xref: /openbmc/linux/drivers/acpi/nfit/core.c (revision 62e59c4e)
1 /*
2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/list_sort.h>
14 #include <linux/libnvdimm.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/ndctl.h>
18 #include <linux/sysfs.h>
19 #include <linux/delay.h>
20 #include <linux/list.h>
21 #include <linux/acpi.h>
22 #include <linux/sort.h>
23 #include <linux/io.h>
24 #include <linux/nd.h>
25 #include <asm/cacheflush.h>
26 #include <acpi/nfit.h>
27 #include "intel.h"
28 #include "nfit.h"
29 
30 /*
31  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
32  * irrelevant.
33  */
34 #include <linux/io-64-nonatomic-hi-lo.h>
35 
36 static bool force_enable_dimms;
37 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
38 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
39 
40 static bool disable_vendor_specific;
41 module_param(disable_vendor_specific, bool, S_IRUGO);
42 MODULE_PARM_DESC(disable_vendor_specific,
43 		"Limit commands to the publicly specified set");
44 
45 static unsigned long override_dsm_mask;
46 module_param(override_dsm_mask, ulong, S_IRUGO);
47 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
48 
49 static int default_dsm_family = -1;
50 module_param(default_dsm_family, int, S_IRUGO);
51 MODULE_PARM_DESC(default_dsm_family,
52 		"Try this DSM type first when identifying NVDIMM family");
53 
54 static bool no_init_ars;
55 module_param(no_init_ars, bool, 0644);
56 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
57 
58 static bool force_labels;
59 module_param(force_labels, bool, 0444);
60 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
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_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
85 {
86 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
87 
88 	/*
89 	 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
90 	 * acpi_device.
91 	 */
92 	if (!nd_desc->provider_name
93 			|| strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
94 		return NULL;
95 
96 	return to_acpi_device(acpi_desc->dev);
97 }
98 
99 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
100 {
101 	struct nd_cmd_clear_error *clear_err;
102 	struct nd_cmd_ars_status *ars_status;
103 	u16 flags;
104 
105 	switch (cmd) {
106 	case ND_CMD_ARS_CAP:
107 		if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
108 			return -ENOTTY;
109 
110 		/* Command failed */
111 		if (status & 0xffff)
112 			return -EIO;
113 
114 		/* No supported scan types for this range */
115 		flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
116 		if ((status >> 16 & flags) == 0)
117 			return -ENOTTY;
118 		return 0;
119 	case ND_CMD_ARS_START:
120 		/* ARS is in progress */
121 		if ((status & 0xffff) == NFIT_ARS_START_BUSY)
122 			return -EBUSY;
123 
124 		/* Command failed */
125 		if (status & 0xffff)
126 			return -EIO;
127 		return 0;
128 	case ND_CMD_ARS_STATUS:
129 		ars_status = buf;
130 		/* Command failed */
131 		if (status & 0xffff)
132 			return -EIO;
133 		/* Check extended status (Upper two bytes) */
134 		if (status == NFIT_ARS_STATUS_DONE)
135 			return 0;
136 
137 		/* ARS is in progress */
138 		if (status == NFIT_ARS_STATUS_BUSY)
139 			return -EBUSY;
140 
141 		/* No ARS performed for the current boot */
142 		if (status == NFIT_ARS_STATUS_NONE)
143 			return -EAGAIN;
144 
145 		/*
146 		 * ARS interrupted, either we overflowed or some other
147 		 * agent wants the scan to stop.  If we didn't overflow
148 		 * then just continue with the returned results.
149 		 */
150 		if (status == NFIT_ARS_STATUS_INTR) {
151 			if (ars_status->out_length >= 40 && (ars_status->flags
152 						& NFIT_ARS_F_OVERFLOW))
153 				return -ENOSPC;
154 			return 0;
155 		}
156 
157 		/* Unknown status */
158 		if (status >> 16)
159 			return -EIO;
160 		return 0;
161 	case ND_CMD_CLEAR_ERROR:
162 		clear_err = buf;
163 		if (status & 0xffff)
164 			return -EIO;
165 		if (!clear_err->cleared)
166 			return -EIO;
167 		if (clear_err->length > clear_err->cleared)
168 			return clear_err->cleared;
169 		return 0;
170 	default:
171 		break;
172 	}
173 
174 	/* all other non-zero status results in an error */
175 	if (status)
176 		return -EIO;
177 	return 0;
178 }
179 
180 #define ACPI_LABELS_LOCKED 3
181 
182 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
183 		u32 status)
184 {
185 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
186 
187 	switch (cmd) {
188 	case ND_CMD_GET_CONFIG_SIZE:
189 		/*
190 		 * In the _LSI, _LSR, _LSW case the locked status is
191 		 * communicated via the read/write commands
192 		 */
193 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
194 			break;
195 
196 		if (status >> 16 & ND_CONFIG_LOCKED)
197 			return -EACCES;
198 		break;
199 	case ND_CMD_GET_CONFIG_DATA:
200 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
201 				&& status == ACPI_LABELS_LOCKED)
202 			return -EACCES;
203 		break;
204 	case ND_CMD_SET_CONFIG_DATA:
205 		if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
206 				&& status == ACPI_LABELS_LOCKED)
207 			return -EACCES;
208 		break;
209 	default:
210 		break;
211 	}
212 
213 	/* all other non-zero status results in an error */
214 	if (status)
215 		return -EIO;
216 	return 0;
217 }
218 
219 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
220 		u32 status)
221 {
222 	if (!nvdimm)
223 		return xlat_bus_status(buf, cmd, status);
224 	return xlat_nvdimm_status(nvdimm, buf, cmd, status);
225 }
226 
227 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
228 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
229 {
230 	int i;
231 	void *dst;
232 	size_t size = 0;
233 	union acpi_object *buf = NULL;
234 
235 	if (pkg->type != ACPI_TYPE_PACKAGE) {
236 		WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
237 				pkg->type);
238 		goto err;
239 	}
240 
241 	for (i = 0; i < pkg->package.count; i++) {
242 		union acpi_object *obj = &pkg->package.elements[i];
243 
244 		if (obj->type == ACPI_TYPE_INTEGER)
245 			size += 4;
246 		else if (obj->type == ACPI_TYPE_BUFFER)
247 			size += obj->buffer.length;
248 		else {
249 			WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
250 					obj->type);
251 			goto err;
252 		}
253 	}
254 
255 	buf = ACPI_ALLOCATE(sizeof(*buf) + size);
256 	if (!buf)
257 		goto err;
258 
259 	dst = buf + 1;
260 	buf->type = ACPI_TYPE_BUFFER;
261 	buf->buffer.length = size;
262 	buf->buffer.pointer = dst;
263 	for (i = 0; i < pkg->package.count; i++) {
264 		union acpi_object *obj = &pkg->package.elements[i];
265 
266 		if (obj->type == ACPI_TYPE_INTEGER) {
267 			memcpy(dst, &obj->integer.value, 4);
268 			dst += 4;
269 		} else if (obj->type == ACPI_TYPE_BUFFER) {
270 			memcpy(dst, obj->buffer.pointer, obj->buffer.length);
271 			dst += obj->buffer.length;
272 		}
273 	}
274 err:
275 	ACPI_FREE(pkg);
276 	return buf;
277 }
278 
279 static union acpi_object *int_to_buf(union acpi_object *integer)
280 {
281 	union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
282 	void *dst = NULL;
283 
284 	if (!buf)
285 		goto err;
286 
287 	if (integer->type != ACPI_TYPE_INTEGER) {
288 		WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
289 				integer->type);
290 		goto err;
291 	}
292 
293 	dst = buf + 1;
294 	buf->type = ACPI_TYPE_BUFFER;
295 	buf->buffer.length = 4;
296 	buf->buffer.pointer = dst;
297 	memcpy(dst, &integer->integer.value, 4);
298 err:
299 	ACPI_FREE(integer);
300 	return buf;
301 }
302 
303 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
304 		u32 len, void *data)
305 {
306 	acpi_status rc;
307 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
308 	struct acpi_object_list input = {
309 		.count = 3,
310 		.pointer = (union acpi_object []) {
311 			[0] = {
312 				.integer.type = ACPI_TYPE_INTEGER,
313 				.integer.value = offset,
314 			},
315 			[1] = {
316 				.integer.type = ACPI_TYPE_INTEGER,
317 				.integer.value = len,
318 			},
319 			[2] = {
320 				.buffer.type = ACPI_TYPE_BUFFER,
321 				.buffer.pointer = data,
322 				.buffer.length = len,
323 			},
324 		},
325 	};
326 
327 	rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
328 	if (ACPI_FAILURE(rc))
329 		return NULL;
330 	return int_to_buf(buf.pointer);
331 }
332 
333 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
334 		u32 len)
335 {
336 	acpi_status rc;
337 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
338 	struct acpi_object_list input = {
339 		.count = 2,
340 		.pointer = (union acpi_object []) {
341 			[0] = {
342 				.integer.type = ACPI_TYPE_INTEGER,
343 				.integer.value = offset,
344 			},
345 			[1] = {
346 				.integer.type = ACPI_TYPE_INTEGER,
347 				.integer.value = len,
348 			},
349 		},
350 	};
351 
352 	rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
353 	if (ACPI_FAILURE(rc))
354 		return NULL;
355 	return pkg_to_buf(buf.pointer);
356 }
357 
358 static union acpi_object *acpi_label_info(acpi_handle handle)
359 {
360 	acpi_status rc;
361 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
362 
363 	rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
364 	if (ACPI_FAILURE(rc))
365 		return NULL;
366 	return pkg_to_buf(buf.pointer);
367 }
368 
369 static u8 nfit_dsm_revid(unsigned family, unsigned func)
370 {
371 	static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
372 		[NVDIMM_FAMILY_INTEL] = {
373 			[NVDIMM_INTEL_GET_MODES] = 2,
374 			[NVDIMM_INTEL_GET_FWINFO] = 2,
375 			[NVDIMM_INTEL_START_FWUPDATE] = 2,
376 			[NVDIMM_INTEL_SEND_FWUPDATE] = 2,
377 			[NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
378 			[NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
379 			[NVDIMM_INTEL_SET_THRESHOLD] = 2,
380 			[NVDIMM_INTEL_INJECT_ERROR] = 2,
381 			[NVDIMM_INTEL_GET_SECURITY_STATE] = 2,
382 			[NVDIMM_INTEL_SET_PASSPHRASE] = 2,
383 			[NVDIMM_INTEL_DISABLE_PASSPHRASE] = 2,
384 			[NVDIMM_INTEL_UNLOCK_UNIT] = 2,
385 			[NVDIMM_INTEL_FREEZE_LOCK] = 2,
386 			[NVDIMM_INTEL_SECURE_ERASE] = 2,
387 			[NVDIMM_INTEL_OVERWRITE] = 2,
388 			[NVDIMM_INTEL_QUERY_OVERWRITE] = 2,
389 			[NVDIMM_INTEL_SET_MASTER_PASSPHRASE] = 2,
390 			[NVDIMM_INTEL_MASTER_SECURE_ERASE] = 2,
391 		},
392 	};
393 	u8 id;
394 
395 	if (family > NVDIMM_FAMILY_MAX)
396 		return 0;
397 	if (func > 31)
398 		return 0;
399 	id = revid_table[family][func];
400 	if (id == 0)
401 		return 1; /* default */
402 	return id;
403 }
404 
405 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
406 {
407 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
408 
409 	if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
410 			&& func >= NVDIMM_INTEL_GET_SECURITY_STATE
411 			&& func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
412 		return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
413 	return true;
414 }
415 
416 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
417 		struct nd_cmd_pkg *call_pkg)
418 {
419 	if (call_pkg) {
420 		int i;
421 
422 		if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
423 			return -ENOTTY;
424 
425 		for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
426 			if (call_pkg->nd_reserved2[i])
427 				return -EINVAL;
428 		return call_pkg->nd_command;
429 	}
430 
431 	/* In the !call_pkg case, bus commands == bus functions */
432 	if (!nfit_mem)
433 		return cmd;
434 
435 	/* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
436 	if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
437 		return cmd;
438 
439 	/*
440 	 * Force function number validation to fail since 0 is never
441 	 * published as a valid function in dsm_mask.
442 	 */
443 	return 0;
444 }
445 
446 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
447 		unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
448 {
449 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
450 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
451 	union acpi_object in_obj, in_buf, *out_obj;
452 	const struct nd_cmd_desc *desc = NULL;
453 	struct device *dev = acpi_desc->dev;
454 	struct nd_cmd_pkg *call_pkg = NULL;
455 	const char *cmd_name, *dimm_name;
456 	unsigned long cmd_mask, dsm_mask;
457 	u32 offset, fw_status = 0;
458 	acpi_handle handle;
459 	const guid_t *guid;
460 	int func, rc, i;
461 
462 	if (cmd_rc)
463 		*cmd_rc = -EINVAL;
464 
465 	if (cmd == ND_CMD_CALL)
466 		call_pkg = buf;
467 	func = cmd_to_func(nfit_mem, cmd, call_pkg);
468 	if (func < 0)
469 		return func;
470 
471 	if (nvdimm) {
472 		struct acpi_device *adev = nfit_mem->adev;
473 
474 		if (!adev)
475 			return -ENOTTY;
476 
477 		dimm_name = nvdimm_name(nvdimm);
478 		cmd_name = nvdimm_cmd_name(cmd);
479 		cmd_mask = nvdimm_cmd_mask(nvdimm);
480 		dsm_mask = nfit_mem->dsm_mask;
481 		desc = nd_cmd_dimm_desc(cmd);
482 		guid = to_nfit_uuid(nfit_mem->family);
483 		handle = adev->handle;
484 	} else {
485 		struct acpi_device *adev = to_acpi_dev(acpi_desc);
486 
487 		cmd_name = nvdimm_bus_cmd_name(cmd);
488 		cmd_mask = nd_desc->cmd_mask;
489 		dsm_mask = nd_desc->bus_dsm_mask;
490 		desc = nd_cmd_bus_desc(cmd);
491 		guid = to_nfit_uuid(NFIT_DEV_BUS);
492 		handle = adev->handle;
493 		dimm_name = "bus";
494 	}
495 
496 	if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
497 		return -ENOTTY;
498 
499 	/*
500 	 * Check for a valid command.  For ND_CMD_CALL, we also have to
501 	 * make sure that the DSM function is supported.
502 	 */
503 	if (cmd == ND_CMD_CALL && !test_bit(func, &dsm_mask))
504 		return -ENOTTY;
505 	else if (!test_bit(cmd, &cmd_mask))
506 		return -ENOTTY;
507 
508 	in_obj.type = ACPI_TYPE_PACKAGE;
509 	in_obj.package.count = 1;
510 	in_obj.package.elements = &in_buf;
511 	in_buf.type = ACPI_TYPE_BUFFER;
512 	in_buf.buffer.pointer = buf;
513 	in_buf.buffer.length = 0;
514 
515 	/* libnvdimm has already validated the input envelope */
516 	for (i = 0; i < desc->in_num; i++)
517 		in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
518 				i, buf);
519 
520 	if (call_pkg) {
521 		/* skip over package wrapper */
522 		in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
523 		in_buf.buffer.length = call_pkg->nd_size_in;
524 	}
525 
526 	dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
527 		dimm_name, cmd, func, in_buf.buffer.length);
528 	if (payload_dumpable(nvdimm, func))
529 		print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
530 				in_buf.buffer.pointer,
531 				min_t(u32, 256, in_buf.buffer.length), true);
532 
533 	/* call the BIOS, prefer the named methods over _DSM if available */
534 	if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
535 			&& test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
536 		out_obj = acpi_label_info(handle);
537 	else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
538 			&& test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
539 		struct nd_cmd_get_config_data_hdr *p = buf;
540 
541 		out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
542 	} else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
543 			&& test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
544 		struct nd_cmd_set_config_hdr *p = buf;
545 
546 		out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
547 				p->in_buf);
548 	} else {
549 		u8 revid;
550 
551 		if (nvdimm)
552 			revid = nfit_dsm_revid(nfit_mem->family, func);
553 		else
554 			revid = 1;
555 		out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
556 	}
557 
558 	if (!out_obj) {
559 		dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
560 		return -EINVAL;
561 	}
562 
563 	if (out_obj->type != ACPI_TYPE_BUFFER) {
564 		dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
565 				dimm_name, cmd_name, out_obj->type);
566 		rc = -EINVAL;
567 		goto out;
568 	}
569 
570 	dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
571 			cmd_name, out_obj->buffer.length);
572 	print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
573 			out_obj->buffer.pointer,
574 			min_t(u32, 128, out_obj->buffer.length), true);
575 
576 	if (call_pkg) {
577 		call_pkg->nd_fw_size = out_obj->buffer.length;
578 		memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
579 			out_obj->buffer.pointer,
580 			min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
581 
582 		ACPI_FREE(out_obj);
583 		/*
584 		 * Need to support FW function w/o known size in advance.
585 		 * Caller can determine required size based upon nd_fw_size.
586 		 * If we return an error (like elsewhere) then caller wouldn't
587 		 * be able to rely upon data returned to make calculation.
588 		 */
589 		if (cmd_rc)
590 			*cmd_rc = 0;
591 		return 0;
592 	}
593 
594 	for (i = 0, offset = 0; i < desc->out_num; i++) {
595 		u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
596 				(u32 *) out_obj->buffer.pointer,
597 				out_obj->buffer.length - offset);
598 
599 		if (offset + out_size > out_obj->buffer.length) {
600 			dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
601 					dimm_name, cmd_name, i);
602 			break;
603 		}
604 
605 		if (in_buf.buffer.length + offset + out_size > buf_len) {
606 			dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
607 					dimm_name, cmd_name, i);
608 			rc = -ENXIO;
609 			goto out;
610 		}
611 		memcpy(buf + in_buf.buffer.length + offset,
612 				out_obj->buffer.pointer + offset, out_size);
613 		offset += out_size;
614 	}
615 
616 	/*
617 	 * Set fw_status for all the commands with a known format to be
618 	 * later interpreted by xlat_status().
619 	 */
620 	if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
621 					&& cmd <= ND_CMD_CLEAR_ERROR)
622 				|| (nvdimm && cmd >= ND_CMD_SMART
623 					&& cmd <= ND_CMD_VENDOR)))
624 		fw_status = *(u32 *) out_obj->buffer.pointer;
625 
626 	if (offset + in_buf.buffer.length < buf_len) {
627 		if (i >= 1) {
628 			/*
629 			 * status valid, return the number of bytes left
630 			 * unfilled in the output buffer
631 			 */
632 			rc = buf_len - offset - in_buf.buffer.length;
633 			if (cmd_rc)
634 				*cmd_rc = xlat_status(nvdimm, buf, cmd,
635 						fw_status);
636 		} else {
637 			dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
638 					__func__, dimm_name, cmd_name, buf_len,
639 					offset);
640 			rc = -ENXIO;
641 		}
642 	} else {
643 		rc = 0;
644 		if (cmd_rc)
645 			*cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
646 	}
647 
648  out:
649 	ACPI_FREE(out_obj);
650 
651 	return rc;
652 }
653 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
654 
655 static const char *spa_type_name(u16 type)
656 {
657 	static const char *to_name[] = {
658 		[NFIT_SPA_VOLATILE] = "volatile",
659 		[NFIT_SPA_PM] = "pmem",
660 		[NFIT_SPA_DCR] = "dimm-control-region",
661 		[NFIT_SPA_BDW] = "block-data-window",
662 		[NFIT_SPA_VDISK] = "volatile-disk",
663 		[NFIT_SPA_VCD] = "volatile-cd",
664 		[NFIT_SPA_PDISK] = "persistent-disk",
665 		[NFIT_SPA_PCD] = "persistent-cd",
666 
667 	};
668 
669 	if (type > NFIT_SPA_PCD)
670 		return "unknown";
671 
672 	return to_name[type];
673 }
674 
675 int nfit_spa_type(struct acpi_nfit_system_address *spa)
676 {
677 	int i;
678 
679 	for (i = 0; i < NFIT_UUID_MAX; i++)
680 		if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
681 			return i;
682 	return -1;
683 }
684 
685 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
686 		struct nfit_table_prev *prev,
687 		struct acpi_nfit_system_address *spa)
688 {
689 	struct device *dev = acpi_desc->dev;
690 	struct nfit_spa *nfit_spa;
691 
692 	if (spa->header.length != sizeof(*spa))
693 		return false;
694 
695 	list_for_each_entry(nfit_spa, &prev->spas, list) {
696 		if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
697 			list_move_tail(&nfit_spa->list, &acpi_desc->spas);
698 			return true;
699 		}
700 	}
701 
702 	nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
703 			GFP_KERNEL);
704 	if (!nfit_spa)
705 		return false;
706 	INIT_LIST_HEAD(&nfit_spa->list);
707 	memcpy(nfit_spa->spa, spa, sizeof(*spa));
708 	list_add_tail(&nfit_spa->list, &acpi_desc->spas);
709 	dev_dbg(dev, "spa index: %d type: %s\n",
710 			spa->range_index,
711 			spa_type_name(nfit_spa_type(spa)));
712 	return true;
713 }
714 
715 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
716 		struct nfit_table_prev *prev,
717 		struct acpi_nfit_memory_map *memdev)
718 {
719 	struct device *dev = acpi_desc->dev;
720 	struct nfit_memdev *nfit_memdev;
721 
722 	if (memdev->header.length != sizeof(*memdev))
723 		return false;
724 
725 	list_for_each_entry(nfit_memdev, &prev->memdevs, list)
726 		if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
727 			list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
728 			return true;
729 		}
730 
731 	nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
732 			GFP_KERNEL);
733 	if (!nfit_memdev)
734 		return false;
735 	INIT_LIST_HEAD(&nfit_memdev->list);
736 	memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
737 	list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
738 	dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
739 			memdev->device_handle, memdev->range_index,
740 			memdev->region_index, memdev->flags);
741 	return true;
742 }
743 
744 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
745 {
746 	struct acpi_nfit_memory_map *memdev;
747 	struct acpi_nfit_desc *acpi_desc;
748 	struct nfit_mem *nfit_mem;
749 	u16 physical_id;
750 
751 	mutex_lock(&acpi_desc_lock);
752 	list_for_each_entry(acpi_desc, &acpi_descs, list) {
753 		mutex_lock(&acpi_desc->init_mutex);
754 		list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
755 			memdev = __to_nfit_memdev(nfit_mem);
756 			if (memdev->device_handle == device_handle) {
757 				*flags = memdev->flags;
758 				physical_id = memdev->physical_id;
759 				mutex_unlock(&acpi_desc->init_mutex);
760 				mutex_unlock(&acpi_desc_lock);
761 				return physical_id;
762 			}
763 		}
764 		mutex_unlock(&acpi_desc->init_mutex);
765 	}
766 	mutex_unlock(&acpi_desc_lock);
767 
768 	return -ENODEV;
769 }
770 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
771 
772 /*
773  * An implementation may provide a truncated control region if no block windows
774  * are defined.
775  */
776 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
777 {
778 	if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
779 				window_size))
780 		return 0;
781 	if (dcr->windows)
782 		return sizeof(*dcr);
783 	return offsetof(struct acpi_nfit_control_region, window_size);
784 }
785 
786 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
787 		struct nfit_table_prev *prev,
788 		struct acpi_nfit_control_region *dcr)
789 {
790 	struct device *dev = acpi_desc->dev;
791 	struct nfit_dcr *nfit_dcr;
792 
793 	if (!sizeof_dcr(dcr))
794 		return false;
795 
796 	list_for_each_entry(nfit_dcr, &prev->dcrs, list)
797 		if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
798 			list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
799 			return true;
800 		}
801 
802 	nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
803 			GFP_KERNEL);
804 	if (!nfit_dcr)
805 		return false;
806 	INIT_LIST_HEAD(&nfit_dcr->list);
807 	memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
808 	list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
809 	dev_dbg(dev, "dcr index: %d windows: %d\n",
810 			dcr->region_index, dcr->windows);
811 	return true;
812 }
813 
814 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
815 		struct nfit_table_prev *prev,
816 		struct acpi_nfit_data_region *bdw)
817 {
818 	struct device *dev = acpi_desc->dev;
819 	struct nfit_bdw *nfit_bdw;
820 
821 	if (bdw->header.length != sizeof(*bdw))
822 		return false;
823 	list_for_each_entry(nfit_bdw, &prev->bdws, list)
824 		if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
825 			list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
826 			return true;
827 		}
828 
829 	nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
830 			GFP_KERNEL);
831 	if (!nfit_bdw)
832 		return false;
833 	INIT_LIST_HEAD(&nfit_bdw->list);
834 	memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
835 	list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
836 	dev_dbg(dev, "bdw dcr: %d windows: %d\n",
837 			bdw->region_index, bdw->windows);
838 	return true;
839 }
840 
841 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
842 {
843 	if (idt->header.length < sizeof(*idt))
844 		return 0;
845 	return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
846 }
847 
848 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
849 		struct nfit_table_prev *prev,
850 		struct acpi_nfit_interleave *idt)
851 {
852 	struct device *dev = acpi_desc->dev;
853 	struct nfit_idt *nfit_idt;
854 
855 	if (!sizeof_idt(idt))
856 		return false;
857 
858 	list_for_each_entry(nfit_idt, &prev->idts, list) {
859 		if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
860 			continue;
861 
862 		if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
863 			list_move_tail(&nfit_idt->list, &acpi_desc->idts);
864 			return true;
865 		}
866 	}
867 
868 	nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
869 			GFP_KERNEL);
870 	if (!nfit_idt)
871 		return false;
872 	INIT_LIST_HEAD(&nfit_idt->list);
873 	memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
874 	list_add_tail(&nfit_idt->list, &acpi_desc->idts);
875 	dev_dbg(dev, "idt index: %d num_lines: %d\n",
876 			idt->interleave_index, idt->line_count);
877 	return true;
878 }
879 
880 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
881 {
882 	if (flush->header.length < sizeof(*flush))
883 		return 0;
884 	return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
885 }
886 
887 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
888 		struct nfit_table_prev *prev,
889 		struct acpi_nfit_flush_address *flush)
890 {
891 	struct device *dev = acpi_desc->dev;
892 	struct nfit_flush *nfit_flush;
893 
894 	if (!sizeof_flush(flush))
895 		return false;
896 
897 	list_for_each_entry(nfit_flush, &prev->flushes, list) {
898 		if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
899 			continue;
900 
901 		if (memcmp(nfit_flush->flush, flush,
902 					sizeof_flush(flush)) == 0) {
903 			list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
904 			return true;
905 		}
906 	}
907 
908 	nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
909 			+ sizeof_flush(flush), GFP_KERNEL);
910 	if (!nfit_flush)
911 		return false;
912 	INIT_LIST_HEAD(&nfit_flush->list);
913 	memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
914 	list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
915 	dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
916 			flush->device_handle, flush->hint_count);
917 	return true;
918 }
919 
920 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
921 		struct acpi_nfit_capabilities *pcap)
922 {
923 	struct device *dev = acpi_desc->dev;
924 	u32 mask;
925 
926 	mask = (1 << (pcap->highest_capability + 1)) - 1;
927 	acpi_desc->platform_cap = pcap->capabilities & mask;
928 	dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
929 	return true;
930 }
931 
932 static void *add_table(struct acpi_nfit_desc *acpi_desc,
933 		struct nfit_table_prev *prev, void *table, const void *end)
934 {
935 	struct device *dev = acpi_desc->dev;
936 	struct acpi_nfit_header *hdr;
937 	void *err = ERR_PTR(-ENOMEM);
938 
939 	if (table >= end)
940 		return NULL;
941 
942 	hdr = table;
943 	if (!hdr->length) {
944 		dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
945 			hdr->type);
946 		return NULL;
947 	}
948 
949 	switch (hdr->type) {
950 	case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
951 		if (!add_spa(acpi_desc, prev, table))
952 			return err;
953 		break;
954 	case ACPI_NFIT_TYPE_MEMORY_MAP:
955 		if (!add_memdev(acpi_desc, prev, table))
956 			return err;
957 		break;
958 	case ACPI_NFIT_TYPE_CONTROL_REGION:
959 		if (!add_dcr(acpi_desc, prev, table))
960 			return err;
961 		break;
962 	case ACPI_NFIT_TYPE_DATA_REGION:
963 		if (!add_bdw(acpi_desc, prev, table))
964 			return err;
965 		break;
966 	case ACPI_NFIT_TYPE_INTERLEAVE:
967 		if (!add_idt(acpi_desc, prev, table))
968 			return err;
969 		break;
970 	case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
971 		if (!add_flush(acpi_desc, prev, table))
972 			return err;
973 		break;
974 	case ACPI_NFIT_TYPE_SMBIOS:
975 		dev_dbg(dev, "smbios\n");
976 		break;
977 	case ACPI_NFIT_TYPE_CAPABILITIES:
978 		if (!add_platform_cap(acpi_desc, table))
979 			return err;
980 		break;
981 	default:
982 		dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
983 		break;
984 	}
985 
986 	return table + hdr->length;
987 }
988 
989 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
990 		struct nfit_mem *nfit_mem)
991 {
992 	u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
993 	u16 dcr = nfit_mem->dcr->region_index;
994 	struct nfit_spa *nfit_spa;
995 
996 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
997 		u16 range_index = nfit_spa->spa->range_index;
998 		int type = nfit_spa_type(nfit_spa->spa);
999 		struct nfit_memdev *nfit_memdev;
1000 
1001 		if (type != NFIT_SPA_BDW)
1002 			continue;
1003 
1004 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1005 			if (nfit_memdev->memdev->range_index != range_index)
1006 				continue;
1007 			if (nfit_memdev->memdev->device_handle != device_handle)
1008 				continue;
1009 			if (nfit_memdev->memdev->region_index != dcr)
1010 				continue;
1011 
1012 			nfit_mem->spa_bdw = nfit_spa->spa;
1013 			return;
1014 		}
1015 	}
1016 
1017 	dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1018 			nfit_mem->spa_dcr->range_index);
1019 	nfit_mem->bdw = NULL;
1020 }
1021 
1022 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1023 		struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1024 {
1025 	u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1026 	struct nfit_memdev *nfit_memdev;
1027 	struct nfit_bdw *nfit_bdw;
1028 	struct nfit_idt *nfit_idt;
1029 	u16 idt_idx, range_index;
1030 
1031 	list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1032 		if (nfit_bdw->bdw->region_index != dcr)
1033 			continue;
1034 		nfit_mem->bdw = nfit_bdw->bdw;
1035 		break;
1036 	}
1037 
1038 	if (!nfit_mem->bdw)
1039 		return;
1040 
1041 	nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1042 
1043 	if (!nfit_mem->spa_bdw)
1044 		return;
1045 
1046 	range_index = nfit_mem->spa_bdw->range_index;
1047 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1048 		if (nfit_memdev->memdev->range_index != range_index ||
1049 				nfit_memdev->memdev->region_index != dcr)
1050 			continue;
1051 		nfit_mem->memdev_bdw = nfit_memdev->memdev;
1052 		idt_idx = nfit_memdev->memdev->interleave_index;
1053 		list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1054 			if (nfit_idt->idt->interleave_index != idt_idx)
1055 				continue;
1056 			nfit_mem->idt_bdw = nfit_idt->idt;
1057 			break;
1058 		}
1059 		break;
1060 	}
1061 }
1062 
1063 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1064 		struct acpi_nfit_system_address *spa)
1065 {
1066 	struct nfit_mem *nfit_mem, *found;
1067 	struct nfit_memdev *nfit_memdev;
1068 	int type = spa ? nfit_spa_type(spa) : 0;
1069 
1070 	switch (type) {
1071 	case NFIT_SPA_DCR:
1072 	case NFIT_SPA_PM:
1073 		break;
1074 	default:
1075 		if (spa)
1076 			return 0;
1077 	}
1078 
1079 	/*
1080 	 * This loop runs in two modes, when a dimm is mapped the loop
1081 	 * adds memdev associations to an existing dimm, or creates a
1082 	 * dimm. In the unmapped dimm case this loop sweeps for memdev
1083 	 * instances with an invalid / zero range_index and adds those
1084 	 * dimms without spa associations.
1085 	 */
1086 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1087 		struct nfit_flush *nfit_flush;
1088 		struct nfit_dcr *nfit_dcr;
1089 		u32 device_handle;
1090 		u16 dcr;
1091 
1092 		if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1093 			continue;
1094 		if (!spa && nfit_memdev->memdev->range_index)
1095 			continue;
1096 		found = NULL;
1097 		dcr = nfit_memdev->memdev->region_index;
1098 		device_handle = nfit_memdev->memdev->device_handle;
1099 		list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1100 			if (__to_nfit_memdev(nfit_mem)->device_handle
1101 					== device_handle) {
1102 				found = nfit_mem;
1103 				break;
1104 			}
1105 
1106 		if (found)
1107 			nfit_mem = found;
1108 		else {
1109 			nfit_mem = devm_kzalloc(acpi_desc->dev,
1110 					sizeof(*nfit_mem), GFP_KERNEL);
1111 			if (!nfit_mem)
1112 				return -ENOMEM;
1113 			INIT_LIST_HEAD(&nfit_mem->list);
1114 			nfit_mem->acpi_desc = acpi_desc;
1115 			list_add(&nfit_mem->list, &acpi_desc->dimms);
1116 		}
1117 
1118 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1119 			if (nfit_dcr->dcr->region_index != dcr)
1120 				continue;
1121 			/*
1122 			 * Record the control region for the dimm.  For
1123 			 * the ACPI 6.1 case, where there are separate
1124 			 * control regions for the pmem vs blk
1125 			 * interfaces, be sure to record the extended
1126 			 * blk details.
1127 			 */
1128 			if (!nfit_mem->dcr)
1129 				nfit_mem->dcr = nfit_dcr->dcr;
1130 			else if (nfit_mem->dcr->windows == 0
1131 					&& nfit_dcr->dcr->windows)
1132 				nfit_mem->dcr = nfit_dcr->dcr;
1133 			break;
1134 		}
1135 
1136 		list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1137 			struct acpi_nfit_flush_address *flush;
1138 			u16 i;
1139 
1140 			if (nfit_flush->flush->device_handle != device_handle)
1141 				continue;
1142 			nfit_mem->nfit_flush = nfit_flush;
1143 			flush = nfit_flush->flush;
1144 			nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1145 					flush->hint_count,
1146 					sizeof(struct resource),
1147 					GFP_KERNEL);
1148 			if (!nfit_mem->flush_wpq)
1149 				return -ENOMEM;
1150 			for (i = 0; i < flush->hint_count; i++) {
1151 				struct resource *res = &nfit_mem->flush_wpq[i];
1152 
1153 				res->start = flush->hint_address[i];
1154 				res->end = res->start + 8 - 1;
1155 			}
1156 			break;
1157 		}
1158 
1159 		if (dcr && !nfit_mem->dcr) {
1160 			dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1161 					spa->range_index, dcr);
1162 			return -ENODEV;
1163 		}
1164 
1165 		if (type == NFIT_SPA_DCR) {
1166 			struct nfit_idt *nfit_idt;
1167 			u16 idt_idx;
1168 
1169 			/* multiple dimms may share a SPA when interleaved */
1170 			nfit_mem->spa_dcr = spa;
1171 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
1172 			idt_idx = nfit_memdev->memdev->interleave_index;
1173 			list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1174 				if (nfit_idt->idt->interleave_index != idt_idx)
1175 					continue;
1176 				nfit_mem->idt_dcr = nfit_idt->idt;
1177 				break;
1178 			}
1179 			nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1180 		} else if (type == NFIT_SPA_PM) {
1181 			/*
1182 			 * A single dimm may belong to multiple SPA-PM
1183 			 * ranges, record at least one in addition to
1184 			 * any SPA-DCR range.
1185 			 */
1186 			nfit_mem->memdev_pmem = nfit_memdev->memdev;
1187 		} else
1188 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
1189 	}
1190 
1191 	return 0;
1192 }
1193 
1194 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1195 {
1196 	struct nfit_mem *a = container_of(_a, typeof(*a), list);
1197 	struct nfit_mem *b = container_of(_b, typeof(*b), list);
1198 	u32 handleA, handleB;
1199 
1200 	handleA = __to_nfit_memdev(a)->device_handle;
1201 	handleB = __to_nfit_memdev(b)->device_handle;
1202 	if (handleA < handleB)
1203 		return -1;
1204 	else if (handleA > handleB)
1205 		return 1;
1206 	return 0;
1207 }
1208 
1209 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1210 {
1211 	struct nfit_spa *nfit_spa;
1212 	int rc;
1213 
1214 
1215 	/*
1216 	 * For each SPA-DCR or SPA-PMEM address range find its
1217 	 * corresponding MEMDEV(s).  From each MEMDEV find the
1218 	 * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1219 	 * try to find a SPA-BDW and a corresponding BDW that references
1220 	 * the DCR.  Throw it all into an nfit_mem object.  Note, that
1221 	 * BDWs are optional.
1222 	 */
1223 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1224 		rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1225 		if (rc)
1226 			return rc;
1227 	}
1228 
1229 	/*
1230 	 * If a DIMM has failed to be mapped into SPA there will be no
1231 	 * SPA entries above. Find and register all the unmapped DIMMs
1232 	 * for reporting and recovery purposes.
1233 	 */
1234 	rc = __nfit_mem_init(acpi_desc, NULL);
1235 	if (rc)
1236 		return rc;
1237 
1238 	list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1239 
1240 	return 0;
1241 }
1242 
1243 static ssize_t bus_dsm_mask_show(struct device *dev,
1244 		struct device_attribute *attr, char *buf)
1245 {
1246 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1247 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1248 
1249 	return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1250 }
1251 static struct device_attribute dev_attr_bus_dsm_mask =
1252 		__ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1253 
1254 static ssize_t revision_show(struct device *dev,
1255 		struct device_attribute *attr, char *buf)
1256 {
1257 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1258 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1259 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1260 
1261 	return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1262 }
1263 static DEVICE_ATTR_RO(revision);
1264 
1265 static ssize_t hw_error_scrub_show(struct device *dev,
1266 		struct device_attribute *attr, char *buf)
1267 {
1268 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1269 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1270 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1271 
1272 	return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1273 }
1274 
1275 /*
1276  * The 'hw_error_scrub' attribute can have the following values written to it:
1277  * '0': Switch to the default mode where an exception will only insert
1278  *      the address of the memory error into the poison and badblocks lists.
1279  * '1': Enable a full scrub to happen if an exception for a memory error is
1280  *      received.
1281  */
1282 static ssize_t hw_error_scrub_store(struct device *dev,
1283 		struct device_attribute *attr, const char *buf, size_t size)
1284 {
1285 	struct nvdimm_bus_descriptor *nd_desc;
1286 	ssize_t rc;
1287 	long val;
1288 
1289 	rc = kstrtol(buf, 0, &val);
1290 	if (rc)
1291 		return rc;
1292 
1293 	device_lock(dev);
1294 	nd_desc = dev_get_drvdata(dev);
1295 	if (nd_desc) {
1296 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1297 
1298 		switch (val) {
1299 		case HW_ERROR_SCRUB_ON:
1300 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1301 			break;
1302 		case HW_ERROR_SCRUB_OFF:
1303 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1304 			break;
1305 		default:
1306 			rc = -EINVAL;
1307 			break;
1308 		}
1309 	}
1310 	device_unlock(dev);
1311 	if (rc)
1312 		return rc;
1313 	return size;
1314 }
1315 static DEVICE_ATTR_RW(hw_error_scrub);
1316 
1317 /*
1318  * This shows the number of full Address Range Scrubs that have been
1319  * completed since driver load time. Userspace can wait on this using
1320  * select/poll etc. A '+' at the end indicates an ARS is in progress
1321  */
1322 static ssize_t scrub_show(struct device *dev,
1323 		struct device_attribute *attr, char *buf)
1324 {
1325 	struct nvdimm_bus_descriptor *nd_desc;
1326 	struct acpi_nfit_desc *acpi_desc;
1327 	ssize_t rc = -ENXIO;
1328 	bool busy;
1329 
1330 	device_lock(dev);
1331 	nd_desc = dev_get_drvdata(dev);
1332 	if (!nd_desc) {
1333 		device_unlock(dev);
1334 		return rc;
1335 	}
1336 	acpi_desc = to_acpi_desc(nd_desc);
1337 
1338 	mutex_lock(&acpi_desc->init_mutex);
1339 	busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1340 		&& !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1341 	rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1342 	/* Allow an admin to poll the busy state at a higher rate */
1343 	if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1344 				&acpi_desc->scrub_flags)) {
1345 		acpi_desc->scrub_tmo = 1;
1346 		mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1347 	}
1348 
1349 	mutex_unlock(&acpi_desc->init_mutex);
1350 	device_unlock(dev);
1351 	return rc;
1352 }
1353 
1354 static ssize_t scrub_store(struct device *dev,
1355 		struct device_attribute *attr, const char *buf, size_t size)
1356 {
1357 	struct nvdimm_bus_descriptor *nd_desc;
1358 	ssize_t rc;
1359 	long val;
1360 
1361 	rc = kstrtol(buf, 0, &val);
1362 	if (rc)
1363 		return rc;
1364 	if (val != 1)
1365 		return -EINVAL;
1366 
1367 	device_lock(dev);
1368 	nd_desc = dev_get_drvdata(dev);
1369 	if (nd_desc) {
1370 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1371 
1372 		rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1373 	}
1374 	device_unlock(dev);
1375 	if (rc)
1376 		return rc;
1377 	return size;
1378 }
1379 static DEVICE_ATTR_RW(scrub);
1380 
1381 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1382 {
1383 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1384 	const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1385 		| 1 << ND_CMD_ARS_STATUS;
1386 
1387 	return (nd_desc->cmd_mask & mask) == mask;
1388 }
1389 
1390 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1391 {
1392 	struct device *dev = container_of(kobj, struct device, kobj);
1393 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1394 
1395 	if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1396 		return 0;
1397 	return a->mode;
1398 }
1399 
1400 static struct attribute *acpi_nfit_attributes[] = {
1401 	&dev_attr_revision.attr,
1402 	&dev_attr_scrub.attr,
1403 	&dev_attr_hw_error_scrub.attr,
1404 	&dev_attr_bus_dsm_mask.attr,
1405 	NULL,
1406 };
1407 
1408 static const struct attribute_group acpi_nfit_attribute_group = {
1409 	.name = "nfit",
1410 	.attrs = acpi_nfit_attributes,
1411 	.is_visible = nfit_visible,
1412 };
1413 
1414 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1415 	&nvdimm_bus_attribute_group,
1416 	&acpi_nfit_attribute_group,
1417 	NULL,
1418 };
1419 
1420 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1421 {
1422 	struct nvdimm *nvdimm = to_nvdimm(dev);
1423 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1424 
1425 	return __to_nfit_memdev(nfit_mem);
1426 }
1427 
1428 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1429 {
1430 	struct nvdimm *nvdimm = to_nvdimm(dev);
1431 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1432 
1433 	return nfit_mem->dcr;
1434 }
1435 
1436 static ssize_t handle_show(struct device *dev,
1437 		struct device_attribute *attr, char *buf)
1438 {
1439 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1440 
1441 	return sprintf(buf, "%#x\n", memdev->device_handle);
1442 }
1443 static DEVICE_ATTR_RO(handle);
1444 
1445 static ssize_t phys_id_show(struct device *dev,
1446 		struct device_attribute *attr, char *buf)
1447 {
1448 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1449 
1450 	return sprintf(buf, "%#x\n", memdev->physical_id);
1451 }
1452 static DEVICE_ATTR_RO(phys_id);
1453 
1454 static ssize_t vendor_show(struct device *dev,
1455 		struct device_attribute *attr, char *buf)
1456 {
1457 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1458 
1459 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1460 }
1461 static DEVICE_ATTR_RO(vendor);
1462 
1463 static ssize_t rev_id_show(struct device *dev,
1464 		struct device_attribute *attr, char *buf)
1465 {
1466 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1467 
1468 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1469 }
1470 static DEVICE_ATTR_RO(rev_id);
1471 
1472 static ssize_t device_show(struct device *dev,
1473 		struct device_attribute *attr, char *buf)
1474 {
1475 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1476 
1477 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1478 }
1479 static DEVICE_ATTR_RO(device);
1480 
1481 static ssize_t subsystem_vendor_show(struct device *dev,
1482 		struct device_attribute *attr, char *buf)
1483 {
1484 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1485 
1486 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1487 }
1488 static DEVICE_ATTR_RO(subsystem_vendor);
1489 
1490 static ssize_t subsystem_rev_id_show(struct device *dev,
1491 		struct device_attribute *attr, char *buf)
1492 {
1493 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1494 
1495 	return sprintf(buf, "0x%04x\n",
1496 			be16_to_cpu(dcr->subsystem_revision_id));
1497 }
1498 static DEVICE_ATTR_RO(subsystem_rev_id);
1499 
1500 static ssize_t subsystem_device_show(struct device *dev,
1501 		struct device_attribute *attr, char *buf)
1502 {
1503 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1504 
1505 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1506 }
1507 static DEVICE_ATTR_RO(subsystem_device);
1508 
1509 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1510 {
1511 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1512 	int formats = 0;
1513 
1514 	if (nfit_mem->memdev_pmem)
1515 		formats++;
1516 	if (nfit_mem->memdev_bdw)
1517 		formats++;
1518 	return formats;
1519 }
1520 
1521 static ssize_t format_show(struct device *dev,
1522 		struct device_attribute *attr, char *buf)
1523 {
1524 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1525 
1526 	return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1527 }
1528 static DEVICE_ATTR_RO(format);
1529 
1530 static ssize_t format1_show(struct device *dev,
1531 		struct device_attribute *attr, char *buf)
1532 {
1533 	u32 handle;
1534 	ssize_t rc = -ENXIO;
1535 	struct nfit_mem *nfit_mem;
1536 	struct nfit_memdev *nfit_memdev;
1537 	struct acpi_nfit_desc *acpi_desc;
1538 	struct nvdimm *nvdimm = to_nvdimm(dev);
1539 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1540 
1541 	nfit_mem = nvdimm_provider_data(nvdimm);
1542 	acpi_desc = nfit_mem->acpi_desc;
1543 	handle = to_nfit_memdev(dev)->device_handle;
1544 
1545 	/* assumes DIMMs have at most 2 published interface codes */
1546 	mutex_lock(&acpi_desc->init_mutex);
1547 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1548 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1549 		struct nfit_dcr *nfit_dcr;
1550 
1551 		if (memdev->device_handle != handle)
1552 			continue;
1553 
1554 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1555 			if (nfit_dcr->dcr->region_index != memdev->region_index)
1556 				continue;
1557 			if (nfit_dcr->dcr->code == dcr->code)
1558 				continue;
1559 			rc = sprintf(buf, "0x%04x\n",
1560 					le16_to_cpu(nfit_dcr->dcr->code));
1561 			break;
1562 		}
1563 		if (rc != ENXIO)
1564 			break;
1565 	}
1566 	mutex_unlock(&acpi_desc->init_mutex);
1567 	return rc;
1568 }
1569 static DEVICE_ATTR_RO(format1);
1570 
1571 static ssize_t formats_show(struct device *dev,
1572 		struct device_attribute *attr, char *buf)
1573 {
1574 	struct nvdimm *nvdimm = to_nvdimm(dev);
1575 
1576 	return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1577 }
1578 static DEVICE_ATTR_RO(formats);
1579 
1580 static ssize_t serial_show(struct device *dev,
1581 		struct device_attribute *attr, char *buf)
1582 {
1583 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1584 
1585 	return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1586 }
1587 static DEVICE_ATTR_RO(serial);
1588 
1589 static ssize_t family_show(struct device *dev,
1590 		struct device_attribute *attr, char *buf)
1591 {
1592 	struct nvdimm *nvdimm = to_nvdimm(dev);
1593 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1594 
1595 	if (nfit_mem->family < 0)
1596 		return -ENXIO;
1597 	return sprintf(buf, "%d\n", nfit_mem->family);
1598 }
1599 static DEVICE_ATTR_RO(family);
1600 
1601 static ssize_t dsm_mask_show(struct device *dev,
1602 		struct device_attribute *attr, char *buf)
1603 {
1604 	struct nvdimm *nvdimm = to_nvdimm(dev);
1605 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1606 
1607 	if (nfit_mem->family < 0)
1608 		return -ENXIO;
1609 	return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1610 }
1611 static DEVICE_ATTR_RO(dsm_mask);
1612 
1613 static ssize_t flags_show(struct device *dev,
1614 		struct device_attribute *attr, char *buf)
1615 {
1616 	struct nvdimm *nvdimm = to_nvdimm(dev);
1617 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1618 	u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1619 
1620 	if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1621 		flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1622 
1623 	return sprintf(buf, "%s%s%s%s%s%s%s\n",
1624 		flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1625 		flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1626 		flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1627 		flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1628 		flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1629 		flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1630 		flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1631 }
1632 static DEVICE_ATTR_RO(flags);
1633 
1634 static ssize_t id_show(struct device *dev,
1635 		struct device_attribute *attr, char *buf)
1636 {
1637 	struct nvdimm *nvdimm = to_nvdimm(dev);
1638 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1639 
1640 	return sprintf(buf, "%s\n", nfit_mem->id);
1641 }
1642 static DEVICE_ATTR_RO(id);
1643 
1644 static ssize_t dirty_shutdown_show(struct device *dev,
1645 		struct device_attribute *attr, char *buf)
1646 {
1647 	struct nvdimm *nvdimm = to_nvdimm(dev);
1648 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1649 
1650 	return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1651 }
1652 static DEVICE_ATTR_RO(dirty_shutdown);
1653 
1654 static struct attribute *acpi_nfit_dimm_attributes[] = {
1655 	&dev_attr_handle.attr,
1656 	&dev_attr_phys_id.attr,
1657 	&dev_attr_vendor.attr,
1658 	&dev_attr_device.attr,
1659 	&dev_attr_rev_id.attr,
1660 	&dev_attr_subsystem_vendor.attr,
1661 	&dev_attr_subsystem_device.attr,
1662 	&dev_attr_subsystem_rev_id.attr,
1663 	&dev_attr_format.attr,
1664 	&dev_attr_formats.attr,
1665 	&dev_attr_format1.attr,
1666 	&dev_attr_serial.attr,
1667 	&dev_attr_flags.attr,
1668 	&dev_attr_id.attr,
1669 	&dev_attr_family.attr,
1670 	&dev_attr_dsm_mask.attr,
1671 	&dev_attr_dirty_shutdown.attr,
1672 	NULL,
1673 };
1674 
1675 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1676 		struct attribute *a, int n)
1677 {
1678 	struct device *dev = container_of(kobj, struct device, kobj);
1679 	struct nvdimm *nvdimm = to_nvdimm(dev);
1680 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1681 
1682 	if (!to_nfit_dcr(dev)) {
1683 		/* Without a dcr only the memdev attributes can be surfaced */
1684 		if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1685 				|| a == &dev_attr_flags.attr
1686 				|| a == &dev_attr_family.attr
1687 				|| a == &dev_attr_dsm_mask.attr)
1688 			return a->mode;
1689 		return 0;
1690 	}
1691 
1692 	if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1693 		return 0;
1694 
1695 	if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1696 			&& a == &dev_attr_dirty_shutdown.attr)
1697 		return 0;
1698 
1699 	return a->mode;
1700 }
1701 
1702 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1703 	.name = "nfit",
1704 	.attrs = acpi_nfit_dimm_attributes,
1705 	.is_visible = acpi_nfit_dimm_attr_visible,
1706 };
1707 
1708 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1709 	&nvdimm_attribute_group,
1710 	&nd_device_attribute_group,
1711 	&acpi_nfit_dimm_attribute_group,
1712 	NULL,
1713 };
1714 
1715 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1716 		u32 device_handle)
1717 {
1718 	struct nfit_mem *nfit_mem;
1719 
1720 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1721 		if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1722 			return nfit_mem->nvdimm;
1723 
1724 	return NULL;
1725 }
1726 
1727 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1728 {
1729 	struct nfit_mem *nfit_mem;
1730 	struct acpi_nfit_desc *acpi_desc;
1731 
1732 	dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1733 			event);
1734 
1735 	if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1736 		dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1737 				event);
1738 		return;
1739 	}
1740 
1741 	acpi_desc = dev_get_drvdata(dev->parent);
1742 	if (!acpi_desc)
1743 		return;
1744 
1745 	/*
1746 	 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1747 	 * is still valid.
1748 	 */
1749 	nfit_mem = dev_get_drvdata(dev);
1750 	if (nfit_mem && nfit_mem->flags_attr)
1751 		sysfs_notify_dirent(nfit_mem->flags_attr);
1752 }
1753 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1754 
1755 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1756 {
1757 	struct acpi_device *adev = data;
1758 	struct device *dev = &adev->dev;
1759 
1760 	device_lock(dev->parent);
1761 	__acpi_nvdimm_notify(dev, event);
1762 	device_unlock(dev->parent);
1763 }
1764 
1765 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1766 {
1767 	acpi_handle handle;
1768 	acpi_status status;
1769 
1770 	status = acpi_get_handle(adev->handle, method, &handle);
1771 
1772 	if (ACPI_SUCCESS(status))
1773 		return true;
1774 	return false;
1775 }
1776 
1777 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1778 {
1779 	struct device *dev = &nfit_mem->adev->dev;
1780 	struct nd_intel_smart smart = { 0 };
1781 	union acpi_object in_buf = {
1782 		.buffer.type = ACPI_TYPE_BUFFER,
1783 		.buffer.length = 0,
1784 	};
1785 	union acpi_object in_obj = {
1786 		.package.type = ACPI_TYPE_PACKAGE,
1787 		.package.count = 1,
1788 		.package.elements = &in_buf,
1789 	};
1790 	const u8 func = ND_INTEL_SMART;
1791 	const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1792 	u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1793 	struct acpi_device *adev = nfit_mem->adev;
1794 	acpi_handle handle = adev->handle;
1795 	union acpi_object *out_obj;
1796 
1797 	if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1798 		return;
1799 
1800 	out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1801 	if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1802 			|| out_obj->buffer.length < sizeof(smart)) {
1803 		dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1804 				dev_name(dev));
1805 		ACPI_FREE(out_obj);
1806 		return;
1807 	}
1808 	memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1809 	ACPI_FREE(out_obj);
1810 
1811 	if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1812 		if (smart.shutdown_state)
1813 			set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1814 	}
1815 
1816 	if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1817 		set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1818 		nfit_mem->dirty_shutdown = smart.shutdown_count;
1819 	}
1820 }
1821 
1822 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1823 {
1824 	/*
1825 	 * For DIMMs that provide a dynamic facility to retrieve a
1826 	 * dirty-shutdown status and/or a dirty-shutdown count, cache
1827 	 * these values in nfit_mem.
1828 	 */
1829 	if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1830 		nfit_intel_shutdown_status(nfit_mem);
1831 }
1832 
1833 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1834 		struct nfit_mem *nfit_mem, u32 device_handle)
1835 {
1836 	struct acpi_device *adev, *adev_dimm;
1837 	struct device *dev = acpi_desc->dev;
1838 	unsigned long dsm_mask, label_mask;
1839 	const guid_t *guid;
1840 	int i;
1841 	int family = -1;
1842 	struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1843 
1844 	/* nfit test assumes 1:1 relationship between commands and dsms */
1845 	nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1846 	nfit_mem->family = NVDIMM_FAMILY_INTEL;
1847 
1848 	if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1849 		sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1850 				be16_to_cpu(dcr->vendor_id),
1851 				dcr->manufacturing_location,
1852 				be16_to_cpu(dcr->manufacturing_date),
1853 				be32_to_cpu(dcr->serial_number));
1854 	else
1855 		sprintf(nfit_mem->id, "%04x-%08x",
1856 				be16_to_cpu(dcr->vendor_id),
1857 				be32_to_cpu(dcr->serial_number));
1858 
1859 	adev = to_acpi_dev(acpi_desc);
1860 	if (!adev) {
1861 		/* unit test case */
1862 		populate_shutdown_status(nfit_mem);
1863 		return 0;
1864 	}
1865 
1866 	adev_dimm = acpi_find_child_device(adev, device_handle, false);
1867 	nfit_mem->adev = adev_dimm;
1868 	if (!adev_dimm) {
1869 		dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1870 				device_handle);
1871 		return force_enable_dimms ? 0 : -ENODEV;
1872 	}
1873 
1874 	if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1875 		ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1876 		dev_err(dev, "%s: notification registration failed\n",
1877 				dev_name(&adev_dimm->dev));
1878 		return -ENXIO;
1879 	}
1880 	/*
1881 	 * Record nfit_mem for the notification path to track back to
1882 	 * the nfit sysfs attributes for this dimm device object.
1883 	 */
1884 	dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1885 
1886 	/*
1887 	 * There are 4 "legacy" NVDIMM command sets
1888 	 * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1889 	 * an EFI working group was established to constrain this
1890 	 * proliferation. The nfit driver probes for the supported command
1891 	 * set by GUID. Note, if you're a platform developer looking to add
1892 	 * a new command set to this probe, consider using an existing set,
1893 	 * or otherwise seek approval to publish the command set at
1894 	 * http://www.uefi.org/RFIC_LIST.
1895 	 *
1896 	 * Note, that checking for function0 (bit0) tells us if any commands
1897 	 * are reachable through this GUID.
1898 	 */
1899 	for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1900 		if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1901 			if (family < 0 || i == default_dsm_family)
1902 				family = i;
1903 
1904 	/* limit the supported commands to those that are publicly documented */
1905 	nfit_mem->family = family;
1906 	if (override_dsm_mask && !disable_vendor_specific)
1907 		dsm_mask = override_dsm_mask;
1908 	else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1909 		dsm_mask = NVDIMM_INTEL_CMDMASK;
1910 		if (disable_vendor_specific)
1911 			dsm_mask &= ~(1 << ND_CMD_VENDOR);
1912 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1913 		dsm_mask = 0x1c3c76;
1914 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1915 		dsm_mask = 0x1fe;
1916 		if (disable_vendor_specific)
1917 			dsm_mask &= ~(1 << 8);
1918 	} else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1919 		dsm_mask = 0xffffffff;
1920 	} else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1921 		dsm_mask = 0x1f;
1922 	} else {
1923 		dev_dbg(dev, "unknown dimm command family\n");
1924 		nfit_mem->family = -1;
1925 		/* DSMs are optional, continue loading the driver... */
1926 		return 0;
1927 	}
1928 
1929 	/*
1930 	 * Function 0 is the command interrogation function, don't
1931 	 * export it to potential userspace use, and enable it to be
1932 	 * used as an error value in acpi_nfit_ctl().
1933 	 */
1934 	dsm_mask &= ~1UL;
1935 
1936 	guid = to_nfit_uuid(nfit_mem->family);
1937 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1938 		if (acpi_check_dsm(adev_dimm->handle, guid,
1939 					nfit_dsm_revid(nfit_mem->family, i),
1940 					1ULL << i))
1941 			set_bit(i, &nfit_mem->dsm_mask);
1942 
1943 	/*
1944 	 * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1945 	 * due to their better semantics handling locked capacity.
1946 	 */
1947 	label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1948 		| 1 << ND_CMD_SET_CONFIG_DATA;
1949 	if (family == NVDIMM_FAMILY_INTEL
1950 			&& (dsm_mask & label_mask) == label_mask)
1951 		/* skip _LS{I,R,W} enabling */;
1952 	else {
1953 		if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1954 				&& acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1955 			dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1956 			set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1957 		}
1958 
1959 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1960 				&& acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1961 			dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1962 			set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1963 		}
1964 
1965 		/*
1966 		 * Quirk read-only label configurations to preserve
1967 		 * access to label-less namespaces by default.
1968 		 */
1969 		if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1970 				&& !force_labels) {
1971 			dev_dbg(dev, "%s: No _LSW, disable labels\n",
1972 					dev_name(&adev_dimm->dev));
1973 			clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1974 		} else
1975 			dev_dbg(dev, "%s: Force enable labels\n",
1976 					dev_name(&adev_dimm->dev));
1977 	}
1978 
1979 	populate_shutdown_status(nfit_mem);
1980 
1981 	return 0;
1982 }
1983 
1984 static void shutdown_dimm_notify(void *data)
1985 {
1986 	struct acpi_nfit_desc *acpi_desc = data;
1987 	struct nfit_mem *nfit_mem;
1988 
1989 	mutex_lock(&acpi_desc->init_mutex);
1990 	/*
1991 	 * Clear out the nfit_mem->flags_attr and shut down dimm event
1992 	 * notifications.
1993 	 */
1994 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1995 		struct acpi_device *adev_dimm = nfit_mem->adev;
1996 
1997 		if (nfit_mem->flags_attr) {
1998 			sysfs_put(nfit_mem->flags_attr);
1999 			nfit_mem->flags_attr = NULL;
2000 		}
2001 		if (adev_dimm) {
2002 			acpi_remove_notify_handler(adev_dimm->handle,
2003 					ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
2004 			dev_set_drvdata(&adev_dimm->dev, NULL);
2005 		}
2006 	}
2007 	mutex_unlock(&acpi_desc->init_mutex);
2008 }
2009 
2010 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2011 {
2012 	switch (family) {
2013 	case NVDIMM_FAMILY_INTEL:
2014 		return intel_security_ops;
2015 	default:
2016 		return NULL;
2017 	}
2018 }
2019 
2020 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2021 {
2022 	struct nfit_mem *nfit_mem;
2023 	int dimm_count = 0, rc;
2024 	struct nvdimm *nvdimm;
2025 
2026 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2027 		struct acpi_nfit_flush_address *flush;
2028 		unsigned long flags = 0, cmd_mask;
2029 		struct nfit_memdev *nfit_memdev;
2030 		u32 device_handle;
2031 		u16 mem_flags;
2032 
2033 		device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2034 		nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2035 		if (nvdimm) {
2036 			dimm_count++;
2037 			continue;
2038 		}
2039 
2040 		if (nfit_mem->bdw && nfit_mem->memdev_pmem)
2041 			set_bit(NDD_ALIASING, &flags);
2042 
2043 		/* collate flags across all memdevs for this dimm */
2044 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2045 			struct acpi_nfit_memory_map *dimm_memdev;
2046 
2047 			dimm_memdev = __to_nfit_memdev(nfit_mem);
2048 			if (dimm_memdev->device_handle
2049 					!= nfit_memdev->memdev->device_handle)
2050 				continue;
2051 			dimm_memdev->flags |= nfit_memdev->memdev->flags;
2052 		}
2053 
2054 		mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2055 		if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2056 			set_bit(NDD_UNARMED, &flags);
2057 
2058 		rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2059 		if (rc)
2060 			continue;
2061 
2062 		/*
2063 		 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2064 		 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2065 		 * userspace interface.
2066 		 */
2067 		cmd_mask = 1UL << ND_CMD_CALL;
2068 		if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2069 			/*
2070 			 * These commands have a 1:1 correspondence
2071 			 * between DSM payload and libnvdimm ioctl
2072 			 * payload format.
2073 			 */
2074 			cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2075 		}
2076 
2077 		/* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2078 		if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2079 			set_bit(NDD_NOBLK, &flags);
2080 
2081 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2082 			set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2083 			set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2084 		}
2085 		if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2086 			set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2087 
2088 		flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2089 			: NULL;
2090 		nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2091 				acpi_nfit_dimm_attribute_groups,
2092 				flags, cmd_mask, flush ? flush->hint_count : 0,
2093 				nfit_mem->flush_wpq, &nfit_mem->id[0],
2094 				acpi_nfit_get_security_ops(nfit_mem->family));
2095 		if (!nvdimm)
2096 			return -ENOMEM;
2097 
2098 		nfit_mem->nvdimm = nvdimm;
2099 		dimm_count++;
2100 
2101 		if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2102 			continue;
2103 
2104 		dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2105 				nvdimm_name(nvdimm),
2106 		  mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2107 		  mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2108 		  mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2109 		  mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2110 		  mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2111 
2112 	}
2113 
2114 	rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2115 	if (rc)
2116 		return rc;
2117 
2118 	/*
2119 	 * Now that dimms are successfully registered, and async registration
2120 	 * is flushed, attempt to enable event notification.
2121 	 */
2122 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2123 		struct kernfs_node *nfit_kernfs;
2124 
2125 		nvdimm = nfit_mem->nvdimm;
2126 		if (!nvdimm)
2127 			continue;
2128 
2129 		nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2130 		if (nfit_kernfs)
2131 			nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2132 					"flags");
2133 		sysfs_put(nfit_kernfs);
2134 		if (!nfit_mem->flags_attr)
2135 			dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2136 					nvdimm_name(nvdimm));
2137 	}
2138 
2139 	return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2140 			acpi_desc);
2141 }
2142 
2143 /*
2144  * These constants are private because there are no kernel consumers of
2145  * these commands.
2146  */
2147 enum nfit_aux_cmds {
2148         NFIT_CMD_TRANSLATE_SPA = 5,
2149         NFIT_CMD_ARS_INJECT_SET = 7,
2150         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2151         NFIT_CMD_ARS_INJECT_GET = 9,
2152 };
2153 
2154 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2155 {
2156 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2157 	const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2158 	struct acpi_device *adev;
2159 	unsigned long dsm_mask;
2160 	int i;
2161 
2162 	nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2163 	nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2164 	adev = to_acpi_dev(acpi_desc);
2165 	if (!adev)
2166 		return;
2167 
2168 	for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2169 		if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2170 			set_bit(i, &nd_desc->cmd_mask);
2171 	set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2172 
2173 	dsm_mask =
2174 		(1 << ND_CMD_ARS_CAP) |
2175 		(1 << ND_CMD_ARS_START) |
2176 		(1 << ND_CMD_ARS_STATUS) |
2177 		(1 << ND_CMD_CLEAR_ERROR) |
2178 		(1 << NFIT_CMD_TRANSLATE_SPA) |
2179 		(1 << NFIT_CMD_ARS_INJECT_SET) |
2180 		(1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2181 		(1 << NFIT_CMD_ARS_INJECT_GET);
2182 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2183 		if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2184 			set_bit(i, &nd_desc->bus_dsm_mask);
2185 }
2186 
2187 static ssize_t range_index_show(struct device *dev,
2188 		struct device_attribute *attr, char *buf)
2189 {
2190 	struct nd_region *nd_region = to_nd_region(dev);
2191 	struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2192 
2193 	return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2194 }
2195 static DEVICE_ATTR_RO(range_index);
2196 
2197 static struct attribute *acpi_nfit_region_attributes[] = {
2198 	&dev_attr_range_index.attr,
2199 	NULL,
2200 };
2201 
2202 static const struct attribute_group acpi_nfit_region_attribute_group = {
2203 	.name = "nfit",
2204 	.attrs = acpi_nfit_region_attributes,
2205 };
2206 
2207 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2208 	&nd_region_attribute_group,
2209 	&nd_mapping_attribute_group,
2210 	&nd_device_attribute_group,
2211 	&nd_numa_attribute_group,
2212 	&acpi_nfit_region_attribute_group,
2213 	NULL,
2214 };
2215 
2216 /* enough info to uniquely specify an interleave set */
2217 struct nfit_set_info {
2218 	struct nfit_set_info_map {
2219 		u64 region_offset;
2220 		u32 serial_number;
2221 		u32 pad;
2222 	} mapping[0];
2223 };
2224 
2225 struct nfit_set_info2 {
2226 	struct nfit_set_info_map2 {
2227 		u64 region_offset;
2228 		u32 serial_number;
2229 		u16 vendor_id;
2230 		u16 manufacturing_date;
2231 		u8  manufacturing_location;
2232 		u8  reserved[31];
2233 	} mapping[0];
2234 };
2235 
2236 static size_t sizeof_nfit_set_info(int num_mappings)
2237 {
2238 	return sizeof(struct nfit_set_info)
2239 		+ num_mappings * sizeof(struct nfit_set_info_map);
2240 }
2241 
2242 static size_t sizeof_nfit_set_info2(int num_mappings)
2243 {
2244 	return sizeof(struct nfit_set_info2)
2245 		+ num_mappings * sizeof(struct nfit_set_info_map2);
2246 }
2247 
2248 static int cmp_map_compat(const void *m0, const void *m1)
2249 {
2250 	const struct nfit_set_info_map *map0 = m0;
2251 	const struct nfit_set_info_map *map1 = m1;
2252 
2253 	return memcmp(&map0->region_offset, &map1->region_offset,
2254 			sizeof(u64));
2255 }
2256 
2257 static int cmp_map(const void *m0, const void *m1)
2258 {
2259 	const struct nfit_set_info_map *map0 = m0;
2260 	const struct nfit_set_info_map *map1 = m1;
2261 
2262 	if (map0->region_offset < map1->region_offset)
2263 		return -1;
2264 	else if (map0->region_offset > map1->region_offset)
2265 		return 1;
2266 	return 0;
2267 }
2268 
2269 static int cmp_map2(const void *m0, const void *m1)
2270 {
2271 	const struct nfit_set_info_map2 *map0 = m0;
2272 	const struct nfit_set_info_map2 *map1 = m1;
2273 
2274 	if (map0->region_offset < map1->region_offset)
2275 		return -1;
2276 	else if (map0->region_offset > map1->region_offset)
2277 		return 1;
2278 	return 0;
2279 }
2280 
2281 /* Retrieve the nth entry referencing this spa */
2282 static struct acpi_nfit_memory_map *memdev_from_spa(
2283 		struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2284 {
2285 	struct nfit_memdev *nfit_memdev;
2286 
2287 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2288 		if (nfit_memdev->memdev->range_index == range_index)
2289 			if (n-- == 0)
2290 				return nfit_memdev->memdev;
2291 	return NULL;
2292 }
2293 
2294 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2295 		struct nd_region_desc *ndr_desc,
2296 		struct acpi_nfit_system_address *spa)
2297 {
2298 	struct device *dev = acpi_desc->dev;
2299 	struct nd_interleave_set *nd_set;
2300 	u16 nr = ndr_desc->num_mappings;
2301 	struct nfit_set_info2 *info2;
2302 	struct nfit_set_info *info;
2303 	int i;
2304 
2305 	nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2306 	if (!nd_set)
2307 		return -ENOMEM;
2308 	guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2309 
2310 	info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2311 	if (!info)
2312 		return -ENOMEM;
2313 
2314 	info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2315 	if (!info2)
2316 		return -ENOMEM;
2317 
2318 	for (i = 0; i < nr; i++) {
2319 		struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2320 		struct nfit_set_info_map *map = &info->mapping[i];
2321 		struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2322 		struct nvdimm *nvdimm = mapping->nvdimm;
2323 		struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2324 		struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2325 				spa->range_index, i);
2326 		struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2327 
2328 		if (!memdev || !nfit_mem->dcr) {
2329 			dev_err(dev, "%s: failed to find DCR\n", __func__);
2330 			return -ENODEV;
2331 		}
2332 
2333 		map->region_offset = memdev->region_offset;
2334 		map->serial_number = dcr->serial_number;
2335 
2336 		map2->region_offset = memdev->region_offset;
2337 		map2->serial_number = dcr->serial_number;
2338 		map2->vendor_id = dcr->vendor_id;
2339 		map2->manufacturing_date = dcr->manufacturing_date;
2340 		map2->manufacturing_location = dcr->manufacturing_location;
2341 	}
2342 
2343 	/* v1.1 namespaces */
2344 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2345 			cmp_map, NULL);
2346 	nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2347 
2348 	/* v1.2 namespaces */
2349 	sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2350 			cmp_map2, NULL);
2351 	nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2352 
2353 	/* support v1.1 namespaces created with the wrong sort order */
2354 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2355 			cmp_map_compat, NULL);
2356 	nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2357 
2358 	/* record the result of the sort for the mapping position */
2359 	for (i = 0; i < nr; i++) {
2360 		struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2361 		int j;
2362 
2363 		for (j = 0; j < nr; j++) {
2364 			struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2365 			struct nvdimm *nvdimm = mapping->nvdimm;
2366 			struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2367 			struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2368 
2369 			if (map2->serial_number == dcr->serial_number &&
2370 			    map2->vendor_id == dcr->vendor_id &&
2371 			    map2->manufacturing_date == dcr->manufacturing_date &&
2372 			    map2->manufacturing_location
2373 				    == dcr->manufacturing_location) {
2374 				mapping->position = i;
2375 				break;
2376 			}
2377 		}
2378 	}
2379 
2380 	ndr_desc->nd_set = nd_set;
2381 	devm_kfree(dev, info);
2382 	devm_kfree(dev, info2);
2383 
2384 	return 0;
2385 }
2386 
2387 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2388 {
2389 	struct acpi_nfit_interleave *idt = mmio->idt;
2390 	u32 sub_line_offset, line_index, line_offset;
2391 	u64 line_no, table_skip_count, table_offset;
2392 
2393 	line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2394 	table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2395 	line_offset = idt->line_offset[line_index]
2396 		* mmio->line_size;
2397 	table_offset = table_skip_count * mmio->table_size;
2398 
2399 	return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2400 }
2401 
2402 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2403 {
2404 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2405 	u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2406 	const u32 STATUS_MASK = 0x80000037;
2407 
2408 	if (mmio->num_lines)
2409 		offset = to_interleave_offset(offset, mmio);
2410 
2411 	return readl(mmio->addr.base + offset) & STATUS_MASK;
2412 }
2413 
2414 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2415 		resource_size_t dpa, unsigned int len, unsigned int write)
2416 {
2417 	u64 cmd, offset;
2418 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2419 
2420 	enum {
2421 		BCW_OFFSET_MASK = (1ULL << 48)-1,
2422 		BCW_LEN_SHIFT = 48,
2423 		BCW_LEN_MASK = (1ULL << 8) - 1,
2424 		BCW_CMD_SHIFT = 56,
2425 	};
2426 
2427 	cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2428 	len = len >> L1_CACHE_SHIFT;
2429 	cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2430 	cmd |= ((u64) write) << BCW_CMD_SHIFT;
2431 
2432 	offset = nfit_blk->cmd_offset + mmio->size * bw;
2433 	if (mmio->num_lines)
2434 		offset = to_interleave_offset(offset, mmio);
2435 
2436 	writeq(cmd, mmio->addr.base + offset);
2437 	nvdimm_flush(nfit_blk->nd_region);
2438 
2439 	if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2440 		readq(mmio->addr.base + offset);
2441 }
2442 
2443 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2444 		resource_size_t dpa, void *iobuf, size_t len, int rw,
2445 		unsigned int lane)
2446 {
2447 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2448 	unsigned int copied = 0;
2449 	u64 base_offset;
2450 	int rc;
2451 
2452 	base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2453 		+ lane * mmio->size;
2454 	write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2455 	while (len) {
2456 		unsigned int c;
2457 		u64 offset;
2458 
2459 		if (mmio->num_lines) {
2460 			u32 line_offset;
2461 
2462 			offset = to_interleave_offset(base_offset + copied,
2463 					mmio);
2464 			div_u64_rem(offset, mmio->line_size, &line_offset);
2465 			c = min_t(size_t, len, mmio->line_size - line_offset);
2466 		} else {
2467 			offset = base_offset + nfit_blk->bdw_offset;
2468 			c = len;
2469 		}
2470 
2471 		if (rw)
2472 			memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2473 		else {
2474 			if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2475 				arch_invalidate_pmem((void __force *)
2476 					mmio->addr.aperture + offset, c);
2477 
2478 			memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2479 		}
2480 
2481 		copied += c;
2482 		len -= c;
2483 	}
2484 
2485 	if (rw)
2486 		nvdimm_flush(nfit_blk->nd_region);
2487 
2488 	rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2489 	return rc;
2490 }
2491 
2492 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2493 		resource_size_t dpa, void *iobuf, u64 len, int rw)
2494 {
2495 	struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2496 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2497 	struct nd_region *nd_region = nfit_blk->nd_region;
2498 	unsigned int lane, copied = 0;
2499 	int rc = 0;
2500 
2501 	lane = nd_region_acquire_lane(nd_region);
2502 	while (len) {
2503 		u64 c = min(len, mmio->size);
2504 
2505 		rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2506 				iobuf + copied, c, rw, lane);
2507 		if (rc)
2508 			break;
2509 
2510 		copied += c;
2511 		len -= c;
2512 	}
2513 	nd_region_release_lane(nd_region, lane);
2514 
2515 	return rc;
2516 }
2517 
2518 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2519 		struct acpi_nfit_interleave *idt, u16 interleave_ways)
2520 {
2521 	if (idt) {
2522 		mmio->num_lines = idt->line_count;
2523 		mmio->line_size = idt->line_size;
2524 		if (interleave_ways == 0)
2525 			return -ENXIO;
2526 		mmio->table_size = mmio->num_lines * interleave_ways
2527 			* mmio->line_size;
2528 	}
2529 
2530 	return 0;
2531 }
2532 
2533 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2534 		struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2535 {
2536 	struct nd_cmd_dimm_flags flags;
2537 	int rc;
2538 
2539 	memset(&flags, 0, sizeof(flags));
2540 	rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2541 			sizeof(flags), NULL);
2542 
2543 	if (rc >= 0 && flags.status == 0)
2544 		nfit_blk->dimm_flags = flags.flags;
2545 	else if (rc == -ENOTTY) {
2546 		/* fall back to a conservative default */
2547 		nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2548 		rc = 0;
2549 	} else
2550 		rc = -ENXIO;
2551 
2552 	return rc;
2553 }
2554 
2555 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2556 		struct device *dev)
2557 {
2558 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2559 	struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2560 	struct nfit_blk_mmio *mmio;
2561 	struct nfit_blk *nfit_blk;
2562 	struct nfit_mem *nfit_mem;
2563 	struct nvdimm *nvdimm;
2564 	int rc;
2565 
2566 	nvdimm = nd_blk_region_to_dimm(ndbr);
2567 	nfit_mem = nvdimm_provider_data(nvdimm);
2568 	if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2569 		dev_dbg(dev, "missing%s%s%s\n",
2570 				nfit_mem ? "" : " nfit_mem",
2571 				(nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2572 				(nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2573 		return -ENXIO;
2574 	}
2575 
2576 	nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2577 	if (!nfit_blk)
2578 		return -ENOMEM;
2579 	nd_blk_region_set_provider_data(ndbr, nfit_blk);
2580 	nfit_blk->nd_region = to_nd_region(dev);
2581 
2582 	/* map block aperture memory */
2583 	nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2584 	mmio = &nfit_blk->mmio[BDW];
2585 	mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2586                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2587 	if (!mmio->addr.base) {
2588 		dev_dbg(dev, "%s failed to map bdw\n",
2589 				nvdimm_name(nvdimm));
2590 		return -ENOMEM;
2591 	}
2592 	mmio->size = nfit_mem->bdw->size;
2593 	mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2594 	mmio->idt = nfit_mem->idt_bdw;
2595 	mmio->spa = nfit_mem->spa_bdw;
2596 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2597 			nfit_mem->memdev_bdw->interleave_ways);
2598 	if (rc) {
2599 		dev_dbg(dev, "%s failed to init bdw interleave\n",
2600 				nvdimm_name(nvdimm));
2601 		return rc;
2602 	}
2603 
2604 	/* map block control memory */
2605 	nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2606 	nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2607 	mmio = &nfit_blk->mmio[DCR];
2608 	mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2609 			nfit_mem->spa_dcr->length);
2610 	if (!mmio->addr.base) {
2611 		dev_dbg(dev, "%s failed to map dcr\n",
2612 				nvdimm_name(nvdimm));
2613 		return -ENOMEM;
2614 	}
2615 	mmio->size = nfit_mem->dcr->window_size;
2616 	mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2617 	mmio->idt = nfit_mem->idt_dcr;
2618 	mmio->spa = nfit_mem->spa_dcr;
2619 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2620 			nfit_mem->memdev_dcr->interleave_ways);
2621 	if (rc) {
2622 		dev_dbg(dev, "%s failed to init dcr interleave\n",
2623 				nvdimm_name(nvdimm));
2624 		return rc;
2625 	}
2626 
2627 	rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2628 	if (rc < 0) {
2629 		dev_dbg(dev, "%s failed get DIMM flags\n",
2630 				nvdimm_name(nvdimm));
2631 		return rc;
2632 	}
2633 
2634 	if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2635 		dev_warn(dev, "unable to guarantee persistence of writes\n");
2636 
2637 	if (mmio->line_size == 0)
2638 		return 0;
2639 
2640 	if ((u32) nfit_blk->cmd_offset % mmio->line_size
2641 			+ 8 > mmio->line_size) {
2642 		dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2643 		return -ENXIO;
2644 	} else if ((u32) nfit_blk->stat_offset % mmio->line_size
2645 			+ 8 > mmio->line_size) {
2646 		dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2647 		return -ENXIO;
2648 	}
2649 
2650 	return 0;
2651 }
2652 
2653 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2654 		struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2655 {
2656 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2657 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2658 	int cmd_rc, rc;
2659 
2660 	cmd->address = spa->address;
2661 	cmd->length = spa->length;
2662 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2663 			sizeof(*cmd), &cmd_rc);
2664 	if (rc < 0)
2665 		return rc;
2666 	return cmd_rc;
2667 }
2668 
2669 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2670 		struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2671 {
2672 	int rc;
2673 	int cmd_rc;
2674 	struct nd_cmd_ars_start ars_start;
2675 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2676 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2677 
2678 	memset(&ars_start, 0, sizeof(ars_start));
2679 	ars_start.address = spa->address;
2680 	ars_start.length = spa->length;
2681 	if (req_type == ARS_REQ_SHORT)
2682 		ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2683 	if (nfit_spa_type(spa) == NFIT_SPA_PM)
2684 		ars_start.type = ND_ARS_PERSISTENT;
2685 	else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2686 		ars_start.type = ND_ARS_VOLATILE;
2687 	else
2688 		return -ENOTTY;
2689 
2690 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2691 			sizeof(ars_start), &cmd_rc);
2692 
2693 	if (rc < 0)
2694 		return rc;
2695 	if (cmd_rc < 0)
2696 		return cmd_rc;
2697 	set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2698 	return 0;
2699 }
2700 
2701 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2702 {
2703 	int rc, cmd_rc;
2704 	struct nd_cmd_ars_start ars_start;
2705 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2706 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2707 
2708 	ars_start = (struct nd_cmd_ars_start) {
2709 		.address = ars_status->restart_address,
2710 		.length = ars_status->restart_length,
2711 		.type = ars_status->type,
2712 	};
2713 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2714 			sizeof(ars_start), &cmd_rc);
2715 	if (rc < 0)
2716 		return rc;
2717 	return cmd_rc;
2718 }
2719 
2720 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2721 {
2722 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2723 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2724 	int rc, cmd_rc;
2725 
2726 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2727 			acpi_desc->max_ars, &cmd_rc);
2728 	if (rc < 0)
2729 		return rc;
2730 	return cmd_rc;
2731 }
2732 
2733 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2734 		struct nfit_spa *nfit_spa)
2735 {
2736 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2737 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2738 	struct nd_region *nd_region = nfit_spa->nd_region;
2739 	struct device *dev;
2740 
2741 	lockdep_assert_held(&acpi_desc->init_mutex);
2742 	/*
2743 	 * Only advance the ARS state for ARS runs initiated by the
2744 	 * kernel, ignore ARS results from BIOS initiated runs for scrub
2745 	 * completion tracking.
2746 	 */
2747 	if (acpi_desc->scrub_spa != nfit_spa)
2748 		return;
2749 
2750 	if ((ars_status->address >= spa->address && ars_status->address
2751 				< spa->address + spa->length)
2752 			|| (ars_status->address < spa->address)) {
2753 		/*
2754 		 * Assume that if a scrub starts at an offset from the
2755 		 * start of nfit_spa that we are in the continuation
2756 		 * case.
2757 		 *
2758 		 * Otherwise, if the scrub covers the spa range, mark
2759 		 * any pending request complete.
2760 		 */
2761 		if (ars_status->address + ars_status->length
2762 				>= spa->address + spa->length)
2763 				/* complete */;
2764 		else
2765 			return;
2766 	} else
2767 		return;
2768 
2769 	acpi_desc->scrub_spa = NULL;
2770 	if (nd_region) {
2771 		dev = nd_region_dev(nd_region);
2772 		nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2773 	} else
2774 		dev = acpi_desc->dev;
2775 	dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2776 }
2777 
2778 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2779 {
2780 	struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2781 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2782 	int rc;
2783 	u32 i;
2784 
2785 	/*
2786 	 * First record starts at 44 byte offset from the start of the
2787 	 * payload.
2788 	 */
2789 	if (ars_status->out_length < 44)
2790 		return 0;
2791 
2792 	/*
2793 	 * Ignore potentially stale results that are only refreshed
2794 	 * after a start-ARS event.
2795 	 */
2796 	if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2797 		dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2798 				ars_status->num_records);
2799 		return 0;
2800 	}
2801 
2802 	for (i = 0; i < ars_status->num_records; i++) {
2803 		/* only process full records */
2804 		if (ars_status->out_length
2805 				< 44 + sizeof(struct nd_ars_record) * (i + 1))
2806 			break;
2807 		rc = nvdimm_bus_add_badrange(nvdimm_bus,
2808 				ars_status->records[i].err_address,
2809 				ars_status->records[i].length);
2810 		if (rc)
2811 			return rc;
2812 	}
2813 	if (i < ars_status->num_records)
2814 		dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2815 
2816 	return 0;
2817 }
2818 
2819 static void acpi_nfit_remove_resource(void *data)
2820 {
2821 	struct resource *res = data;
2822 
2823 	remove_resource(res);
2824 }
2825 
2826 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2827 		struct nd_region_desc *ndr_desc)
2828 {
2829 	struct resource *res, *nd_res = ndr_desc->res;
2830 	int is_pmem, ret;
2831 
2832 	/* No operation if the region is already registered as PMEM */
2833 	is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2834 				IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2835 	if (is_pmem == REGION_INTERSECTS)
2836 		return 0;
2837 
2838 	res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2839 	if (!res)
2840 		return -ENOMEM;
2841 
2842 	res->name = "Persistent Memory";
2843 	res->start = nd_res->start;
2844 	res->end = nd_res->end;
2845 	res->flags = IORESOURCE_MEM;
2846 	res->desc = IORES_DESC_PERSISTENT_MEMORY;
2847 
2848 	ret = insert_resource(&iomem_resource, res);
2849 	if (ret)
2850 		return ret;
2851 
2852 	ret = devm_add_action_or_reset(acpi_desc->dev,
2853 					acpi_nfit_remove_resource,
2854 					res);
2855 	if (ret)
2856 		return ret;
2857 
2858 	return 0;
2859 }
2860 
2861 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2862 		struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2863 		struct acpi_nfit_memory_map *memdev,
2864 		struct nfit_spa *nfit_spa)
2865 {
2866 	struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2867 			memdev->device_handle);
2868 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2869 	struct nd_blk_region_desc *ndbr_desc;
2870 	struct nfit_mem *nfit_mem;
2871 	int rc;
2872 
2873 	if (!nvdimm) {
2874 		dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2875 				spa->range_index, memdev->device_handle);
2876 		return -ENODEV;
2877 	}
2878 
2879 	mapping->nvdimm = nvdimm;
2880 	switch (nfit_spa_type(spa)) {
2881 	case NFIT_SPA_PM:
2882 	case NFIT_SPA_VOLATILE:
2883 		mapping->start = memdev->address;
2884 		mapping->size = memdev->region_size;
2885 		break;
2886 	case NFIT_SPA_DCR:
2887 		nfit_mem = nvdimm_provider_data(nvdimm);
2888 		if (!nfit_mem || !nfit_mem->bdw) {
2889 			dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2890 					spa->range_index, nvdimm_name(nvdimm));
2891 			break;
2892 		}
2893 
2894 		mapping->size = nfit_mem->bdw->capacity;
2895 		mapping->start = nfit_mem->bdw->start_address;
2896 		ndr_desc->num_lanes = nfit_mem->bdw->windows;
2897 		ndr_desc->mapping = mapping;
2898 		ndr_desc->num_mappings = 1;
2899 		ndbr_desc = to_blk_region_desc(ndr_desc);
2900 		ndbr_desc->enable = acpi_nfit_blk_region_enable;
2901 		ndbr_desc->do_io = acpi_desc->blk_do_io;
2902 		rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2903 		if (rc)
2904 			return rc;
2905 		nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2906 				ndr_desc);
2907 		if (!nfit_spa->nd_region)
2908 			return -ENOMEM;
2909 		break;
2910 	}
2911 
2912 	return 0;
2913 }
2914 
2915 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2916 {
2917 	return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2918 		nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2919 		nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2920 		nfit_spa_type(spa) == NFIT_SPA_PCD);
2921 }
2922 
2923 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2924 {
2925 	return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2926 		nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2927 		nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2928 }
2929 
2930 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2931 		struct nfit_spa *nfit_spa)
2932 {
2933 	static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2934 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2935 	struct nd_blk_region_desc ndbr_desc;
2936 	struct nd_region_desc *ndr_desc;
2937 	struct nfit_memdev *nfit_memdev;
2938 	struct nvdimm_bus *nvdimm_bus;
2939 	struct resource res;
2940 	int count = 0, rc;
2941 
2942 	if (nfit_spa->nd_region)
2943 		return 0;
2944 
2945 	if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2946 		dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2947 		return 0;
2948 	}
2949 
2950 	memset(&res, 0, sizeof(res));
2951 	memset(&mappings, 0, sizeof(mappings));
2952 	memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2953 	res.start = spa->address;
2954 	res.end = res.start + spa->length - 1;
2955 	ndr_desc = &ndbr_desc.ndr_desc;
2956 	ndr_desc->res = &res;
2957 	ndr_desc->provider_data = nfit_spa;
2958 	ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2959 	if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2960 		ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2961 						spa->proximity_domain);
2962 		ndr_desc->target_node = acpi_map_pxm_to_node(
2963 				spa->proximity_domain);
2964 	} else {
2965 		ndr_desc->numa_node = NUMA_NO_NODE;
2966 		ndr_desc->target_node = NUMA_NO_NODE;
2967 	}
2968 
2969 	/*
2970 	 * Persistence domain bits are hierarchical, if
2971 	 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2972 	 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2973 	 */
2974 	if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2975 		set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2976 	else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2977 		set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2978 
2979 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2980 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2981 		struct nd_mapping_desc *mapping;
2982 
2983 		if (memdev->range_index != spa->range_index)
2984 			continue;
2985 		if (count >= ND_MAX_MAPPINGS) {
2986 			dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2987 					spa->range_index, ND_MAX_MAPPINGS);
2988 			return -ENXIO;
2989 		}
2990 		mapping = &mappings[count++];
2991 		rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2992 				memdev, nfit_spa);
2993 		if (rc)
2994 			goto out;
2995 	}
2996 
2997 	ndr_desc->mapping = mappings;
2998 	ndr_desc->num_mappings = count;
2999 	rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
3000 	if (rc)
3001 		goto out;
3002 
3003 	nvdimm_bus = acpi_desc->nvdimm_bus;
3004 	if (nfit_spa_type(spa) == NFIT_SPA_PM) {
3005 		rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
3006 		if (rc) {
3007 			dev_warn(acpi_desc->dev,
3008 				"failed to insert pmem resource to iomem: %d\n",
3009 				rc);
3010 			goto out;
3011 		}
3012 
3013 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3014 				ndr_desc);
3015 		if (!nfit_spa->nd_region)
3016 			rc = -ENOMEM;
3017 	} else if (nfit_spa_is_volatile(spa)) {
3018 		nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3019 				ndr_desc);
3020 		if (!nfit_spa->nd_region)
3021 			rc = -ENOMEM;
3022 	} else if (nfit_spa_is_virtual(spa)) {
3023 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3024 				ndr_desc);
3025 		if (!nfit_spa->nd_region)
3026 			rc = -ENOMEM;
3027 	}
3028 
3029  out:
3030 	if (rc)
3031 		dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3032 				nfit_spa->spa->range_index);
3033 	return rc;
3034 }
3035 
3036 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3037 {
3038 	struct device *dev = acpi_desc->dev;
3039 	struct nd_cmd_ars_status *ars_status;
3040 
3041 	if (acpi_desc->ars_status) {
3042 		memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3043 		return 0;
3044 	}
3045 
3046 	ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3047 	if (!ars_status)
3048 		return -ENOMEM;
3049 	acpi_desc->ars_status = ars_status;
3050 	return 0;
3051 }
3052 
3053 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3054 {
3055 	int rc;
3056 
3057 	if (ars_status_alloc(acpi_desc))
3058 		return -ENOMEM;
3059 
3060 	rc = ars_get_status(acpi_desc);
3061 
3062 	if (rc < 0 && rc != -ENOSPC)
3063 		return rc;
3064 
3065 	if (ars_status_process_records(acpi_desc))
3066 		dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3067 
3068 	return rc;
3069 }
3070 
3071 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3072 		struct nfit_spa *nfit_spa)
3073 {
3074 	int rc;
3075 
3076 	if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3077 		return acpi_nfit_register_region(acpi_desc, nfit_spa);
3078 
3079 	set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3080 	if (!no_init_ars)
3081 		set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3082 
3083 	switch (acpi_nfit_query_poison(acpi_desc)) {
3084 	case 0:
3085 	case -ENOSPC:
3086 	case -EAGAIN:
3087 		rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3088 		/* shouldn't happen, try again later */
3089 		if (rc == -EBUSY)
3090 			break;
3091 		if (rc) {
3092 			set_bit(ARS_FAILED, &nfit_spa->ars_state);
3093 			break;
3094 		}
3095 		clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3096 		rc = acpi_nfit_query_poison(acpi_desc);
3097 		if (rc)
3098 			break;
3099 		acpi_desc->scrub_spa = nfit_spa;
3100 		ars_complete(acpi_desc, nfit_spa);
3101 		/*
3102 		 * If ars_complete() says we didn't complete the
3103 		 * short scrub, we'll try again with a long
3104 		 * request.
3105 		 */
3106 		acpi_desc->scrub_spa = NULL;
3107 		break;
3108 	case -EBUSY:
3109 	case -ENOMEM:
3110 		/*
3111 		 * BIOS was using ARS, wait for it to complete (or
3112 		 * resources to become available) and then perform our
3113 		 * own scrubs.
3114 		 */
3115 		break;
3116 	default:
3117 		set_bit(ARS_FAILED, &nfit_spa->ars_state);
3118 		break;
3119 	}
3120 
3121 	return acpi_nfit_register_region(acpi_desc, nfit_spa);
3122 }
3123 
3124 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3125 {
3126 	struct nfit_spa *nfit_spa;
3127 
3128 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3129 		if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3130 			continue;
3131 		ars_complete(acpi_desc, nfit_spa);
3132 	}
3133 }
3134 
3135 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3136 		int query_rc)
3137 {
3138 	unsigned int tmo = acpi_desc->scrub_tmo;
3139 	struct device *dev = acpi_desc->dev;
3140 	struct nfit_spa *nfit_spa;
3141 
3142 	lockdep_assert_held(&acpi_desc->init_mutex);
3143 
3144 	if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3145 		return 0;
3146 
3147 	if (query_rc == -EBUSY) {
3148 		dev_dbg(dev, "ARS: ARS busy\n");
3149 		return min(30U * 60U, tmo * 2);
3150 	}
3151 	if (query_rc == -ENOSPC) {
3152 		dev_dbg(dev, "ARS: ARS continue\n");
3153 		ars_continue(acpi_desc);
3154 		return 1;
3155 	}
3156 	if (query_rc && query_rc != -EAGAIN) {
3157 		unsigned long long addr, end;
3158 
3159 		addr = acpi_desc->ars_status->address;
3160 		end = addr + acpi_desc->ars_status->length;
3161 		dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3162 				query_rc);
3163 	}
3164 
3165 	ars_complete_all(acpi_desc);
3166 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3167 		enum nfit_ars_state req_type;
3168 		int rc;
3169 
3170 		if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3171 			continue;
3172 
3173 		/* prefer short ARS requests first */
3174 		if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3175 			req_type = ARS_REQ_SHORT;
3176 		else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3177 			req_type = ARS_REQ_LONG;
3178 		else
3179 			continue;
3180 		rc = ars_start(acpi_desc, nfit_spa, req_type);
3181 
3182 		dev = nd_region_dev(nfit_spa->nd_region);
3183 		dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3184 				nfit_spa->spa->range_index,
3185 				req_type == ARS_REQ_SHORT ? "short" : "long",
3186 				rc);
3187 		/*
3188 		 * Hmm, we raced someone else starting ARS? Try again in
3189 		 * a bit.
3190 		 */
3191 		if (rc == -EBUSY)
3192 			return 1;
3193 		if (rc == 0) {
3194 			dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3195 					"scrub start while range %d active\n",
3196 					acpi_desc->scrub_spa->spa->range_index);
3197 			clear_bit(req_type, &nfit_spa->ars_state);
3198 			acpi_desc->scrub_spa = nfit_spa;
3199 			/*
3200 			 * Consider this spa last for future scrub
3201 			 * requests
3202 			 */
3203 			list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3204 			return 1;
3205 		}
3206 
3207 		dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3208 				nfit_spa->spa->range_index, rc);
3209 		set_bit(ARS_FAILED, &nfit_spa->ars_state);
3210 	}
3211 	return 0;
3212 }
3213 
3214 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3215 {
3216 	lockdep_assert_held(&acpi_desc->init_mutex);
3217 
3218 	set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3219 	/* note this should only be set from within the workqueue */
3220 	if (tmo)
3221 		acpi_desc->scrub_tmo = tmo;
3222 	queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3223 }
3224 
3225 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3226 {
3227 	__sched_ars(acpi_desc, 0);
3228 }
3229 
3230 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3231 {
3232 	lockdep_assert_held(&acpi_desc->init_mutex);
3233 
3234 	clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3235 	acpi_desc->scrub_count++;
3236 	if (acpi_desc->scrub_count_state)
3237 		sysfs_notify_dirent(acpi_desc->scrub_count_state);
3238 }
3239 
3240 static void acpi_nfit_scrub(struct work_struct *work)
3241 {
3242 	struct acpi_nfit_desc *acpi_desc;
3243 	unsigned int tmo;
3244 	int query_rc;
3245 
3246 	acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3247 	mutex_lock(&acpi_desc->init_mutex);
3248 	query_rc = acpi_nfit_query_poison(acpi_desc);
3249 	tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3250 	if (tmo)
3251 		__sched_ars(acpi_desc, tmo);
3252 	else
3253 		notify_ars_done(acpi_desc);
3254 	memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3255 	clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3256 	mutex_unlock(&acpi_desc->init_mutex);
3257 }
3258 
3259 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3260 		struct nfit_spa *nfit_spa)
3261 {
3262 	int type = nfit_spa_type(nfit_spa->spa);
3263 	struct nd_cmd_ars_cap ars_cap;
3264 	int rc;
3265 
3266 	set_bit(ARS_FAILED, &nfit_spa->ars_state);
3267 	memset(&ars_cap, 0, sizeof(ars_cap));
3268 	rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3269 	if (rc < 0)
3270 		return;
3271 	/* check that the supported scrub types match the spa type */
3272 	if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3273 				& ND_ARS_VOLATILE) == 0)
3274 		return;
3275 	if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3276 				& ND_ARS_PERSISTENT) == 0)
3277 		return;
3278 
3279 	nfit_spa->max_ars = ars_cap.max_ars_out;
3280 	nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3281 	acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3282 	clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3283 }
3284 
3285 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3286 {
3287 	struct nfit_spa *nfit_spa;
3288 	int rc;
3289 
3290 	set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3291 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3292 		switch (nfit_spa_type(nfit_spa->spa)) {
3293 		case NFIT_SPA_VOLATILE:
3294 		case NFIT_SPA_PM:
3295 			acpi_nfit_init_ars(acpi_desc, nfit_spa);
3296 			break;
3297 		}
3298 	}
3299 
3300 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3301 		switch (nfit_spa_type(nfit_spa->spa)) {
3302 		case NFIT_SPA_VOLATILE:
3303 		case NFIT_SPA_PM:
3304 			/* register regions and kick off initial ARS run */
3305 			rc = ars_register(acpi_desc, nfit_spa);
3306 			if (rc)
3307 				return rc;
3308 			break;
3309 		case NFIT_SPA_BDW:
3310 			/* nothing to register */
3311 			break;
3312 		case NFIT_SPA_DCR:
3313 		case NFIT_SPA_VDISK:
3314 		case NFIT_SPA_VCD:
3315 		case NFIT_SPA_PDISK:
3316 		case NFIT_SPA_PCD:
3317 			/* register known regions that don't support ARS */
3318 			rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3319 			if (rc)
3320 				return rc;
3321 			break;
3322 		default:
3323 			/* don't register unknown regions */
3324 			break;
3325 		}
3326 
3327 	sched_ars(acpi_desc);
3328 	return 0;
3329 }
3330 
3331 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3332 		struct nfit_table_prev *prev)
3333 {
3334 	struct device *dev = acpi_desc->dev;
3335 
3336 	if (!list_empty(&prev->spas) ||
3337 			!list_empty(&prev->memdevs) ||
3338 			!list_empty(&prev->dcrs) ||
3339 			!list_empty(&prev->bdws) ||
3340 			!list_empty(&prev->idts) ||
3341 			!list_empty(&prev->flushes)) {
3342 		dev_err(dev, "new nfit deletes entries (unsupported)\n");
3343 		return -ENXIO;
3344 	}
3345 	return 0;
3346 }
3347 
3348 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3349 {
3350 	struct device *dev = acpi_desc->dev;
3351 	struct kernfs_node *nfit;
3352 	struct device *bus_dev;
3353 
3354 	if (!ars_supported(acpi_desc->nvdimm_bus))
3355 		return 0;
3356 
3357 	bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3358 	nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3359 	if (!nfit) {
3360 		dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3361 		return -ENODEV;
3362 	}
3363 	acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3364 	sysfs_put(nfit);
3365 	if (!acpi_desc->scrub_count_state) {
3366 		dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3367 		return -ENODEV;
3368 	}
3369 
3370 	return 0;
3371 }
3372 
3373 static void acpi_nfit_unregister(void *data)
3374 {
3375 	struct acpi_nfit_desc *acpi_desc = data;
3376 
3377 	nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3378 }
3379 
3380 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3381 {
3382 	struct device *dev = acpi_desc->dev;
3383 	struct nfit_table_prev prev;
3384 	const void *end;
3385 	int rc;
3386 
3387 	if (!acpi_desc->nvdimm_bus) {
3388 		acpi_nfit_init_dsms(acpi_desc);
3389 
3390 		acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3391 				&acpi_desc->nd_desc);
3392 		if (!acpi_desc->nvdimm_bus)
3393 			return -ENOMEM;
3394 
3395 		rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3396 				acpi_desc);
3397 		if (rc)
3398 			return rc;
3399 
3400 		rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3401 		if (rc)
3402 			return rc;
3403 
3404 		/* register this acpi_desc for mce notifications */
3405 		mutex_lock(&acpi_desc_lock);
3406 		list_add_tail(&acpi_desc->list, &acpi_descs);
3407 		mutex_unlock(&acpi_desc_lock);
3408 	}
3409 
3410 	mutex_lock(&acpi_desc->init_mutex);
3411 
3412 	INIT_LIST_HEAD(&prev.spas);
3413 	INIT_LIST_HEAD(&prev.memdevs);
3414 	INIT_LIST_HEAD(&prev.dcrs);
3415 	INIT_LIST_HEAD(&prev.bdws);
3416 	INIT_LIST_HEAD(&prev.idts);
3417 	INIT_LIST_HEAD(&prev.flushes);
3418 
3419 	list_cut_position(&prev.spas, &acpi_desc->spas,
3420 				acpi_desc->spas.prev);
3421 	list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3422 				acpi_desc->memdevs.prev);
3423 	list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3424 				acpi_desc->dcrs.prev);
3425 	list_cut_position(&prev.bdws, &acpi_desc->bdws,
3426 				acpi_desc->bdws.prev);
3427 	list_cut_position(&prev.idts, &acpi_desc->idts,
3428 				acpi_desc->idts.prev);
3429 	list_cut_position(&prev.flushes, &acpi_desc->flushes,
3430 				acpi_desc->flushes.prev);
3431 
3432 	end = data + sz;
3433 	while (!IS_ERR_OR_NULL(data))
3434 		data = add_table(acpi_desc, &prev, data, end);
3435 
3436 	if (IS_ERR(data)) {
3437 		dev_dbg(dev, "nfit table parsing error: %ld\n",	PTR_ERR(data));
3438 		rc = PTR_ERR(data);
3439 		goto out_unlock;
3440 	}
3441 
3442 	rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3443 	if (rc)
3444 		goto out_unlock;
3445 
3446 	rc = nfit_mem_init(acpi_desc);
3447 	if (rc)
3448 		goto out_unlock;
3449 
3450 	rc = acpi_nfit_register_dimms(acpi_desc);
3451 	if (rc)
3452 		goto out_unlock;
3453 
3454 	rc = acpi_nfit_register_regions(acpi_desc);
3455 
3456  out_unlock:
3457 	mutex_unlock(&acpi_desc->init_mutex);
3458 	return rc;
3459 }
3460 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3461 
3462 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3463 {
3464 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3465 	struct device *dev = acpi_desc->dev;
3466 
3467 	/* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3468 	device_lock(dev);
3469 	device_unlock(dev);
3470 
3471 	/* Bounce the init_mutex to complete initial registration */
3472 	mutex_lock(&acpi_desc->init_mutex);
3473 	mutex_unlock(&acpi_desc->init_mutex);
3474 
3475 	return 0;
3476 }
3477 
3478 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3479 		struct nvdimm *nvdimm, unsigned int cmd)
3480 {
3481 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3482 
3483 	if (nvdimm)
3484 		return 0;
3485 	if (cmd != ND_CMD_ARS_START)
3486 		return 0;
3487 
3488 	/*
3489 	 * The kernel and userspace may race to initiate a scrub, but
3490 	 * the scrub thread is prepared to lose that initial race.  It
3491 	 * just needs guarantees that any ARS it initiates are not
3492 	 * interrupted by any intervening start requests from userspace.
3493 	 */
3494 	if (work_busy(&acpi_desc->dwork.work))
3495 		return -EBUSY;
3496 
3497 	return 0;
3498 }
3499 
3500 /* prevent security commands from being issued via ioctl */
3501 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3502 		struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3503 {
3504 	struct nd_cmd_pkg *call_pkg = buf;
3505 	unsigned int func;
3506 
3507 	if (nvdimm && cmd == ND_CMD_CALL &&
3508 			call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3509 		func = call_pkg->nd_command;
3510 		if ((1 << func) & NVDIMM_INTEL_SECURITY_CMDMASK)
3511 			return -EOPNOTSUPP;
3512 	}
3513 
3514 	return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3515 }
3516 
3517 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3518 		enum nfit_ars_state req_type)
3519 {
3520 	struct device *dev = acpi_desc->dev;
3521 	int scheduled = 0, busy = 0;
3522 	struct nfit_spa *nfit_spa;
3523 
3524 	mutex_lock(&acpi_desc->init_mutex);
3525 	if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3526 		mutex_unlock(&acpi_desc->init_mutex);
3527 		return 0;
3528 	}
3529 
3530 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3531 		int type = nfit_spa_type(nfit_spa->spa);
3532 
3533 		if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3534 			continue;
3535 		if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3536 			continue;
3537 
3538 		if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3539 			busy++;
3540 		else
3541 			scheduled++;
3542 	}
3543 	if (scheduled) {
3544 		sched_ars(acpi_desc);
3545 		dev_dbg(dev, "ars_scan triggered\n");
3546 	}
3547 	mutex_unlock(&acpi_desc->init_mutex);
3548 
3549 	if (scheduled)
3550 		return 0;
3551 	if (busy)
3552 		return -EBUSY;
3553 	return -ENOTTY;
3554 }
3555 
3556 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3557 {
3558 	struct nvdimm_bus_descriptor *nd_desc;
3559 
3560 	dev_set_drvdata(dev, acpi_desc);
3561 	acpi_desc->dev = dev;
3562 	acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3563 	nd_desc = &acpi_desc->nd_desc;
3564 	nd_desc->provider_name = "ACPI.NFIT";
3565 	nd_desc->module = THIS_MODULE;
3566 	nd_desc->ndctl = acpi_nfit_ctl;
3567 	nd_desc->flush_probe = acpi_nfit_flush_probe;
3568 	nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3569 	nd_desc->attr_groups = acpi_nfit_attribute_groups;
3570 
3571 	INIT_LIST_HEAD(&acpi_desc->spas);
3572 	INIT_LIST_HEAD(&acpi_desc->dcrs);
3573 	INIT_LIST_HEAD(&acpi_desc->bdws);
3574 	INIT_LIST_HEAD(&acpi_desc->idts);
3575 	INIT_LIST_HEAD(&acpi_desc->flushes);
3576 	INIT_LIST_HEAD(&acpi_desc->memdevs);
3577 	INIT_LIST_HEAD(&acpi_desc->dimms);
3578 	INIT_LIST_HEAD(&acpi_desc->list);
3579 	mutex_init(&acpi_desc->init_mutex);
3580 	acpi_desc->scrub_tmo = 1;
3581 	INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3582 }
3583 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3584 
3585 static void acpi_nfit_put_table(void *table)
3586 {
3587 	acpi_put_table(table);
3588 }
3589 
3590 void acpi_nfit_shutdown(void *data)
3591 {
3592 	struct acpi_nfit_desc *acpi_desc = data;
3593 	struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3594 
3595 	/*
3596 	 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3597 	 * race teardown
3598 	 */
3599 	mutex_lock(&acpi_desc_lock);
3600 	list_del(&acpi_desc->list);
3601 	mutex_unlock(&acpi_desc_lock);
3602 
3603 	mutex_lock(&acpi_desc->init_mutex);
3604 	set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3605 	cancel_delayed_work_sync(&acpi_desc->dwork);
3606 	mutex_unlock(&acpi_desc->init_mutex);
3607 
3608 	/*
3609 	 * Bounce the nvdimm bus lock to make sure any in-flight
3610 	 * acpi_nfit_ars_rescan() submissions have had a chance to
3611 	 * either submit or see ->cancel set.
3612 	 */
3613 	device_lock(bus_dev);
3614 	device_unlock(bus_dev);
3615 
3616 	flush_workqueue(nfit_wq);
3617 }
3618 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3619 
3620 static int acpi_nfit_add(struct acpi_device *adev)
3621 {
3622 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3623 	struct acpi_nfit_desc *acpi_desc;
3624 	struct device *dev = &adev->dev;
3625 	struct acpi_table_header *tbl;
3626 	acpi_status status = AE_OK;
3627 	acpi_size sz;
3628 	int rc = 0;
3629 
3630 	status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3631 	if (ACPI_FAILURE(status)) {
3632 		/* The NVDIMM root device allows OS to trigger enumeration of
3633 		 * NVDIMMs through NFIT at boot time and re-enumeration at
3634 		 * root level via the _FIT method during runtime.
3635 		 * This is ok to return 0 here, we could have an nvdimm
3636 		 * hotplugged later and evaluate _FIT method which returns
3637 		 * data in the format of a series of NFIT Structures.
3638 		 */
3639 		dev_dbg(dev, "failed to find NFIT at startup\n");
3640 		return 0;
3641 	}
3642 
3643 	rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3644 	if (rc)
3645 		return rc;
3646 	sz = tbl->length;
3647 
3648 	acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3649 	if (!acpi_desc)
3650 		return -ENOMEM;
3651 	acpi_nfit_desc_init(acpi_desc, &adev->dev);
3652 
3653 	/* Save the acpi header for exporting the revision via sysfs */
3654 	acpi_desc->acpi_header = *tbl;
3655 
3656 	/* Evaluate _FIT and override with that if present */
3657 	status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3658 	if (ACPI_SUCCESS(status) && buf.length > 0) {
3659 		union acpi_object *obj = buf.pointer;
3660 
3661 		if (obj->type == ACPI_TYPE_BUFFER)
3662 			rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3663 					obj->buffer.length);
3664 		else
3665 			dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3666 				(int) obj->type);
3667 		kfree(buf.pointer);
3668 	} else
3669 		/* skip over the lead-in header table */
3670 		rc = acpi_nfit_init(acpi_desc, (void *) tbl
3671 				+ sizeof(struct acpi_table_nfit),
3672 				sz - sizeof(struct acpi_table_nfit));
3673 
3674 	if (rc)
3675 		return rc;
3676 	return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3677 }
3678 
3679 static int acpi_nfit_remove(struct acpi_device *adev)
3680 {
3681 	/* see acpi_nfit_unregister */
3682 	return 0;
3683 }
3684 
3685 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3686 {
3687 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3688 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3689 	union acpi_object *obj;
3690 	acpi_status status;
3691 	int ret;
3692 
3693 	if (!dev->driver) {
3694 		/* dev->driver may be null if we're being removed */
3695 		dev_dbg(dev, "no driver found for dev\n");
3696 		return;
3697 	}
3698 
3699 	if (!acpi_desc) {
3700 		acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3701 		if (!acpi_desc)
3702 			return;
3703 		acpi_nfit_desc_init(acpi_desc, dev);
3704 	} else {
3705 		/*
3706 		 * Finish previous registration before considering new
3707 		 * regions.
3708 		 */
3709 		flush_workqueue(nfit_wq);
3710 	}
3711 
3712 	/* Evaluate _FIT */
3713 	status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3714 	if (ACPI_FAILURE(status)) {
3715 		dev_err(dev, "failed to evaluate _FIT\n");
3716 		return;
3717 	}
3718 
3719 	obj = buf.pointer;
3720 	if (obj->type == ACPI_TYPE_BUFFER) {
3721 		ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3722 				obj->buffer.length);
3723 		if (ret)
3724 			dev_err(dev, "failed to merge updated NFIT\n");
3725 	} else
3726 		dev_err(dev, "Invalid _FIT\n");
3727 	kfree(buf.pointer);
3728 }
3729 
3730 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3731 {
3732 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3733 
3734 	if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3735 		acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3736 	else
3737 		acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3738 }
3739 
3740 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3741 {
3742 	dev_dbg(dev, "event: 0x%x\n", event);
3743 
3744 	switch (event) {
3745 	case NFIT_NOTIFY_UPDATE:
3746 		return acpi_nfit_update_notify(dev, handle);
3747 	case NFIT_NOTIFY_UC_MEMORY_ERROR:
3748 		return acpi_nfit_uc_error_notify(dev, handle);
3749 	default:
3750 		return;
3751 	}
3752 }
3753 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3754 
3755 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3756 {
3757 	device_lock(&adev->dev);
3758 	__acpi_nfit_notify(&adev->dev, adev->handle, event);
3759 	device_unlock(&adev->dev);
3760 }
3761 
3762 static const struct acpi_device_id acpi_nfit_ids[] = {
3763 	{ "ACPI0012", 0 },
3764 	{ "", 0 },
3765 };
3766 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3767 
3768 static struct acpi_driver acpi_nfit_driver = {
3769 	.name = KBUILD_MODNAME,
3770 	.ids = acpi_nfit_ids,
3771 	.ops = {
3772 		.add = acpi_nfit_add,
3773 		.remove = acpi_nfit_remove,
3774 		.notify = acpi_nfit_notify,
3775 	},
3776 };
3777 
3778 static __init int nfit_init(void)
3779 {
3780 	int ret;
3781 
3782 	BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3783 	BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3784 	BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3785 	BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3786 	BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3787 	BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3788 	BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3789 	BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3790 
3791 	guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3792 	guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3793 	guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3794 	guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3795 	guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3796 	guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3797 	guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3798 	guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3799 	guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3800 	guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3801 	guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3802 	guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3803 	guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3804 	guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3805 
3806 	nfit_wq = create_singlethread_workqueue("nfit");
3807 	if (!nfit_wq)
3808 		return -ENOMEM;
3809 
3810 	nfit_mce_register();
3811 	ret = acpi_bus_register_driver(&acpi_nfit_driver);
3812 	if (ret) {
3813 		nfit_mce_unregister();
3814 		destroy_workqueue(nfit_wq);
3815 	}
3816 
3817 	return ret;
3818 
3819 }
3820 
3821 static __exit void nfit_exit(void)
3822 {
3823 	nfit_mce_unregister();
3824 	acpi_bus_unregister_driver(&acpi_nfit_driver);
3825 	destroy_workqueue(nfit_wq);
3826 	WARN_ON(!list_empty(&acpi_descs));
3827 }
3828 
3829 module_init(nfit_init);
3830 module_exit(nfit_exit);
3831 MODULE_LICENSE("GPL v2");
3832 MODULE_AUTHOR("Intel Corporation");
3833