xref: /linux/drivers/acpi/nfit/core.c (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4  */
5 #include <linux/list_sort.h>
6 #include <linux/libnvdimm.h>
7 #include <linux/module.h>
8 #include <linux/mutex.h>
9 #include <linux/ndctl.h>
10 #include <linux/sysfs.h>
11 #include <linux/delay.h>
12 #include <linux/list.h>
13 #include <linux/acpi.h>
14 #include <linux/sort.h>
15 #include <linux/io.h>
16 #include <linux/nd.h>
17 #include <asm/cacheflush.h>
18 #include <acpi/nfit.h>
19 #include "intel.h"
20 #include "nfit.h"
21 
22 /*
23  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
24  * irrelevant.
25  */
26 #include <linux/io-64-nonatomic-hi-lo.h>
27 
28 static bool force_enable_dimms;
29 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
30 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
31 
32 static bool disable_vendor_specific;
33 module_param(disable_vendor_specific, bool, S_IRUGO);
34 MODULE_PARM_DESC(disable_vendor_specific,
35 		"Limit commands to the publicly specified set");
36 
37 static unsigned long override_dsm_mask;
38 module_param(override_dsm_mask, ulong, S_IRUGO);
39 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
40 
41 static int default_dsm_family = -1;
42 module_param(default_dsm_family, int, S_IRUGO);
43 MODULE_PARM_DESC(default_dsm_family,
44 		"Try this DSM type first when identifying NVDIMM family");
45 
46 static bool no_init_ars;
47 module_param(no_init_ars, bool, 0644);
48 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
49 
50 static bool force_labels;
51 module_param(force_labels, bool, 0444);
52 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
53 
54 LIST_HEAD(acpi_descs);
55 DEFINE_MUTEX(acpi_desc_lock);
56 
57 static struct workqueue_struct *nfit_wq;
58 
59 struct nfit_table_prev {
60 	struct list_head spas;
61 	struct list_head memdevs;
62 	struct list_head dcrs;
63 	struct list_head bdws;
64 	struct list_head idts;
65 	struct list_head flushes;
66 };
67 
68 static guid_t nfit_uuid[NFIT_UUID_MAX];
69 
70 const guid_t *to_nfit_uuid(enum nfit_uuids id)
71 {
72 	return &nfit_uuid[id];
73 }
74 EXPORT_SYMBOL(to_nfit_uuid);
75 
76 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
77 {
78 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
79 
80 	/*
81 	 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
82 	 * acpi_device.
83 	 */
84 	if (!nd_desc->provider_name
85 			|| strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
86 		return NULL;
87 
88 	return to_acpi_device(acpi_desc->dev);
89 }
90 
91 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
92 {
93 	struct nd_cmd_clear_error *clear_err;
94 	struct nd_cmd_ars_status *ars_status;
95 	u16 flags;
96 
97 	switch (cmd) {
98 	case ND_CMD_ARS_CAP:
99 		if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
100 			return -ENOTTY;
101 
102 		/* Command failed */
103 		if (status & 0xffff)
104 			return -EIO;
105 
106 		/* No supported scan types for this range */
107 		flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
108 		if ((status >> 16 & flags) == 0)
109 			return -ENOTTY;
110 		return 0;
111 	case ND_CMD_ARS_START:
112 		/* ARS is in progress */
113 		if ((status & 0xffff) == NFIT_ARS_START_BUSY)
114 			return -EBUSY;
115 
116 		/* Command failed */
117 		if (status & 0xffff)
118 			return -EIO;
119 		return 0;
120 	case ND_CMD_ARS_STATUS:
121 		ars_status = buf;
122 		/* Command failed */
123 		if (status & 0xffff)
124 			return -EIO;
125 		/* Check extended status (Upper two bytes) */
126 		if (status == NFIT_ARS_STATUS_DONE)
127 			return 0;
128 
129 		/* ARS is in progress */
130 		if (status == NFIT_ARS_STATUS_BUSY)
131 			return -EBUSY;
132 
133 		/* No ARS performed for the current boot */
134 		if (status == NFIT_ARS_STATUS_NONE)
135 			return -EAGAIN;
136 
137 		/*
138 		 * ARS interrupted, either we overflowed or some other
139 		 * agent wants the scan to stop.  If we didn't overflow
140 		 * then just continue with the returned results.
141 		 */
142 		if (status == NFIT_ARS_STATUS_INTR) {
143 			if (ars_status->out_length >= 40 && (ars_status->flags
144 						& NFIT_ARS_F_OVERFLOW))
145 				return -ENOSPC;
146 			return 0;
147 		}
148 
149 		/* Unknown status */
150 		if (status >> 16)
151 			return -EIO;
152 		return 0;
153 	case ND_CMD_CLEAR_ERROR:
154 		clear_err = buf;
155 		if (status & 0xffff)
156 			return -EIO;
157 		if (!clear_err->cleared)
158 			return -EIO;
159 		if (clear_err->length > clear_err->cleared)
160 			return clear_err->cleared;
161 		return 0;
162 	default:
163 		break;
164 	}
165 
166 	/* all other non-zero status results in an error */
167 	if (status)
168 		return -EIO;
169 	return 0;
170 }
171 
172 #define ACPI_LABELS_LOCKED 3
173 
174 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
175 		u32 status)
176 {
177 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
178 
179 	switch (cmd) {
180 	case ND_CMD_GET_CONFIG_SIZE:
181 		/*
182 		 * In the _LSI, _LSR, _LSW case the locked status is
183 		 * communicated via the read/write commands
184 		 */
185 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
186 			break;
187 
188 		if (status >> 16 & ND_CONFIG_LOCKED)
189 			return -EACCES;
190 		break;
191 	case ND_CMD_GET_CONFIG_DATA:
192 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
193 				&& status == ACPI_LABELS_LOCKED)
194 			return -EACCES;
195 		break;
196 	case ND_CMD_SET_CONFIG_DATA:
197 		if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
198 				&& status == ACPI_LABELS_LOCKED)
199 			return -EACCES;
200 		break;
201 	default:
202 		break;
203 	}
204 
205 	/* all other non-zero status results in an error */
206 	if (status)
207 		return -EIO;
208 	return 0;
209 }
210 
211 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
212 		u32 status)
213 {
214 	if (!nvdimm)
215 		return xlat_bus_status(buf, cmd, status);
216 	return xlat_nvdimm_status(nvdimm, buf, cmd, status);
217 }
218 
219 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
220 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
221 {
222 	int i;
223 	void *dst;
224 	size_t size = 0;
225 	union acpi_object *buf = NULL;
226 
227 	if (pkg->type != ACPI_TYPE_PACKAGE) {
228 		WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
229 				pkg->type);
230 		goto err;
231 	}
232 
233 	for (i = 0; i < pkg->package.count; i++) {
234 		union acpi_object *obj = &pkg->package.elements[i];
235 
236 		if (obj->type == ACPI_TYPE_INTEGER)
237 			size += 4;
238 		else if (obj->type == ACPI_TYPE_BUFFER)
239 			size += obj->buffer.length;
240 		else {
241 			WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
242 					obj->type);
243 			goto err;
244 		}
245 	}
246 
247 	buf = ACPI_ALLOCATE(sizeof(*buf) + size);
248 	if (!buf)
249 		goto err;
250 
251 	dst = buf + 1;
252 	buf->type = ACPI_TYPE_BUFFER;
253 	buf->buffer.length = size;
254 	buf->buffer.pointer = dst;
255 	for (i = 0; i < pkg->package.count; i++) {
256 		union acpi_object *obj = &pkg->package.elements[i];
257 
258 		if (obj->type == ACPI_TYPE_INTEGER) {
259 			memcpy(dst, &obj->integer.value, 4);
260 			dst += 4;
261 		} else if (obj->type == ACPI_TYPE_BUFFER) {
262 			memcpy(dst, obj->buffer.pointer, obj->buffer.length);
263 			dst += obj->buffer.length;
264 		}
265 	}
266 err:
267 	ACPI_FREE(pkg);
268 	return buf;
269 }
270 
271 static union acpi_object *int_to_buf(union acpi_object *integer)
272 {
273 	union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
274 	void *dst = NULL;
275 
276 	if (!buf)
277 		goto err;
278 
279 	if (integer->type != ACPI_TYPE_INTEGER) {
280 		WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
281 				integer->type);
282 		goto err;
283 	}
284 
285 	dst = buf + 1;
286 	buf->type = ACPI_TYPE_BUFFER;
287 	buf->buffer.length = 4;
288 	buf->buffer.pointer = dst;
289 	memcpy(dst, &integer->integer.value, 4);
290 err:
291 	ACPI_FREE(integer);
292 	return buf;
293 }
294 
295 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
296 		u32 len, void *data)
297 {
298 	acpi_status rc;
299 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
300 	struct acpi_object_list input = {
301 		.count = 3,
302 		.pointer = (union acpi_object []) {
303 			[0] = {
304 				.integer.type = ACPI_TYPE_INTEGER,
305 				.integer.value = offset,
306 			},
307 			[1] = {
308 				.integer.type = ACPI_TYPE_INTEGER,
309 				.integer.value = len,
310 			},
311 			[2] = {
312 				.buffer.type = ACPI_TYPE_BUFFER,
313 				.buffer.pointer = data,
314 				.buffer.length = len,
315 			},
316 		},
317 	};
318 
319 	rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
320 	if (ACPI_FAILURE(rc))
321 		return NULL;
322 	return int_to_buf(buf.pointer);
323 }
324 
325 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
326 		u32 len)
327 {
328 	acpi_status rc;
329 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
330 	struct acpi_object_list input = {
331 		.count = 2,
332 		.pointer = (union acpi_object []) {
333 			[0] = {
334 				.integer.type = ACPI_TYPE_INTEGER,
335 				.integer.value = offset,
336 			},
337 			[1] = {
338 				.integer.type = ACPI_TYPE_INTEGER,
339 				.integer.value = len,
340 			},
341 		},
342 	};
343 
344 	rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
345 	if (ACPI_FAILURE(rc))
346 		return NULL;
347 	return pkg_to_buf(buf.pointer);
348 }
349 
350 static union acpi_object *acpi_label_info(acpi_handle handle)
351 {
352 	acpi_status rc;
353 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
354 
355 	rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
356 	if (ACPI_FAILURE(rc))
357 		return NULL;
358 	return pkg_to_buf(buf.pointer);
359 }
360 
361 static u8 nfit_dsm_revid(unsigned family, unsigned func)
362 {
363 	static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
364 		[NVDIMM_FAMILY_INTEL] = {
365 			[NVDIMM_INTEL_GET_MODES] = 2,
366 			[NVDIMM_INTEL_GET_FWINFO] = 2,
367 			[NVDIMM_INTEL_START_FWUPDATE] = 2,
368 			[NVDIMM_INTEL_SEND_FWUPDATE] = 2,
369 			[NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
370 			[NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
371 			[NVDIMM_INTEL_SET_THRESHOLD] = 2,
372 			[NVDIMM_INTEL_INJECT_ERROR] = 2,
373 			[NVDIMM_INTEL_GET_SECURITY_STATE] = 2,
374 			[NVDIMM_INTEL_SET_PASSPHRASE] = 2,
375 			[NVDIMM_INTEL_DISABLE_PASSPHRASE] = 2,
376 			[NVDIMM_INTEL_UNLOCK_UNIT] = 2,
377 			[NVDIMM_INTEL_FREEZE_LOCK] = 2,
378 			[NVDIMM_INTEL_SECURE_ERASE] = 2,
379 			[NVDIMM_INTEL_OVERWRITE] = 2,
380 			[NVDIMM_INTEL_QUERY_OVERWRITE] = 2,
381 			[NVDIMM_INTEL_SET_MASTER_PASSPHRASE] = 2,
382 			[NVDIMM_INTEL_MASTER_SECURE_ERASE] = 2,
383 		},
384 	};
385 	u8 id;
386 
387 	if (family > NVDIMM_FAMILY_MAX)
388 		return 0;
389 	if (func > 31)
390 		return 0;
391 	id = revid_table[family][func];
392 	if (id == 0)
393 		return 1; /* default */
394 	return id;
395 }
396 
397 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
398 {
399 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
400 
401 	if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
402 			&& func >= NVDIMM_INTEL_GET_SECURITY_STATE
403 			&& func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
404 		return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
405 	return true;
406 }
407 
408 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
409 		struct nd_cmd_pkg *call_pkg)
410 {
411 	if (call_pkg) {
412 		int i;
413 
414 		if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
415 			return -ENOTTY;
416 
417 		for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
418 			if (call_pkg->nd_reserved2[i])
419 				return -EINVAL;
420 		return call_pkg->nd_command;
421 	}
422 
423 	/* In the !call_pkg case, bus commands == bus functions */
424 	if (!nfit_mem)
425 		return cmd;
426 
427 	/* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
428 	if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
429 		return cmd;
430 
431 	/*
432 	 * Force function number validation to fail since 0 is never
433 	 * published as a valid function in dsm_mask.
434 	 */
435 	return 0;
436 }
437 
438 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
439 		unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
440 {
441 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
442 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
443 	union acpi_object in_obj, in_buf, *out_obj;
444 	const struct nd_cmd_desc *desc = NULL;
445 	struct device *dev = acpi_desc->dev;
446 	struct nd_cmd_pkg *call_pkg = NULL;
447 	const char *cmd_name, *dimm_name;
448 	unsigned long cmd_mask, dsm_mask;
449 	u32 offset, fw_status = 0;
450 	acpi_handle handle;
451 	const guid_t *guid;
452 	int func, rc, i;
453 
454 	if (cmd_rc)
455 		*cmd_rc = -EINVAL;
456 
457 	if (cmd == ND_CMD_CALL)
458 		call_pkg = buf;
459 	func = cmd_to_func(nfit_mem, cmd, call_pkg);
460 	if (func < 0)
461 		return func;
462 
463 	if (nvdimm) {
464 		struct acpi_device *adev = nfit_mem->adev;
465 
466 		if (!adev)
467 			return -ENOTTY;
468 
469 		dimm_name = nvdimm_name(nvdimm);
470 		cmd_name = nvdimm_cmd_name(cmd);
471 		cmd_mask = nvdimm_cmd_mask(nvdimm);
472 		dsm_mask = nfit_mem->dsm_mask;
473 		desc = nd_cmd_dimm_desc(cmd);
474 		guid = to_nfit_uuid(nfit_mem->family);
475 		handle = adev->handle;
476 	} else {
477 		struct acpi_device *adev = to_acpi_dev(acpi_desc);
478 
479 		cmd_name = nvdimm_bus_cmd_name(cmd);
480 		cmd_mask = nd_desc->cmd_mask;
481 		dsm_mask = nd_desc->bus_dsm_mask;
482 		desc = nd_cmd_bus_desc(cmd);
483 		guid = to_nfit_uuid(NFIT_DEV_BUS);
484 		handle = adev->handle;
485 		dimm_name = "bus";
486 	}
487 
488 	if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
489 		return -ENOTTY;
490 
491 	/*
492 	 * Check for a valid command.  For ND_CMD_CALL, we also have to
493 	 * make sure that the DSM function is supported.
494 	 */
495 	if (cmd == ND_CMD_CALL && !test_bit(func, &dsm_mask))
496 		return -ENOTTY;
497 	else if (!test_bit(cmd, &cmd_mask))
498 		return -ENOTTY;
499 
500 	in_obj.type = ACPI_TYPE_PACKAGE;
501 	in_obj.package.count = 1;
502 	in_obj.package.elements = &in_buf;
503 	in_buf.type = ACPI_TYPE_BUFFER;
504 	in_buf.buffer.pointer = buf;
505 	in_buf.buffer.length = 0;
506 
507 	/* libnvdimm has already validated the input envelope */
508 	for (i = 0; i < desc->in_num; i++)
509 		in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
510 				i, buf);
511 
512 	if (call_pkg) {
513 		/* skip over package wrapper */
514 		in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
515 		in_buf.buffer.length = call_pkg->nd_size_in;
516 	}
517 
518 	dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
519 		dimm_name, cmd, func, in_buf.buffer.length);
520 	if (payload_dumpable(nvdimm, func))
521 		print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
522 				in_buf.buffer.pointer,
523 				min_t(u32, 256, in_buf.buffer.length), true);
524 
525 	/* call the BIOS, prefer the named methods over _DSM if available */
526 	if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
527 			&& test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
528 		out_obj = acpi_label_info(handle);
529 	else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
530 			&& test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
531 		struct nd_cmd_get_config_data_hdr *p = buf;
532 
533 		out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
534 	} else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
535 			&& test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
536 		struct nd_cmd_set_config_hdr *p = buf;
537 
538 		out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
539 				p->in_buf);
540 	} else {
541 		u8 revid;
542 
543 		if (nvdimm)
544 			revid = nfit_dsm_revid(nfit_mem->family, func);
545 		else
546 			revid = 1;
547 		out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
548 	}
549 
550 	if (!out_obj) {
551 		dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
552 		return -EINVAL;
553 	}
554 
555 	if (out_obj->type != ACPI_TYPE_BUFFER) {
556 		dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
557 				dimm_name, cmd_name, out_obj->type);
558 		rc = -EINVAL;
559 		goto out;
560 	}
561 
562 	dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
563 			cmd_name, out_obj->buffer.length);
564 	print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
565 			out_obj->buffer.pointer,
566 			min_t(u32, 128, out_obj->buffer.length), true);
567 
568 	if (call_pkg) {
569 		call_pkg->nd_fw_size = out_obj->buffer.length;
570 		memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
571 			out_obj->buffer.pointer,
572 			min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
573 
574 		ACPI_FREE(out_obj);
575 		/*
576 		 * Need to support FW function w/o known size in advance.
577 		 * Caller can determine required size based upon nd_fw_size.
578 		 * If we return an error (like elsewhere) then caller wouldn't
579 		 * be able to rely upon data returned to make calculation.
580 		 */
581 		if (cmd_rc)
582 			*cmd_rc = 0;
583 		return 0;
584 	}
585 
586 	for (i = 0, offset = 0; i < desc->out_num; i++) {
587 		u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
588 				(u32 *) out_obj->buffer.pointer,
589 				out_obj->buffer.length - offset);
590 
591 		if (offset + out_size > out_obj->buffer.length) {
592 			dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
593 					dimm_name, cmd_name, i);
594 			break;
595 		}
596 
597 		if (in_buf.buffer.length + offset + out_size > buf_len) {
598 			dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
599 					dimm_name, cmd_name, i);
600 			rc = -ENXIO;
601 			goto out;
602 		}
603 		memcpy(buf + in_buf.buffer.length + offset,
604 				out_obj->buffer.pointer + offset, out_size);
605 		offset += out_size;
606 	}
607 
608 	/*
609 	 * Set fw_status for all the commands with a known format to be
610 	 * later interpreted by xlat_status().
611 	 */
612 	if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
613 					&& cmd <= ND_CMD_CLEAR_ERROR)
614 				|| (nvdimm && cmd >= ND_CMD_SMART
615 					&& cmd <= ND_CMD_VENDOR)))
616 		fw_status = *(u32 *) out_obj->buffer.pointer;
617 
618 	if (offset + in_buf.buffer.length < buf_len) {
619 		if (i >= 1) {
620 			/*
621 			 * status valid, return the number of bytes left
622 			 * unfilled in the output buffer
623 			 */
624 			rc = buf_len - offset - in_buf.buffer.length;
625 			if (cmd_rc)
626 				*cmd_rc = xlat_status(nvdimm, buf, cmd,
627 						fw_status);
628 		} else {
629 			dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
630 					__func__, dimm_name, cmd_name, buf_len,
631 					offset);
632 			rc = -ENXIO;
633 		}
634 	} else {
635 		rc = 0;
636 		if (cmd_rc)
637 			*cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
638 	}
639 
640  out:
641 	ACPI_FREE(out_obj);
642 
643 	return rc;
644 }
645 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
646 
647 static const char *spa_type_name(u16 type)
648 {
649 	static const char *to_name[] = {
650 		[NFIT_SPA_VOLATILE] = "volatile",
651 		[NFIT_SPA_PM] = "pmem",
652 		[NFIT_SPA_DCR] = "dimm-control-region",
653 		[NFIT_SPA_BDW] = "block-data-window",
654 		[NFIT_SPA_VDISK] = "volatile-disk",
655 		[NFIT_SPA_VCD] = "volatile-cd",
656 		[NFIT_SPA_PDISK] = "persistent-disk",
657 		[NFIT_SPA_PCD] = "persistent-cd",
658 
659 	};
660 
661 	if (type > NFIT_SPA_PCD)
662 		return "unknown";
663 
664 	return to_name[type];
665 }
666 
667 int nfit_spa_type(struct acpi_nfit_system_address *spa)
668 {
669 	int i;
670 
671 	for (i = 0; i < NFIT_UUID_MAX; i++)
672 		if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
673 			return i;
674 	return -1;
675 }
676 
677 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
678 		struct nfit_table_prev *prev,
679 		struct acpi_nfit_system_address *spa)
680 {
681 	struct device *dev = acpi_desc->dev;
682 	struct nfit_spa *nfit_spa;
683 
684 	if (spa->header.length != sizeof(*spa))
685 		return false;
686 
687 	list_for_each_entry(nfit_spa, &prev->spas, list) {
688 		if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
689 			list_move_tail(&nfit_spa->list, &acpi_desc->spas);
690 			return true;
691 		}
692 	}
693 
694 	nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
695 			GFP_KERNEL);
696 	if (!nfit_spa)
697 		return false;
698 	INIT_LIST_HEAD(&nfit_spa->list);
699 	memcpy(nfit_spa->spa, spa, sizeof(*spa));
700 	list_add_tail(&nfit_spa->list, &acpi_desc->spas);
701 	dev_dbg(dev, "spa index: %d type: %s\n",
702 			spa->range_index,
703 			spa_type_name(nfit_spa_type(spa)));
704 	return true;
705 }
706 
707 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
708 		struct nfit_table_prev *prev,
709 		struct acpi_nfit_memory_map *memdev)
710 {
711 	struct device *dev = acpi_desc->dev;
712 	struct nfit_memdev *nfit_memdev;
713 
714 	if (memdev->header.length != sizeof(*memdev))
715 		return false;
716 
717 	list_for_each_entry(nfit_memdev, &prev->memdevs, list)
718 		if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
719 			list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
720 			return true;
721 		}
722 
723 	nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
724 			GFP_KERNEL);
725 	if (!nfit_memdev)
726 		return false;
727 	INIT_LIST_HEAD(&nfit_memdev->list);
728 	memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
729 	list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
730 	dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
731 			memdev->device_handle, memdev->range_index,
732 			memdev->region_index, memdev->flags);
733 	return true;
734 }
735 
736 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
737 {
738 	struct acpi_nfit_memory_map *memdev;
739 	struct acpi_nfit_desc *acpi_desc;
740 	struct nfit_mem *nfit_mem;
741 	u16 physical_id;
742 
743 	mutex_lock(&acpi_desc_lock);
744 	list_for_each_entry(acpi_desc, &acpi_descs, list) {
745 		mutex_lock(&acpi_desc->init_mutex);
746 		list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
747 			memdev = __to_nfit_memdev(nfit_mem);
748 			if (memdev->device_handle == device_handle) {
749 				*flags = memdev->flags;
750 				physical_id = memdev->physical_id;
751 				mutex_unlock(&acpi_desc->init_mutex);
752 				mutex_unlock(&acpi_desc_lock);
753 				return physical_id;
754 			}
755 		}
756 		mutex_unlock(&acpi_desc->init_mutex);
757 	}
758 	mutex_unlock(&acpi_desc_lock);
759 
760 	return -ENODEV;
761 }
762 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
763 
764 /*
765  * An implementation may provide a truncated control region if no block windows
766  * are defined.
767  */
768 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
769 {
770 	if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
771 				window_size))
772 		return 0;
773 	if (dcr->windows)
774 		return sizeof(*dcr);
775 	return offsetof(struct acpi_nfit_control_region, window_size);
776 }
777 
778 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
779 		struct nfit_table_prev *prev,
780 		struct acpi_nfit_control_region *dcr)
781 {
782 	struct device *dev = acpi_desc->dev;
783 	struct nfit_dcr *nfit_dcr;
784 
785 	if (!sizeof_dcr(dcr))
786 		return false;
787 
788 	list_for_each_entry(nfit_dcr, &prev->dcrs, list)
789 		if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
790 			list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
791 			return true;
792 		}
793 
794 	nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
795 			GFP_KERNEL);
796 	if (!nfit_dcr)
797 		return false;
798 	INIT_LIST_HEAD(&nfit_dcr->list);
799 	memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
800 	list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
801 	dev_dbg(dev, "dcr index: %d windows: %d\n",
802 			dcr->region_index, dcr->windows);
803 	return true;
804 }
805 
806 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
807 		struct nfit_table_prev *prev,
808 		struct acpi_nfit_data_region *bdw)
809 {
810 	struct device *dev = acpi_desc->dev;
811 	struct nfit_bdw *nfit_bdw;
812 
813 	if (bdw->header.length != sizeof(*bdw))
814 		return false;
815 	list_for_each_entry(nfit_bdw, &prev->bdws, list)
816 		if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
817 			list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
818 			return true;
819 		}
820 
821 	nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
822 			GFP_KERNEL);
823 	if (!nfit_bdw)
824 		return false;
825 	INIT_LIST_HEAD(&nfit_bdw->list);
826 	memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
827 	list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
828 	dev_dbg(dev, "bdw dcr: %d windows: %d\n",
829 			bdw->region_index, bdw->windows);
830 	return true;
831 }
832 
833 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
834 {
835 	if (idt->header.length < sizeof(*idt))
836 		return 0;
837 	return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
838 }
839 
840 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
841 		struct nfit_table_prev *prev,
842 		struct acpi_nfit_interleave *idt)
843 {
844 	struct device *dev = acpi_desc->dev;
845 	struct nfit_idt *nfit_idt;
846 
847 	if (!sizeof_idt(idt))
848 		return false;
849 
850 	list_for_each_entry(nfit_idt, &prev->idts, list) {
851 		if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
852 			continue;
853 
854 		if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
855 			list_move_tail(&nfit_idt->list, &acpi_desc->idts);
856 			return true;
857 		}
858 	}
859 
860 	nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
861 			GFP_KERNEL);
862 	if (!nfit_idt)
863 		return false;
864 	INIT_LIST_HEAD(&nfit_idt->list);
865 	memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
866 	list_add_tail(&nfit_idt->list, &acpi_desc->idts);
867 	dev_dbg(dev, "idt index: %d num_lines: %d\n",
868 			idt->interleave_index, idt->line_count);
869 	return true;
870 }
871 
872 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
873 {
874 	if (flush->header.length < sizeof(*flush))
875 		return 0;
876 	return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
877 }
878 
879 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
880 		struct nfit_table_prev *prev,
881 		struct acpi_nfit_flush_address *flush)
882 {
883 	struct device *dev = acpi_desc->dev;
884 	struct nfit_flush *nfit_flush;
885 
886 	if (!sizeof_flush(flush))
887 		return false;
888 
889 	list_for_each_entry(nfit_flush, &prev->flushes, list) {
890 		if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
891 			continue;
892 
893 		if (memcmp(nfit_flush->flush, flush,
894 					sizeof_flush(flush)) == 0) {
895 			list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
896 			return true;
897 		}
898 	}
899 
900 	nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
901 			+ sizeof_flush(flush), GFP_KERNEL);
902 	if (!nfit_flush)
903 		return false;
904 	INIT_LIST_HEAD(&nfit_flush->list);
905 	memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
906 	list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
907 	dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
908 			flush->device_handle, flush->hint_count);
909 	return true;
910 }
911 
912 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
913 		struct acpi_nfit_capabilities *pcap)
914 {
915 	struct device *dev = acpi_desc->dev;
916 	u32 mask;
917 
918 	mask = (1 << (pcap->highest_capability + 1)) - 1;
919 	acpi_desc->platform_cap = pcap->capabilities & mask;
920 	dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
921 	return true;
922 }
923 
924 static void *add_table(struct acpi_nfit_desc *acpi_desc,
925 		struct nfit_table_prev *prev, void *table, const void *end)
926 {
927 	struct device *dev = acpi_desc->dev;
928 	struct acpi_nfit_header *hdr;
929 	void *err = ERR_PTR(-ENOMEM);
930 
931 	if (table >= end)
932 		return NULL;
933 
934 	hdr = table;
935 	if (!hdr->length) {
936 		dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
937 			hdr->type);
938 		return NULL;
939 	}
940 
941 	switch (hdr->type) {
942 	case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
943 		if (!add_spa(acpi_desc, prev, table))
944 			return err;
945 		break;
946 	case ACPI_NFIT_TYPE_MEMORY_MAP:
947 		if (!add_memdev(acpi_desc, prev, table))
948 			return err;
949 		break;
950 	case ACPI_NFIT_TYPE_CONTROL_REGION:
951 		if (!add_dcr(acpi_desc, prev, table))
952 			return err;
953 		break;
954 	case ACPI_NFIT_TYPE_DATA_REGION:
955 		if (!add_bdw(acpi_desc, prev, table))
956 			return err;
957 		break;
958 	case ACPI_NFIT_TYPE_INTERLEAVE:
959 		if (!add_idt(acpi_desc, prev, table))
960 			return err;
961 		break;
962 	case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
963 		if (!add_flush(acpi_desc, prev, table))
964 			return err;
965 		break;
966 	case ACPI_NFIT_TYPE_SMBIOS:
967 		dev_dbg(dev, "smbios\n");
968 		break;
969 	case ACPI_NFIT_TYPE_CAPABILITIES:
970 		if (!add_platform_cap(acpi_desc, table))
971 			return err;
972 		break;
973 	default:
974 		dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
975 		break;
976 	}
977 
978 	return table + hdr->length;
979 }
980 
981 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
982 		struct nfit_mem *nfit_mem)
983 {
984 	u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
985 	u16 dcr = nfit_mem->dcr->region_index;
986 	struct nfit_spa *nfit_spa;
987 
988 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
989 		u16 range_index = nfit_spa->spa->range_index;
990 		int type = nfit_spa_type(nfit_spa->spa);
991 		struct nfit_memdev *nfit_memdev;
992 
993 		if (type != NFIT_SPA_BDW)
994 			continue;
995 
996 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
997 			if (nfit_memdev->memdev->range_index != range_index)
998 				continue;
999 			if (nfit_memdev->memdev->device_handle != device_handle)
1000 				continue;
1001 			if (nfit_memdev->memdev->region_index != dcr)
1002 				continue;
1003 
1004 			nfit_mem->spa_bdw = nfit_spa->spa;
1005 			return;
1006 		}
1007 	}
1008 
1009 	dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
1010 			nfit_mem->spa_dcr->range_index);
1011 	nfit_mem->bdw = NULL;
1012 }
1013 
1014 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
1015 		struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
1016 {
1017 	u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
1018 	struct nfit_memdev *nfit_memdev;
1019 	struct nfit_bdw *nfit_bdw;
1020 	struct nfit_idt *nfit_idt;
1021 	u16 idt_idx, range_index;
1022 
1023 	list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
1024 		if (nfit_bdw->bdw->region_index != dcr)
1025 			continue;
1026 		nfit_mem->bdw = nfit_bdw->bdw;
1027 		break;
1028 	}
1029 
1030 	if (!nfit_mem->bdw)
1031 		return;
1032 
1033 	nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
1034 
1035 	if (!nfit_mem->spa_bdw)
1036 		return;
1037 
1038 	range_index = nfit_mem->spa_bdw->range_index;
1039 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1040 		if (nfit_memdev->memdev->range_index != range_index ||
1041 				nfit_memdev->memdev->region_index != dcr)
1042 			continue;
1043 		nfit_mem->memdev_bdw = nfit_memdev->memdev;
1044 		idt_idx = nfit_memdev->memdev->interleave_index;
1045 		list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1046 			if (nfit_idt->idt->interleave_index != idt_idx)
1047 				continue;
1048 			nfit_mem->idt_bdw = nfit_idt->idt;
1049 			break;
1050 		}
1051 		break;
1052 	}
1053 }
1054 
1055 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1056 		struct acpi_nfit_system_address *spa)
1057 {
1058 	struct nfit_mem *nfit_mem, *found;
1059 	struct nfit_memdev *nfit_memdev;
1060 	int type = spa ? nfit_spa_type(spa) : 0;
1061 
1062 	switch (type) {
1063 	case NFIT_SPA_DCR:
1064 	case NFIT_SPA_PM:
1065 		break;
1066 	default:
1067 		if (spa)
1068 			return 0;
1069 	}
1070 
1071 	/*
1072 	 * This loop runs in two modes, when a dimm is mapped the loop
1073 	 * adds memdev associations to an existing dimm, or creates a
1074 	 * dimm. In the unmapped dimm case this loop sweeps for memdev
1075 	 * instances with an invalid / zero range_index and adds those
1076 	 * dimms without spa associations.
1077 	 */
1078 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1079 		struct nfit_flush *nfit_flush;
1080 		struct nfit_dcr *nfit_dcr;
1081 		u32 device_handle;
1082 		u16 dcr;
1083 
1084 		if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1085 			continue;
1086 		if (!spa && nfit_memdev->memdev->range_index)
1087 			continue;
1088 		found = NULL;
1089 		dcr = nfit_memdev->memdev->region_index;
1090 		device_handle = nfit_memdev->memdev->device_handle;
1091 		list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1092 			if (__to_nfit_memdev(nfit_mem)->device_handle
1093 					== device_handle) {
1094 				found = nfit_mem;
1095 				break;
1096 			}
1097 
1098 		if (found)
1099 			nfit_mem = found;
1100 		else {
1101 			nfit_mem = devm_kzalloc(acpi_desc->dev,
1102 					sizeof(*nfit_mem), GFP_KERNEL);
1103 			if (!nfit_mem)
1104 				return -ENOMEM;
1105 			INIT_LIST_HEAD(&nfit_mem->list);
1106 			nfit_mem->acpi_desc = acpi_desc;
1107 			list_add(&nfit_mem->list, &acpi_desc->dimms);
1108 		}
1109 
1110 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1111 			if (nfit_dcr->dcr->region_index != dcr)
1112 				continue;
1113 			/*
1114 			 * Record the control region for the dimm.  For
1115 			 * the ACPI 6.1 case, where there are separate
1116 			 * control regions for the pmem vs blk
1117 			 * interfaces, be sure to record the extended
1118 			 * blk details.
1119 			 */
1120 			if (!nfit_mem->dcr)
1121 				nfit_mem->dcr = nfit_dcr->dcr;
1122 			else if (nfit_mem->dcr->windows == 0
1123 					&& nfit_dcr->dcr->windows)
1124 				nfit_mem->dcr = nfit_dcr->dcr;
1125 			break;
1126 		}
1127 
1128 		list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1129 			struct acpi_nfit_flush_address *flush;
1130 			u16 i;
1131 
1132 			if (nfit_flush->flush->device_handle != device_handle)
1133 				continue;
1134 			nfit_mem->nfit_flush = nfit_flush;
1135 			flush = nfit_flush->flush;
1136 			nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1137 					flush->hint_count,
1138 					sizeof(struct resource),
1139 					GFP_KERNEL);
1140 			if (!nfit_mem->flush_wpq)
1141 				return -ENOMEM;
1142 			for (i = 0; i < flush->hint_count; i++) {
1143 				struct resource *res = &nfit_mem->flush_wpq[i];
1144 
1145 				res->start = flush->hint_address[i];
1146 				res->end = res->start + 8 - 1;
1147 			}
1148 			break;
1149 		}
1150 
1151 		if (dcr && !nfit_mem->dcr) {
1152 			dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1153 					spa->range_index, dcr);
1154 			return -ENODEV;
1155 		}
1156 
1157 		if (type == NFIT_SPA_DCR) {
1158 			struct nfit_idt *nfit_idt;
1159 			u16 idt_idx;
1160 
1161 			/* multiple dimms may share a SPA when interleaved */
1162 			nfit_mem->spa_dcr = spa;
1163 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
1164 			idt_idx = nfit_memdev->memdev->interleave_index;
1165 			list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1166 				if (nfit_idt->idt->interleave_index != idt_idx)
1167 					continue;
1168 				nfit_mem->idt_dcr = nfit_idt->idt;
1169 				break;
1170 			}
1171 			nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1172 		} else if (type == NFIT_SPA_PM) {
1173 			/*
1174 			 * A single dimm may belong to multiple SPA-PM
1175 			 * ranges, record at least one in addition to
1176 			 * any SPA-DCR range.
1177 			 */
1178 			nfit_mem->memdev_pmem = nfit_memdev->memdev;
1179 		} else
1180 			nfit_mem->memdev_dcr = nfit_memdev->memdev;
1181 	}
1182 
1183 	return 0;
1184 }
1185 
1186 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1187 {
1188 	struct nfit_mem *a = container_of(_a, typeof(*a), list);
1189 	struct nfit_mem *b = container_of(_b, typeof(*b), list);
1190 	u32 handleA, handleB;
1191 
1192 	handleA = __to_nfit_memdev(a)->device_handle;
1193 	handleB = __to_nfit_memdev(b)->device_handle;
1194 	if (handleA < handleB)
1195 		return -1;
1196 	else if (handleA > handleB)
1197 		return 1;
1198 	return 0;
1199 }
1200 
1201 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1202 {
1203 	struct nfit_spa *nfit_spa;
1204 	int rc;
1205 
1206 
1207 	/*
1208 	 * For each SPA-DCR or SPA-PMEM address range find its
1209 	 * corresponding MEMDEV(s).  From each MEMDEV find the
1210 	 * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1211 	 * try to find a SPA-BDW and a corresponding BDW that references
1212 	 * the DCR.  Throw it all into an nfit_mem object.  Note, that
1213 	 * BDWs are optional.
1214 	 */
1215 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1216 		rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1217 		if (rc)
1218 			return rc;
1219 	}
1220 
1221 	/*
1222 	 * If a DIMM has failed to be mapped into SPA there will be no
1223 	 * SPA entries above. Find and register all the unmapped DIMMs
1224 	 * for reporting and recovery purposes.
1225 	 */
1226 	rc = __nfit_mem_init(acpi_desc, NULL);
1227 	if (rc)
1228 		return rc;
1229 
1230 	list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1231 
1232 	return 0;
1233 }
1234 
1235 static ssize_t bus_dsm_mask_show(struct device *dev,
1236 		struct device_attribute *attr, char *buf)
1237 {
1238 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1239 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1240 
1241 	return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1242 }
1243 static struct device_attribute dev_attr_bus_dsm_mask =
1244 		__ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1245 
1246 static ssize_t revision_show(struct device *dev,
1247 		struct device_attribute *attr, char *buf)
1248 {
1249 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1250 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1251 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1252 
1253 	return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1254 }
1255 static DEVICE_ATTR_RO(revision);
1256 
1257 static ssize_t hw_error_scrub_show(struct device *dev,
1258 		struct device_attribute *attr, char *buf)
1259 {
1260 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1261 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1262 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1263 
1264 	return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1265 }
1266 
1267 /*
1268  * The 'hw_error_scrub' attribute can have the following values written to it:
1269  * '0': Switch to the default mode where an exception will only insert
1270  *      the address of the memory error into the poison and badblocks lists.
1271  * '1': Enable a full scrub to happen if an exception for a memory error is
1272  *      received.
1273  */
1274 static ssize_t hw_error_scrub_store(struct device *dev,
1275 		struct device_attribute *attr, const char *buf, size_t size)
1276 {
1277 	struct nvdimm_bus_descriptor *nd_desc;
1278 	ssize_t rc;
1279 	long val;
1280 
1281 	rc = kstrtol(buf, 0, &val);
1282 	if (rc)
1283 		return rc;
1284 
1285 	nfit_device_lock(dev);
1286 	nd_desc = dev_get_drvdata(dev);
1287 	if (nd_desc) {
1288 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1289 
1290 		switch (val) {
1291 		case HW_ERROR_SCRUB_ON:
1292 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1293 			break;
1294 		case HW_ERROR_SCRUB_OFF:
1295 			acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1296 			break;
1297 		default:
1298 			rc = -EINVAL;
1299 			break;
1300 		}
1301 	}
1302 	nfit_device_unlock(dev);
1303 	if (rc)
1304 		return rc;
1305 	return size;
1306 }
1307 static DEVICE_ATTR_RW(hw_error_scrub);
1308 
1309 /*
1310  * This shows the number of full Address Range Scrubs that have been
1311  * completed since driver load time. Userspace can wait on this using
1312  * select/poll etc. A '+' at the end indicates an ARS is in progress
1313  */
1314 static ssize_t scrub_show(struct device *dev,
1315 		struct device_attribute *attr, char *buf)
1316 {
1317 	struct nvdimm_bus_descriptor *nd_desc;
1318 	struct acpi_nfit_desc *acpi_desc;
1319 	ssize_t rc = -ENXIO;
1320 	bool busy;
1321 
1322 	nfit_device_lock(dev);
1323 	nd_desc = dev_get_drvdata(dev);
1324 	if (!nd_desc) {
1325 		device_unlock(dev);
1326 		return rc;
1327 	}
1328 	acpi_desc = to_acpi_desc(nd_desc);
1329 
1330 	mutex_lock(&acpi_desc->init_mutex);
1331 	busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1332 		&& !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1333 	rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1334 	/* Allow an admin to poll the busy state at a higher rate */
1335 	if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1336 				&acpi_desc->scrub_flags)) {
1337 		acpi_desc->scrub_tmo = 1;
1338 		mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1339 	}
1340 
1341 	mutex_unlock(&acpi_desc->init_mutex);
1342 	nfit_device_unlock(dev);
1343 	return rc;
1344 }
1345 
1346 static ssize_t scrub_store(struct device *dev,
1347 		struct device_attribute *attr, const char *buf, size_t size)
1348 {
1349 	struct nvdimm_bus_descriptor *nd_desc;
1350 	ssize_t rc;
1351 	long val;
1352 
1353 	rc = kstrtol(buf, 0, &val);
1354 	if (rc)
1355 		return rc;
1356 	if (val != 1)
1357 		return -EINVAL;
1358 
1359 	nfit_device_lock(dev);
1360 	nd_desc = dev_get_drvdata(dev);
1361 	if (nd_desc) {
1362 		struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1363 
1364 		rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1365 	}
1366 	nfit_device_unlock(dev);
1367 	if (rc)
1368 		return rc;
1369 	return size;
1370 }
1371 static DEVICE_ATTR_RW(scrub);
1372 
1373 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1374 {
1375 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1376 	const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1377 		| 1 << ND_CMD_ARS_STATUS;
1378 
1379 	return (nd_desc->cmd_mask & mask) == mask;
1380 }
1381 
1382 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1383 {
1384 	struct device *dev = container_of(kobj, struct device, kobj);
1385 	struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1386 
1387 	if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1388 		return 0;
1389 	return a->mode;
1390 }
1391 
1392 static struct attribute *acpi_nfit_attributes[] = {
1393 	&dev_attr_revision.attr,
1394 	&dev_attr_scrub.attr,
1395 	&dev_attr_hw_error_scrub.attr,
1396 	&dev_attr_bus_dsm_mask.attr,
1397 	NULL,
1398 };
1399 
1400 static const struct attribute_group acpi_nfit_attribute_group = {
1401 	.name = "nfit",
1402 	.attrs = acpi_nfit_attributes,
1403 	.is_visible = nfit_visible,
1404 };
1405 
1406 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1407 	&nvdimm_bus_attribute_group,
1408 	&acpi_nfit_attribute_group,
1409 	NULL,
1410 };
1411 
1412 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1413 {
1414 	struct nvdimm *nvdimm = to_nvdimm(dev);
1415 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1416 
1417 	return __to_nfit_memdev(nfit_mem);
1418 }
1419 
1420 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1421 {
1422 	struct nvdimm *nvdimm = to_nvdimm(dev);
1423 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1424 
1425 	return nfit_mem->dcr;
1426 }
1427 
1428 static ssize_t handle_show(struct device *dev,
1429 		struct device_attribute *attr, char *buf)
1430 {
1431 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1432 
1433 	return sprintf(buf, "%#x\n", memdev->device_handle);
1434 }
1435 static DEVICE_ATTR_RO(handle);
1436 
1437 static ssize_t phys_id_show(struct device *dev,
1438 		struct device_attribute *attr, char *buf)
1439 {
1440 	struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1441 
1442 	return sprintf(buf, "%#x\n", memdev->physical_id);
1443 }
1444 static DEVICE_ATTR_RO(phys_id);
1445 
1446 static ssize_t vendor_show(struct device *dev,
1447 		struct device_attribute *attr, char *buf)
1448 {
1449 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1450 
1451 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1452 }
1453 static DEVICE_ATTR_RO(vendor);
1454 
1455 static ssize_t rev_id_show(struct device *dev,
1456 		struct device_attribute *attr, char *buf)
1457 {
1458 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1459 
1460 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1461 }
1462 static DEVICE_ATTR_RO(rev_id);
1463 
1464 static ssize_t device_show(struct device *dev,
1465 		struct device_attribute *attr, char *buf)
1466 {
1467 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1468 
1469 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1470 }
1471 static DEVICE_ATTR_RO(device);
1472 
1473 static ssize_t subsystem_vendor_show(struct device *dev,
1474 		struct device_attribute *attr, char *buf)
1475 {
1476 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1477 
1478 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1479 }
1480 static DEVICE_ATTR_RO(subsystem_vendor);
1481 
1482 static ssize_t subsystem_rev_id_show(struct device *dev,
1483 		struct device_attribute *attr, char *buf)
1484 {
1485 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1486 
1487 	return sprintf(buf, "0x%04x\n",
1488 			be16_to_cpu(dcr->subsystem_revision_id));
1489 }
1490 static DEVICE_ATTR_RO(subsystem_rev_id);
1491 
1492 static ssize_t subsystem_device_show(struct device *dev,
1493 		struct device_attribute *attr, char *buf)
1494 {
1495 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1496 
1497 	return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1498 }
1499 static DEVICE_ATTR_RO(subsystem_device);
1500 
1501 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1502 {
1503 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1504 	int formats = 0;
1505 
1506 	if (nfit_mem->memdev_pmem)
1507 		formats++;
1508 	if (nfit_mem->memdev_bdw)
1509 		formats++;
1510 	return formats;
1511 }
1512 
1513 static ssize_t format_show(struct device *dev,
1514 		struct device_attribute *attr, char *buf)
1515 {
1516 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1517 
1518 	return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1519 }
1520 static DEVICE_ATTR_RO(format);
1521 
1522 static ssize_t format1_show(struct device *dev,
1523 		struct device_attribute *attr, char *buf)
1524 {
1525 	u32 handle;
1526 	ssize_t rc = -ENXIO;
1527 	struct nfit_mem *nfit_mem;
1528 	struct nfit_memdev *nfit_memdev;
1529 	struct acpi_nfit_desc *acpi_desc;
1530 	struct nvdimm *nvdimm = to_nvdimm(dev);
1531 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1532 
1533 	nfit_mem = nvdimm_provider_data(nvdimm);
1534 	acpi_desc = nfit_mem->acpi_desc;
1535 	handle = to_nfit_memdev(dev)->device_handle;
1536 
1537 	/* assumes DIMMs have at most 2 published interface codes */
1538 	mutex_lock(&acpi_desc->init_mutex);
1539 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1540 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1541 		struct nfit_dcr *nfit_dcr;
1542 
1543 		if (memdev->device_handle != handle)
1544 			continue;
1545 
1546 		list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1547 			if (nfit_dcr->dcr->region_index != memdev->region_index)
1548 				continue;
1549 			if (nfit_dcr->dcr->code == dcr->code)
1550 				continue;
1551 			rc = sprintf(buf, "0x%04x\n",
1552 					le16_to_cpu(nfit_dcr->dcr->code));
1553 			break;
1554 		}
1555 		if (rc != ENXIO)
1556 			break;
1557 	}
1558 	mutex_unlock(&acpi_desc->init_mutex);
1559 	return rc;
1560 }
1561 static DEVICE_ATTR_RO(format1);
1562 
1563 static ssize_t formats_show(struct device *dev,
1564 		struct device_attribute *attr, char *buf)
1565 {
1566 	struct nvdimm *nvdimm = to_nvdimm(dev);
1567 
1568 	return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1569 }
1570 static DEVICE_ATTR_RO(formats);
1571 
1572 static ssize_t serial_show(struct device *dev,
1573 		struct device_attribute *attr, char *buf)
1574 {
1575 	struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1576 
1577 	return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1578 }
1579 static DEVICE_ATTR_RO(serial);
1580 
1581 static ssize_t family_show(struct device *dev,
1582 		struct device_attribute *attr, char *buf)
1583 {
1584 	struct nvdimm *nvdimm = to_nvdimm(dev);
1585 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1586 
1587 	if (nfit_mem->family < 0)
1588 		return -ENXIO;
1589 	return sprintf(buf, "%d\n", nfit_mem->family);
1590 }
1591 static DEVICE_ATTR_RO(family);
1592 
1593 static ssize_t dsm_mask_show(struct device *dev,
1594 		struct device_attribute *attr, char *buf)
1595 {
1596 	struct nvdimm *nvdimm = to_nvdimm(dev);
1597 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1598 
1599 	if (nfit_mem->family < 0)
1600 		return -ENXIO;
1601 	return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1602 }
1603 static DEVICE_ATTR_RO(dsm_mask);
1604 
1605 static ssize_t flags_show(struct device *dev,
1606 		struct device_attribute *attr, char *buf)
1607 {
1608 	struct nvdimm *nvdimm = to_nvdimm(dev);
1609 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1610 	u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1611 
1612 	if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1613 		flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1614 
1615 	return sprintf(buf, "%s%s%s%s%s%s%s\n",
1616 		flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1617 		flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1618 		flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1619 		flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1620 		flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1621 		flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1622 		flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1623 }
1624 static DEVICE_ATTR_RO(flags);
1625 
1626 static ssize_t id_show(struct device *dev,
1627 		struct device_attribute *attr, char *buf)
1628 {
1629 	struct nvdimm *nvdimm = to_nvdimm(dev);
1630 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1631 
1632 	return sprintf(buf, "%s\n", nfit_mem->id);
1633 }
1634 static DEVICE_ATTR_RO(id);
1635 
1636 static ssize_t dirty_shutdown_show(struct device *dev,
1637 		struct device_attribute *attr, char *buf)
1638 {
1639 	struct nvdimm *nvdimm = to_nvdimm(dev);
1640 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1641 
1642 	return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1643 }
1644 static DEVICE_ATTR_RO(dirty_shutdown);
1645 
1646 static struct attribute *acpi_nfit_dimm_attributes[] = {
1647 	&dev_attr_handle.attr,
1648 	&dev_attr_phys_id.attr,
1649 	&dev_attr_vendor.attr,
1650 	&dev_attr_device.attr,
1651 	&dev_attr_rev_id.attr,
1652 	&dev_attr_subsystem_vendor.attr,
1653 	&dev_attr_subsystem_device.attr,
1654 	&dev_attr_subsystem_rev_id.attr,
1655 	&dev_attr_format.attr,
1656 	&dev_attr_formats.attr,
1657 	&dev_attr_format1.attr,
1658 	&dev_attr_serial.attr,
1659 	&dev_attr_flags.attr,
1660 	&dev_attr_id.attr,
1661 	&dev_attr_family.attr,
1662 	&dev_attr_dsm_mask.attr,
1663 	&dev_attr_dirty_shutdown.attr,
1664 	NULL,
1665 };
1666 
1667 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1668 		struct attribute *a, int n)
1669 {
1670 	struct device *dev = container_of(kobj, struct device, kobj);
1671 	struct nvdimm *nvdimm = to_nvdimm(dev);
1672 	struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1673 
1674 	if (!to_nfit_dcr(dev)) {
1675 		/* Without a dcr only the memdev attributes can be surfaced */
1676 		if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1677 				|| a == &dev_attr_flags.attr
1678 				|| a == &dev_attr_family.attr
1679 				|| a == &dev_attr_dsm_mask.attr)
1680 			return a->mode;
1681 		return 0;
1682 	}
1683 
1684 	if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1685 		return 0;
1686 
1687 	if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1688 			&& a == &dev_attr_dirty_shutdown.attr)
1689 		return 0;
1690 
1691 	return a->mode;
1692 }
1693 
1694 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1695 	.name = "nfit",
1696 	.attrs = acpi_nfit_dimm_attributes,
1697 	.is_visible = acpi_nfit_dimm_attr_visible,
1698 };
1699 
1700 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1701 	&nvdimm_attribute_group,
1702 	&nd_device_attribute_group,
1703 	&acpi_nfit_dimm_attribute_group,
1704 	NULL,
1705 };
1706 
1707 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1708 		u32 device_handle)
1709 {
1710 	struct nfit_mem *nfit_mem;
1711 
1712 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1713 		if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1714 			return nfit_mem->nvdimm;
1715 
1716 	return NULL;
1717 }
1718 
1719 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1720 {
1721 	struct nfit_mem *nfit_mem;
1722 	struct acpi_nfit_desc *acpi_desc;
1723 
1724 	dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1725 			event);
1726 
1727 	if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1728 		dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1729 				event);
1730 		return;
1731 	}
1732 
1733 	acpi_desc = dev_get_drvdata(dev->parent);
1734 	if (!acpi_desc)
1735 		return;
1736 
1737 	/*
1738 	 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1739 	 * is still valid.
1740 	 */
1741 	nfit_mem = dev_get_drvdata(dev);
1742 	if (nfit_mem && nfit_mem->flags_attr)
1743 		sysfs_notify_dirent(nfit_mem->flags_attr);
1744 }
1745 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1746 
1747 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1748 {
1749 	struct acpi_device *adev = data;
1750 	struct device *dev = &adev->dev;
1751 
1752 	nfit_device_lock(dev->parent);
1753 	__acpi_nvdimm_notify(dev, event);
1754 	nfit_device_unlock(dev->parent);
1755 }
1756 
1757 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1758 {
1759 	acpi_handle handle;
1760 	acpi_status status;
1761 
1762 	status = acpi_get_handle(adev->handle, method, &handle);
1763 
1764 	if (ACPI_SUCCESS(status))
1765 		return true;
1766 	return false;
1767 }
1768 
1769 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1770 {
1771 	struct device *dev = &nfit_mem->adev->dev;
1772 	struct nd_intel_smart smart = { 0 };
1773 	union acpi_object in_buf = {
1774 		.buffer.type = ACPI_TYPE_BUFFER,
1775 		.buffer.length = 0,
1776 	};
1777 	union acpi_object in_obj = {
1778 		.package.type = ACPI_TYPE_PACKAGE,
1779 		.package.count = 1,
1780 		.package.elements = &in_buf,
1781 	};
1782 	const u8 func = ND_INTEL_SMART;
1783 	const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1784 	u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1785 	struct acpi_device *adev = nfit_mem->adev;
1786 	acpi_handle handle = adev->handle;
1787 	union acpi_object *out_obj;
1788 
1789 	if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1790 		return;
1791 
1792 	out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1793 	if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1794 			|| out_obj->buffer.length < sizeof(smart)) {
1795 		dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1796 				dev_name(dev));
1797 		ACPI_FREE(out_obj);
1798 		return;
1799 	}
1800 	memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1801 	ACPI_FREE(out_obj);
1802 
1803 	if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1804 		if (smart.shutdown_state)
1805 			set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1806 	}
1807 
1808 	if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1809 		set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1810 		nfit_mem->dirty_shutdown = smart.shutdown_count;
1811 	}
1812 }
1813 
1814 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1815 {
1816 	/*
1817 	 * For DIMMs that provide a dynamic facility to retrieve a
1818 	 * dirty-shutdown status and/or a dirty-shutdown count, cache
1819 	 * these values in nfit_mem.
1820 	 */
1821 	if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1822 		nfit_intel_shutdown_status(nfit_mem);
1823 }
1824 
1825 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1826 		struct nfit_mem *nfit_mem, u32 device_handle)
1827 {
1828 	struct acpi_device *adev, *adev_dimm;
1829 	struct device *dev = acpi_desc->dev;
1830 	unsigned long dsm_mask, label_mask;
1831 	const guid_t *guid;
1832 	int i;
1833 	int family = -1;
1834 	struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1835 
1836 	/* nfit test assumes 1:1 relationship between commands and dsms */
1837 	nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1838 	nfit_mem->family = NVDIMM_FAMILY_INTEL;
1839 
1840 	if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1841 		sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1842 				be16_to_cpu(dcr->vendor_id),
1843 				dcr->manufacturing_location,
1844 				be16_to_cpu(dcr->manufacturing_date),
1845 				be32_to_cpu(dcr->serial_number));
1846 	else
1847 		sprintf(nfit_mem->id, "%04x-%08x",
1848 				be16_to_cpu(dcr->vendor_id),
1849 				be32_to_cpu(dcr->serial_number));
1850 
1851 	adev = to_acpi_dev(acpi_desc);
1852 	if (!adev) {
1853 		/* unit test case */
1854 		populate_shutdown_status(nfit_mem);
1855 		return 0;
1856 	}
1857 
1858 	adev_dimm = acpi_find_child_device(adev, device_handle, false);
1859 	nfit_mem->adev = adev_dimm;
1860 	if (!adev_dimm) {
1861 		dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1862 				device_handle);
1863 		return force_enable_dimms ? 0 : -ENODEV;
1864 	}
1865 
1866 	if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1867 		ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1868 		dev_err(dev, "%s: notification registration failed\n",
1869 				dev_name(&adev_dimm->dev));
1870 		return -ENXIO;
1871 	}
1872 	/*
1873 	 * Record nfit_mem for the notification path to track back to
1874 	 * the nfit sysfs attributes for this dimm device object.
1875 	 */
1876 	dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1877 
1878 	/*
1879 	 * There are 4 "legacy" NVDIMM command sets
1880 	 * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1881 	 * an EFI working group was established to constrain this
1882 	 * proliferation. The nfit driver probes for the supported command
1883 	 * set by GUID. Note, if you're a platform developer looking to add
1884 	 * a new command set to this probe, consider using an existing set,
1885 	 * or otherwise seek approval to publish the command set at
1886 	 * http://www.uefi.org/RFIC_LIST.
1887 	 *
1888 	 * Note, that checking for function0 (bit0) tells us if any commands
1889 	 * are reachable through this GUID.
1890 	 */
1891 	for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1892 		if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1893 			if (family < 0 || i == default_dsm_family)
1894 				family = i;
1895 
1896 	/* limit the supported commands to those that are publicly documented */
1897 	nfit_mem->family = family;
1898 	if (override_dsm_mask && !disable_vendor_specific)
1899 		dsm_mask = override_dsm_mask;
1900 	else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1901 		dsm_mask = NVDIMM_INTEL_CMDMASK;
1902 		if (disable_vendor_specific)
1903 			dsm_mask &= ~(1 << ND_CMD_VENDOR);
1904 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1905 		dsm_mask = 0x1c3c76;
1906 	} else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1907 		dsm_mask = 0x1fe;
1908 		if (disable_vendor_specific)
1909 			dsm_mask &= ~(1 << 8);
1910 	} else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1911 		dsm_mask = 0xffffffff;
1912 	} else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1913 		dsm_mask = 0x1f;
1914 	} else {
1915 		dev_dbg(dev, "unknown dimm command family\n");
1916 		nfit_mem->family = -1;
1917 		/* DSMs are optional, continue loading the driver... */
1918 		return 0;
1919 	}
1920 
1921 	/*
1922 	 * Function 0 is the command interrogation function, don't
1923 	 * export it to potential userspace use, and enable it to be
1924 	 * used as an error value in acpi_nfit_ctl().
1925 	 */
1926 	dsm_mask &= ~1UL;
1927 
1928 	guid = to_nfit_uuid(nfit_mem->family);
1929 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1930 		if (acpi_check_dsm(adev_dimm->handle, guid,
1931 					nfit_dsm_revid(nfit_mem->family, i),
1932 					1ULL << i))
1933 			set_bit(i, &nfit_mem->dsm_mask);
1934 
1935 	/*
1936 	 * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1937 	 * due to their better semantics handling locked capacity.
1938 	 */
1939 	label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1940 		| 1 << ND_CMD_SET_CONFIG_DATA;
1941 	if (family == NVDIMM_FAMILY_INTEL
1942 			&& (dsm_mask & label_mask) == label_mask)
1943 		/* skip _LS{I,R,W} enabling */;
1944 	else {
1945 		if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1946 				&& acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1947 			dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1948 			set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1949 		}
1950 
1951 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1952 				&& acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1953 			dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1954 			set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1955 		}
1956 
1957 		/*
1958 		 * Quirk read-only label configurations to preserve
1959 		 * access to label-less namespaces by default.
1960 		 */
1961 		if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1962 				&& !force_labels) {
1963 			dev_dbg(dev, "%s: No _LSW, disable labels\n",
1964 					dev_name(&adev_dimm->dev));
1965 			clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1966 		} else
1967 			dev_dbg(dev, "%s: Force enable labels\n",
1968 					dev_name(&adev_dimm->dev));
1969 	}
1970 
1971 	populate_shutdown_status(nfit_mem);
1972 
1973 	return 0;
1974 }
1975 
1976 static void shutdown_dimm_notify(void *data)
1977 {
1978 	struct acpi_nfit_desc *acpi_desc = data;
1979 	struct nfit_mem *nfit_mem;
1980 
1981 	mutex_lock(&acpi_desc->init_mutex);
1982 	/*
1983 	 * Clear out the nfit_mem->flags_attr and shut down dimm event
1984 	 * notifications.
1985 	 */
1986 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1987 		struct acpi_device *adev_dimm = nfit_mem->adev;
1988 
1989 		if (nfit_mem->flags_attr) {
1990 			sysfs_put(nfit_mem->flags_attr);
1991 			nfit_mem->flags_attr = NULL;
1992 		}
1993 		if (adev_dimm) {
1994 			acpi_remove_notify_handler(adev_dimm->handle,
1995 					ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1996 			dev_set_drvdata(&adev_dimm->dev, NULL);
1997 		}
1998 	}
1999 	mutex_unlock(&acpi_desc->init_mutex);
2000 }
2001 
2002 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
2003 {
2004 	switch (family) {
2005 	case NVDIMM_FAMILY_INTEL:
2006 		return intel_security_ops;
2007 	default:
2008 		return NULL;
2009 	}
2010 }
2011 
2012 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
2013 {
2014 	struct nfit_mem *nfit_mem;
2015 	int dimm_count = 0, rc;
2016 	struct nvdimm *nvdimm;
2017 
2018 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2019 		struct acpi_nfit_flush_address *flush;
2020 		unsigned long flags = 0, cmd_mask;
2021 		struct nfit_memdev *nfit_memdev;
2022 		u32 device_handle;
2023 		u16 mem_flags;
2024 
2025 		device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2026 		nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2027 		if (nvdimm) {
2028 			dimm_count++;
2029 			continue;
2030 		}
2031 
2032 		if (nfit_mem->bdw && nfit_mem->memdev_pmem)
2033 			set_bit(NDD_ALIASING, &flags);
2034 
2035 		/* collate flags across all memdevs for this dimm */
2036 		list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2037 			struct acpi_nfit_memory_map *dimm_memdev;
2038 
2039 			dimm_memdev = __to_nfit_memdev(nfit_mem);
2040 			if (dimm_memdev->device_handle
2041 					!= nfit_memdev->memdev->device_handle)
2042 				continue;
2043 			dimm_memdev->flags |= nfit_memdev->memdev->flags;
2044 		}
2045 
2046 		mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2047 		if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2048 			set_bit(NDD_UNARMED, &flags);
2049 
2050 		rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2051 		if (rc)
2052 			continue;
2053 
2054 		/*
2055 		 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2056 		 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2057 		 * userspace interface.
2058 		 */
2059 		cmd_mask = 1UL << ND_CMD_CALL;
2060 		if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2061 			/*
2062 			 * These commands have a 1:1 correspondence
2063 			 * between DSM payload and libnvdimm ioctl
2064 			 * payload format.
2065 			 */
2066 			cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2067 		}
2068 
2069 		/* Quirk to ignore LOCAL for labels on HYPERV DIMMs */
2070 		if (nfit_mem->family == NVDIMM_FAMILY_HYPERV)
2071 			set_bit(NDD_NOBLK, &flags);
2072 
2073 		if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2074 			set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2075 			set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2076 		}
2077 		if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2078 			set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2079 
2080 		flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2081 			: NULL;
2082 		nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2083 				acpi_nfit_dimm_attribute_groups,
2084 				flags, cmd_mask, flush ? flush->hint_count : 0,
2085 				nfit_mem->flush_wpq, &nfit_mem->id[0],
2086 				acpi_nfit_get_security_ops(nfit_mem->family));
2087 		if (!nvdimm)
2088 			return -ENOMEM;
2089 
2090 		nfit_mem->nvdimm = nvdimm;
2091 		dimm_count++;
2092 
2093 		if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2094 			continue;
2095 
2096 		dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2097 				nvdimm_name(nvdimm),
2098 		  mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2099 		  mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2100 		  mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2101 		  mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2102 		  mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2103 
2104 	}
2105 
2106 	rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2107 	if (rc)
2108 		return rc;
2109 
2110 	/*
2111 	 * Now that dimms are successfully registered, and async registration
2112 	 * is flushed, attempt to enable event notification.
2113 	 */
2114 	list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2115 		struct kernfs_node *nfit_kernfs;
2116 
2117 		nvdimm = nfit_mem->nvdimm;
2118 		if (!nvdimm)
2119 			continue;
2120 
2121 		nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2122 		if (nfit_kernfs)
2123 			nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2124 					"flags");
2125 		sysfs_put(nfit_kernfs);
2126 		if (!nfit_mem->flags_attr)
2127 			dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2128 					nvdimm_name(nvdimm));
2129 	}
2130 
2131 	return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2132 			acpi_desc);
2133 }
2134 
2135 /*
2136  * These constants are private because there are no kernel consumers of
2137  * these commands.
2138  */
2139 enum nfit_aux_cmds {
2140         NFIT_CMD_TRANSLATE_SPA = 5,
2141         NFIT_CMD_ARS_INJECT_SET = 7,
2142         NFIT_CMD_ARS_INJECT_CLEAR = 8,
2143         NFIT_CMD_ARS_INJECT_GET = 9,
2144 };
2145 
2146 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2147 {
2148 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2149 	const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2150 	struct acpi_device *adev;
2151 	unsigned long dsm_mask;
2152 	int i;
2153 
2154 	nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2155 	nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
2156 	adev = to_acpi_dev(acpi_desc);
2157 	if (!adev)
2158 		return;
2159 
2160 	for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2161 		if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2162 			set_bit(i, &nd_desc->cmd_mask);
2163 	set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2164 
2165 	dsm_mask =
2166 		(1 << ND_CMD_ARS_CAP) |
2167 		(1 << ND_CMD_ARS_START) |
2168 		(1 << ND_CMD_ARS_STATUS) |
2169 		(1 << ND_CMD_CLEAR_ERROR) |
2170 		(1 << NFIT_CMD_TRANSLATE_SPA) |
2171 		(1 << NFIT_CMD_ARS_INJECT_SET) |
2172 		(1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2173 		(1 << NFIT_CMD_ARS_INJECT_GET);
2174 	for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2175 		if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2176 			set_bit(i, &nd_desc->bus_dsm_mask);
2177 }
2178 
2179 static ssize_t range_index_show(struct device *dev,
2180 		struct device_attribute *attr, char *buf)
2181 {
2182 	struct nd_region *nd_region = to_nd_region(dev);
2183 	struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2184 
2185 	return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2186 }
2187 static DEVICE_ATTR_RO(range_index);
2188 
2189 static struct attribute *acpi_nfit_region_attributes[] = {
2190 	&dev_attr_range_index.attr,
2191 	NULL,
2192 };
2193 
2194 static const struct attribute_group acpi_nfit_region_attribute_group = {
2195 	.name = "nfit",
2196 	.attrs = acpi_nfit_region_attributes,
2197 };
2198 
2199 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2200 	&nd_region_attribute_group,
2201 	&nd_mapping_attribute_group,
2202 	&nd_device_attribute_group,
2203 	&nd_numa_attribute_group,
2204 	&acpi_nfit_region_attribute_group,
2205 	NULL,
2206 };
2207 
2208 /* enough info to uniquely specify an interleave set */
2209 struct nfit_set_info {
2210 	struct nfit_set_info_map {
2211 		u64 region_offset;
2212 		u32 serial_number;
2213 		u32 pad;
2214 	} mapping[0];
2215 };
2216 
2217 struct nfit_set_info2 {
2218 	struct nfit_set_info_map2 {
2219 		u64 region_offset;
2220 		u32 serial_number;
2221 		u16 vendor_id;
2222 		u16 manufacturing_date;
2223 		u8  manufacturing_location;
2224 		u8  reserved[31];
2225 	} mapping[0];
2226 };
2227 
2228 static size_t sizeof_nfit_set_info(int num_mappings)
2229 {
2230 	return sizeof(struct nfit_set_info)
2231 		+ num_mappings * sizeof(struct nfit_set_info_map);
2232 }
2233 
2234 static size_t sizeof_nfit_set_info2(int num_mappings)
2235 {
2236 	return sizeof(struct nfit_set_info2)
2237 		+ num_mappings * sizeof(struct nfit_set_info_map2);
2238 }
2239 
2240 static int cmp_map_compat(const void *m0, const void *m1)
2241 {
2242 	const struct nfit_set_info_map *map0 = m0;
2243 	const struct nfit_set_info_map *map1 = m1;
2244 
2245 	return memcmp(&map0->region_offset, &map1->region_offset,
2246 			sizeof(u64));
2247 }
2248 
2249 static int cmp_map(const void *m0, const void *m1)
2250 {
2251 	const struct nfit_set_info_map *map0 = m0;
2252 	const struct nfit_set_info_map *map1 = m1;
2253 
2254 	if (map0->region_offset < map1->region_offset)
2255 		return -1;
2256 	else if (map0->region_offset > map1->region_offset)
2257 		return 1;
2258 	return 0;
2259 }
2260 
2261 static int cmp_map2(const void *m0, const void *m1)
2262 {
2263 	const struct nfit_set_info_map2 *map0 = m0;
2264 	const struct nfit_set_info_map2 *map1 = m1;
2265 
2266 	if (map0->region_offset < map1->region_offset)
2267 		return -1;
2268 	else if (map0->region_offset > map1->region_offset)
2269 		return 1;
2270 	return 0;
2271 }
2272 
2273 /* Retrieve the nth entry referencing this spa */
2274 static struct acpi_nfit_memory_map *memdev_from_spa(
2275 		struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2276 {
2277 	struct nfit_memdev *nfit_memdev;
2278 
2279 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2280 		if (nfit_memdev->memdev->range_index == range_index)
2281 			if (n-- == 0)
2282 				return nfit_memdev->memdev;
2283 	return NULL;
2284 }
2285 
2286 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2287 		struct nd_region_desc *ndr_desc,
2288 		struct acpi_nfit_system_address *spa)
2289 {
2290 	struct device *dev = acpi_desc->dev;
2291 	struct nd_interleave_set *nd_set;
2292 	u16 nr = ndr_desc->num_mappings;
2293 	struct nfit_set_info2 *info2;
2294 	struct nfit_set_info *info;
2295 	int i;
2296 
2297 	nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2298 	if (!nd_set)
2299 		return -ENOMEM;
2300 	guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2301 
2302 	info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2303 	if (!info)
2304 		return -ENOMEM;
2305 
2306 	info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2307 	if (!info2)
2308 		return -ENOMEM;
2309 
2310 	for (i = 0; i < nr; i++) {
2311 		struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2312 		struct nfit_set_info_map *map = &info->mapping[i];
2313 		struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2314 		struct nvdimm *nvdimm = mapping->nvdimm;
2315 		struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2316 		struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2317 				spa->range_index, i);
2318 		struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2319 
2320 		if (!memdev || !nfit_mem->dcr) {
2321 			dev_err(dev, "%s: failed to find DCR\n", __func__);
2322 			return -ENODEV;
2323 		}
2324 
2325 		map->region_offset = memdev->region_offset;
2326 		map->serial_number = dcr->serial_number;
2327 
2328 		map2->region_offset = memdev->region_offset;
2329 		map2->serial_number = dcr->serial_number;
2330 		map2->vendor_id = dcr->vendor_id;
2331 		map2->manufacturing_date = dcr->manufacturing_date;
2332 		map2->manufacturing_location = dcr->manufacturing_location;
2333 	}
2334 
2335 	/* v1.1 namespaces */
2336 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2337 			cmp_map, NULL);
2338 	nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2339 
2340 	/* v1.2 namespaces */
2341 	sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2342 			cmp_map2, NULL);
2343 	nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2344 
2345 	/* support v1.1 namespaces created with the wrong sort order */
2346 	sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2347 			cmp_map_compat, NULL);
2348 	nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2349 
2350 	/* record the result of the sort for the mapping position */
2351 	for (i = 0; i < nr; i++) {
2352 		struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2353 		int j;
2354 
2355 		for (j = 0; j < nr; j++) {
2356 			struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2357 			struct nvdimm *nvdimm = mapping->nvdimm;
2358 			struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2359 			struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2360 
2361 			if (map2->serial_number == dcr->serial_number &&
2362 			    map2->vendor_id == dcr->vendor_id &&
2363 			    map2->manufacturing_date == dcr->manufacturing_date &&
2364 			    map2->manufacturing_location
2365 				    == dcr->manufacturing_location) {
2366 				mapping->position = i;
2367 				break;
2368 			}
2369 		}
2370 	}
2371 
2372 	ndr_desc->nd_set = nd_set;
2373 	devm_kfree(dev, info);
2374 	devm_kfree(dev, info2);
2375 
2376 	return 0;
2377 }
2378 
2379 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2380 {
2381 	struct acpi_nfit_interleave *idt = mmio->idt;
2382 	u32 sub_line_offset, line_index, line_offset;
2383 	u64 line_no, table_skip_count, table_offset;
2384 
2385 	line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2386 	table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2387 	line_offset = idt->line_offset[line_index]
2388 		* mmio->line_size;
2389 	table_offset = table_skip_count * mmio->table_size;
2390 
2391 	return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2392 }
2393 
2394 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2395 {
2396 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2397 	u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2398 	const u32 STATUS_MASK = 0x80000037;
2399 
2400 	if (mmio->num_lines)
2401 		offset = to_interleave_offset(offset, mmio);
2402 
2403 	return readl(mmio->addr.base + offset) & STATUS_MASK;
2404 }
2405 
2406 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2407 		resource_size_t dpa, unsigned int len, unsigned int write)
2408 {
2409 	u64 cmd, offset;
2410 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2411 
2412 	enum {
2413 		BCW_OFFSET_MASK = (1ULL << 48)-1,
2414 		BCW_LEN_SHIFT = 48,
2415 		BCW_LEN_MASK = (1ULL << 8) - 1,
2416 		BCW_CMD_SHIFT = 56,
2417 	};
2418 
2419 	cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2420 	len = len >> L1_CACHE_SHIFT;
2421 	cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2422 	cmd |= ((u64) write) << BCW_CMD_SHIFT;
2423 
2424 	offset = nfit_blk->cmd_offset + mmio->size * bw;
2425 	if (mmio->num_lines)
2426 		offset = to_interleave_offset(offset, mmio);
2427 
2428 	writeq(cmd, mmio->addr.base + offset);
2429 	nvdimm_flush(nfit_blk->nd_region, NULL);
2430 
2431 	if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2432 		readq(mmio->addr.base + offset);
2433 }
2434 
2435 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2436 		resource_size_t dpa, void *iobuf, size_t len, int rw,
2437 		unsigned int lane)
2438 {
2439 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2440 	unsigned int copied = 0;
2441 	u64 base_offset;
2442 	int rc;
2443 
2444 	base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2445 		+ lane * mmio->size;
2446 	write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2447 	while (len) {
2448 		unsigned int c;
2449 		u64 offset;
2450 
2451 		if (mmio->num_lines) {
2452 			u32 line_offset;
2453 
2454 			offset = to_interleave_offset(base_offset + copied,
2455 					mmio);
2456 			div_u64_rem(offset, mmio->line_size, &line_offset);
2457 			c = min_t(size_t, len, mmio->line_size - line_offset);
2458 		} else {
2459 			offset = base_offset + nfit_blk->bdw_offset;
2460 			c = len;
2461 		}
2462 
2463 		if (rw)
2464 			memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2465 		else {
2466 			if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2467 				arch_invalidate_pmem((void __force *)
2468 					mmio->addr.aperture + offset, c);
2469 
2470 			memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2471 		}
2472 
2473 		copied += c;
2474 		len -= c;
2475 	}
2476 
2477 	if (rw)
2478 		nvdimm_flush(nfit_blk->nd_region, NULL);
2479 
2480 	rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2481 	return rc;
2482 }
2483 
2484 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2485 		resource_size_t dpa, void *iobuf, u64 len, int rw)
2486 {
2487 	struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2488 	struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2489 	struct nd_region *nd_region = nfit_blk->nd_region;
2490 	unsigned int lane, copied = 0;
2491 	int rc = 0;
2492 
2493 	lane = nd_region_acquire_lane(nd_region);
2494 	while (len) {
2495 		u64 c = min(len, mmio->size);
2496 
2497 		rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2498 				iobuf + copied, c, rw, lane);
2499 		if (rc)
2500 			break;
2501 
2502 		copied += c;
2503 		len -= c;
2504 	}
2505 	nd_region_release_lane(nd_region, lane);
2506 
2507 	return rc;
2508 }
2509 
2510 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2511 		struct acpi_nfit_interleave *idt, u16 interleave_ways)
2512 {
2513 	if (idt) {
2514 		mmio->num_lines = idt->line_count;
2515 		mmio->line_size = idt->line_size;
2516 		if (interleave_ways == 0)
2517 			return -ENXIO;
2518 		mmio->table_size = mmio->num_lines * interleave_ways
2519 			* mmio->line_size;
2520 	}
2521 
2522 	return 0;
2523 }
2524 
2525 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2526 		struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2527 {
2528 	struct nd_cmd_dimm_flags flags;
2529 	int rc;
2530 
2531 	memset(&flags, 0, sizeof(flags));
2532 	rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2533 			sizeof(flags), NULL);
2534 
2535 	if (rc >= 0 && flags.status == 0)
2536 		nfit_blk->dimm_flags = flags.flags;
2537 	else if (rc == -ENOTTY) {
2538 		/* fall back to a conservative default */
2539 		nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2540 		rc = 0;
2541 	} else
2542 		rc = -ENXIO;
2543 
2544 	return rc;
2545 }
2546 
2547 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2548 		struct device *dev)
2549 {
2550 	struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2551 	struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2552 	struct nfit_blk_mmio *mmio;
2553 	struct nfit_blk *nfit_blk;
2554 	struct nfit_mem *nfit_mem;
2555 	struct nvdimm *nvdimm;
2556 	int rc;
2557 
2558 	nvdimm = nd_blk_region_to_dimm(ndbr);
2559 	nfit_mem = nvdimm_provider_data(nvdimm);
2560 	if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2561 		dev_dbg(dev, "missing%s%s%s\n",
2562 				nfit_mem ? "" : " nfit_mem",
2563 				(nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2564 				(nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2565 		return -ENXIO;
2566 	}
2567 
2568 	nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2569 	if (!nfit_blk)
2570 		return -ENOMEM;
2571 	nd_blk_region_set_provider_data(ndbr, nfit_blk);
2572 	nfit_blk->nd_region = to_nd_region(dev);
2573 
2574 	/* map block aperture memory */
2575 	nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2576 	mmio = &nfit_blk->mmio[BDW];
2577 	mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2578                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2579 	if (!mmio->addr.base) {
2580 		dev_dbg(dev, "%s failed to map bdw\n",
2581 				nvdimm_name(nvdimm));
2582 		return -ENOMEM;
2583 	}
2584 	mmio->size = nfit_mem->bdw->size;
2585 	mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2586 	mmio->idt = nfit_mem->idt_bdw;
2587 	mmio->spa = nfit_mem->spa_bdw;
2588 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2589 			nfit_mem->memdev_bdw->interleave_ways);
2590 	if (rc) {
2591 		dev_dbg(dev, "%s failed to init bdw interleave\n",
2592 				nvdimm_name(nvdimm));
2593 		return rc;
2594 	}
2595 
2596 	/* map block control memory */
2597 	nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2598 	nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2599 	mmio = &nfit_blk->mmio[DCR];
2600 	mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2601 			nfit_mem->spa_dcr->length);
2602 	if (!mmio->addr.base) {
2603 		dev_dbg(dev, "%s failed to map dcr\n",
2604 				nvdimm_name(nvdimm));
2605 		return -ENOMEM;
2606 	}
2607 	mmio->size = nfit_mem->dcr->window_size;
2608 	mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2609 	mmio->idt = nfit_mem->idt_dcr;
2610 	mmio->spa = nfit_mem->spa_dcr;
2611 	rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2612 			nfit_mem->memdev_dcr->interleave_ways);
2613 	if (rc) {
2614 		dev_dbg(dev, "%s failed to init dcr interleave\n",
2615 				nvdimm_name(nvdimm));
2616 		return rc;
2617 	}
2618 
2619 	rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2620 	if (rc < 0) {
2621 		dev_dbg(dev, "%s failed get DIMM flags\n",
2622 				nvdimm_name(nvdimm));
2623 		return rc;
2624 	}
2625 
2626 	if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2627 		dev_warn(dev, "unable to guarantee persistence of writes\n");
2628 
2629 	if (mmio->line_size == 0)
2630 		return 0;
2631 
2632 	if ((u32) nfit_blk->cmd_offset % mmio->line_size
2633 			+ 8 > mmio->line_size) {
2634 		dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2635 		return -ENXIO;
2636 	} else if ((u32) nfit_blk->stat_offset % mmio->line_size
2637 			+ 8 > mmio->line_size) {
2638 		dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2639 		return -ENXIO;
2640 	}
2641 
2642 	return 0;
2643 }
2644 
2645 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2646 		struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2647 {
2648 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2649 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2650 	int cmd_rc, rc;
2651 
2652 	cmd->address = spa->address;
2653 	cmd->length = spa->length;
2654 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2655 			sizeof(*cmd), &cmd_rc);
2656 	if (rc < 0)
2657 		return rc;
2658 	return cmd_rc;
2659 }
2660 
2661 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2662 		struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2663 {
2664 	int rc;
2665 	int cmd_rc;
2666 	struct nd_cmd_ars_start ars_start;
2667 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2668 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2669 
2670 	memset(&ars_start, 0, sizeof(ars_start));
2671 	ars_start.address = spa->address;
2672 	ars_start.length = spa->length;
2673 	if (req_type == ARS_REQ_SHORT)
2674 		ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2675 	if (nfit_spa_type(spa) == NFIT_SPA_PM)
2676 		ars_start.type = ND_ARS_PERSISTENT;
2677 	else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2678 		ars_start.type = ND_ARS_VOLATILE;
2679 	else
2680 		return -ENOTTY;
2681 
2682 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2683 			sizeof(ars_start), &cmd_rc);
2684 
2685 	if (rc < 0)
2686 		return rc;
2687 	if (cmd_rc < 0)
2688 		return cmd_rc;
2689 	set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2690 	return 0;
2691 }
2692 
2693 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2694 {
2695 	int rc, cmd_rc;
2696 	struct nd_cmd_ars_start ars_start;
2697 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2698 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2699 
2700 	ars_start = (struct nd_cmd_ars_start) {
2701 		.address = ars_status->restart_address,
2702 		.length = ars_status->restart_length,
2703 		.type = ars_status->type,
2704 	};
2705 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2706 			sizeof(ars_start), &cmd_rc);
2707 	if (rc < 0)
2708 		return rc;
2709 	return cmd_rc;
2710 }
2711 
2712 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2713 {
2714 	struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2715 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2716 	int rc, cmd_rc;
2717 
2718 	rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2719 			acpi_desc->max_ars, &cmd_rc);
2720 	if (rc < 0)
2721 		return rc;
2722 	return cmd_rc;
2723 }
2724 
2725 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2726 		struct nfit_spa *nfit_spa)
2727 {
2728 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2729 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2730 	struct nd_region *nd_region = nfit_spa->nd_region;
2731 	struct device *dev;
2732 
2733 	lockdep_assert_held(&acpi_desc->init_mutex);
2734 	/*
2735 	 * Only advance the ARS state for ARS runs initiated by the
2736 	 * kernel, ignore ARS results from BIOS initiated runs for scrub
2737 	 * completion tracking.
2738 	 */
2739 	if (acpi_desc->scrub_spa != nfit_spa)
2740 		return;
2741 
2742 	if ((ars_status->address >= spa->address && ars_status->address
2743 				< spa->address + spa->length)
2744 			|| (ars_status->address < spa->address)) {
2745 		/*
2746 		 * Assume that if a scrub starts at an offset from the
2747 		 * start of nfit_spa that we are in the continuation
2748 		 * case.
2749 		 *
2750 		 * Otherwise, if the scrub covers the spa range, mark
2751 		 * any pending request complete.
2752 		 */
2753 		if (ars_status->address + ars_status->length
2754 				>= spa->address + spa->length)
2755 				/* complete */;
2756 		else
2757 			return;
2758 	} else
2759 		return;
2760 
2761 	acpi_desc->scrub_spa = NULL;
2762 	if (nd_region) {
2763 		dev = nd_region_dev(nd_region);
2764 		nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2765 	} else
2766 		dev = acpi_desc->dev;
2767 	dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2768 }
2769 
2770 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2771 {
2772 	struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2773 	struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2774 	int rc;
2775 	u32 i;
2776 
2777 	/*
2778 	 * First record starts at 44 byte offset from the start of the
2779 	 * payload.
2780 	 */
2781 	if (ars_status->out_length < 44)
2782 		return 0;
2783 
2784 	/*
2785 	 * Ignore potentially stale results that are only refreshed
2786 	 * after a start-ARS event.
2787 	 */
2788 	if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2789 		dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2790 				ars_status->num_records);
2791 		return 0;
2792 	}
2793 
2794 	for (i = 0; i < ars_status->num_records; i++) {
2795 		/* only process full records */
2796 		if (ars_status->out_length
2797 				< 44 + sizeof(struct nd_ars_record) * (i + 1))
2798 			break;
2799 		rc = nvdimm_bus_add_badrange(nvdimm_bus,
2800 				ars_status->records[i].err_address,
2801 				ars_status->records[i].length);
2802 		if (rc)
2803 			return rc;
2804 	}
2805 	if (i < ars_status->num_records)
2806 		dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2807 
2808 	return 0;
2809 }
2810 
2811 static void acpi_nfit_remove_resource(void *data)
2812 {
2813 	struct resource *res = data;
2814 
2815 	remove_resource(res);
2816 }
2817 
2818 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2819 		struct nd_region_desc *ndr_desc)
2820 {
2821 	struct resource *res, *nd_res = ndr_desc->res;
2822 	int is_pmem, ret;
2823 
2824 	/* No operation if the region is already registered as PMEM */
2825 	is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2826 				IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2827 	if (is_pmem == REGION_INTERSECTS)
2828 		return 0;
2829 
2830 	res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2831 	if (!res)
2832 		return -ENOMEM;
2833 
2834 	res->name = "Persistent Memory";
2835 	res->start = nd_res->start;
2836 	res->end = nd_res->end;
2837 	res->flags = IORESOURCE_MEM;
2838 	res->desc = IORES_DESC_PERSISTENT_MEMORY;
2839 
2840 	ret = insert_resource(&iomem_resource, res);
2841 	if (ret)
2842 		return ret;
2843 
2844 	ret = devm_add_action_or_reset(acpi_desc->dev,
2845 					acpi_nfit_remove_resource,
2846 					res);
2847 	if (ret)
2848 		return ret;
2849 
2850 	return 0;
2851 }
2852 
2853 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2854 		struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2855 		struct acpi_nfit_memory_map *memdev,
2856 		struct nfit_spa *nfit_spa)
2857 {
2858 	struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2859 			memdev->device_handle);
2860 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2861 	struct nd_blk_region_desc *ndbr_desc;
2862 	struct nfit_mem *nfit_mem;
2863 	int rc;
2864 
2865 	if (!nvdimm) {
2866 		dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2867 				spa->range_index, memdev->device_handle);
2868 		return -ENODEV;
2869 	}
2870 
2871 	mapping->nvdimm = nvdimm;
2872 	switch (nfit_spa_type(spa)) {
2873 	case NFIT_SPA_PM:
2874 	case NFIT_SPA_VOLATILE:
2875 		mapping->start = memdev->address;
2876 		mapping->size = memdev->region_size;
2877 		break;
2878 	case NFIT_SPA_DCR:
2879 		nfit_mem = nvdimm_provider_data(nvdimm);
2880 		if (!nfit_mem || !nfit_mem->bdw) {
2881 			dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2882 					spa->range_index, nvdimm_name(nvdimm));
2883 			break;
2884 		}
2885 
2886 		mapping->size = nfit_mem->bdw->capacity;
2887 		mapping->start = nfit_mem->bdw->start_address;
2888 		ndr_desc->num_lanes = nfit_mem->bdw->windows;
2889 		ndr_desc->mapping = mapping;
2890 		ndr_desc->num_mappings = 1;
2891 		ndbr_desc = to_blk_region_desc(ndr_desc);
2892 		ndbr_desc->enable = acpi_nfit_blk_region_enable;
2893 		ndbr_desc->do_io = acpi_desc->blk_do_io;
2894 		rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2895 		if (rc)
2896 			return rc;
2897 		nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2898 				ndr_desc);
2899 		if (!nfit_spa->nd_region)
2900 			return -ENOMEM;
2901 		break;
2902 	}
2903 
2904 	return 0;
2905 }
2906 
2907 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2908 {
2909 	return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2910 		nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2911 		nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2912 		nfit_spa_type(spa) == NFIT_SPA_PCD);
2913 }
2914 
2915 static bool nfit_spa_is_volatile(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_VOLATILE);
2920 }
2921 
2922 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2923 		struct nfit_spa *nfit_spa)
2924 {
2925 	static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2926 	struct acpi_nfit_system_address *spa = nfit_spa->spa;
2927 	struct nd_blk_region_desc ndbr_desc;
2928 	struct nd_region_desc *ndr_desc;
2929 	struct nfit_memdev *nfit_memdev;
2930 	struct nvdimm_bus *nvdimm_bus;
2931 	struct resource res;
2932 	int count = 0, rc;
2933 
2934 	if (nfit_spa->nd_region)
2935 		return 0;
2936 
2937 	if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2938 		dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2939 		return 0;
2940 	}
2941 
2942 	memset(&res, 0, sizeof(res));
2943 	memset(&mappings, 0, sizeof(mappings));
2944 	memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2945 	res.start = spa->address;
2946 	res.end = res.start + spa->length - 1;
2947 	ndr_desc = &ndbr_desc.ndr_desc;
2948 	ndr_desc->res = &res;
2949 	ndr_desc->provider_data = nfit_spa;
2950 	ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2951 	if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2952 		ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2953 						spa->proximity_domain);
2954 		ndr_desc->target_node = acpi_map_pxm_to_node(
2955 				spa->proximity_domain);
2956 	} else {
2957 		ndr_desc->numa_node = NUMA_NO_NODE;
2958 		ndr_desc->target_node = NUMA_NO_NODE;
2959 	}
2960 
2961 	/*
2962 	 * Persistence domain bits are hierarchical, if
2963 	 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2964 	 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2965 	 */
2966 	if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2967 		set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2968 	else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2969 		set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2970 
2971 	list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2972 		struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2973 		struct nd_mapping_desc *mapping;
2974 
2975 		if (memdev->range_index != spa->range_index)
2976 			continue;
2977 		if (count >= ND_MAX_MAPPINGS) {
2978 			dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2979 					spa->range_index, ND_MAX_MAPPINGS);
2980 			return -ENXIO;
2981 		}
2982 		mapping = &mappings[count++];
2983 		rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2984 				memdev, nfit_spa);
2985 		if (rc)
2986 			goto out;
2987 	}
2988 
2989 	ndr_desc->mapping = mappings;
2990 	ndr_desc->num_mappings = count;
2991 	rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2992 	if (rc)
2993 		goto out;
2994 
2995 	nvdimm_bus = acpi_desc->nvdimm_bus;
2996 	if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2997 		rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2998 		if (rc) {
2999 			dev_warn(acpi_desc->dev,
3000 				"failed to insert pmem resource to iomem: %d\n",
3001 				rc);
3002 			goto out;
3003 		}
3004 
3005 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3006 				ndr_desc);
3007 		if (!nfit_spa->nd_region)
3008 			rc = -ENOMEM;
3009 	} else if (nfit_spa_is_volatile(spa)) {
3010 		nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
3011 				ndr_desc);
3012 		if (!nfit_spa->nd_region)
3013 			rc = -ENOMEM;
3014 	} else if (nfit_spa_is_virtual(spa)) {
3015 		nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
3016 				ndr_desc);
3017 		if (!nfit_spa->nd_region)
3018 			rc = -ENOMEM;
3019 	}
3020 
3021  out:
3022 	if (rc)
3023 		dev_err(acpi_desc->dev, "failed to register spa range %d\n",
3024 				nfit_spa->spa->range_index);
3025 	return rc;
3026 }
3027 
3028 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
3029 {
3030 	struct device *dev = acpi_desc->dev;
3031 	struct nd_cmd_ars_status *ars_status;
3032 
3033 	if (acpi_desc->ars_status) {
3034 		memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3035 		return 0;
3036 	}
3037 
3038 	ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
3039 	if (!ars_status)
3040 		return -ENOMEM;
3041 	acpi_desc->ars_status = ars_status;
3042 	return 0;
3043 }
3044 
3045 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
3046 {
3047 	int rc;
3048 
3049 	if (ars_status_alloc(acpi_desc))
3050 		return -ENOMEM;
3051 
3052 	rc = ars_get_status(acpi_desc);
3053 
3054 	if (rc < 0 && rc != -ENOSPC)
3055 		return rc;
3056 
3057 	if (ars_status_process_records(acpi_desc))
3058 		dev_err(acpi_desc->dev, "Failed to process ARS records\n");
3059 
3060 	return rc;
3061 }
3062 
3063 static int ars_register(struct acpi_nfit_desc *acpi_desc,
3064 		struct nfit_spa *nfit_spa)
3065 {
3066 	int rc;
3067 
3068 	if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3069 		return acpi_nfit_register_region(acpi_desc, nfit_spa);
3070 
3071 	set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3072 	if (!no_init_ars)
3073 		set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
3074 
3075 	switch (acpi_nfit_query_poison(acpi_desc)) {
3076 	case 0:
3077 	case -ENOSPC:
3078 	case -EAGAIN:
3079 		rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
3080 		/* shouldn't happen, try again later */
3081 		if (rc == -EBUSY)
3082 			break;
3083 		if (rc) {
3084 			set_bit(ARS_FAILED, &nfit_spa->ars_state);
3085 			break;
3086 		}
3087 		clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
3088 		rc = acpi_nfit_query_poison(acpi_desc);
3089 		if (rc)
3090 			break;
3091 		acpi_desc->scrub_spa = nfit_spa;
3092 		ars_complete(acpi_desc, nfit_spa);
3093 		/*
3094 		 * If ars_complete() says we didn't complete the
3095 		 * short scrub, we'll try again with a long
3096 		 * request.
3097 		 */
3098 		acpi_desc->scrub_spa = NULL;
3099 		break;
3100 	case -EBUSY:
3101 	case -ENOMEM:
3102 		/*
3103 		 * BIOS was using ARS, wait for it to complete (or
3104 		 * resources to become available) and then perform our
3105 		 * own scrubs.
3106 		 */
3107 		break;
3108 	default:
3109 		set_bit(ARS_FAILED, &nfit_spa->ars_state);
3110 		break;
3111 	}
3112 
3113 	return acpi_nfit_register_region(acpi_desc, nfit_spa);
3114 }
3115 
3116 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
3117 {
3118 	struct nfit_spa *nfit_spa;
3119 
3120 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3121 		if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3122 			continue;
3123 		ars_complete(acpi_desc, nfit_spa);
3124 	}
3125 }
3126 
3127 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
3128 		int query_rc)
3129 {
3130 	unsigned int tmo = acpi_desc->scrub_tmo;
3131 	struct device *dev = acpi_desc->dev;
3132 	struct nfit_spa *nfit_spa;
3133 
3134 	lockdep_assert_held(&acpi_desc->init_mutex);
3135 
3136 	if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
3137 		return 0;
3138 
3139 	if (query_rc == -EBUSY) {
3140 		dev_dbg(dev, "ARS: ARS busy\n");
3141 		return min(30U * 60U, tmo * 2);
3142 	}
3143 	if (query_rc == -ENOSPC) {
3144 		dev_dbg(dev, "ARS: ARS continue\n");
3145 		ars_continue(acpi_desc);
3146 		return 1;
3147 	}
3148 	if (query_rc && query_rc != -EAGAIN) {
3149 		unsigned long long addr, end;
3150 
3151 		addr = acpi_desc->ars_status->address;
3152 		end = addr + acpi_desc->ars_status->length;
3153 		dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
3154 				query_rc);
3155 	}
3156 
3157 	ars_complete_all(acpi_desc);
3158 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3159 		enum nfit_ars_state req_type;
3160 		int rc;
3161 
3162 		if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3163 			continue;
3164 
3165 		/* prefer short ARS requests first */
3166 		if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
3167 			req_type = ARS_REQ_SHORT;
3168 		else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
3169 			req_type = ARS_REQ_LONG;
3170 		else
3171 			continue;
3172 		rc = ars_start(acpi_desc, nfit_spa, req_type);
3173 
3174 		dev = nd_region_dev(nfit_spa->nd_region);
3175 		dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
3176 				nfit_spa->spa->range_index,
3177 				req_type == ARS_REQ_SHORT ? "short" : "long",
3178 				rc);
3179 		/*
3180 		 * Hmm, we raced someone else starting ARS? Try again in
3181 		 * a bit.
3182 		 */
3183 		if (rc == -EBUSY)
3184 			return 1;
3185 		if (rc == 0) {
3186 			dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
3187 					"scrub start while range %d active\n",
3188 					acpi_desc->scrub_spa->spa->range_index);
3189 			clear_bit(req_type, &nfit_spa->ars_state);
3190 			acpi_desc->scrub_spa = nfit_spa;
3191 			/*
3192 			 * Consider this spa last for future scrub
3193 			 * requests
3194 			 */
3195 			list_move_tail(&nfit_spa->list, &acpi_desc->spas);
3196 			return 1;
3197 		}
3198 
3199 		dev_err(dev, "ARS: range %d ARS failed (%d)\n",
3200 				nfit_spa->spa->range_index, rc);
3201 		set_bit(ARS_FAILED, &nfit_spa->ars_state);
3202 	}
3203 	return 0;
3204 }
3205 
3206 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
3207 {
3208 	lockdep_assert_held(&acpi_desc->init_mutex);
3209 
3210 	set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3211 	/* note this should only be set from within the workqueue */
3212 	if (tmo)
3213 		acpi_desc->scrub_tmo = tmo;
3214 	queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
3215 }
3216 
3217 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
3218 {
3219 	__sched_ars(acpi_desc, 0);
3220 }
3221 
3222 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
3223 {
3224 	lockdep_assert_held(&acpi_desc->init_mutex);
3225 
3226 	clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
3227 	acpi_desc->scrub_count++;
3228 	if (acpi_desc->scrub_count_state)
3229 		sysfs_notify_dirent(acpi_desc->scrub_count_state);
3230 }
3231 
3232 static void acpi_nfit_scrub(struct work_struct *work)
3233 {
3234 	struct acpi_nfit_desc *acpi_desc;
3235 	unsigned int tmo;
3236 	int query_rc;
3237 
3238 	acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
3239 	mutex_lock(&acpi_desc->init_mutex);
3240 	query_rc = acpi_nfit_query_poison(acpi_desc);
3241 	tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
3242 	if (tmo)
3243 		__sched_ars(acpi_desc, tmo);
3244 	else
3245 		notify_ars_done(acpi_desc);
3246 	memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
3247 	clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
3248 	mutex_unlock(&acpi_desc->init_mutex);
3249 }
3250 
3251 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
3252 		struct nfit_spa *nfit_spa)
3253 {
3254 	int type = nfit_spa_type(nfit_spa->spa);
3255 	struct nd_cmd_ars_cap ars_cap;
3256 	int rc;
3257 
3258 	set_bit(ARS_FAILED, &nfit_spa->ars_state);
3259 	memset(&ars_cap, 0, sizeof(ars_cap));
3260 	rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
3261 	if (rc < 0)
3262 		return;
3263 	/* check that the supported scrub types match the spa type */
3264 	if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3265 				& ND_ARS_VOLATILE) == 0)
3266 		return;
3267 	if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3268 				& ND_ARS_PERSISTENT) == 0)
3269 		return;
3270 
3271 	nfit_spa->max_ars = ars_cap.max_ars_out;
3272 	nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3273 	acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3274 	clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3275 }
3276 
3277 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3278 {
3279 	struct nfit_spa *nfit_spa;
3280 	int rc;
3281 
3282 	set_bit(ARS_VALID, &acpi_desc->scrub_flags);
3283 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3284 		switch (nfit_spa_type(nfit_spa->spa)) {
3285 		case NFIT_SPA_VOLATILE:
3286 		case NFIT_SPA_PM:
3287 			acpi_nfit_init_ars(acpi_desc, nfit_spa);
3288 			break;
3289 		}
3290 	}
3291 
3292 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3293 		switch (nfit_spa_type(nfit_spa->spa)) {
3294 		case NFIT_SPA_VOLATILE:
3295 		case NFIT_SPA_PM:
3296 			/* register regions and kick off initial ARS run */
3297 			rc = ars_register(acpi_desc, nfit_spa);
3298 			if (rc)
3299 				return rc;
3300 			break;
3301 		case NFIT_SPA_BDW:
3302 			/* nothing to register */
3303 			break;
3304 		case NFIT_SPA_DCR:
3305 		case NFIT_SPA_VDISK:
3306 		case NFIT_SPA_VCD:
3307 		case NFIT_SPA_PDISK:
3308 		case NFIT_SPA_PCD:
3309 			/* register known regions that don't support ARS */
3310 			rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3311 			if (rc)
3312 				return rc;
3313 			break;
3314 		default:
3315 			/* don't register unknown regions */
3316 			break;
3317 		}
3318 
3319 	sched_ars(acpi_desc);
3320 	return 0;
3321 }
3322 
3323 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3324 		struct nfit_table_prev *prev)
3325 {
3326 	struct device *dev = acpi_desc->dev;
3327 
3328 	if (!list_empty(&prev->spas) ||
3329 			!list_empty(&prev->memdevs) ||
3330 			!list_empty(&prev->dcrs) ||
3331 			!list_empty(&prev->bdws) ||
3332 			!list_empty(&prev->idts) ||
3333 			!list_empty(&prev->flushes)) {
3334 		dev_err(dev, "new nfit deletes entries (unsupported)\n");
3335 		return -ENXIO;
3336 	}
3337 	return 0;
3338 }
3339 
3340 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3341 {
3342 	struct device *dev = acpi_desc->dev;
3343 	struct kernfs_node *nfit;
3344 	struct device *bus_dev;
3345 
3346 	if (!ars_supported(acpi_desc->nvdimm_bus))
3347 		return 0;
3348 
3349 	bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3350 	nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3351 	if (!nfit) {
3352 		dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3353 		return -ENODEV;
3354 	}
3355 	acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3356 	sysfs_put(nfit);
3357 	if (!acpi_desc->scrub_count_state) {
3358 		dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3359 		return -ENODEV;
3360 	}
3361 
3362 	return 0;
3363 }
3364 
3365 static void acpi_nfit_unregister(void *data)
3366 {
3367 	struct acpi_nfit_desc *acpi_desc = data;
3368 
3369 	nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3370 }
3371 
3372 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3373 {
3374 	struct device *dev = acpi_desc->dev;
3375 	struct nfit_table_prev prev;
3376 	const void *end;
3377 	int rc;
3378 
3379 	if (!acpi_desc->nvdimm_bus) {
3380 		acpi_nfit_init_dsms(acpi_desc);
3381 
3382 		acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3383 				&acpi_desc->nd_desc);
3384 		if (!acpi_desc->nvdimm_bus)
3385 			return -ENOMEM;
3386 
3387 		rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3388 				acpi_desc);
3389 		if (rc)
3390 			return rc;
3391 
3392 		rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3393 		if (rc)
3394 			return rc;
3395 
3396 		/* register this acpi_desc for mce notifications */
3397 		mutex_lock(&acpi_desc_lock);
3398 		list_add_tail(&acpi_desc->list, &acpi_descs);
3399 		mutex_unlock(&acpi_desc_lock);
3400 	}
3401 
3402 	mutex_lock(&acpi_desc->init_mutex);
3403 
3404 	INIT_LIST_HEAD(&prev.spas);
3405 	INIT_LIST_HEAD(&prev.memdevs);
3406 	INIT_LIST_HEAD(&prev.dcrs);
3407 	INIT_LIST_HEAD(&prev.bdws);
3408 	INIT_LIST_HEAD(&prev.idts);
3409 	INIT_LIST_HEAD(&prev.flushes);
3410 
3411 	list_cut_position(&prev.spas, &acpi_desc->spas,
3412 				acpi_desc->spas.prev);
3413 	list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3414 				acpi_desc->memdevs.prev);
3415 	list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3416 				acpi_desc->dcrs.prev);
3417 	list_cut_position(&prev.bdws, &acpi_desc->bdws,
3418 				acpi_desc->bdws.prev);
3419 	list_cut_position(&prev.idts, &acpi_desc->idts,
3420 				acpi_desc->idts.prev);
3421 	list_cut_position(&prev.flushes, &acpi_desc->flushes,
3422 				acpi_desc->flushes.prev);
3423 
3424 	end = data + sz;
3425 	while (!IS_ERR_OR_NULL(data))
3426 		data = add_table(acpi_desc, &prev, data, end);
3427 
3428 	if (IS_ERR(data)) {
3429 		dev_dbg(dev, "nfit table parsing error: %ld\n",	PTR_ERR(data));
3430 		rc = PTR_ERR(data);
3431 		goto out_unlock;
3432 	}
3433 
3434 	rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3435 	if (rc)
3436 		goto out_unlock;
3437 
3438 	rc = nfit_mem_init(acpi_desc);
3439 	if (rc)
3440 		goto out_unlock;
3441 
3442 	rc = acpi_nfit_register_dimms(acpi_desc);
3443 	if (rc)
3444 		goto out_unlock;
3445 
3446 	rc = acpi_nfit_register_regions(acpi_desc);
3447 
3448  out_unlock:
3449 	mutex_unlock(&acpi_desc->init_mutex);
3450 	return rc;
3451 }
3452 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3453 
3454 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3455 {
3456 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3457 	struct device *dev = acpi_desc->dev;
3458 
3459 	/* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3460 	nfit_device_lock(dev);
3461 	nfit_device_unlock(dev);
3462 
3463 	/* Bounce the init_mutex to complete initial registration */
3464 	mutex_lock(&acpi_desc->init_mutex);
3465 	mutex_unlock(&acpi_desc->init_mutex);
3466 
3467 	return 0;
3468 }
3469 
3470 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3471 		struct nvdimm *nvdimm, unsigned int cmd)
3472 {
3473 	struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3474 
3475 	if (nvdimm)
3476 		return 0;
3477 	if (cmd != ND_CMD_ARS_START)
3478 		return 0;
3479 
3480 	/*
3481 	 * The kernel and userspace may race to initiate a scrub, but
3482 	 * the scrub thread is prepared to lose that initial race.  It
3483 	 * just needs guarantees that any ARS it initiates are not
3484 	 * interrupted by any intervening start requests from userspace.
3485 	 */
3486 	if (work_busy(&acpi_desc->dwork.work))
3487 		return -EBUSY;
3488 
3489 	return 0;
3490 }
3491 
3492 /* prevent security commands from being issued via ioctl */
3493 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3494 		struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3495 {
3496 	struct nd_cmd_pkg *call_pkg = buf;
3497 	unsigned int func;
3498 
3499 	if (nvdimm && cmd == ND_CMD_CALL &&
3500 			call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3501 		func = call_pkg->nd_command;
3502 		if ((1 << func) & NVDIMM_INTEL_SECURITY_CMDMASK)
3503 			return -EOPNOTSUPP;
3504 	}
3505 
3506 	return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3507 }
3508 
3509 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3510 		enum nfit_ars_state req_type)
3511 {
3512 	struct device *dev = acpi_desc->dev;
3513 	int scheduled = 0, busy = 0;
3514 	struct nfit_spa *nfit_spa;
3515 
3516 	mutex_lock(&acpi_desc->init_mutex);
3517 	if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3518 		mutex_unlock(&acpi_desc->init_mutex);
3519 		return 0;
3520 	}
3521 
3522 	list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3523 		int type = nfit_spa_type(nfit_spa->spa);
3524 
3525 		if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3526 			continue;
3527 		if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3528 			continue;
3529 
3530 		if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3531 			busy++;
3532 		else
3533 			scheduled++;
3534 	}
3535 	if (scheduled) {
3536 		sched_ars(acpi_desc);
3537 		dev_dbg(dev, "ars_scan triggered\n");
3538 	}
3539 	mutex_unlock(&acpi_desc->init_mutex);
3540 
3541 	if (scheduled)
3542 		return 0;
3543 	if (busy)
3544 		return -EBUSY;
3545 	return -ENOTTY;
3546 }
3547 
3548 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3549 {
3550 	struct nvdimm_bus_descriptor *nd_desc;
3551 
3552 	dev_set_drvdata(dev, acpi_desc);
3553 	acpi_desc->dev = dev;
3554 	acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3555 	nd_desc = &acpi_desc->nd_desc;
3556 	nd_desc->provider_name = "ACPI.NFIT";
3557 	nd_desc->module = THIS_MODULE;
3558 	nd_desc->ndctl = acpi_nfit_ctl;
3559 	nd_desc->flush_probe = acpi_nfit_flush_probe;
3560 	nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3561 	nd_desc->attr_groups = acpi_nfit_attribute_groups;
3562 
3563 	INIT_LIST_HEAD(&acpi_desc->spas);
3564 	INIT_LIST_HEAD(&acpi_desc->dcrs);
3565 	INIT_LIST_HEAD(&acpi_desc->bdws);
3566 	INIT_LIST_HEAD(&acpi_desc->idts);
3567 	INIT_LIST_HEAD(&acpi_desc->flushes);
3568 	INIT_LIST_HEAD(&acpi_desc->memdevs);
3569 	INIT_LIST_HEAD(&acpi_desc->dimms);
3570 	INIT_LIST_HEAD(&acpi_desc->list);
3571 	mutex_init(&acpi_desc->init_mutex);
3572 	acpi_desc->scrub_tmo = 1;
3573 	INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3574 }
3575 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3576 
3577 static void acpi_nfit_put_table(void *table)
3578 {
3579 	acpi_put_table(table);
3580 }
3581 
3582 void acpi_nfit_shutdown(void *data)
3583 {
3584 	struct acpi_nfit_desc *acpi_desc = data;
3585 	struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3586 
3587 	/*
3588 	 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3589 	 * race teardown
3590 	 */
3591 	mutex_lock(&acpi_desc_lock);
3592 	list_del(&acpi_desc->list);
3593 	mutex_unlock(&acpi_desc_lock);
3594 
3595 	mutex_lock(&acpi_desc->init_mutex);
3596 	set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3597 	cancel_delayed_work_sync(&acpi_desc->dwork);
3598 	mutex_unlock(&acpi_desc->init_mutex);
3599 
3600 	/*
3601 	 * Bounce the nvdimm bus lock to make sure any in-flight
3602 	 * acpi_nfit_ars_rescan() submissions have had a chance to
3603 	 * either submit or see ->cancel set.
3604 	 */
3605 	nfit_device_lock(bus_dev);
3606 	nfit_device_unlock(bus_dev);
3607 
3608 	flush_workqueue(nfit_wq);
3609 }
3610 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3611 
3612 static int acpi_nfit_add(struct acpi_device *adev)
3613 {
3614 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3615 	struct acpi_nfit_desc *acpi_desc;
3616 	struct device *dev = &adev->dev;
3617 	struct acpi_table_header *tbl;
3618 	acpi_status status = AE_OK;
3619 	acpi_size sz;
3620 	int rc = 0;
3621 
3622 	status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3623 	if (ACPI_FAILURE(status)) {
3624 		/* The NVDIMM root device allows OS to trigger enumeration of
3625 		 * NVDIMMs through NFIT at boot time and re-enumeration at
3626 		 * root level via the _FIT method during runtime.
3627 		 * This is ok to return 0 here, we could have an nvdimm
3628 		 * hotplugged later and evaluate _FIT method which returns
3629 		 * data in the format of a series of NFIT Structures.
3630 		 */
3631 		dev_dbg(dev, "failed to find NFIT at startup\n");
3632 		return 0;
3633 	}
3634 
3635 	rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3636 	if (rc)
3637 		return rc;
3638 	sz = tbl->length;
3639 
3640 	acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3641 	if (!acpi_desc)
3642 		return -ENOMEM;
3643 	acpi_nfit_desc_init(acpi_desc, &adev->dev);
3644 
3645 	/* Save the acpi header for exporting the revision via sysfs */
3646 	acpi_desc->acpi_header = *tbl;
3647 
3648 	/* Evaluate _FIT and override with that if present */
3649 	status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3650 	if (ACPI_SUCCESS(status) && buf.length > 0) {
3651 		union acpi_object *obj = buf.pointer;
3652 
3653 		if (obj->type == ACPI_TYPE_BUFFER)
3654 			rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3655 					obj->buffer.length);
3656 		else
3657 			dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3658 				(int) obj->type);
3659 		kfree(buf.pointer);
3660 	} else
3661 		/* skip over the lead-in header table */
3662 		rc = acpi_nfit_init(acpi_desc, (void *) tbl
3663 				+ sizeof(struct acpi_table_nfit),
3664 				sz - sizeof(struct acpi_table_nfit));
3665 
3666 	if (rc)
3667 		return rc;
3668 	return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3669 }
3670 
3671 static int acpi_nfit_remove(struct acpi_device *adev)
3672 {
3673 	/* see acpi_nfit_unregister */
3674 	return 0;
3675 }
3676 
3677 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3678 {
3679 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3680 	struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3681 	union acpi_object *obj;
3682 	acpi_status status;
3683 	int ret;
3684 
3685 	if (!dev->driver) {
3686 		/* dev->driver may be null if we're being removed */
3687 		dev_dbg(dev, "no driver found for dev\n");
3688 		return;
3689 	}
3690 
3691 	if (!acpi_desc) {
3692 		acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3693 		if (!acpi_desc)
3694 			return;
3695 		acpi_nfit_desc_init(acpi_desc, dev);
3696 	} else {
3697 		/*
3698 		 * Finish previous registration before considering new
3699 		 * regions.
3700 		 */
3701 		flush_workqueue(nfit_wq);
3702 	}
3703 
3704 	/* Evaluate _FIT */
3705 	status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3706 	if (ACPI_FAILURE(status)) {
3707 		dev_err(dev, "failed to evaluate _FIT\n");
3708 		return;
3709 	}
3710 
3711 	obj = buf.pointer;
3712 	if (obj->type == ACPI_TYPE_BUFFER) {
3713 		ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3714 				obj->buffer.length);
3715 		if (ret)
3716 			dev_err(dev, "failed to merge updated NFIT\n");
3717 	} else
3718 		dev_err(dev, "Invalid _FIT\n");
3719 	kfree(buf.pointer);
3720 }
3721 
3722 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3723 {
3724 	struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3725 
3726 	if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3727 		acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3728 	else
3729 		acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3730 }
3731 
3732 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3733 {
3734 	dev_dbg(dev, "event: 0x%x\n", event);
3735 
3736 	switch (event) {
3737 	case NFIT_NOTIFY_UPDATE:
3738 		return acpi_nfit_update_notify(dev, handle);
3739 	case NFIT_NOTIFY_UC_MEMORY_ERROR:
3740 		return acpi_nfit_uc_error_notify(dev, handle);
3741 	default:
3742 		return;
3743 	}
3744 }
3745 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3746 
3747 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3748 {
3749 	nfit_device_lock(&adev->dev);
3750 	__acpi_nfit_notify(&adev->dev, adev->handle, event);
3751 	nfit_device_unlock(&adev->dev);
3752 }
3753 
3754 static const struct acpi_device_id acpi_nfit_ids[] = {
3755 	{ "ACPI0012", 0 },
3756 	{ "", 0 },
3757 };
3758 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3759 
3760 static struct acpi_driver acpi_nfit_driver = {
3761 	.name = KBUILD_MODNAME,
3762 	.ids = acpi_nfit_ids,
3763 	.ops = {
3764 		.add = acpi_nfit_add,
3765 		.remove = acpi_nfit_remove,
3766 		.notify = acpi_nfit_notify,
3767 	},
3768 };
3769 
3770 static __init int nfit_init(void)
3771 {
3772 	int ret;
3773 
3774 	BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3775 	BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3776 	BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3777 	BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3778 	BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3779 	BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3780 	BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3781 	BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3782 
3783 	guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3784 	guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3785 	guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3786 	guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3787 	guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3788 	guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3789 	guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3790 	guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3791 	guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3792 	guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3793 	guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3794 	guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3795 	guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3796 	guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3797 
3798 	nfit_wq = create_singlethread_workqueue("nfit");
3799 	if (!nfit_wq)
3800 		return -ENOMEM;
3801 
3802 	nfit_mce_register();
3803 	ret = acpi_bus_register_driver(&acpi_nfit_driver);
3804 	if (ret) {
3805 		nfit_mce_unregister();
3806 		destroy_workqueue(nfit_wq);
3807 	}
3808 
3809 	return ret;
3810 
3811 }
3812 
3813 static __exit void nfit_exit(void)
3814 {
3815 	nfit_mce_unregister();
3816 	acpi_bus_unregister_driver(&acpi_nfit_driver);
3817 	destroy_workqueue(nfit_wq);
3818 	WARN_ON(!list_empty(&acpi_descs));
3819 }
3820 
3821 module_init(nfit_init);
3822 module_exit(nfit_exit);
3823 MODULE_LICENSE("GPL v2");
3824 MODULE_AUTHOR("Intel Corporation");
3825