acpi.h>
22 #include <linux/sort.h>
23 #include <linux/io.h>
24 #include <linux/nd.h>
25 #include <asm/cacheflush.h>
26 #include "nfit.h"
27 
28 /*
29  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
30  * irrelevant.
31  */
32 #include <linux/io-64-nonatomic-hi-lo.h>
33 
34 static bool force_enable_dimms;
35 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
36 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
37 
38 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
39 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
40 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
41 
42 /* after three payloads of overflow, it's dead jim */
43 static unsigned int scrub_overflow_abort = 3;
44 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
45 MODULE_PARM_DESC(scrub_overflow_abort,
46 		"Number of times we overflow ARS results before abort");
47 
48 static bool disable_vendor_specific;
49 module_param(disable_vendor_specific, bool, S_IRUGO);
50 MODULE_PARM_DESC(disable_vendor_specific,
51 		"Limit commands to the publicly specified set");
52 
53 static unsigned long override_dsm_mask;
54 module_param(override_dsm_mask, ulong, S_IRUGO);
55 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
56 
57 static int default_dsm_family = -1;
58 module_param(default_dsm_family, int, S_IRUGO);
59 MODULE_PARM_DESC(default_dsm_family,
60 		"Try this DSM type first when identifying NVDIMM family");
61 
62 LIST_HEAD(acpi_descs);
63 DEFINE_MUTEX(acpi_desc_lock);
64 
65 static struct workqueue_struct *nfit_wq;
66 
67 struct nfit_table_prev {
68 	struct list_head spas;
69 	struct list_head memdevs;
70 	struct list_head dcrs;
71 	struct list_head bdws;
72 	struct list_head idts;
73 	struct list_head flushes;
74 };
75 
76 static guid_t nfit_uuid[NFIT_UUID_MAX];
77 
78 const guid_t *to_nfit_uuid(enum nfit_uuids id)
79 {
80 	return &nfit_uuid[id];
81 }
82 EXPORT_SYMBOL(to_nfit_uuid);
83 
84 static struct acpi_nfit_desc *to_acpi_nfit_desc(
85 		struct nvdimm_bus_descriptor *nd_desc)
86 {
87 	return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
88 }
89 
90 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
91 {
92 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
93 
94 	/*
95 	 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
96 	 * acpi_device.
97 	 */
98 	if (!nd_desc->provider_name
99 			|| strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
100 		return NULL;
101 
102 	return to_acpi_device(acpi_desc->dev);
103 }
104 
105 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
106 {
107 	struct nd_cmd_clear_error *clear_err;
108 	struct nd_cmd_ars_status *ars_status;
109 	u16 flags;
110 
111 	switch (cmd) {
112 	case ND_CMD_ARS_CAP:
113 		if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
114 			return -ENOTTY;
115 
116 		/* Command failed */
117 		if (status & 0xffff)
118 			return -EIO;
119 
120 		/* No supported scan types for this range */
121 		flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
122 		if ((status >> 16 & flags) == 0)
123 			return -ENOTTY;
124 		return 0;
125 	case ND_CMD_ARS_START:
126 		/* ARS is in progress */
127 		if ((status & 0xffff) == NFIT_ARS_START_BUSY)
128 			return -EBUSY;
129 
130 		/* Command failed */
131 		if (status & 0xffff)
132 			return -EIO;
133 		return 0;
134 	case ND_CMD_ARS_STATUS:
135 		ars_status = buf;
136 		/* Command failed */
137 		if (status & 0xffff)
138 			return -EIO;
139 		/* Check extended status (Upper two bytes) */
140 		if (status == NFIT_ARS_STATUS_DONE)
141 			return 0;
142 
143 		/* ARS is in progress */
144 		if (status == NFIT_ARS_STATUS_BUSY)
145 			return -EBUSY;
146 
147 		/* No ARS performed for the current boot */
148 		if (status == NFIT_ARS_STATUS_NONE)
149 			return -EAGAIN;
150 
151 		/*
152 		 * ARS interrupted, either we overflowed or some other
153 		 * agent wants the scan to stop.  If we didn't overflow
154 		 * then just continue with the returned results.
155 		 */
156 		if (status == NFIT_ARS_STATUS_INTR) {
157 			if (ars_status->out_length >= 40 && (ars_status->flags
158 						& NFIT_ARS_F_OVERFLOW))
159 				return -ENOSPC;
160 			return 0;
161 		}
162 
163 		/* Unknown status */
164 		if (status >> 16)
165 			return -EIO;
166 		return 0;
167 	case ND_CMD_CLEAR_ERROR:
168 		clear_err = buf;
169 		if (status & 0xffff)
170 			return -EIO;
171 		if (!clear_err->cleared)
172 			return -EIO;
173 		if (clear_err->length > clear_err->cleared)
174 			return clear_err->cleared;
175 		return 0;
176 	default:
177 		break;
178 	}
179 
180 	/* all other non-zero status results in an error */
181 	if (status)
182 		return -EIO;
183 	return 0;
184 }
185 
186 static int xlat_nvdimm_status(void *buf, unsigned int cmd, u32 status)
187 {
188 	switch (cmd) {
189 	case ND_CMD_GET_CONFIG_SIZE:
190 		if (status >> 16 & ND_CONFIG_LOCKED)
191 			return -EACCES;
192 		break;
193 	default:
194 		break;
195 	}
196 
197 	/* all other non-zero status results in an error */
198 	if (status)
199 		return -EIO;
200 	return 0;
201 }
202 
203 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
204 		u32 status)
205 {
206 	if (!nvdimm)
207 		return xlat_bus_status(buf, cmd, status);
208 	return xlat_nvdimm_status(buf, cmd, status);
209 }
210 
211 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
212 		unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
213 {
214 	struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
215 	union acpi_object in_obj, in_buf, *out_obj;
216 	const struct nd_cmd_desc *desc = NULL;
217 	struct device *dev = acpi_desc->dev;
218 	struct nd_cmd_pkg *call_pkg = NULL;
219 	const char *cmd_name, *dimm_name;
220 	unsigned long cmd_mask, dsm_mask;
221 	u32 offset, fw_status = 0;
222 	acpi_handle handle;
223 	unsigned int func;
224 	const guid_t *guid;
225 	int rc, i;
226 
227 	func = cmd;
228 	if (cmd == ND_CMD_CALL) {
229 		call_pkg = buf;
230 		func = call_pkg->nd_command;
231 
232 		for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
233 			if (call_pkg->nd_reserved2[i])
234 				return -EINVAL;
235 	}
236 
237 	if (nvdimm) {
238 		struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
239 		struct acpi_device *adev = nfit_mem->adev;
240 
241 		if (!adev)
242 			return -ENOTTY;
243 		if (call_pkg && nfit_mem->family != call_pkg->nd_family)
244 			return -ENOTTY;
245 
246 		dimm_name = nvdimm_name(nvdimm);
247 		cmd_name = nvdimm_cmd_name(cmd);
248 		cmd_mask = nvdimm_cmd_mask(nvdimm);
249 		dsm_mask = nfit_mem->dsm_mask;
250 		desc = nd_cmd_dimm_desc(cmd);
251 		guid = to_nfit_uuid(nfit_mem->family);
252 		handle = adev->handle;
253 	} else {
254 		struct acpi_device *adev = to_acpi_dev(acpi_desc);
255 
256 		cmd_name = nvdimm_bus_cmd_name(cmd);
257 		cmd_mask = nd_desc->cmd_mask;
258 		dsm_mask = cmd_mask;
259 		if (cmd == ND_CMD_CALL)
260 			dsm_mask = nd_desc->bus_dsm_mask;
261 		desc = nd_cmd_bus_desc(cmd);
262 		guid = to_nfit_uuid(NFIT_DEV_BUS);
263 		handle = adev->handle;
264 		dimm_name = "bus";
265 	}
266 
267 	if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
268 		return -ENOTTY;
269 
270 	if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
271 		return -ENOTTY;
272 
273 	in_obj.type = ACPI_TYPE_PACKAGE;
274 	in_obj.package.count = 1;
275 	in_obj.package.elements = &in_buf;
276 	in_buf.type = ACPI_TYPE_BUFFER;
277 	in_buf.buffer.pointer = buf;
278 	in_buf.buffer.length = 0;
279 
280 	/* libnvdimm has already validated the input envelope */
281 	for (i = 0; i < desc->in_num; i++)
282 		in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
283 				i, buf);
284 
285 	if (call_pkg) {
286 		/* skip over package wrapper */
287 		in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
288 		in_buf.buffer.length = call_pkg->nd_size_in;
289 	}
290 
291 	dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
292 			__func__, dimm_name, cmd, func, in_buf.buffer.length);
293 	print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
294 			in_buf.buffer.pointer,
295 			min_t(u32, 256, in_buf.buffer.length), true);
296 
297 	out_obj = acpi_evaluate_dsm(handle, guid, 1, func, &in_obj);
298 	if (!out_obj) {
299 		dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
300 				cmd_name);
301 		return -EINVAL;
302 	}
303 
304 	if (call_pkg) {
305 		call_pkg->nd_fw_size = out_obj->buffer.length;
306 		memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
307 			out_obj->buffer.pointer,
308 			min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
309 
310 		ACPI_FREE(out_obj);
311 		/*
312 		 * Need to support FW function w/o known size in advance.
313 		 * Caller can determine required size based upon nd_fw_size.
314 		 * If we return an error (like elsewhere) then caller wouldn't
315 		 * be able to rely upon data returned to make calculation.
316 		 */
317 		return 0;
318 	}
319 
320 	if (out_obj->package.type != ACPI_TYPE_BUFFER) {
321 		dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
322 				__func__, dimm_name, cmd_name, out_obj->type);
323 		rc = -EINVAL;
324 		goto out;
325 	}
326 
327 	dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__, dimm_name,
328 			cmd_name, out_obj->buffer.length);
329 	print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
330 			out_obj->buffer.pointer,
331 			min_t(u32, 128, out_obj->buffer.length), true);
332 
333 	for (i = 0, offset = 0; i < desc->out_num; i++) {
334 		u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
335 				(u32 *) out_obj->buffer.pointer,
336 				out_obj->buffer.length - offset);
337 
338 		if (offset + out_size > out_obj->buffer.length) {
339 			dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
340 					__func__, dimm_name, cmd_name, i);
341 			break;
342 		}
343 
344 		if (in_buf.buffer.length + offset + out_size > buf_len) {
345 			dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
346 					__func__, dimm_name, cmd_name, i);
347 			rc = -ENXIO;
348 			goto out;
349 		}
350 		memcpy(buf + in_buf.buffer.length + offset,
351 				out_obj->buffer.pointer + offset, out_size);
352 		offset += out_size;
353 	}
354 
355 	/*
356 	 * Set fw_status for all the commands with a known format to be
357 	 * later interpreted by xlat_status().
358 	 */
359 	if (i >= 1 && ((cmd >= ND_CMD_ARS_CAP && cmd <= ND_CMD_CLEAR_ERROR)
360 			|| (cmd >= ND_CMD_SMART && cmd <= ND_CMD_VENDOR)))
361 		fw_status = *(u32 *) out_obj->buffer.pointer;
362 
363 	if (offset + in_buf.buffer.length < buf_len) {
364 		if (i >= 1) {
365 			/*
366 			 * status valid, return the number of bytes left
367 			 * unfilled in the output buffer
368 			 */
369 			rc = buf_len - offset - in_buf.buffer.length;
370 			if (cmd_rc)
371 				*cmd_rc = xlat_status(nvdimm, buf, cmd,
372 						fw_status);
373 		} else {
374 			dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
375 					__func__, dimm_name, cmd_name, buf_len,
376 					offset);
377 			rc = -ENXIO;
378 		}
379 	} else {
380 		rc = 0;
381 		if (cmd_rc)
382 			*cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
383 	}
384 
385  out:
386 	ACPI_FREE(out_obj);
387 
388 	return rc;
389 }
390 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
391 
392 static const char *spa_type_name(u16 type)
393 {
394 	static const char *to_name[] = {
395 		[NFIT_SPA_VOLATILE] = "volatile",
396 		[NFIT_SPA_PM] = "pmem",
397 		[NFIT_SPA_DCR] = "dimm-control-region",
398 		[NFIT_SPA_BDW] = "block-data-window",
399 		[NFIT_SPA_VDISK] = "volatile-disk",
400 		[NFIT_SPA_VCD] = "volatile-cd",
401 		[NFIT_SPA_PDISK] = "persistent-disk",
402 		[NFIT_SPA_PCD] = "persistent-cd",
403 
404 	};
405 
406 	if (type > NFIT_SPA_PCD)
407 		return "unknown";
408 
409 	return to_name[type];
410 }
411 
412 int nfit_spa_type(struct acpi_nfit_system_address *spa)
413 {
414 	int i;
415 
416 	for (i = 0; i < NFIT_UUID_MAX; i++)
417 		if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
418 			return i;
419 	return -1;
420 }
421 
422 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
423 		struct nfit_table_prev *prev,
424 		struct acpi_nfit_system_address *spa)
425 {
426 	struct device *dev = acpi_desc->dev;
427 	struct nfit_spa *nfit_spa;
428 
429 	if (spa->header.length != sizeof(*spa))
430 		return false;
431 
432 	list_for_each_entry(nfit_spa, &prev->spas, list) {
433 		if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
434 			list_move_tail(&nfit_spa->list, &acpi_desc->spas);
435 			return true;
436 		}
437 	}
438 
439 	nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
440 			GFP_KERNEL);
441 	if (!nfit_spa)
442 		return false;
443 	INIT_LIST_HEAD(&nfit_spa->list);
444 	memcpy(nfit_spa->spa, spa, sizeof(*spa));
445 	list_add_tail(&nfit_spa->list, &acpi_desc->spas);
446 	dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
447 			spa->range_index,
448 			spa_type_name(nfit_spa_type(spa)));
449 	return true;
450 }
451 
452 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
453 		struct nfit_table_prev *prev,
454 		struct acpi_nfit_memory_map *memdev)
455 {
456 	struct device *dev = acpi_desc->dev;
457 	struct nfit_memdev *nfit_memdev;
458 
459 	if (memdev->header.length != sizeof(*memdev))
460 		return false;
461 
462 	list_for_each_entry(nfit_memdev, &prev->memdevs, list)
463 		if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
464 			list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
465 			return true;
466 		}
467 
468 	nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
469 			GFP_KERNEL);
470 	if (!nfit_memdev)
471 		return false;
472 	INIT_LIST_HEAD(&nfit_memdev->list);
473 	memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
474 	list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
475 	dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
476 			__func__, memdev->device_handle, memdev->range_index,
477 			memdev->region_index, memdev->flags);
478 	return true;
479 }
480 
481 /*
482  * An implementation may provide a truncated control region if no block windows
483  * are defined.
484  */
485 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
486 {
487 	if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
488 				window_size))
489 		return 0;
490 	if (dcr->windows)
491 		return sizeof(*dcr);
492 	return offsetof(struct acpi_nfit_control_region, window_size);
493 }
494 
495 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
496 		struct nfit_table_prev *prev,
497 		struct acpi_nfit_control_region *dcr)
498 {
499 	struct device *dev = acpi_desc->dev;
500 	struct nfit_dcr *nfit_dcr;
501 
502 	if (!sizeof_dcr(dcr))
503 		return false;
504 
505 	list_for_each_entry(nfit_dcr, &prev->dcrs, list)
506 		if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
507 			list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
508 			return true;
509 		}
510 
511 	nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
512 			GFP_KERNEL);
513 	if (!nfit_dcr)
514 		return false;
515 	INIT_LIST_HEAD(&nfit_dcr->list);
516 	memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
517 	list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
518 	dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
519 			dcr->region_index, dcr->windows);
520 	return true;
521 }
522 
523 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
524 		struct nfit_table_prev *prev,
525 		struct acpi_nfit_data_region *bdw)
526 {
527 	struct device *dev = acpi_desc->dev;
528 	struct nfit_bdw *nfit_bdw;
529 
530 	if (bdw->header.length != sizeof(*bdw))
531 		return false;
532 	list_for_each_entry(nfit_bdw, &prev->bdws, list)
533 		if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
534 			list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
535 			return true;
536 		}
537 
538 	nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
539 			GFP_KERNEL);
540 	if (!nfit_bdw)
541 		return false;
542 	INIT_LIST_HEAD(&nfit_bdw->list);
543 	memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
544 	list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
545 	dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
546 			bdw->region_index, bdw->windows);
547 	return true;
548 }
549 
550 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
551 {
552 	if (idt->header.length < sizeof(*idt))
553 		return 0;
554 	return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
555 }
556 
557 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
558 		struct nfit_table_prev *prev,
559 		struct acpi_nfit_interleave *idt)
560 {
561 	struct device *dev = acpi_desc->dev;
562 	struct nfit_idt *nfit_idt;
563 
564 	if (!sizeof_idt(idt))
565 		return false;
566 
567 	list_for_each_entry(nfit_idt, &prev->idts, list) {
568 		if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
569 			continue;
570 
571 		if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
572 			list_move_tail(&nfit_idt->list, &acpi_desc->idts);
573 			return true;
574 		}
575 	}
576 
577 	nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
578 			GFP_KERNEL);
579 	if (!nfit_idt)
580 		return false;
581 	INIT_LIST_HEAD(&nfit_idt->list);
582 	memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
583 	list_add_tail(&nfit_idt->list, &acpi_desc->idts);
584 	dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
585 			idt->interleave_index, idt->line_count);
586 	return true;
587 }
588 
589 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
590 {
591 	if (flush->header.length < sizeof(*flush))
592 		return 0;
593 	return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
594 }
595 
596 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
597 		struct nfit_table_prev *prev,
598 		struct acpi_nfit_flush_address *flush)
599 {
600 	struct device *dev = acpi_desc->dev;
601 	struct nfit_flush *nfit_flush;
602 
603 	if (!sizeof_flush(flush))
604 		return false;
605 
606 	list_for_each_entry(nfit_flush, &prev->flushes, list) {
607 		if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
608 			continue;
609 
610 		if (memcmp(nfit_flush->flush, flush,
611 					sizeof_flush(flush)) == 0) {
612 			list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
613 			return true;
614 		}
615 	}
616 
617 	nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
618 			+ sizeof_flush(flush), GFP_KERNEL);
619 	if (!nfit_flush)
620 		return false;
621 	INIT_LIST_HEAD(&nfit_flush->list);
622 	memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
623 	list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
624 	dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
625 			flush->device_handle, flush->hint_count);
626 	return true;
627 }
628 
629 static void *add_table(struct acpi_nfit_desc *acpi_desc,
630 		struct nfit_table_prev *prev, void *table, const void *end)
631 {
632 	struct device *dev = acpi_desc->dev;
633 	struct acpi_nfit_header *hdr;
634 	void *err = ERR_PTR(-ENOMEM);
635 
636 	if (table >= end)
637 		return NULL;
638 
639 	hdr = table;
640 	if (!hdr->length) {
641 		dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
642 			hdr->type);
643 		return NULL;
644 	}
645 
646 	switch (hdr->type) {
647 	case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
648 		if (!add_spa(acpi_desc, prev, table))
649 			return err;
650 		break;
651 	case ACPI_NFIT_TYPE_MEMORY_MAP:
652 		if (!add_memdev(acpi_desc, prev, table))
653 			return err;
654 		break;
655 	case ACPI_NFIT_TYPE_CONTROL_REGION:
656 		if (!add_dcr(acpi_desc, prev, table))
657 			return err;
658 		break;
659 	case ACPI_NFIT_TYPE_DATA_REGION:
660 		if (!add_bdw(acpi_desc, prev, table))
661 			return err;
662 		break;
663 	case ACPI_NFIT_TYPE_INTERLEAVE:
664 		if (!add_idt(acpi_desc, prev, table))
665 			return err;
666 		break;
667 	case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
668 		if (!add_flush(acpi_desc, prev, table))
669 			return err;
670 		break;
671 	case ACPI_NFIT_TYPE_SMBIOS:
672 		dev_dbg(dev, "%s: smbios\n", __func__);
673 		break;
674 	default:
675 		dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
676 		break;
677 	}
678 
679 	return table + hdr->length;
680 }
681 
682 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
683 		struct nfit_mem *nfit_mem)
684 {
685 	u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
686 	u16 dcr = nfit_mem->dcr->region_index;
687 	struct nfit_spa *nfit_spa;
688 
689 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
690 		u16 range_index = nfit_spa->spa->range_index;
691 		int type = nfit_spa_type(nfit_spa->spa);
692 		struct nfit_memdev *nfit_memdev;
693 
694 		if (type != NFIT_SPA_BDW)
695 			continue;
696 
697 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
698 			if (nfit_memdev->memdev->range_index != range_index)
699 				continue;
700 			if (nfit_memdev->memdev->device_handle != device_handle)
701 				continue;
702 			if (nfit_memdev->memdev->region_index != dcr)
703 				continue;
704 
705 			nfit_mem->spa_bdw = nfit_spa->spa;
706 			return;
707 		}
708 	}
709 
710 	dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
711 			nfit_mem->spa_dcr->range_index);
712 	nfit_mem->bdw = NULL;
713 }
714 
715 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
716 		struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
717 {
718 	u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
719 	struct nfit_memdev *nfit_memdev;
720 	struct nfit_bdw *nfit_bdw;
721 	struct nfit_idt *nfit_idt;
722 	u16 idt_idx, range_index;
723 
724 	list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
725 		if (nfit_bdw->bdw->region_index != dcr)
726 			continue;
727 		nfit_mem->bdw = nfit_bdw->bdw;
728 		break;
729 	}
730 
731 	if (!nfit_mem->bdw)
732 		return;
733 
734 	nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
735 
736 	if (!nfit_mem->spa_bdw)
737 		return;
738 
739 	range_index = nfit_mem->spa_bdw->range_index;
740 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
741 		if (nfit_memdev->memdev->range_index != range_index ||
742 				nfit_memdev->memdev->region_index != dcr)
743 			continue;
744 		nfit_mem->memdev_bdw = nfit_memdev->memdev;
745 		idt_idx = nfit_memdev->memdev->interleave_index;
746 		list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
747 			if (nfit_idt->idt->interleave_index != idt_idx)
748 				continue;
749 			nfit_mem->idt_bdw = nfit_idt->idt;
750 			break;
751 		}
752 		break;
753 	}
754 }
755 
756 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
757 		struct acpi_nfit_system_address *spa)
758 {
759 	struct nfit_mem *nfit_mem, *found;
760 	struct nfit_memdev *nfit_memdev;
761 	int type = spa ? nfit_spa_type(spa) : 0;
762 
763 	switch (type) {
764 	case NFIT_SPA_DCR:
765 	case NFIT_SPA_PM:
766 		break;
767 	default:
768 		if (spa)
769 			return 0;
770 	}
771 
772 	/*
773 	 * This loop runs in two modes, when a dimm is mapped the loop
774 	 * adds memdev associations to an existing dimm, or creates a
775 	 * dimm. In the unmapped dimm case this loop sweeps for memdev
776 	 * instances with an invalid / zero range_index and adds those
777 	 * dimms without spa associations.
778 	 */
779 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
780 		struct nfit_flush *nfit_flush;
781 		struct nfit_dcr *nfit_dcr;
782 		u32 device_handle;
783 		u16 dcr;
784 
785 		if (spa && nfit_memdev->memdev->range_index != spa->range_index)
786 			continue;
787 		if (!spa && nfit_memdev->memdev->range_index)
788 			continue;
789 		found = NULL;
790 		dcr = nfit_memdev->memdev->region_index;
791 		device_handle = nfit_memdev->memdev->device_handle;
792 		list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
793 			if (__to_nfit_memdev(nfit_mem)->device_handle
794 					== device_handle) {
795 				found = nfit_mem;
796 				break;
797 			}
798 
799 		if (found)
800 			nfit_mem = found;
801 		else {
802 			nfit_mem = devm_kzalloc(acpi_desc->dev,
803 					sizeof(*nfit_mem), GFP_KERNEL);
804 			if (!nfit_mem)
805 				return -ENOMEM;
806 			INIT_LIST_HEAD(&nfit_mem->list);
807 			nfit_mem->acpi_desc = acpi_desc;
808 			list_add(&nfit_mem->list, &acpi_desc->dimms);
809 		}
810 
811 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
812 			if (nfit_dcr->dcr->region_index != dcr)
813 				continue;
814 			/*
815 			 * Record the control region for the dimm.  For
816 			 * the ACPI 6.1 case, where there are separate
817 			 * control regions for the pmem vs blk
818 			 * interfaces, be sure to record the extended
819 			 * blk details.
820 			 */
821 			if (!nfit_mem->dcr)
822 				nfit_mem->dcr = nfit_dcr->dcr;
823 			else if (nfit_mem->dcr->windows == 0
824 					&& nfit_dcr->dcr->windows)
825 				nfit_mem->dcr = nfit_dcr->dcr;
826 			break;
827 		}
828 
829 		list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
830 			struct acpi_nfit_flush_address *flush;
831 			u16 i;
832 
833 			if (nfit_flush->flush->device_handle != device_handle)
834 				continue;
835 			nfit_mem->nfit_flush = nfit_flush;
836 			flush = nfit_flush->flush;
837 			nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
838 					flush->hint_count
839 					* sizeof(struct resource), GFP_KERNEL);
840 			if (!nfit_mem->flush_wpq)
841 				return -ENOMEM;
842 			for (i = 0; i < flush->hint_count; i++) {
843 				struct resource *res = &nfit_mem->flush_wpq[i];
844 
845 				res->start = flush->hint_address[i];
846 				res->end = res->start + 8 - 1;
847 			}
848 			break;
849 		}
850 
851 		if (dcr && !nfit_mem->dcr) {
852 			dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
853 					spa->range_index, dcr);
854 			return -ENODEV;
855 		}
856 
857 		if (type == NFIT_SPA_DCR) {
858 			struct nfit_idt *nfit_idt;
859 			u16 idt_idx;
860 
861 			/* multiple dimms may share a SPA when interleaved */
862 			nfit_mem->spa_dcr = spa;
863 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
864 			idt_idx = nfit_memdev->memdev->interleave_index;
865 			list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
866 				if (nfit_idt->idt->interleave_index != idt_idx)
867 					continue;
868 				nfit_mem->idt_dcr = nfit_idt->idt;
869 				break;
870 			}
871 			nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
872 		} else if (type == NFIT_SPA_PM) {
873 			/*
874 			 * A single dimm may belong to multiple SPA-PM
875 			 * ranges, record at least one in addition to
876 			 * any SPA-DCR range.
877 			 */
878 			nfit_mem->memdev_pmem = nfit_memdev->memdev;
879 		} else
880 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
881 	}
882 
883 	return 0;
884 }
885 
886 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
887 {
888 	struct nfit_mem *a = container_of(_a, typeof(*a), list);
889 	struct nfit_mem *b = container_of(_b, typeof(*b), list);
890 	u32 handleA, handleB;
891 
892 	handleA = __to_nfit_memdev(a)->device_handle;
893 	handleB = __to_nfit_memdev(b)->device_handle;
894 	if (handleA < handleB)
895 		return -1;
896 	else if (handleA > handleB)
897 		return 1;
898 	return 0;
899 }
900 
901 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
902 {
903 	struct nfit_spa *nfit_spa;
904 	int rc;
905 
906 
907 	/*
908 	 * For each SPA-DCR or SPA-PMEM address range find its
909 	 * corresponding MEMDEV(s).  From each MEMDEV find the
910 	 * corresponding DCR.  Then, if we're operating on a SPA-DCR,
911 	 * try to find a SPA-BDW and a corresponding BDW that references
912 	 * the DCR.  Throw it all into an nfit_mem object.  Note, that
913 	 * BDWs are optional.
914 	 */
915 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
916 		rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
917 		if (rc)
918 			return rc;
919 	}
920 
921 	/*
922 	 * If a DIMM has failed to be mapped into SPA there will be no
923 	 * SPA entries above. Find and register all the unmapped DIMMs
924 	 * for reporting and recovery purposes.
925 	 */
926 	rc = __nfit_mem_init(acpi_desc, NULL);
927 	if (rc)
928 		return rc;
929 
930 	list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
931 
932 	return 0;
933 }
934 
935 static ssize_t bus_dsm_mask_show(struct device *dev,
936 		struct device_attribute *attr, char *buf)
937 {
938 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
939 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
940 
941 	return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
942 }
943 static struct device_attribute dev_attr_bus_dsm_mask =
944 		__ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
945 
946 static ssize_t revision_show(struct device *dev,
947 		struct device_attribute *attr, char *buf)
948 {
949 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
950 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
951 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
952 
953 	return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
954 }
955 static DEVICE_ATTR_RO(revision);
956 
957 static ssize_t hw_error_scrub_show(struct device *dev,
958 		struct device_attribute *attr, char *buf)
959 {
960 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
961 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
962 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
963 
964 	return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
965 }
966 
967 /*
968  * The 'hw_error_scrub' attribute can have the following values written to it:
969  * '0': Switch to the default mode where an exception will only insert
970  *      the address of the memory error into the poison and badblocks lists.
971  * '1': Enable a full scrub to happen if an exception for a memory error is
972  *      received.
973  */
974 static ssize_t hw_error_scrub_store(struct device *dev,
975 		struct device_attribute *attr, const char *buf, size_t size)
976 {
977 	struct nvdimm_bus_descriptor *nd_desc;
978 	ssize_t rc;
979 	long val;
980 
981 	rc = kstrtol(buf, 0, &val);
982 	if (rc)
983 		return rc;
984 
985 	device_lock(dev);
986 	nd_desc = dev_get_drvdata(dev);
987 	if (nd_desc) {
988 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
989 
990 		switch (val) {
991 		case HW_ERROR_SCRUB_ON:
992 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
993 			break;
994 		case HW_ERROR_SCRUB_OFF:
995 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
996 			break;
997 		default:
998 			rc = -EINVAL;
999 			break;
1000 		}
1001 	}
1002 	device_unlock(dev);
1003 	if (rc)
1004 		return rc;
1005 	return size;
1006 }
1007 static DEVICE_ATTR_RW(hw_error_scrub);
1008 
1009 /*
1010  * This shows the number of full Address Range Scrubs that have been
1011  * completed since driver load time. Userspace can wait on this using
1012  * select/poll etc. A '+' at the end indicates an ARS is in progress
1013  */
1014 static ssize_t scrub_show(struct device *dev,
1015 		struct device_attribute *attr, char *buf)
1016 {
1017 	struct nvdimm_bus_descriptor *nd_desc;
1018 	ssize_t rc = -ENXIO;
1019 
1020 	device_lock(dev);
1021 	nd_desc = dev_get_drvdata(dev);
1022 	if (nd_desc) {
1023 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1024 
1025 		rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1026 				(work_busy(&acpi_desc->work)) ? "+\n" : "\n");
1027 	}
1028 	device_unlock(dev);
1029 	return rc;
1030 }
1031 
1032 static ssize_t scrub_store(struct device *dev,
1033 		struct device_attribute *attr, const char *buf, size_t size)
1034 {
1035 	struct nvdimm_bus_descriptor *nd_desc;
1036 	ssize_t rc;
1037 	long val;
1038 
1039 	rc = kstrtol(buf, 0, &val);
1040 	if (rc)
1041 		return rc;
1042 	if (val != 1)
1043 		return -EINVAL;
1044 
1045 	device_lock(dev);
1046 	nd_desc = dev_get_drvdata(dev);
1047 	if (nd_desc) {
1048 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1049 
1050 		rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1051 	}
1052 	device_unlock(dev);
1053 	if (rc)
1054 		return rc;
1055 	return size;
1056 }
1057 static DEVICE_ATTR_RW(scrub);
1058 
1059 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1060 {
1061 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1062 	const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1063 		| 1 << ND_CMD_ARS_STATUS;
1064 
1065 	return (nd_desc->cmd_mask & mask) == mask;
1066 }
1067 
1068 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1069 {
1070 	struct device *dev = container_of(kobj, struct device, kobj);
1071 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1072 
1073 	if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1074 		return 0;
1075 	return a->mode;
1076 }
1077 
1078 static struct attribute *acpi_nfit_attributes[] = {
1079 	&dev_attr_revision.attr,
1080 	&dev_attr_scrub.attr,
1081 	&dev_attr_hw_error_scrub.attr,
1082 	&dev_attr_bus_dsm_mask.attr,
1083 	NULL,
1084 };
1085 
1086 static const struct attribute_group acpi_nfit_attribute_group = {
1087 	.name = "nfit",
1088 	.attrs = acpi_nfit_attributes,
1089 	.is_visible = nfit_visible,
1090 };
1091 
1092 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1093 	&nvdimm_bus_attribute_group,
1094 	&acpi_nfit_attribute_group,
1095 	NULL,
1096 };
1097 
1098 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1099 {
1100 	struct nvdimm *nvdimm = to_nvdimm(dev);
1101 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1102 
1103 	return __to_nfit_memdev(nfit_mem);
1104 }
1105 
1106 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1107 {
1108 	struct nvdimm *nvdimm = to_nvdimm(dev);
1109 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1110 
1111 	return nfit_mem->dcr;
1112 }
1113 
1114 static ssize_t handle_show(struct device *dev,
1115 		struct device_attribute *attr, char *buf)
1116 {
1117 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1118 
1119 	return sprintf(buf, "%#x\n", memdev->device_handle);
1120 }
1121 static DEVICE_ATTR_RO(handle);
1122 
1123 static ssize_t phys_id_show(struct device *dev,
1124 		struct device_attribute *attr, char *buf)
1125 {
1126 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1127 
1128 	return sprintf(buf, "%#x\n", memdev->physical_id);
1129 }
1130 static DEVICE_ATTR_RO(phys_id);
1131 
1132 static ssize_t vendor_show(struct device *dev,
1133 		struct device_attribute *attr, char *buf)
1134 {
1135 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1136 
1137 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1138 }
1139 static DEVICE_ATTR_RO(vendor);
1140 
1141 static ssize_t rev_id_show(struct device *dev,
1142 		struct device_attribute *attr, char *buf)
1143 {
1144 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1145 
1146 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1147 }
1148 static DEVICE_ATTR_RO(rev_id);
1149 
1150 static ssize_t device_show(struct device *dev,
1151 		struct device_attribute *attr, char *buf)
1152 {
1153 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1154 
1155 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1156 }
1157 static DEVICE_ATTR_RO(device);
1158 
1159 static ssize_t subsystem_vendor_show(struct device *dev,
1160 		struct device_attribute *attr, char *buf)
1161 {
1162 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1163 
1164 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1165 }
1166 static DEVICE_ATTR_RO(subsystem_vendor);
1167 
1168 static ssize_t subsystem_rev_id_show(struct device *dev,
1169 		struct device_attribute *attr, char *buf)
1170 {
1171 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1172 
1173 	return sprintf(buf, "0x%04x\n",
1174 			be16_to_cpu(dcr->subsystem_revision_id));
1175 }
1176 static DEVICE_ATTR_RO(subsystem_rev_id);
1177 
1178 static ssize_t subsystem_device_show(struct device *dev,
1179 		struct device_attribute *attr, char *buf)
1180 {
1181 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1182 
1183 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1184 }
1185 static DEVICE_ATTR_RO(subsystem_device);
1186 
1187 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1188 {
1189 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1190 	int formats = 0;
1191 
1192 	if (nfit_mem->memdev_pmem)
1193 		formats++;
1194 	if (nfit_mem->memdev_bdw)
1195 		formats++;
1196 	return formats;
1197 }
1198 
1199 static ssize_t format_show(struct device *dev,
1200 		struct device_attribute *attr, char *buf)
1201 {
1202 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1203 
1204 	return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1205 }
1206 static DEVICE_ATTR_RO(format);
1207 
1208 static ssize_t format1_show(struct device *dev,
1209 		struct device_attribute *attr, char *buf)
1210 {
1211 	u32 handle;
1212 	ssize_t rc = -ENXIO;
1213 	struct nfit_mem *nfit_mem;
1214 	struct nfit_memdev *nfit_memdev;
1215 	struct acpi_nfit_desc *acpi_desc;
1216 	struct nvdimm *nvdimm = to_nvdimm(dev);
1217 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1218 
1219 	nfit_mem = nvdimm_provider_data(nvdimm);
1220 	acpi_desc = nfit_mem->acpi_desc;
1221 	handle = to_nfit_memdev(dev)->device_handle;
1222 
1223 	/* assumes DIMMs have at most 2 published interface codes */
1224 	mutex_lock(&acpi_desc->init_mutex);
1225 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1226 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1227 		struct nfit_dcr *nfit_dcr;
1228 
1229 		if (memdev->device_handle != handle)
1230 			continue;
1231 
1232 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1233 			if (nfit_dcr->dcr->region_index != memdev->region_index)
1234 				continue;
1235 			if (nfit_dcr->dcr->code == dcr->code)
1236 				continue;
1237 			rc = sprintf(buf, "0x%04x\n",
1238 					le16_to_cpu(nfit_dcr->dcr->code));
1239 			break;
1240 		}
1241 		if (rc != ENXIO)
1242 			break;
1243 	}
1244 	mutex_unlock(&acpi_desc->init_mutex);
1245 	return rc;
1246 }
1247 static DEVICE_ATTR_RO(format1);
1248 
1249 static ssize_t formats_show(struct device *dev,
1250 		struct device_attribute *attr, char *buf)
1251 {
1252 	struct nvdimm *nvdimm = to_nvdimm(dev);
1253 
1254 	return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1255 }
1256 static DEVICE_ATTR_RO(formats);
1257 
1258 static ssize_t serial_show(struct device *dev,
1259 		struct device_attribute *attr, char *buf)
1260 {
1261 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1262 
1263 	return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1264 }
1265 static DEVICE_ATTR_RO(serial);
1266 
1267 static ssize_t family_show(struct device *dev,
1268 		struct device_attribute *attr, char *buf)
1269 {
1270 	struct nvdimm *nvdimm = to_nvdimm(dev);
1271 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1272 
1273 	if (nfit_mem->family < 0)
1274 		return -ENXIO;
1275 	return sprintf(buf, "%d\n", nfit_mem->family);
1276 }
1277 static DEVICE_ATTR_RO(family);
1278 
1279 static ssize_t dsm_mask_show(struct device *dev,
1280 		struct device_attribute *attr, char *buf)
1281 {
1282 	struct nvdimm *nvdimm = to_nvdimm(dev);
1283 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1284 
1285 	if (nfit_mem->family < 0)
1286 		return -ENXIO;
1287 	return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1288 }
1289 static DEVICE_ATTR_RO(dsm_mask);
1290 
1291 static ssize_t flags_show(struct device *dev,
1292 		struct device_attribute *attr, char *buf)
1293 {
1294 	u16 flags = to_nfit_memdev(dev)->flags;
1295 
1296 	return sprintf(buf, "%s%s%s%s%s%s%s\n",
1297 		flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1298 		flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1299 		flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1300 		flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1301 		flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1302 		flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1303 		flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1304 }
1305 static DEVICE_ATTR_RO(flags);
1306 
1307 static ssize_t id_show(struct device *dev,
1308 		struct device_attribute *attr, char *buf)
1309 {
1310 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1311 
1312 	if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1313 		return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1314 				be16_to_cpu(dcr->vendor_id),
1315 				dcr->manufacturing_location,
1316 				be16_to_cpu(dcr->manufacturing_date),
1317 				be32_to_cpu(dcr->serial_number));
1318 	else
1319 		return sprintf(buf, "%04x-%08x\n",
1320 				be16_to_cpu(dcr->vendor_id),
1321 				be32_to_cpu(dcr->serial_number));
1322 }
1323 static DEVICE_ATTR_RO(id);
1324 
1325 static struct attribute *acpi_nfit_dimm_attributes[] = {
1326 	&dev_attr_handle.attr,
1327 	&dev_attr_phys_id.attr,
1328 	&dev_attr_vendor.attr,
1329 	&dev_attr_device.attr,
1330 	&dev_attr_rev_id.attr,
1331 	&dev_attr_subsystem_vendor.attr,
1332 	&dev_attr_subsystem_device.attr,
1333 	&dev_attr_subsystem_rev_id.attr,
1334 	&dev_attr_format.attr,
1335 	&dev_attr_formats.attr,
1336 	&dev_attr_format1.attr,
1337 	&dev_attr_serial.attr,
1338 	&dev_attr_flags.attr,
1339 	&dev_attr_id.attr,
1340 	&dev_attr_family.attr,
1341 	&dev_attr_dsm_mask.attr,
1342 	NULL,
1343 };
1344 
1345 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1346 		struct attribute *a, int n)
1347 {
1348 	struct device *dev = container_of(kobj, struct device, kobj);
1349 	struct nvdimm *nvdimm = to_nvdimm(dev);
1350 
1351 	if (!to_nfit_dcr(dev)) {
1352 		/* Without a dcr only the memdev attributes can be surfaced */
1353 		if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1354 				|| a == &dev_attr_flags.attr
1355 				|| a == &dev_attr_family.attr
1356 				|| a == &dev_attr_dsm_mask.attr)
1357 			return a->mode;
1358 		return 0;
1359 	}
1360 
1361 	if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1362 		return 0;
1363 	return a->mode;
1364 }
1365 
1366 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1367 	.name = "nfit",
1368 	.attrs = acpi_nfit_dimm_attributes,
1369 	.is_visible = acpi_nfit_dimm_attr_visible,
1370 };
1371 
1372 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1373 	&nvdimm_attribute_group,
1374 	&nd_device_attribute_group,
1375 	&acpi_nfit_dimm_attribute_group,
1376 	NULL,
1377 };
1378 
1379 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1380 		u32 device_handle)
1381 {
1382 	struct nfit_mem *nfit_mem;
1383 
1384 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1385 		if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1386 			return nfit_mem->nvdimm;
1387 
1388 	return NULL;
1389 }
1390 
1391 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1392 {
1393 	struct nfit_mem *nfit_mem;
1394 	struct acpi_nfit_desc *acpi_desc;
1395 
1396 	dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1397 			event);
1398 
1399 	if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1400 		dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1401 				event);
1402 		return;
1403 	}
1404 
1405 	acpi_desc = dev_get_drvdata(dev->parent);
1406 	if (!acpi_desc)
1407 		return;
1408 
1409 	/*
1410 	 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1411 	 * is still valid.
1412 	 */
1413 	nfit_mem = dev_get_drvdata(dev);
1414 	if (nfit_mem && nfit_mem->flags_attr)
1415 		sysfs_notify_dirent(nfit_mem->flags_attr);
1416 }
1417 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1418 
1419 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1420 {
1421 	struct acpi_device *adev = data;
1422 	struct device *dev = &adev->dev;
1423 
1424 	device_lock(dev->parent);
1425 	__acpi_nvdimm_notify(dev, event);
1426 	device_unlock(dev->parent);
1427 }
1428 
1429 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1430 		struct nfit_mem *nfit_mem, u32 device_handle)
1431 {
1432 	struct acpi_device *adev, *adev_dimm;
1433 	struct device *dev = acpi_desc->dev;
1434 	unsigned long dsm_mask;
1435 	const guid_t *guid;
1436 	int i;
1437 	int family = -1;
1438 
1439 	/* nfit test assumes 1:1 relationship between commands and dsms */
1440 	nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1441 	nfit_mem->family = NVDIMM_FAMILY_INTEL;
1442 	adev = to_acpi_dev(acpi_desc);
1443 	if (!adev)
1444 		return 0;
1445 
1446 	adev_dimm = acpi_find_child_device(adev, device_handle, false);
1447 	nfit_mem->adev = adev_dimm;
1448 	if (!adev_dimm) {
1449 		dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1450 				device_handle);
1451 		return force_enable_dimms ? 0 : -ENODEV;
1452 	}
1453 
1454 	if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1455 		ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1456 		dev_err(dev, "%s: notification registration failed\n",
1457 				dev_name(&adev_dimm->dev));
1458 		return -ENXIO;
1459 	}
1460 
1461 	/*
1462 	 * Until standardization materializes we need to consider 4
1463 	 * different command sets.  Note, that checking for function0 (bit0)
1464 	 * tells us if any commands are reachable through this GUID.
1465 	 */
1466 	for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1467 		if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1468 			if (family < 0 || i == default_dsm_family)
1469 				family = i;
1470 
1471 	/* limit the supported commands to those that are publicly documented */
1472 	nfit_mem->family = family;
1473 	if (override_dsm_mask && !disable_vendor_specific)
1474 		dsm_mask = override_dsm_mask;
1475 	else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1476 		dsm_mask = 0x3fe;
1477 		if (disable_vendor_specific)
1478 			dsm_mask &= ~(1 << ND_CMD_VENDOR);
1479 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1480 		dsm_mask = 0x1c3c76;
1481 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1482 		dsm_mask = 0x1fe;
1483 		if (disable_vendor_specific)
1484 			dsm_mask &= ~(1 << 8);
1485 	} else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1486 		dsm_mask = 0xffffffff;
1487 	} else {
1488 		dev_dbg(dev, "unknown dimm command family\n");
1489 		nfit_mem->family = -1;
1490 		/* DSMs are optional, continue loading the driver... */
1491 		return 0;
1492 	}
1493 
1494 	guid = to_nfit_uuid(nfit_mem->family);
1495 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1496 		if (acpi_check_dsm(adev_dimm->handle, guid, 1, 1ULL << i))
1497 			set_bit(i, &nfit_mem->dsm_mask);
1498 
1499 	return 0;
1500 }
1501 
1502 static void shutdown_dimm_notify(void *data)
1503 {
1504 	struct acpi_nfit_desc *acpi_desc = data;
1505 	struct nfit_mem *nfit_mem;
1506 
1507 	mutex_lock(&acpi_desc->init_mutex);
1508 	/*
1509 	 * Clear out the nfit_mem->flags_attr and shut down dimm event
1510 	 * notifications.
1511 	 */
1512 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1513 		struct acpi_device *adev_dimm = nfit_mem->adev;
1514 
1515 		if (nfit_mem->flags_attr) {
1516 			sysfs_put(nfit_mem->flags_attr);
1517 			nfit_mem->flags_attr = NULL;
1518 		}
1519 		if (adev_dimm)
1520 			acpi_remove_notify_handler(adev_dimm->handle,
1521 					ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1522 	}
1523 	mutex_unlock(&acpi_desc->init_mutex);
1524 }
1525 
1526 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1527 {
1528 	struct nfit_mem *nfit_mem;
1529 	int dimm_count = 0, rc;
1530 	struct nvdimm *nvdimm;
1531 
1532 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1533 		struct acpi_nfit_flush_address *flush;
1534 		unsigned long flags = 0, cmd_mask;
1535 		struct nfit_memdev *nfit_memdev;
1536 		u32 device_handle;
1537 		u16 mem_flags;
1538 
1539 		device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1540 		nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1541 		if (nvdimm) {
1542 			dimm_count++;
1543 			continue;
1544 		}
1545 
1546 		if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1547 			set_bit(NDD_ALIASING, &flags);
1548 
1549 		/* collate flags across all memdevs for this dimm */
1550 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1551 			struct acpi_nfit_memory_map *dimm_memdev;
1552 
1553 			dimm_memdev = __to_nfit_memdev(nfit_mem);
1554 			if (dimm_memdev->device_handle
1555 					!= nfit_memdev->memdev->device_handle)
1556 				continue;
1557 			dimm_memdev->flags |= nfit_memdev->memdev->flags;
1558 		}
1559 
1560 		mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1561 		if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1562 			set_bit(NDD_UNARMED, &flags);
1563 
1564 		rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1565 		if (rc)
1566 			continue;
1567 
1568 		/*
1569 		 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1570 		 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1571 		 * userspace interface.
1572 		 */
1573 		cmd_mask = 1UL << ND_CMD_CALL;
1574 		if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1575 			cmd_mask |= nfit_mem->dsm_mask;
1576 
1577 		flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1578 			: NULL;
1579 		nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1580 				acpi_nfit_dimm_attribute_groups,
1581 				flags, cmd_mask, flush ? flush->hint_count : 0,
1582 				nfit_mem->flush_wpq);
1583 		if (!nvdimm)
1584 			return -ENOMEM;
1585 
1586 		nfit_mem->nvdimm = nvdimm;
1587 		dimm_count++;
1588 
1589 		if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1590 			continue;
1591 
1592 		dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1593 				nvdimm_name(nvdimm),
1594 		  mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1595 		  mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1596 		  mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1597 		  mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1598 		  mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1599 
1600 	}
1601 
1602 	rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1603 	if (rc)
1604 		return rc;
1605 
1606 	/*
1607 	 * Now that dimms are successfully registered, and async registration
1608 	 * is flushed, attempt to enable event notification.
1609 	 */
1610 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1611 		struct kernfs_node *nfit_kernfs;
1612 
1613 		nvdimm = nfit_mem->nvdimm;
1614 		nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1615 		if (nfit_kernfs)
1616 			nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1617 					"flags");
1618 		sysfs_put(nfit_kernfs);
1619 		if (!nfit_mem->flags_attr)
1620 			dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1621 					nvdimm_name(nvdimm));
1622 	}
1623 
1624 	return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1625 			acpi_desc);
1626 }
1627 
1628 /*
1629  * These constants are private because there are no kernel consumers of
1630  * these commands.
1631  */
1632 enum nfit_aux_cmds {
1633         NFIT_CMD_TRANSLATE_SPA = 5,
1634         NFIT_CMD_ARS_INJECT_SET = 7,
1635         NFIT_CMD_ARS_INJECT_CLEAR = 8,
1636         NFIT_CMD_ARS_INJECT_GET = 9,
1637 };
1638 
1639 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1640 {
1641 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1642 	const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1643 	struct acpi_device *adev;
1644 	unsigned long dsm_mask;
1645 	int i;
1646 
1647 	nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1648 	adev = to_acpi_dev(acpi_desc);
1649 	if (!adev)
1650 		return;
1651 
1652 	for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1653 		if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1654 			set_bit(i, &nd_desc->cmd_mask);
1655 	set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1656 
1657 	dsm_mask =
1658 		(1 << ND_CMD_ARS_CAP) |
1659 		(1 << ND_CMD_ARS_START) |
1660 		(1 << ND_CMD_ARS_STATUS) |
1661 		(1 << ND_CMD_CLEAR_ERROR) |
1662 		(1 << NFIT_CMD_TRANSLATE_SPA) |
1663 		(1 << NFIT_CMD_ARS_INJECT_SET) |
1664 		(1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1665 		(1 << NFIT_CMD_ARS_INJECT_GET);
1666 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1667 		if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1668 			set_bit(i, &nd_desc->bus_dsm_mask);
1669 }
1670 
1671 static ssize_t range_index_show(struct device *dev,
1672 		struct device_attribute *attr, char *buf)
1673 {
1674 	struct nd_region *nd_region = to_nd_region(dev);
1675 	struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1676 
1677 	return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1678 }
1679 static DEVICE_ATTR_RO(range_index);
1680 
1681 static ssize_t ecc_unit_size_show(struct device *dev,
1682 		struct device_attribute *attr, char *buf)
1683 {
1684 	struct nd_region *nd_region = to_nd_region(dev);
1685 	struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1686 
1687 	return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1688 }
1689 static DEVICE_ATTR_RO(ecc_unit_size);
1690 
1691 static struct attribute *acpi_nfit_region_attributes[] = {
1692 	&dev_attr_range_index.attr,
1693 	&dev_attr_ecc_unit_size.attr,
1694 	NULL,
1695 };
1696 
1697 static const struct attribute_group acpi_nfit_region_attribute_group = {
1698 	.name = "nfit",
1699 	.attrs = acpi_nfit_region_attributes,
1700 };
1701 
1702 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1703 	&nd_region_attribute_group,
1704 	&nd_mapping_attribute_group,
1705 	&nd_device_attribute_group,
1706 	&nd_numa_attribute_group,
1707 	&acpi_nfit_region_attribute_group,
1708 	NULL,
1709 };
1710 
1711 /* enough info to uniquely specify an interleave set */
1712 struct nfit_set_info {
1713 	struct nfit_set_info_map {
1714 		u64 region_offset;
1715 		u32 serial_number;
1716 		u32 pad;
1717 	} mapping[0];
1718 };
1719 
1720 struct nfit_set_info2 {
1721 	struct nfit_set_info_map2 {
1722 		u64 region_offset;
1723 		u32 serial_number;
1724 		u16 vendor_id;
1725 		u16 manufacturing_date;
1726 		u8  manufacturing_location;
1727 		u8  reserved[31];
1728 	} mapping[0];
1729 };
1730 
1731 static size_t sizeof_nfit_set_info(int num_mappings)
1732 {
1733 	return sizeof(struct nfit_set_info)
1734 		+ num_mappings * sizeof(struct nfit_set_info_map);
1735 }
1736 
1737 static size_t sizeof_nfit_set_info2(int num_mappings)
1738 {
1739 	return sizeof(struct nfit_set_info2)
1740 		+ num_mappings * sizeof(struct nfit_set_info_map2);
1741 }
1742 
1743 static int cmp_map_compat(const void *m0, const void *m1)
1744 {
1745 	const struct nfit_set_info_map *map0 = m0;
1746 	const struct nfit_set_info_map *map1 = m1;
1747 
1748 	return memcmp(&map0->region_offset, &map1->region_offset,
1749 			sizeof(u64));
1750 }
1751 
1752 static int cmp_map(const void *m0, const void *m1)
1753 {
1754 	const struct nfit_set_info_map *map0 = m0;
1755 	const struct nfit_set_info_map *map1 = m1;
1756 
1757 	if (map0->region_offset < map1->region_offset)
1758 		return -1;
1759 	else if (map0->region_offset > map1->region_offset)
1760 		return 1;
1761 	return 0;
1762 }
1763 
1764 static int cmp_map2(const void *m0, const void *m1)
1765 {
1766 	const struct nfit_set_info_map2 *map0 = m0;
1767 	const struct nfit_set_info_map2 *map1 = m1;
1768 
1769 	if (map0->region_offset < map1->region_offset)
1770 		return -1;
1771 	else if (map0->region_offset > map1->region_offset)
1772 		return 1;
1773 	return 0;
1774 }
1775 
1776 /* Retrieve the nth entry referencing this spa */
1777 static struct acpi_nfit_memory_map *memdev_from_spa(
1778 		struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1779 {
1780 	struct nfit_memdev *nfit_memdev;
1781 
1782 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1783 		if (nfit_memdev->memdev->range_index == range_index)
1784 			if (n-- == 0)
1785 				return nfit_memdev->memdev;
1786 	return NULL;
1787 }
1788 
1789 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1790 		struct nd_region_desc *ndr_desc,
1791 		struct acpi_nfit_system_address *spa)
1792 {
1793 	struct device *dev = acpi_desc->dev;
1794 	struct nd_interleave_set *nd_set;
1795 	u16 nr = ndr_desc->num_mappings;
1796 	struct nfit_set_info2 *info2;
1797 	struct nfit_set_info *info;
1798 	int i;
1799 
1800 	nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1801 	if (!nd_set)
1802 		return -ENOMEM;
1803 	ndr_desc->nd_set = nd_set;
1804 	guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
1805 
1806 	info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1807 	if (!info)
1808 		return -ENOMEM;
1809 
1810 	info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
1811 	if (!info2)
1812 		return -ENOMEM;
1813 
1814 	for (i = 0; i < nr; i++) {
1815 		struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1816 		struct nfit_set_info_map *map = &info->mapping[i];
1817 		struct nfit_set_info_map2 *map2 = &info2->mapping[i];
1818 		struct nvdimm *nvdimm = mapping->nvdimm;
1819 		struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1820 		struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1821 				spa->range_index, i);
1822 		struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1823 
1824 		if (!memdev || !nfit_mem->dcr) {
1825 			dev_err(dev, "%s: failed to find DCR\n", __func__);
1826 			return -ENODEV;
1827 		}
1828 
1829 		map->region_offset = memdev->region_offset;
1830 		map->serial_number = dcr->serial_number;
1831 
1832 		map2->region_offset = memdev->region_offset;
1833 		map2->serial_number = dcr->serial_number;
1834 		map2->vendor_id = dcr->vendor_id;
1835 		map2->manufacturing_date = dcr->manufacturing_date;
1836 		map2->manufacturing_location = dcr->manufacturing_location;
1837 	}
1838 
1839 	/* v1.1 namespaces */
1840 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1841 			cmp_map, NULL);
1842 	nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1843 
1844 	/* v1.2 namespaces */
1845 	sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
1846 			cmp_map2, NULL);
1847 	nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
1848 
1849 	/* support v1.1 namespaces created with the wrong sort order */
1850 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1851 			cmp_map_compat, NULL);
1852 	nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1853 
1854 	/* record the result of the sort for the mapping position */
1855 	for (i = 0; i < nr; i++) {
1856 		struct nfit_set_info_map2 *map2 = &info2->mapping[i];
1857 		int j;
1858 
1859 		for (j = 0; j < nr; j++) {
1860 			struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
1861 			struct nvdimm *nvdimm = mapping->nvdimm;
1862 			struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1863 			struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1864 
1865 			if (map2->serial_number == dcr->serial_number &&
1866 			    map2->vendor_id == dcr->vendor_id &&
1867 			    map2->manufacturing_date == dcr->manufacturing_date &&
1868 			    map2->manufacturing_location
1869 				    == dcr->manufacturing_location) {
1870 				mapping->position = i;
1871 				break;
1872 			}
1873 		}
1874 	}
1875 
1876 	ndr_desc->nd_set = nd_set;
1877 	devm_kfree(dev, info);
1878 	devm_kfree(dev, info2);
1879 
1880 	return 0;
1881 }
1882 
1883 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1884 {
1885 	struct acpi_nfit_interleave *idt = mmio->idt;
1886 	u32 sub_line_offset, line_index, line_offset;
1887 	u64 line_no, table_skip_count, table_offset;
1888 
1889 	line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1890 	table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1891 	line_offset = idt->line_offset[line_index]
1892 		* mmio->line_size;
1893 	table_offset = table_skip_count * mmio->table_size;
1894 
1895 	return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1896 }
1897 
1898 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1899 {
1900 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1901 	u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1902 	const u32 STATUS_MASK = 0x80000037;
1903 
1904 	if (mmio->num_lines)
1905 		offset = to_interleave_offset(offset, mmio);
1906 
1907 	return readl(mmio->addr.base + offset) & STATUS_MASK;
1908 }
1909 
1910 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1911 		resource_size_t dpa, unsigned int len, unsigned int write)
1912 {
1913 	u64 cmd, offset;
1914 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1915 
1916 	enum {
1917 		BCW_OFFSET_MASK = (1ULL << 48)-1,
1918 		BCW_LEN_SHIFT = 48,
1919 		BCW_LEN_MASK = (1ULL << 8) - 1,
1920 		BCW_CMD_SHIFT = 56,
1921 	};
1922 
1923 	cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1924 	len = len >> L1_CACHE_SHIFT;
1925 	cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1926 	cmd |= ((u64) write) << BCW_CMD_SHIFT;
1927 
1928 	offset = nfit_blk->cmd_offset + mmio->size * bw;
1929 	if (mmio->num_lines)
1930 		offset = to_interleave_offset(offset, mmio);
1931 
1932 	writeq(cmd, mmio->addr.base + offset);
1933 	nvdimm_flush(nfit_blk->nd_region);
1934 
1935 	if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1936 		readq(mmio->addr.base + offset);
1937 }
1938 
1939 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1940 		resource_size_t dpa, void *iobuf, size_t len, int rw,
1941 		unsigned int lane)
1942 {
1943 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1944 	unsigned int copied = 0;
1945 	u64 base_offset;
1946 	int rc;
1947 
1948 	base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1949 		+ lane * mmio->size;
1950 	write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1951 	while (len) {
1952 		unsigned int c;
1953 		u64 offset;
1954 
1955 		if (mmio->num_lines) {
1956 			u32 line_offset;
1957 
1958 			offset = to_interleave_offset(base_offset + copied,
1959 					mmio);
1960 			div_u64_rem(offset, mmio->line_size, &line_offset);
1961 			c = min_t(size_t, len, mmio->line_size - line_offset);
1962 		} else {
1963 			offset = base_offset + nfit_blk->bdw_offset;
1964 			c = len;
1965 		}
1966 
1967 		if (rw)
1968 			memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
1969 		else {
1970 			if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1971 				arch_invalidate_pmem((void __force *)
1972 					mmio->addr.aperture + offset, c);
1973 
1974 			memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
1975 		}
1976 
1977 		copied += c;
1978 		len -= c;
1979 	}
1980 
1981 	if (rw)
1982 		nvdimm_flush(nfit_blk->nd_region);
1983 
1984 	rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1985 	return rc;
1986 }
1987 
1988 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1989 		resource_size_t dpa, void *iobuf, u64 len, int rw)
1990 {
1991 	struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1992 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1993 	struct nd_region *nd_region = nfit_blk->nd_region;
1994 	unsigned int lane, copied = 0;
1995 	int rc = 0;
1996 
1997 	lane = nd_region_acquire_lane(nd_region);
1998 	while (len) {
1999 		u64 c = min(len, mmio->size);
2000 
2001 		rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2002 				iobuf + copied, c, rw, lane);
2003 		if (rc)
2004 			break;
2005 
2006 		copied += c;
2007 		len -= c;
2008 	}
2009 	nd_region_release_lane(nd_region, lane);
2010 
2011 	return rc;
2012 }
2013 
2014 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2015 		struct acpi_nfit_interleave *idt, u16 interleave_ways)
2016 {
2017 	if (idt) {
2018 		mmio->num_lines = idt->line_count;
2019 		mmio->line_size = idt->line_size;
2020 		if (interleave_ways == 0)
2021 			return -ENXIO;
2022 		mmio->table_size = mmio->num_lines * interleave_ways
2023 			* mmio->line_size;
2024 	}
2025 
2026 	return 0;
2027 }
2028 
2029 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2030 		struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2031 {
2032 	struct nd_cmd_dimm_flags flags;
2033 	int rc;
2034 
2035 	memset(&flags, 0, sizeof(flags));
2036 	rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2037 			sizeof(flags), NULL);
2038 
2039 	if (rc >= 0 && flags.status == 0)
2040 		nfit_blk->dimm_flags = flags.flags;
2041 	else if (rc == -ENOTTY) {
2042 		/* fall back to a conservative default */
2043 		nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2044 		rc = 0;
2045 	} else
2046 		rc = -ENXIO;
2047 
2048 	return rc;
2049 }
2050 
2051 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2052 		struct device *dev)
2053 {
2054 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2055 	struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2056 	struct nfit_blk_mmio *mmio;
2057 	struct nfit_blk *nfit_blk;
2058 	struct nfit_mem *nfit_mem;
2059 	struct nvdimm *nvdimm;
2060 	int rc;
2061 
2062 	nvdimm = nd_blk_region_to_dimm(ndbr);
2063 	nfit_mem = nvdimm_provider_data(nvdimm);
2064 	if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2065 		dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2066 				nfit_mem ? "" : " nfit_mem",
2067 				(nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2068 				(nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2069 		return -ENXIO;
2070 	}
2071 
2072 	nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2073 	if (!nfit_blk)
2074 		return -ENOMEM;
2075 	nd_blk_region_set_provider_data(ndbr, nfit_blk);
2076 	nfit_blk->nd_region = to_nd_region(dev);
2077 
2078 	/* map block aperture memory */
2079 	nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2080 	mmio = &nfit_blk->mmio[BDW];
2081 	mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2082                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2083 	if (!mmio->addr.base) {
2084 		dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2085 				nvdimm_name(nvdimm));
2086 		return -ENOMEM;
2087 	}
2088 	mmio->size = nfit_mem->bdw->size;
2089 	mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2090 	mmio->idt = nfit_mem->idt_bdw;
2091 	mmio->spa = nfit_mem->spa_bdw;
2092 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2093 			nfit_mem->memdev_bdw->interleave_ways);
2094 	if (rc) {
2095 		dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2096 				__func__, nvdimm_name(nvdimm));
2097 		return rc;
2098 	}
2099 
2100 	/* map block control memory */
2101 	nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2102 	nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2103 	mmio = &nfit_blk->mmio[DCR];
2104 	mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2105 			nfit_mem->spa_dcr->length);
2106 	if (!mmio->addr.base) {
2107 		dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2108 				nvdimm_name(nvdimm));
2109 		return -ENOMEM;
2110 	}
2111 	mmio->size = nfit_mem->dcr->window_size;
2112 	mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2113 	mmio->idt = nfit_mem->idt_dcr;
2114 	mmio->spa = nfit_mem->spa_dcr;
2115 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2116 			nfit_mem->memdev_dcr->interleave_ways);
2117 	if (rc) {
2118 		dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2119 				__func__, nvdimm_name(nvdimm));
2120 		return rc;
2121 	}
2122 
2123 	rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2124 	if (rc < 0) {
2125 		dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2126 				__func__, nvdimm_name(nvdimm));
2127 		return rc;
2128 	}
2129 
2130 	if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2131 		dev_warn(dev, "unable to guarantee persistence of writes\n");
2132 
2133 	if (mmio->line_size == 0)
2134 		return 0;
2135 
2136 	if ((u32) nfit_blk->cmd_offset % mmio->line_size
2137 			+ 8 > mmio->line_size) {
2138 		dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2139 		return -ENXIO;
2140 	} else if ((u32) nfit_blk->stat_offset % mmio->line_size
2141 			+ 8 > mmio->line_size) {
2142 		dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2143 		return -ENXIO;
2144 	}
2145 
2146 	return 0;
2147 }
2148 
2149 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2150 		struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2151 {
2152 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2153 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2154 	int cmd_rc, rc;
2155 
2156 	cmd->address = spa->address;
2157 	cmd->length = spa->length;
2158 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2159 			sizeof(*cmd), &cmd_rc);
2160 	if (rc < 0)
2161 		return rc;
2162 	return cmd_rc;
2163 }
2164 
2165 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2166 {
2167 	int rc;
2168 	int cmd_rc;
2169 	struct nd_cmd_ars_start ars_start;
2170 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2171 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2172 
2173 	memset(&ars_start, 0, sizeof(ars_start));
2174 	ars_start.address = spa->address;
2175 	ars_start.length = spa->length;
2176 	ars_start.flags = acpi_desc->ars_start_flags;
2177 	if (nfit_spa_type(spa) == NFIT_SPA_PM)
2178 		ars_start.type = ND_ARS_PERSISTENT;
2179 	else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2180 		ars_start.type = ND_ARS_VOLATILE;
2181 	else
2182 		return -ENOTTY;
2183 
2184 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2185 			sizeof(ars_start), &cmd_rc);
2186 
2187 	if (rc < 0)
2188 		return rc;
2189 	return cmd_rc;
2190 }
2191 
2192 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2193 {
2194 	int rc, cmd_rc;
2195 	struct nd_cmd_ars_start ars_start;
2196 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2197 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2198 
2199 	memset(&ars_start, 0, sizeof(ars_start));
2200 	ars_start.address = ars_status->restart_address;
2201 	ars_start.length = ars_status->restart_length;
2202 	ars_start.type = ars_status->type;
2203 	ars_start.flags = acpi_desc->ars_start_flags;
2204 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2205 			sizeof(ars_start), &cmd_rc);
2206 	if (rc < 0)
2207 		return rc;
2208 	return cmd_rc;
2209 }
2210 
2211 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2212 {
2213 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2214 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2215 	int rc, cmd_rc;
2216 
2217 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2218 			acpi_desc->ars_status_size, &cmd_rc);
2219 	if (rc < 0)
2220 		return rc;
2221 	return cmd_rc;
2222 }
2223 
2224 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2225 		struct nd_cmd_ars_status *ars_status)
2226 {
2227 	struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2228 	int rc;
2229 	u32 i;
2230 
2231 	/*
2232 	 * First record starts at 44 byte offset from the start of the
2233 	 * payload.
2234 	 */
2235 	if (ars_status->out_length < 44)
2236 		return 0;
2237 	for (i = 0; i < ars_status->num_records; i++) {
2238 		/* only process full records */
2239 		if (ars_status->out_length
2240 				< 44 + sizeof(struct nd_ars_record) * (i + 1))
2241 			break;
2242 		rc = nvdimm_bus_add_poison(nvdimm_bus,
2243 				ars_status->records[i].err_address,
2244 				ars_status->records[i].length);
2245 		if (rc)
2246 			return rc;
2247 	}
2248 	if (i < ars_status->num_records)
2249 		dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2250 
2251 	return 0;
2252 }
2253 
2254 static void acpi_nfit_remove_resource(void *data)
2255 {
2256 	struct resource *res = data;
2257 
2258 	remove_resource(res);
2259 }
2260 
2261 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2262 		struct nd_region_desc *ndr_desc)
2263 {
2264 	struct resource *res, *nd_res = ndr_desc->res;
2265 	int is_pmem, ret;
2266 
2267 	/* No operation if the region is already registered as PMEM */
2268 	is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2269 				IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2270 	if (is_pmem == REGION_INTERSECTS)
2271 		return 0;
2272 
2273 	res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2274 	if (!res)
2275 		return -ENOMEM;
2276 
2277 	res->name = "Persistent Memory";
2278 	res->start = nd_res->start;
2279 	res->end = nd_res->end;
2280 	res->flags = IORESOURCE_MEM;
2281 	res->desc = IORES_DESC_PERSISTENT_MEMORY;
2282 
2283 	ret = insert_resource(&iomem_resource, res);
2284 	if (ret)
2285 		return ret;
2286 
2287 	ret = devm_add_action_or_reset(acpi_desc->dev,
2288 					acpi_nfit_remove_resource,
2289 					res);
2290 	if (ret)
2291 		return ret;
2292 
2293 	return 0;
2294 }
2295 
2296 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2297 		struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2298 		struct acpi_nfit_memory_map *memdev,
2299 		struct nfit_spa *nfit_spa)
2300 {
2301 	struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2302 			memdev->device_handle);
2303 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2304 	struct nd_blk_region_desc *ndbr_desc;
2305 	struct nfit_mem *nfit_mem;
2306 	int blk_valid = 0, rc;
2307 
2308 	if (!nvdimm) {
2309 		dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2310 				spa->range_index, memdev->device_handle);
2311 		return -ENODEV;
2312 	}
2313 
2314 	mapping->nvdimm = nvdimm;
2315 	switch (nfit_spa_type(spa)) {
2316 	case NFIT_SPA_PM:
2317 	case NFIT_SPA_VOLATILE:
2318 		mapping->start = memdev->address;
2319 		mapping->size = memdev->region_size;
2320 		break;
2321 	case NFIT_SPA_DCR:
2322 		nfit_mem = nvdimm_provider_data(nvdimm);
2323 		if (!nfit_mem || !nfit_mem->bdw) {
2324 			dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2325 					spa->range_index, nvdimm_name(nvdimm));
2326 		} else {
2327 			mapping->size = nfit_mem->bdw->capacity;
2328 			mapping->start = nfit_mem->bdw->start_address;
2329 			ndr_desc->num_lanes = nfit_mem->bdw->windows;
2330 			blk_valid = 1;
2331 		}
2332 
2333 		ndr_desc->mapping = mapping;
2334 		ndr_desc->num_mappings = blk_valid;
2335 		ndbr_desc = to_blk_region_desc(ndr_desc);
2336 		ndbr_desc->enable = acpi_nfit_blk_region_enable;
2337 		ndbr_desc->do_io = acpi_desc->blk_do_io;
2338 		rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2339 		if (rc)
2340 			return rc;
2341 		nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2342 				ndr_desc);
2343 		if (!nfit_spa->nd_region)
2344 			return -ENOMEM;
2345 		break;
2346 	}
2347 
2348 	return 0;
2349 }
2350 
2351 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2352 {
2353 	return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2354 		nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2355 		nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2356 		nfit_spa_type(spa) == NFIT_SPA_PCD);
2357 }
2358 
2359 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2360 {
2361 	return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2362 		nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2363 		nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2364 }
2365 
2366 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2367 		struct nfit_spa *nfit_spa)
2368 {
2369 	static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2370 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2371 	struct nd_blk_region_desc ndbr_desc;
2372 	struct nd_region_desc *ndr_desc;
2373 	struct nfit_memdev *nfit_memdev;
2374 	struct nvdimm_bus *nvdimm_bus;
2375 	struct resource res;
2376 	int count = 0, rc;
2377 
2378 	if (nfit_spa->nd_region)
2379 		return 0;
2380 
2381 	if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2382 		dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2383 				__func__);
2384 		return 0;
2385 	}
2386 
2387 	memset(&res, 0, sizeof(res));
2388 	memset(&mappings, 0, sizeof(mappings));
2389 	memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2390 	res.start = spa->address;
2391 	res.end = res.start + spa->length - 1;
2392 	ndr_desc = &ndbr_desc.ndr_desc;
2393 	ndr_desc->res = &res;
2394 	ndr_desc->provider_data = nfit_spa;
2395 	ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2396 	if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2397 		ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2398 						spa->proximity_domain);
2399 	else
2400 		ndr_desc->numa_node = NUMA_NO_NODE;
2401 
2402 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2403 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2404 		struct nd_mapping_desc *mapping;
2405 
2406 		if (memdev->range_index != spa->range_index)
2407 			continue;
2408 		if (count >= ND_MAX_MAPPINGS) {
2409 			dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2410 					spa->range_index, ND_MAX_MAPPINGS);
2411 			return -ENXIO;
2412 		}
2413 		mapping = &mappings[count++];
2414 		rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2415 				memdev, nfit_spa);
2416 		if (rc)
2417 			goto out;
2418 	}
2419 
2420 	ndr_desc->mapping = mappings;
2421 	ndr_desc->num_mappings = count;
2422 	rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2423 	if (rc)
2424 		goto out;
2425 
2426 	nvdimm_bus = acpi_desc->nvdimm_bus;
2427 	if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2428 		rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2429 		if (rc) {
2430 			dev_warn(acpi_desc->dev,
2431 				"failed to insert pmem resource to iomem: %d\n",
2432 				rc);
2433 			goto out;
2434 		}
2435 
2436 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2437 				ndr_desc);
2438 		if (!nfit_spa->nd_region)
2439 			rc = -ENOMEM;
2440 	} else if (nfit_spa_is_volatile(spa)) {
2441 		nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2442 				ndr_desc);
2443 		if (!nfit_spa->nd_region)
2444 			rc = -ENOMEM;
2445 	} else if (nfit_spa_is_virtual(spa)) {
2446 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2447 				ndr_desc);
2448 		if (!nfit_spa->nd_region)
2449 			rc = -ENOMEM;
2450 	}
2451 
2452  out:
2453 	if (rc)
2454 		dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2455 				nfit_spa->spa->range_index);
2456 	return rc;
2457 }
2458 
2459 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2460 		u32 max_ars)
2461 {
2462 	struct device *dev = acpi_desc->dev;
2463 	struct nd_cmd_ars_status *ars_status;
2464 
2465 	if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2466 		memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2467 		return 0;
2468 	}
2469 
2470 	if (acpi_desc->ars_status)
2471 		devm_kfree(dev, acpi_desc->ars_status);
2472 	acpi_desc->ars_status = NULL;
2473 	ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2474 	if (!ars_status)
2475 		return -ENOMEM;
2476 	acpi_desc->ars_status = ars_status;
2477 	acpi_desc->ars_status_size = max_ars;
2478 	return 0;
2479 }
2480 
2481 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2482 		struct nfit_spa *nfit_spa)
2483 {
2484 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2485 	int rc;
2486 
2487 	if (!nfit_spa->max_ars) {
2488 		struct nd_cmd_ars_cap ars_cap;
2489 
2490 		memset(&ars_cap, 0, sizeof(ars_cap));
2491 		rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2492 		if (rc < 0)
2493 			return rc;
2494 		nfit_spa->max_ars = ars_cap.max_ars_out;
2495 		nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2496 		/* check that the supported scrub types match the spa type */
2497 		if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2498 				((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2499 			return -ENOTTY;
2500 		else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2501 				((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2502 			return -ENOTTY;
2503 	}
2504 
2505 	if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2506 		return -ENOMEM;
2507 
2508 	rc = ars_get_status(acpi_desc);
2509 	if (rc < 0 && rc != -ENOSPC)
2510 		return rc;
2511 
2512 	if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2513 		return -ENOMEM;
2514 
2515 	return 0;
2516 }
2517 
2518 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2519 		struct nfit_spa *nfit_spa)
2520 {
2521 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2522 	unsigned int overflow_retry = scrub_overflow_abort;
2523 	u64 init_ars_start = 0, init_ars_len = 0;
2524 	struct device *dev = acpi_desc->dev;
2525 	unsigned int tmo = scrub_timeout;
2526 	int rc;
2527 
2528 	if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2529 		return;
2530 
2531 	rc = ars_start(acpi_desc, nfit_spa);
2532 	/*
2533 	 * If we timed out the initial scan we'll still be busy here,
2534 	 * and will wait another timeout before giving up permanently.
2535 	 */
2536 	if (rc < 0 && rc != -EBUSY)
2537 		return;
2538 
2539 	do {
2540 		u64 ars_start, ars_len;
2541 
2542 		if (acpi_desc->cancel)
2543 			break;
2544 		rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2545 		if (rc == -ENOTTY)
2546 			break;
2547 		if (rc == -EBUSY && !tmo) {
2548 			dev_warn(dev, "range %d ars timeout, aborting\n",
2549 					spa->range_index);
2550 			break;
2551 		}
2552 
2553 		if (rc == -EBUSY) {
2554 			/*
2555 			 * Note, entries may be appended to the list
2556 			 * while the lock is dropped, but the workqueue
2557 			 * being active prevents entries being deleted /
2558 			 * freed.
2559 			 */
2560 			mutex_unlock(&acpi_desc->init_mutex);
2561 			ssleep(1);
2562 			tmo--;
2563 			mutex_lock(&acpi_desc->init_mutex);
2564 			continue;
2565 		}
2566 
2567 		/* we got some results, but there are more pending... */
2568 		if (rc == -ENOSPC && overflow_retry--) {
2569 			if (!init_ars_len) {
2570 				init_ars_len = acpi_desc->ars_status->length;
2571 				init_ars_start = acpi_desc->ars_status->address;
2572 			}
2573 			rc = ars_continue(acpi_desc);
2574 		}
2575 
2576 		if (rc < 0) {
2577 			dev_warn(dev, "range %d ars continuation failed\n",
2578 					spa->range_index);
2579 			break;
2580 		}
2581 
2582 		if (init_ars_len) {
2583 			ars_start = init_ars_start;
2584 			ars_len = init_ars_len;
2585 		} else {
2586 			ars_start = acpi_desc->ars_status->address;
2587 			ars_len = acpi_desc->ars_status->length;
2588 		}
2589 		dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2590 				spa->range_index, ars_start, ars_len);
2591 		/* notify the region about new poison entries */
2592 		nvdimm_region_notify(nfit_spa->nd_region,
2593 				NVDIMM_REVALIDATE_POISON);
2594 		break;
2595 	} while (1);
2596 }
2597 
2598 static void acpi_nfit_scrub(struct work_struct *work)
2599 {
2600 	struct device *dev;
2601 	u64 init_scrub_length = 0;
2602 	struct nfit_spa *nfit_spa;
2603 	u64 init_scrub_address = 0;
2604 	bool init_ars_done = false;
2605 	struct acpi_nfit_desc *acpi_desc;
2606 	unsigned int tmo = scrub_timeout;
2607 	unsigned int overflow_retry = scrub_overflow_abort;
2608 
2609 	acpi_desc = container_of(work, typeof(*acpi_desc), work);
2610 	dev = acpi_desc->dev;
2611 
2612 	/*
2613 	 * We scrub in 2 phases.  The first phase waits for any platform
2614 	 * firmware initiated scrubs to complete and then we go search for the
2615 	 * affected spa regions to mark them scanned.  In the second phase we
2616 	 * initiate a directed scrub for every range that was not scrubbed in
2617 	 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2618 	 * the first phase, but really only care about running phase 2, where
2619 	 * regions can be notified of new poison.
2620 	 */
2621 
2622 	/* process platform firmware initiated scrubs */
2623  retry:
2624 	mutex_lock(&acpi_desc->init_mutex);
2625 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2626 		struct nd_cmd_ars_status *ars_status;
2627 		struct acpi_nfit_system_address *spa;
2628 		u64 ars_start, ars_len;
2629 		int rc;
2630 
2631 		if (acpi_desc->cancel)
2632 			break;
2633 
2634 		if (nfit_spa->nd_region)
2635 			continue;
2636 
2637 		if (init_ars_done) {
2638 			/*
2639 			 * No need to re-query, we're now just
2640 			 * reconciling all the ranges covered by the
2641 			 * initial scrub
2642 			 */
2643 			rc = 0;
2644 		} else
2645 			rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2646 
2647 		if (rc == -ENOTTY) {
2648 			/* no ars capability, just register spa and move on */
2649 			acpi_nfit_register_region(acpi_desc, nfit_spa);
2650 			continue;
2651 		}
2652 
2653 		if (rc == -EBUSY && !tmo) {
2654 			/* fallthrough to directed scrub in phase 2 */
2655 			dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2656 			break;
2657 		} else if (rc == -EBUSY) {
2658 			mutex_unlock(&acpi_desc->init_mutex);
2659 			ssleep(1);
2660 			tmo--;
2661 			goto retry;
2662 		}
2663 
2664 		/* we got some results, but there are more pending... */
2665 		if (rc == -ENOSPC && overflow_retry--) {
2666 			ars_status = acpi_desc->ars_status;
2667 			/*
2668 			 * Record the original scrub range, so that we
2669 			 * can recall all the ranges impacted by the
2670 			 * initial scrub.
2671 			 */
2672 			if (!init_scrub_length) {
2673 				init_scrub_length = ars_status->length;
2674 				init_scrub_address = ars_status->address;
2675 			}
2676 			rc = ars_continue(acpi_desc);
2677 			if (rc == 0) {
2678 				mutex_unlock(&acpi_desc->init_mutex);
2679 				goto retry;
2680 			}
2681 		}
2682 
2683 		if (rc < 0) {
2684 			/*
2685 			 * Initial scrub failed, we'll give it one more
2686 			 * try below...
2687 			 */
2688 			break;
2689 		}
2690 
2691 		/* We got some final results, record completed ranges */
2692 		ars_status = acpi_desc->ars_status;
2693 		if (init_scrub_length) {
2694 			ars_start = init_scrub_address;
2695 			ars_len = ars_start + init_scrub_length;
2696 		} else {
2697 			ars_start = ars_status->address;
2698 			ars_len = ars_status->length;
2699 		}
2700 		spa = nfit_spa->spa;
2701 
2702 		if (!init_ars_done) {
2703 			init_ars_done = true;
2704 			dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2705 					ars_start, ars_len);
2706 		}
2707 		if (ars_start <= spa->address && ars_start + ars_len
2708 				>= spa->address + spa->length)
2709 			acpi_nfit_register_region(acpi_desc, nfit_spa);
2710 	}
2711 
2712 	/*
2713 	 * For all the ranges not covered by an initial scrub we still
2714 	 * want to see if there are errors, but it's ok to discover them
2715 	 * asynchronously.
2716 	 */
2717 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2718 		/*
2719 		 * Flag all the ranges that still need scrubbing, but
2720 		 * register them now to make data available.
2721 		 */
2722 		if (!nfit_spa->nd_region) {
2723 			nfit_spa->ars_required = 1;
2724 			acpi_nfit_register_region(acpi_desc, nfit_spa);
2725 		}
2726 	}
2727 	acpi_desc->init_complete = 1;
2728 
2729 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2730 		acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2731 	acpi_desc->scrub_count++;
2732 	acpi_desc->ars_start_flags = 0;
2733 	if (acpi_desc->scrub_count_state)
2734 		sysfs_notify_dirent(acpi_desc->scrub_count_state);
2735 	mutex_unlock(&acpi_desc->init_mutex);
2736 }
2737 
2738 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2739 {
2740 	struct nfit_spa *nfit_spa;
2741 	int rc;
2742 
2743 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2744 		if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2745 			/* BLK regions don't need to wait for ars results */
2746 			rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2747 			if (rc)
2748 				return rc;
2749 		}
2750 
2751 	acpi_desc->ars_start_flags = 0;
2752 	if (!acpi_desc->cancel)
2753 		queue_work(nfit_wq, &acpi_desc->work);
2754 	return 0;
2755 }
2756 
2757 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2758 		struct nfit_table_prev *prev)
2759 {
2760 	struct device *dev = acpi_desc->dev;
2761 
2762 	if (!list_empty(&prev->spas) ||
2763 			!list_empty(&prev->memdevs) ||
2764 			!list_empty(&prev->dcrs) ||
2765 			!list_empty(&prev->bdws) ||
2766 			!list_empty(&prev->idts) ||
2767 			!list_empty(&prev->flushes)) {
2768 		dev_err(dev, "new nfit deletes entries (unsupported)\n");
2769 		return -ENXIO;
2770 	}
2771 	return 0;
2772 }
2773 
2774 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
2775 {
2776 	struct device *dev = acpi_desc->dev;
2777 	struct kernfs_node *nfit;
2778 	struct device *bus_dev;
2779 
2780 	if (!ars_supported(acpi_desc->nvdimm_bus))
2781 		return 0;
2782 
2783 	bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2784 	nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
2785 	if (!nfit) {
2786 		dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
2787 		return -ENODEV;
2788 	}
2789 	acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
2790 	sysfs_put(nfit);
2791 	if (!acpi_desc->scrub_count_state) {
2792 		dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
2793 		return -ENODEV;
2794 	}
2795 
2796 	return 0;
2797 }
2798 
2799 static void acpi_nfit_unregister(void *data)
2800 {
2801 	struct acpi_nfit_desc *acpi_desc = data;
2802 
2803 	nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2804 }
2805 
2806 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
2807 {
2808 	struct device *dev = acpi_desc->dev;
2809 	struct nfit_table_prev prev;
2810 	const void *end;
2811 	int rc;
2812 
2813 	if (!acpi_desc->nvdimm_bus) {
2814 		acpi_nfit_init_dsms(acpi_desc);
2815 
2816 		acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
2817 				&acpi_desc->nd_desc);
2818 		if (!acpi_desc->nvdimm_bus)
2819 			return -ENOMEM;
2820 
2821 		rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
2822 				acpi_desc);
2823 		if (rc)
2824 			return rc;
2825 
2826 		rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
2827 		if (rc)
2828 			return rc;
2829 
2830 		/* register this acpi_desc for mce notifications */
2831 		mutex_lock(&acpi_desc_lock);
2832 		list_add_tail(&acpi_desc->list, &acpi_descs);
2833 		mutex_unlock(&acpi_desc_lock);
2834 	}
2835 
2836 	mutex_lock(&acpi_desc->init_mutex);
2837 
2838 	INIT_LIST_HEAD(&prev.spas);
2839 	INIT_LIST_HEAD(&prev.memdevs);
2840 	INIT_LIST_HEAD(&prev.dcrs);
2841 	INIT_LIST_HEAD(&prev.bdws);
2842 	INIT_LIST_HEAD(&prev.idts);
2843 	INIT_LIST_HEAD(&prev.flushes);
2844 
2845 	list_cut_position(&prev.spas, &acpi_desc->spas,
2846 				acpi_desc->spas.prev);
2847 	list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2848 				acpi_desc->memdevs.prev);
2849 	list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2850 				acpi_desc->dcrs.prev);
2851 	list_cut_position(&prev.bdws, &acpi_desc->bdws,
2852 				acpi_desc->bdws.prev);
2853 	list_cut_position(&prev.idts, &acpi_desc->idts,
2854 				acpi_desc->idts.prev);
2855 	list_cut_position(&prev.flushes, &acpi_desc->flushes,
2856 				acpi_desc->flushes.prev);
2857 
2858 	end = data + sz;
2859 	while (!IS_ERR_OR_NULL(data))
2860 		data = add_table(acpi_desc, &prev, data, end);
2861 
2862 	if (IS_ERR(data)) {
2863 		dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2864 				PTR_ERR(data));
2865 		rc = PTR_ERR(data);
2866 		goto out_unlock;
2867 	}
2868 
2869 	rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2870 	if (rc)
2871 		goto out_unlock;
2872 
2873 	rc = nfit_mem_init(acpi_desc);
2874 	if (rc)
2875 		goto out_unlock;
2876 
2877 	rc = acpi_nfit_register_dimms(acpi_desc);
2878 	if (rc)
2879 		goto out_unlock;
2880 
2881 	rc = acpi_nfit_register_regions(acpi_desc);
2882 
2883  out_unlock:
2884 	mutex_unlock(&acpi_desc->init_mutex);
2885 	return rc;
2886 }
2887 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2888 
2889 struct acpi_nfit_flush_work {
2890 	struct work_struct work;
2891 	struct completion cmp;
2892 };
2893 
2894 static void flush_probe(struct work_struct *work)
2895 {
2896 	struct acpi_nfit_flush_work *flush;
2897 
2898 	flush = container_of(work, typeof(*flush), work);
2899 	complete(&flush->cmp);
2900 }
2901 
2902 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2903 {
2904 	struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2905 	struct device *dev = acpi_desc->dev;
2906 	struct acpi_nfit_flush_work flush;
2907 	int rc;
2908 
2909 	/* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2910 	device_lock(dev);
2911 	device_unlock(dev);
2912 
2913 	/* bounce the init_mutex to make init_complete valid */
2914 	mutex_lock(&acpi_desc->init_mutex);
2915 	if (acpi_desc->cancel || acpi_desc->init_complete) {
2916 		mutex_unlock(&acpi_desc->init_mutex);
2917 		return 0;
2918 	}
2919 
2920 	/*
2921 	 * Scrub work could take 10s of seconds, userspace may give up so we
2922 	 * need to be interruptible while waiting.
2923 	 */
2924 	INIT_WORK_ONSTACK(&flush.work, flush_probe);
2925 	init_completion(&flush.cmp);
2926 	queue_work(nfit_wq, &flush.work);
2927 	mutex_unlock(&acpi_desc->init_mutex);
2928 
2929 	rc = wait_for_completion_interruptible(&flush.cmp);
2930 	cancel_work_sync(&flush.work);
2931 	return rc;
2932 }
2933 
2934 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2935 		struct nvdimm *nvdimm, unsigned int cmd)
2936 {
2937 	struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2938 
2939 	if (nvdimm)
2940 		return 0;
2941 	if (cmd != ND_CMD_ARS_START)
2942 		return 0;
2943 
2944 	/*
2945 	 * The kernel and userspace may race to initiate a scrub, but
2946 	 * the scrub thread is prepared to lose that initial race.  It
2947 	 * just needs guarantees that any ars it initiates are not
2948 	 * interrupted by any intervening start reqeusts from userspace.
2949 	 */
2950 	if (work_busy(&acpi_desc->work))
2951 		return -EBUSY;
2952 
2953 	return 0;
2954 }
2955 
2956 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
2957 {
2958 	struct device *dev = acpi_desc->dev;
2959 	struct nfit_spa *nfit_spa;
2960 
2961 	if (work_busy(&acpi_desc->work))
2962 		return -EBUSY;
2963 
2964 	mutex_lock(&acpi_desc->init_mutex);
2965 	if (acpi_desc->cancel) {
2966 		mutex_unlock(&acpi_desc->init_mutex);
2967 		return 0;
2968 	}
2969 
2970 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2971 		struct acpi_nfit_system_address *spa = nfit_spa->spa;
2972 
2973 		if (nfit_spa_type(spa) != NFIT_SPA_PM)
2974 			continue;
2975 
2976 		nfit_spa->ars_required = 1;
2977 	}
2978 	acpi_desc->ars_start_flags = flags;
2979 	queue_work(nfit_wq, &acpi_desc->work);
2980 	dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
2981 	mutex_unlock(&acpi_desc->init_mutex);
2982 
2983 	return 0;
2984 }
2985 
2986 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2987 {
2988 	struct nvdimm_bus_descriptor *nd_desc;
2989 
2990 	dev_set_drvdata(dev, acpi_desc);
2991 	acpi_desc->dev = dev;
2992 	acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2993 	nd_desc = &acpi_desc->nd_desc;
2994 	nd_desc->provider_name = "ACPI.NFIT";
2995 	nd_desc->module = THIS_MODULE;
2996 	nd_desc->ndctl = acpi_nfit_ctl;
2997 	nd_desc->flush_probe = acpi_nfit_flush_probe;
2998 	nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2999 	nd_desc->attr_groups = acpi_nfit_attribute_groups;
3000 
3001 	INIT_LIST_HEAD(&acpi_desc->spas);
3002 	INIT_LIST_HEAD(&acpi_desc->dcrs);
3003 	INIT_LIST_HEAD(&acpi_desc->bdws);
3004 	INIT_LIST_HEAD(&acpi_desc->idts);
3005 	INIT_LIST_HEAD(&acpi_desc->flushes);
3006 	INIT_LIST_HEAD(&acpi_desc->memdevs);
3007 	INIT_LIST_HEAD(&acpi_desc->dimms);
3008 	INIT_LIST_HEAD(&acpi_desc->list);
3009 	mutex_init(&acpi_desc->init_mutex);
3010 	INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3011 }
3012 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3013 
3014 static void acpi_nfit_put_table(void *table)
3015 {
3016 	acpi_put_table(table);
3017 }
3018 
3019 void acpi_nfit_shutdown(void *data)
3020 {
3021 	struct acpi_nfit_desc *acpi_desc = data;
3022 	struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3023 
3024 	/*
3025 	 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3026 	 * race teardown
3027 	 */
3028 	mutex_lock(&acpi_desc_lock);
3029 	list_del(&acpi_desc->list);
3030 	mutex_unlock(&acpi_desc_lock);
3031 
3032 	mutex_lock(&acpi_desc->init_mutex);
3033 	acpi_desc->cancel = 1;
3034 	mutex_unlock(&acpi_desc->init_mutex);
3035 
3036 	/*
3037 	 * Bounce the nvdimm bus lock to make sure any in-flight
3038 	 * acpi_nfit_ars_rescan() submissions have had a chance to
3039 	 * either submit or see ->cancel set.
3040 	 */
3041 	device_lock(bus_dev);
3042 	device_unlock(bus_dev);
3043 
3044 	flush_workqueue(nfit_wq);
3045 }
3046 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3047 
3048 static int acpi_nfit_add(struct acpi_device *adev)
3049 {
3050 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3051 	struct acpi_nfit_desc *acpi_desc;
3052 	struct device *dev = &adev->dev;
3053 	struct acpi_table_header *tbl;
3054 	acpi_status status = AE_OK;
3055 	acpi_size sz;
3056 	int rc = 0;
3057 
3058 	status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3059 	if (ACPI_FAILURE(status)) {
3060 		/* This is ok, we could have an nvdimm hotplugged later */
3061 		dev_dbg(dev, "failed to find NFIT at startup\n");
3062 		return 0;
3063 	}
3064 
3065 	rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3066 	if (rc)
3067 		return rc;
3068 	sz = tbl->length;
3069 
3070 	acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3071 	if (!acpi_desc)
3072 		return -ENOMEM;
3073 	acpi_nfit_desc_init(acpi_desc, &adev->dev);
3074 
3075 	/* Save the acpi header for exporting the revision via sysfs */
3076 	acpi_desc->acpi_header = *tbl;
3077 
3078 	/* Evaluate _FIT and override with that if present */
3079 	status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3080 	if (ACPI_SUCCESS(status) && buf.length > 0) {
3081 		union acpi_object *obj = buf.pointer;
3082 
3083 		if (obj->type == ACPI_TYPE_BUFFER)
3084 			rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3085 					obj->buffer.length);
3086 		else
3087 			dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
3088 				 __func__, (int) obj->type);
3089 		kfree(buf.pointer);
3090 	} else
3091 		/* skip over the lead-in header table */
3092 		rc = acpi_nfit_init(acpi_desc, (void *) tbl
3093 				+ sizeof(struct acpi_table_nfit),
3094 				sz - sizeof(struct acpi_table_nfit));
3095 
3096 	if (rc)
3097 		return rc;
3098 	return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3099 }
3100 
3101 static int acpi_nfit_remove(struct acpi_device *adev)
3102 {
3103 	/* see acpi_nfit_unregister */
3104 	return 0;
3105 }
3106 
3107 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3108 {
3109 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3110 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3111 	union acpi_object *obj;
3112 	acpi_status status;
3113 	int ret;
3114 
3115 	if (!dev->driver) {
3116 		/* dev->driver may be null if we're being removed */
3117 		dev_dbg(dev, "%s: no driver found for dev\n", __func__);
3118 		return;
3119 	}
3120 
3121 	if (!acpi_desc) {
3122 		acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3123 		if (!acpi_desc)
3124 			return;
3125 		acpi_nfit_desc_init(acpi_desc, dev);
3126 	} else {
3127 		/*
3128 		 * Finish previous registration before considering new
3129 		 * regions.
3130 		 */
3131 		flush_workqueue(nfit_wq);
3132 	}
3133 
3134 	/* Evaluate _FIT */
3135 	status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3136 	if (ACPI_FAILURE(status)) {
3137 		dev_err(dev, "failed to evaluate _FIT\n");
3138 		return;
3139 	}
3140 
3141 	obj = buf.pointer;
3142 	if (obj->type == ACPI_TYPE_BUFFER) {
3143 		ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3144 				obj->buffer.length);
3145 		if (ret)
3146 			dev_err(dev, "failed to merge updated NFIT\n");
3147 	} else
3148 		dev_err(dev, "Invalid _FIT\n");
3149 	kfree(buf.pointer);
3150 }
3151 
3152 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3153 {
3154 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3155 	u8 flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3156 			0 : ND_ARS_RETURN_PREV_DATA;
3157 
3158 	acpi_nfit_ars_rescan(acpi_desc, flags);
3159 }
3160 
3161 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3162 {
3163 	dev_dbg(dev, "%s: event: 0x%x\n", __func__, event);
3164 
3165 	switch (event) {
3166 	case NFIT_NOTIFY_UPDATE:
3167 		return acpi_nfit_update_notify(dev, handle);
3168 	case NFIT_NOTIFY_UC_MEMORY_ERROR:
3169 		return acpi_nfit_uc_error_notify(dev, handle);
3170 	default:
3171 		return;
3172 	}
3173 }
3174 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3175 
3176 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3177 {
3178 	device_lock(&adev->dev);
3179 	__acpi_nfit_notify(&adev->dev, adev->handle, event);
3180 	device_unlock(&adev->dev);
3181 }
3182 
3183 static const struct acpi_device_id acpi_nfit_ids[] = {
3184 	{ "ACPI0012", 0 },
3185 	{ "", 0 },
3186 };
3187 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3188 
3189 static struct acpi_driver acpi_nfit_driver = {
3190 	.name = KBUILD_MODNAME,
3191 	.ids = acpi_nfit_ids,
3192 	.ops = {
3193 		.add = acpi_nfit_add,
3194 		.remove = acpi_nfit_remove,
3195 		.notify = acpi_nfit_notify,
3196 	},
3197 };
3198 
3199 static __init int nfit_init(void)
3200 {
3201 	int ret;
3202 
3203 	BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3204 	BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3205 	BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3206 	BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3207 	BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3208 	BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3209 	BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3210 
3211 	guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3212 	guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3213 	guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3214 	guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3215 	guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3216 	guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3217 	guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3218 	guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3219 	guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3220 	guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3221 	guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3222 	guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3223 	guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3224 
3225 	nfit_wq = create_singlethread_workqueue("nfit");
3226 	if (!nfit_wq)
3227 		return -ENOMEM;
3228 
3229 	nfit_mce_register();
3230 	ret = acpi_bus_register_driver(&acpi_nfit_driver);
3231 	if (ret) {
3232 		nfit_mce_unregister();
3233 		destroy_workqueue(nfit_wq);
3234 	}
3235 
3236 	return ret;
3237 
3238 }
3239 
3240 static __exit void nfit_exit(void)
3241 {
3242 	nfit_mce_unregister();
3243 	acpi_bus_unregister_driver(&acpi_nfit_driver);
3244 	destroy_workqueue(nfit_wq);
3245 	WARN_ON(!list_empty(&acpi_descs));
3246 }
3247 
3248 module_init(nfit_init);
3249 module_exit(nfit_exit);
3250 MODULE_LICENSE("GPL v2");
3251 MODULE_AUTHOR("Intel Corporation");
3252