1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
5 #include <linux/list_sort.h>
6 #include <linux/libnvdimm.h>
7 #include <linux/module.h>
8 #include <linux/nospec.h>
9 #include <linux/mutex.h>
10 #include <linux/ndctl.h>
11 #include <linux/sysfs.h>
12 #include <linux/delay.h>
13 #include <linux/list.h>
14 #include <linux/acpi.h>
15 #include <linux/sort.h>
18 #include <asm/cacheflush.h>
19 #include <acpi/nfit.h>
24 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
27 #include <linux/io-64-nonatomic-hi-lo.h>
29 static bool force_enable_dimms;
30 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
31 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
33 static bool disable_vendor_specific;
34 module_param(disable_vendor_specific, bool, S_IRUGO);
35 MODULE_PARM_DESC(disable_vendor_specific,
36 "Limit commands to the publicly specified set");
38 static unsigned long override_dsm_mask;
39 module_param(override_dsm_mask, ulong, S_IRUGO);
40 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
42 static int default_dsm_family = -1;
43 module_param(default_dsm_family, int, S_IRUGO);
44 MODULE_PARM_DESC(default_dsm_family,
45 "Try this DSM type first when identifying NVDIMM family");
47 static bool no_init_ars;
48 module_param(no_init_ars, bool, 0644);
49 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
51 static bool force_labels;
52 module_param(force_labels, bool, 0444);
53 MODULE_PARM_DESC(force_labels, "Opt-in to labels despite missing methods");
55 LIST_HEAD(acpi_descs);
56 DEFINE_MUTEX(acpi_desc_lock);
58 static struct workqueue_struct *nfit_wq;
60 struct nfit_table_prev {
61 struct list_head spas;
62 struct list_head memdevs;
63 struct list_head dcrs;
64 struct list_head bdws;
65 struct list_head idts;
66 struct list_head flushes;
69 static guid_t nfit_uuid[NFIT_UUID_MAX];
71 const guid_t *to_nfit_uuid(enum nfit_uuids id)
73 return &nfit_uuid[id];
75 EXPORT_SYMBOL(to_nfit_uuid);
77 static const guid_t *to_nfit_bus_uuid(int family)
79 if (WARN_ONCE(family == NVDIMM_BUS_FAMILY_NFIT,
80 "only secondary bus families can be translated\n"))
83 * The index of bus UUIDs starts immediately following the last
86 return to_nfit_uuid(family + NVDIMM_FAMILY_MAX);
89 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
91 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
94 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
97 if (!nd_desc->provider_name
98 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
101 return to_acpi_device(acpi_desc->dev);
104 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
106 struct nd_cmd_clear_error *clear_err;
107 struct nd_cmd_ars_status *ars_status;
112 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
119 /* No supported scan types for this range */
120 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
121 if ((status >> 16 & flags) == 0)
124 case ND_CMD_ARS_START:
125 /* ARS is in progress */
126 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
133 case ND_CMD_ARS_STATUS:
138 /* Check extended status (Upper two bytes) */
139 if (status == NFIT_ARS_STATUS_DONE)
142 /* ARS is in progress */
143 if (status == NFIT_ARS_STATUS_BUSY)
146 /* No ARS performed for the current boot */
147 if (status == NFIT_ARS_STATUS_NONE)
151 * ARS interrupted, either we overflowed or some other
152 * agent wants the scan to stop. If we didn't overflow
153 * then just continue with the returned results.
155 if (status == NFIT_ARS_STATUS_INTR) {
156 if (ars_status->out_length >= 40 && (ars_status->flags
157 & NFIT_ARS_F_OVERFLOW))
166 case ND_CMD_CLEAR_ERROR:
170 if (!clear_err->cleared)
172 if (clear_err->length > clear_err->cleared)
173 return clear_err->cleared;
179 /* all other non-zero status results in an error */
185 #define ACPI_LABELS_LOCKED 3
187 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
190 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
193 case ND_CMD_GET_CONFIG_SIZE:
195 * In the _LSI, _LSR, _LSW case the locked status is
196 * communicated via the read/write commands
198 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
201 if (status >> 16 & ND_CONFIG_LOCKED)
204 case ND_CMD_GET_CONFIG_DATA:
205 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
206 && status == ACPI_LABELS_LOCKED)
209 case ND_CMD_SET_CONFIG_DATA:
210 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
211 && status == ACPI_LABELS_LOCKED)
218 /* all other non-zero status results in an error */
224 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
228 return xlat_bus_status(buf, cmd, status);
229 return xlat_nvdimm_status(nvdimm, buf, cmd, status);
232 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
233 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
238 union acpi_object *buf = NULL;
240 if (pkg->type != ACPI_TYPE_PACKAGE) {
241 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
246 for (i = 0; i < pkg->package.count; i++) {
247 union acpi_object *obj = &pkg->package.elements[i];
249 if (obj->type == ACPI_TYPE_INTEGER)
251 else if (obj->type == ACPI_TYPE_BUFFER)
252 size += obj->buffer.length;
254 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
260 buf = ACPI_ALLOCATE(sizeof(*buf) + size);
265 buf->type = ACPI_TYPE_BUFFER;
266 buf->buffer.length = size;
267 buf->buffer.pointer = dst;
268 for (i = 0; i < pkg->package.count; i++) {
269 union acpi_object *obj = &pkg->package.elements[i];
271 if (obj->type == ACPI_TYPE_INTEGER) {
272 memcpy(dst, &obj->integer.value, 4);
274 } else if (obj->type == ACPI_TYPE_BUFFER) {
275 memcpy(dst, obj->buffer.pointer, obj->buffer.length);
276 dst += obj->buffer.length;
284 static union acpi_object *int_to_buf(union acpi_object *integer)
286 union acpi_object *buf = NULL;
289 if (integer->type != ACPI_TYPE_INTEGER) {
290 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
295 buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
300 buf->type = ACPI_TYPE_BUFFER;
301 buf->buffer.length = 4;
302 buf->buffer.pointer = dst;
303 memcpy(dst, &integer->integer.value, 4);
309 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
313 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
314 struct acpi_object_list input = {
316 .pointer = (union acpi_object []) {
318 .integer.type = ACPI_TYPE_INTEGER,
319 .integer.value = offset,
322 .integer.type = ACPI_TYPE_INTEGER,
323 .integer.value = len,
326 .buffer.type = ACPI_TYPE_BUFFER,
327 .buffer.pointer = data,
328 .buffer.length = len,
333 rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
334 if (ACPI_FAILURE(rc))
336 return int_to_buf(buf.pointer);
339 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
343 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
344 struct acpi_object_list input = {
346 .pointer = (union acpi_object []) {
348 .integer.type = ACPI_TYPE_INTEGER,
349 .integer.value = offset,
352 .integer.type = ACPI_TYPE_INTEGER,
353 .integer.value = len,
358 rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
359 if (ACPI_FAILURE(rc))
361 return pkg_to_buf(buf.pointer);
364 static union acpi_object *acpi_label_info(acpi_handle handle)
367 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
369 rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
370 if (ACPI_FAILURE(rc))
372 return pkg_to_buf(buf.pointer);
375 static u8 nfit_dsm_revid(unsigned family, unsigned func)
377 static const u8 revid_table[NVDIMM_FAMILY_MAX+1][NVDIMM_CMD_MAX+1] = {
378 [NVDIMM_FAMILY_INTEL] = {
379 [NVDIMM_INTEL_GET_MODES ...
380 NVDIMM_INTEL_FW_ACTIVATE_ARM] = 2,
385 if (family > NVDIMM_FAMILY_MAX)
387 if (func > NVDIMM_CMD_MAX)
389 id = revid_table[family][func];
391 return 1; /* default */
395 static bool payload_dumpable(struct nvdimm *nvdimm, unsigned int func)
397 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
399 if (nfit_mem && nfit_mem->family == NVDIMM_FAMILY_INTEL
400 && func >= NVDIMM_INTEL_GET_SECURITY_STATE
401 && func <= NVDIMM_INTEL_MASTER_SECURE_ERASE)
402 return IS_ENABLED(CONFIG_NFIT_SECURITY_DEBUG);
406 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
407 struct nd_cmd_pkg *call_pkg, int *family)
412 if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
415 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
416 if (call_pkg->nd_reserved2[i])
418 *family = call_pkg->nd_family;
419 return call_pkg->nd_command;
422 /* In the !call_pkg case, bus commands == bus functions */
426 /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
427 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
431 * Force function number validation to fail since 0 is never
432 * published as a valid function in dsm_mask.
437 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
438 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
440 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
441 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
442 union acpi_object in_obj, in_buf, *out_obj;
443 const struct nd_cmd_desc *desc = NULL;
444 struct device *dev = acpi_desc->dev;
445 struct nd_cmd_pkg *call_pkg = NULL;
446 const char *cmd_name, *dimm_name;
447 unsigned long cmd_mask, dsm_mask;
448 u32 offset, fw_status = 0;
457 if (cmd == ND_CMD_CALL) {
458 if (!buf || buf_len < sizeof(*call_pkg))
464 func = cmd_to_func(nfit_mem, cmd, call_pkg, &family);
469 struct acpi_device *adev = nfit_mem->adev;
474 dimm_name = nvdimm_name(nvdimm);
475 cmd_name = nvdimm_cmd_name(cmd);
476 cmd_mask = nvdimm_cmd_mask(nvdimm);
477 dsm_mask = nfit_mem->dsm_mask;
478 desc = nd_cmd_dimm_desc(cmd);
479 guid = to_nfit_uuid(nfit_mem->family);
480 handle = adev->handle;
482 struct acpi_device *adev = to_acpi_dev(acpi_desc);
484 cmd_name = nvdimm_bus_cmd_name(cmd);
485 cmd_mask = nd_desc->cmd_mask;
486 if (cmd == ND_CMD_CALL && call_pkg->nd_family) {
487 family = call_pkg->nd_family;
488 if (family > NVDIMM_BUS_FAMILY_MAX ||
489 !test_bit(family, &nd_desc->bus_family_mask))
491 family = array_index_nospec(family,
492 NVDIMM_BUS_FAMILY_MAX + 1);
493 dsm_mask = acpi_desc->family_dsm_mask[family];
494 guid = to_nfit_bus_uuid(family);
496 dsm_mask = acpi_desc->bus_dsm_mask;
497 guid = to_nfit_uuid(NFIT_DEV_BUS);
499 desc = nd_cmd_bus_desc(cmd);
500 handle = adev->handle;
504 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
508 * Check for a valid command. For ND_CMD_CALL, we also have to
509 * make sure that the DSM function is supported.
511 if (cmd == ND_CMD_CALL &&
512 (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask)))
514 else if (!test_bit(cmd, &cmd_mask))
517 in_obj.type = ACPI_TYPE_PACKAGE;
518 in_obj.package.count = 1;
519 in_obj.package.elements = &in_buf;
520 in_buf.type = ACPI_TYPE_BUFFER;
521 in_buf.buffer.pointer = buf;
522 in_buf.buffer.length = 0;
524 /* libnvdimm has already validated the input envelope */
525 for (i = 0; i < desc->in_num; i++)
526 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
530 /* skip over package wrapper */
531 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
532 in_buf.buffer.length = call_pkg->nd_size_in;
535 dev_dbg(dev, "%s cmd: %d: family: %d func: %d input length: %d\n",
536 dimm_name, cmd, family, func, in_buf.buffer.length);
537 if (payload_dumpable(nvdimm, func))
538 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
539 in_buf.buffer.pointer,
540 min_t(u32, 256, in_buf.buffer.length), true);
542 /* call the BIOS, prefer the named methods over _DSM if available */
543 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
544 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
545 out_obj = acpi_label_info(handle);
546 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
547 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
548 struct nd_cmd_get_config_data_hdr *p = buf;
550 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
551 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
552 && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
553 struct nd_cmd_set_config_hdr *p = buf;
555 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
561 revid = nfit_dsm_revid(nfit_mem->family, func);
564 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
568 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
572 if (out_obj->type != ACPI_TYPE_BUFFER) {
573 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
574 dimm_name, cmd_name, out_obj->type);
579 dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
580 cmd_name, out_obj->buffer.length);
581 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
582 out_obj->buffer.pointer,
583 min_t(u32, 128, out_obj->buffer.length), true);
586 call_pkg->nd_fw_size = out_obj->buffer.length;
587 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
588 out_obj->buffer.pointer,
589 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
593 * Need to support FW function w/o known size in advance.
594 * Caller can determine required size based upon nd_fw_size.
595 * If we return an error (like elsewhere) then caller wouldn't
596 * be able to rely upon data returned to make calculation.
603 for (i = 0, offset = 0; i < desc->out_num; i++) {
604 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
605 (u32 *) out_obj->buffer.pointer,
606 out_obj->buffer.length - offset);
608 if (offset + out_size > out_obj->buffer.length) {
609 dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
610 dimm_name, cmd_name, i);
614 if (in_buf.buffer.length + offset + out_size > buf_len) {
615 dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
616 dimm_name, cmd_name, i);
620 memcpy(buf + in_buf.buffer.length + offset,
621 out_obj->buffer.pointer + offset, out_size);
626 * Set fw_status for all the commands with a known format to be
627 * later interpreted by xlat_status().
629 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
630 && cmd <= ND_CMD_CLEAR_ERROR)
631 || (nvdimm && cmd >= ND_CMD_SMART
632 && cmd <= ND_CMD_VENDOR)))
633 fw_status = *(u32 *) out_obj->buffer.pointer;
635 if (offset + in_buf.buffer.length < buf_len) {
638 * status valid, return the number of bytes left
639 * unfilled in the output buffer
641 rc = buf_len - offset - in_buf.buffer.length;
643 *cmd_rc = xlat_status(nvdimm, buf, cmd,
646 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
647 __func__, dimm_name, cmd_name, buf_len,
654 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
662 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
664 static const char *spa_type_name(u16 type)
666 static const char *to_name[] = {
667 [NFIT_SPA_VOLATILE] = "volatile",
668 [NFIT_SPA_PM] = "pmem",
669 [NFIT_SPA_DCR] = "dimm-control-region",
670 [NFIT_SPA_BDW] = "block-data-window",
671 [NFIT_SPA_VDISK] = "volatile-disk",
672 [NFIT_SPA_VCD] = "volatile-cd",
673 [NFIT_SPA_PDISK] = "persistent-disk",
674 [NFIT_SPA_PCD] = "persistent-cd",
678 if (type > NFIT_SPA_PCD)
681 return to_name[type];
684 int nfit_spa_type(struct acpi_nfit_system_address *spa)
689 import_guid(&guid, spa->range_guid);
690 for (i = 0; i < NFIT_UUID_MAX; i++)
691 if (guid_equal(to_nfit_uuid(i), &guid))
696 static size_t sizeof_spa(struct acpi_nfit_system_address *spa)
698 if (spa->flags & ACPI_NFIT_LOCATION_COOKIE_VALID)
700 return sizeof(*spa) - 8;
703 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
704 struct nfit_table_prev *prev,
705 struct acpi_nfit_system_address *spa)
707 struct device *dev = acpi_desc->dev;
708 struct nfit_spa *nfit_spa;
710 if (spa->header.length != sizeof_spa(spa))
713 list_for_each_entry(nfit_spa, &prev->spas, list) {
714 if (memcmp(nfit_spa->spa, spa, sizeof_spa(spa)) == 0) {
715 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
720 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof_spa(spa),
724 INIT_LIST_HEAD(&nfit_spa->list);
725 memcpy(nfit_spa->spa, spa, sizeof_spa(spa));
726 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
727 dev_dbg(dev, "spa index: %d type: %s\n",
729 spa_type_name(nfit_spa_type(spa)));
733 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
734 struct nfit_table_prev *prev,
735 struct acpi_nfit_memory_map *memdev)
737 struct device *dev = acpi_desc->dev;
738 struct nfit_memdev *nfit_memdev;
740 if (memdev->header.length != sizeof(*memdev))
743 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
744 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
745 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
749 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
753 INIT_LIST_HEAD(&nfit_memdev->list);
754 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
755 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
756 dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
757 memdev->device_handle, memdev->range_index,
758 memdev->region_index, memdev->flags);
762 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
764 struct acpi_nfit_memory_map *memdev;
765 struct acpi_nfit_desc *acpi_desc;
766 struct nfit_mem *nfit_mem;
769 mutex_lock(&acpi_desc_lock);
770 list_for_each_entry(acpi_desc, &acpi_descs, list) {
771 mutex_lock(&acpi_desc->init_mutex);
772 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
773 memdev = __to_nfit_memdev(nfit_mem);
774 if (memdev->device_handle == device_handle) {
775 *flags = memdev->flags;
776 physical_id = memdev->physical_id;
777 mutex_unlock(&acpi_desc->init_mutex);
778 mutex_unlock(&acpi_desc_lock);
782 mutex_unlock(&acpi_desc->init_mutex);
784 mutex_unlock(&acpi_desc_lock);
788 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
791 * An implementation may provide a truncated control region if no block windows
794 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
796 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
801 return offsetof(struct acpi_nfit_control_region, window_size);
804 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
805 struct nfit_table_prev *prev,
806 struct acpi_nfit_control_region *dcr)
808 struct device *dev = acpi_desc->dev;
809 struct nfit_dcr *nfit_dcr;
811 if (!sizeof_dcr(dcr))
814 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
815 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
816 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
820 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
824 INIT_LIST_HEAD(&nfit_dcr->list);
825 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
826 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
827 dev_dbg(dev, "dcr index: %d windows: %d\n",
828 dcr->region_index, dcr->windows);
832 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
833 struct nfit_table_prev *prev,
834 struct acpi_nfit_data_region *bdw)
836 struct device *dev = acpi_desc->dev;
837 struct nfit_bdw *nfit_bdw;
839 if (bdw->header.length != sizeof(*bdw))
841 list_for_each_entry(nfit_bdw, &prev->bdws, list)
842 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
843 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
847 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
851 INIT_LIST_HEAD(&nfit_bdw->list);
852 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
853 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
854 dev_dbg(dev, "bdw dcr: %d windows: %d\n",
855 bdw->region_index, bdw->windows);
859 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
861 if (idt->header.length < sizeof(*idt))
863 return sizeof(*idt) + sizeof(u32) * idt->line_count;
866 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
867 struct nfit_table_prev *prev,
868 struct acpi_nfit_interleave *idt)
870 struct device *dev = acpi_desc->dev;
871 struct nfit_idt *nfit_idt;
873 if (!sizeof_idt(idt))
876 list_for_each_entry(nfit_idt, &prev->idts, list) {
877 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
880 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
881 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
886 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
890 INIT_LIST_HEAD(&nfit_idt->list);
891 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
892 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
893 dev_dbg(dev, "idt index: %d num_lines: %d\n",
894 idt->interleave_index, idt->line_count);
898 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
900 if (flush->header.length < sizeof(*flush))
902 return struct_size(flush, hint_address, flush->hint_count);
905 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
906 struct nfit_table_prev *prev,
907 struct acpi_nfit_flush_address *flush)
909 struct device *dev = acpi_desc->dev;
910 struct nfit_flush *nfit_flush;
912 if (!sizeof_flush(flush))
915 list_for_each_entry(nfit_flush, &prev->flushes, list) {
916 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
919 if (memcmp(nfit_flush->flush, flush,
920 sizeof_flush(flush)) == 0) {
921 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
926 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
927 + sizeof_flush(flush), GFP_KERNEL);
930 INIT_LIST_HEAD(&nfit_flush->list);
931 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
932 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
933 dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
934 flush->device_handle, flush->hint_count);
938 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
939 struct acpi_nfit_capabilities *pcap)
941 struct device *dev = acpi_desc->dev;
944 mask = (1 << (pcap->highest_capability + 1)) - 1;
945 acpi_desc->platform_cap = pcap->capabilities & mask;
946 dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
950 static void *add_table(struct acpi_nfit_desc *acpi_desc,
951 struct nfit_table_prev *prev, void *table, const void *end)
953 struct device *dev = acpi_desc->dev;
954 struct acpi_nfit_header *hdr;
955 void *err = ERR_PTR(-ENOMEM);
962 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
968 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
969 if (!add_spa(acpi_desc, prev, table))
972 case ACPI_NFIT_TYPE_MEMORY_MAP:
973 if (!add_memdev(acpi_desc, prev, table))
976 case ACPI_NFIT_TYPE_CONTROL_REGION:
977 if (!add_dcr(acpi_desc, prev, table))
980 case ACPI_NFIT_TYPE_DATA_REGION:
981 if (!add_bdw(acpi_desc, prev, table))
984 case ACPI_NFIT_TYPE_INTERLEAVE:
985 if (!add_idt(acpi_desc, prev, table))
988 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
989 if (!add_flush(acpi_desc, prev, table))
992 case ACPI_NFIT_TYPE_SMBIOS:
993 dev_dbg(dev, "smbios\n");
995 case ACPI_NFIT_TYPE_CAPABILITIES:
996 if (!add_platform_cap(acpi_desc, table))
1000 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
1004 return table + hdr->length;
1007 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1008 struct acpi_nfit_system_address *spa)
1010 struct nfit_mem *nfit_mem, *found;
1011 struct nfit_memdev *nfit_memdev;
1012 int type = spa ? nfit_spa_type(spa) : 0;
1024 * This loop runs in two modes, when a dimm is mapped the loop
1025 * adds memdev associations to an existing dimm, or creates a
1026 * dimm. In the unmapped dimm case this loop sweeps for memdev
1027 * instances with an invalid / zero range_index and adds those
1028 * dimms without spa associations.
1030 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1031 struct nfit_flush *nfit_flush;
1032 struct nfit_dcr *nfit_dcr;
1036 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1038 if (!spa && nfit_memdev->memdev->range_index)
1041 dcr = nfit_memdev->memdev->region_index;
1042 device_handle = nfit_memdev->memdev->device_handle;
1043 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1044 if (__to_nfit_memdev(nfit_mem)->device_handle
1053 nfit_mem = devm_kzalloc(acpi_desc->dev,
1054 sizeof(*nfit_mem), GFP_KERNEL);
1057 INIT_LIST_HEAD(&nfit_mem->list);
1058 nfit_mem->acpi_desc = acpi_desc;
1059 list_add(&nfit_mem->list, &acpi_desc->dimms);
1062 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1063 if (nfit_dcr->dcr->region_index != dcr)
1066 * Record the control region for the dimm. For
1067 * the ACPI 6.1 case, where there are separate
1068 * control regions for the pmem vs blk
1069 * interfaces, be sure to record the extended
1073 nfit_mem->dcr = nfit_dcr->dcr;
1074 else if (nfit_mem->dcr->windows == 0
1075 && nfit_dcr->dcr->windows)
1076 nfit_mem->dcr = nfit_dcr->dcr;
1080 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1081 struct acpi_nfit_flush_address *flush;
1084 if (nfit_flush->flush->device_handle != device_handle)
1086 nfit_mem->nfit_flush = nfit_flush;
1087 flush = nfit_flush->flush;
1088 nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1090 sizeof(struct resource),
1092 if (!nfit_mem->flush_wpq)
1094 for (i = 0; i < flush->hint_count; i++) {
1095 struct resource *res = &nfit_mem->flush_wpq[i];
1097 res->start = flush->hint_address[i];
1098 res->end = res->start + 8 - 1;
1103 if (dcr && !nfit_mem->dcr) {
1104 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1105 spa->range_index, dcr);
1109 if (type == NFIT_SPA_DCR) {
1110 struct nfit_idt *nfit_idt;
1113 /* multiple dimms may share a SPA when interleaved */
1114 nfit_mem->spa_dcr = spa;
1115 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1116 idt_idx = nfit_memdev->memdev->interleave_index;
1117 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1118 if (nfit_idt->idt->interleave_index != idt_idx)
1120 nfit_mem->idt_dcr = nfit_idt->idt;
1123 } else if (type == NFIT_SPA_PM) {
1125 * A single dimm may belong to multiple SPA-PM
1126 * ranges, record at least one in addition to
1127 * any SPA-DCR range.
1129 nfit_mem->memdev_pmem = nfit_memdev->memdev;
1131 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1137 static int nfit_mem_cmp(void *priv, const struct list_head *_a,
1138 const struct list_head *_b)
1140 struct nfit_mem *a = container_of(_a, typeof(*a), list);
1141 struct nfit_mem *b = container_of(_b, typeof(*b), list);
1142 u32 handleA, handleB;
1144 handleA = __to_nfit_memdev(a)->device_handle;
1145 handleB = __to_nfit_memdev(b)->device_handle;
1146 if (handleA < handleB)
1148 else if (handleA > handleB)
1153 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1155 struct nfit_spa *nfit_spa;
1160 * For each SPA-DCR or SPA-PMEM address range find its
1161 * corresponding MEMDEV(s). From each MEMDEV find the
1162 * corresponding DCR. Then, if we're operating on a SPA-DCR,
1163 * try to find a SPA-BDW and a corresponding BDW that references
1164 * the DCR. Throw it all into an nfit_mem object. Note, that
1165 * BDWs are optional.
1167 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1168 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1174 * If a DIMM has failed to be mapped into SPA there will be no
1175 * SPA entries above. Find and register all the unmapped DIMMs
1176 * for reporting and recovery purposes.
1178 rc = __nfit_mem_init(acpi_desc, NULL);
1182 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1187 static ssize_t bus_dsm_mask_show(struct device *dev,
1188 struct device_attribute *attr, char *buf)
1190 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1191 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1192 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1194 return sysfs_emit(buf, "%#lx\n", acpi_desc->bus_dsm_mask);
1196 static struct device_attribute dev_attr_bus_dsm_mask =
1197 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1199 static ssize_t revision_show(struct device *dev,
1200 struct device_attribute *attr, char *buf)
1202 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1203 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1204 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1206 return sysfs_emit(buf, "%d\n", acpi_desc->acpi_header.revision);
1208 static DEVICE_ATTR_RO(revision);
1210 static ssize_t hw_error_scrub_show(struct device *dev,
1211 struct device_attribute *attr, char *buf)
1213 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1214 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1215 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1217 return sysfs_emit(buf, "%d\n", acpi_desc->scrub_mode);
1221 * The 'hw_error_scrub' attribute can have the following values written to it:
1222 * '0': Switch to the default mode where an exception will only insert
1223 * the address of the memory error into the poison and badblocks lists.
1224 * '1': Enable a full scrub to happen if an exception for a memory error is
1227 static ssize_t hw_error_scrub_store(struct device *dev,
1228 struct device_attribute *attr, const char *buf, size_t size)
1230 struct nvdimm_bus_descriptor *nd_desc;
1234 rc = kstrtol(buf, 0, &val);
1239 nd_desc = dev_get_drvdata(dev);
1241 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1244 case HW_ERROR_SCRUB_ON:
1245 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1247 case HW_ERROR_SCRUB_OFF:
1248 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1260 static DEVICE_ATTR_RW(hw_error_scrub);
1263 * This shows the number of full Address Range Scrubs that have been
1264 * completed since driver load time. Userspace can wait on this using
1265 * select/poll etc. A '+' at the end indicates an ARS is in progress
1267 static ssize_t scrub_show(struct device *dev,
1268 struct device_attribute *attr, char *buf)
1270 struct nvdimm_bus_descriptor *nd_desc;
1271 struct acpi_nfit_desc *acpi_desc;
1272 ssize_t rc = -ENXIO;
1276 nd_desc = dev_get_drvdata(dev);
1281 acpi_desc = to_acpi_desc(nd_desc);
1283 mutex_lock(&acpi_desc->init_mutex);
1284 busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1285 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1286 rc = sysfs_emit(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1287 /* Allow an admin to poll the busy state at a higher rate */
1288 if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1289 &acpi_desc->scrub_flags)) {
1290 acpi_desc->scrub_tmo = 1;
1291 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1294 mutex_unlock(&acpi_desc->init_mutex);
1299 static ssize_t scrub_store(struct device *dev,
1300 struct device_attribute *attr, const char *buf, size_t size)
1302 struct nvdimm_bus_descriptor *nd_desc;
1306 rc = kstrtol(buf, 0, &val);
1313 nd_desc = dev_get_drvdata(dev);
1315 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1317 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1324 static DEVICE_ATTR_RW(scrub);
1326 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1328 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1329 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1330 | 1 << ND_CMD_ARS_STATUS;
1332 return (nd_desc->cmd_mask & mask) == mask;
1335 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1337 struct device *dev = kobj_to_dev(kobj);
1338 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1340 if (a == &dev_attr_scrub.attr)
1341 return ars_supported(nvdimm_bus) ? a->mode : 0;
1343 if (a == &dev_attr_firmware_activate_noidle.attr)
1344 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0;
1349 static struct attribute *acpi_nfit_attributes[] = {
1350 &dev_attr_revision.attr,
1351 &dev_attr_scrub.attr,
1352 &dev_attr_hw_error_scrub.attr,
1353 &dev_attr_bus_dsm_mask.attr,
1354 &dev_attr_firmware_activate_noidle.attr,
1358 static const struct attribute_group acpi_nfit_attribute_group = {
1360 .attrs = acpi_nfit_attributes,
1361 .is_visible = nfit_visible,
1364 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1365 &acpi_nfit_attribute_group,
1369 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1371 struct nvdimm *nvdimm = to_nvdimm(dev);
1372 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1374 return __to_nfit_memdev(nfit_mem);
1377 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1379 struct nvdimm *nvdimm = to_nvdimm(dev);
1380 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1382 return nfit_mem->dcr;
1385 static ssize_t handle_show(struct device *dev,
1386 struct device_attribute *attr, char *buf)
1388 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1390 return sysfs_emit(buf, "%#x\n", memdev->device_handle);
1392 static DEVICE_ATTR_RO(handle);
1394 static ssize_t phys_id_show(struct device *dev,
1395 struct device_attribute *attr, char *buf)
1397 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1399 return sysfs_emit(buf, "%#x\n", memdev->physical_id);
1401 static DEVICE_ATTR_RO(phys_id);
1403 static ssize_t vendor_show(struct device *dev,
1404 struct device_attribute *attr, char *buf)
1406 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1408 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1410 static DEVICE_ATTR_RO(vendor);
1412 static ssize_t rev_id_show(struct device *dev,
1413 struct device_attribute *attr, char *buf)
1415 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1417 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1419 static DEVICE_ATTR_RO(rev_id);
1421 static ssize_t device_show(struct device *dev,
1422 struct device_attribute *attr, char *buf)
1424 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1426 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1428 static DEVICE_ATTR_RO(device);
1430 static ssize_t subsystem_vendor_show(struct device *dev,
1431 struct device_attribute *attr, char *buf)
1433 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1435 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1437 static DEVICE_ATTR_RO(subsystem_vendor);
1439 static ssize_t subsystem_rev_id_show(struct device *dev,
1440 struct device_attribute *attr, char *buf)
1442 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1444 return sysfs_emit(buf, "0x%04x\n",
1445 be16_to_cpu(dcr->subsystem_revision_id));
1447 static DEVICE_ATTR_RO(subsystem_rev_id);
1449 static ssize_t subsystem_device_show(struct device *dev,
1450 struct device_attribute *attr, char *buf)
1452 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1454 return sysfs_emit(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1456 static DEVICE_ATTR_RO(subsystem_device);
1458 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1460 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1463 if (nfit_mem->memdev_pmem)
1468 static ssize_t format_show(struct device *dev,
1469 struct device_attribute *attr, char *buf)
1471 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1473 return sysfs_emit(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1475 static DEVICE_ATTR_RO(format);
1477 static ssize_t format1_show(struct device *dev,
1478 struct device_attribute *attr, char *buf)
1481 ssize_t rc = -ENXIO;
1482 struct nfit_mem *nfit_mem;
1483 struct nfit_memdev *nfit_memdev;
1484 struct acpi_nfit_desc *acpi_desc;
1485 struct nvdimm *nvdimm = to_nvdimm(dev);
1486 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1488 nfit_mem = nvdimm_provider_data(nvdimm);
1489 acpi_desc = nfit_mem->acpi_desc;
1490 handle = to_nfit_memdev(dev)->device_handle;
1492 /* assumes DIMMs have at most 2 published interface codes */
1493 mutex_lock(&acpi_desc->init_mutex);
1494 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1495 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1496 struct nfit_dcr *nfit_dcr;
1498 if (memdev->device_handle != handle)
1501 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1502 if (nfit_dcr->dcr->region_index != memdev->region_index)
1504 if (nfit_dcr->dcr->code == dcr->code)
1506 rc = sysfs_emit(buf, "0x%04x\n",
1507 le16_to_cpu(nfit_dcr->dcr->code));
1513 mutex_unlock(&acpi_desc->init_mutex);
1516 static DEVICE_ATTR_RO(format1);
1518 static ssize_t formats_show(struct device *dev,
1519 struct device_attribute *attr, char *buf)
1521 struct nvdimm *nvdimm = to_nvdimm(dev);
1523 return sysfs_emit(buf, "%d\n", num_nvdimm_formats(nvdimm));
1525 static DEVICE_ATTR_RO(formats);
1527 static ssize_t serial_show(struct device *dev,
1528 struct device_attribute *attr, char *buf)
1530 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1532 return sysfs_emit(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1534 static DEVICE_ATTR_RO(serial);
1536 static ssize_t family_show(struct device *dev,
1537 struct device_attribute *attr, char *buf)
1539 struct nvdimm *nvdimm = to_nvdimm(dev);
1540 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1542 if (nfit_mem->family < 0)
1544 return sysfs_emit(buf, "%d\n", nfit_mem->family);
1546 static DEVICE_ATTR_RO(family);
1548 static ssize_t dsm_mask_show(struct device *dev,
1549 struct device_attribute *attr, char *buf)
1551 struct nvdimm *nvdimm = to_nvdimm(dev);
1552 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1554 if (nfit_mem->family < 0)
1556 return sysfs_emit(buf, "%#lx\n", nfit_mem->dsm_mask);
1558 static DEVICE_ATTR_RO(dsm_mask);
1560 static ssize_t flags_show(struct device *dev,
1561 struct device_attribute *attr, char *buf)
1563 struct nvdimm *nvdimm = to_nvdimm(dev);
1564 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1565 u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1567 if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1568 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1570 return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
1571 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1572 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1573 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1574 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1575 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1576 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1577 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1579 static DEVICE_ATTR_RO(flags);
1581 static ssize_t id_show(struct device *dev,
1582 struct device_attribute *attr, char *buf)
1584 struct nvdimm *nvdimm = to_nvdimm(dev);
1585 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1587 return sysfs_emit(buf, "%s\n", nfit_mem->id);
1589 static DEVICE_ATTR_RO(id);
1591 static ssize_t dirty_shutdown_show(struct device *dev,
1592 struct device_attribute *attr, char *buf)
1594 struct nvdimm *nvdimm = to_nvdimm(dev);
1595 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1597 return sysfs_emit(buf, "%d\n", nfit_mem->dirty_shutdown);
1599 static DEVICE_ATTR_RO(dirty_shutdown);
1601 static struct attribute *acpi_nfit_dimm_attributes[] = {
1602 &dev_attr_handle.attr,
1603 &dev_attr_phys_id.attr,
1604 &dev_attr_vendor.attr,
1605 &dev_attr_device.attr,
1606 &dev_attr_rev_id.attr,
1607 &dev_attr_subsystem_vendor.attr,
1608 &dev_attr_subsystem_device.attr,
1609 &dev_attr_subsystem_rev_id.attr,
1610 &dev_attr_format.attr,
1611 &dev_attr_formats.attr,
1612 &dev_attr_format1.attr,
1613 &dev_attr_serial.attr,
1614 &dev_attr_flags.attr,
1616 &dev_attr_family.attr,
1617 &dev_attr_dsm_mask.attr,
1618 &dev_attr_dirty_shutdown.attr,
1622 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1623 struct attribute *a, int n)
1625 struct device *dev = kobj_to_dev(kobj);
1626 struct nvdimm *nvdimm = to_nvdimm(dev);
1627 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1629 if (!to_nfit_dcr(dev)) {
1630 /* Without a dcr only the memdev attributes can be surfaced */
1631 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1632 || a == &dev_attr_flags.attr
1633 || a == &dev_attr_family.attr
1634 || a == &dev_attr_dsm_mask.attr)
1639 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1642 if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1643 && a == &dev_attr_dirty_shutdown.attr)
1649 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1651 .attrs = acpi_nfit_dimm_attributes,
1652 .is_visible = acpi_nfit_dimm_attr_visible,
1655 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1656 &acpi_nfit_dimm_attribute_group,
1660 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1663 struct nfit_mem *nfit_mem;
1665 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1666 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1667 return nfit_mem->nvdimm;
1672 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1674 struct nfit_mem *nfit_mem;
1675 struct acpi_nfit_desc *acpi_desc;
1677 dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1680 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1681 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1686 acpi_desc = dev_get_drvdata(dev->parent);
1691 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1694 nfit_mem = dev_get_drvdata(dev);
1695 if (nfit_mem && nfit_mem->flags_attr)
1696 sysfs_notify_dirent(nfit_mem->flags_attr);
1698 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1700 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1702 struct acpi_device *adev = data;
1703 struct device *dev = &adev->dev;
1705 device_lock(dev->parent);
1706 __acpi_nvdimm_notify(dev, event);
1707 device_unlock(dev->parent);
1710 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1715 status = acpi_get_handle(adev->handle, method, &handle);
1717 if (ACPI_SUCCESS(status))
1722 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1724 struct device *dev = &nfit_mem->adev->dev;
1725 struct nd_intel_smart smart = { 0 };
1726 union acpi_object in_buf = {
1727 .buffer.type = ACPI_TYPE_BUFFER,
1730 union acpi_object in_obj = {
1731 .package.type = ACPI_TYPE_PACKAGE,
1733 .package.elements = &in_buf,
1735 const u8 func = ND_INTEL_SMART;
1736 const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1737 u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1738 struct acpi_device *adev = nfit_mem->adev;
1739 acpi_handle handle = adev->handle;
1740 union acpi_object *out_obj;
1742 if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1745 out_obj = acpi_evaluate_dsm_typed(handle, guid, revid, func, &in_obj, ACPI_TYPE_BUFFER);
1746 if (!out_obj || out_obj->buffer.length < sizeof(smart)) {
1747 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1752 memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1755 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1756 if (smart.shutdown_state)
1757 set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1760 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1761 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1762 nfit_mem->dirty_shutdown = smart.shutdown_count;
1766 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1769 * For DIMMs that provide a dynamic facility to retrieve a
1770 * dirty-shutdown status and/or a dirty-shutdown count, cache
1771 * these values in nfit_mem.
1773 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1774 nfit_intel_shutdown_status(nfit_mem);
1777 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1778 struct nfit_mem *nfit_mem, u32 device_handle)
1780 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1781 struct acpi_device *adev, *adev_dimm;
1782 struct device *dev = acpi_desc->dev;
1783 unsigned long dsm_mask, label_mask;
1787 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1789 /* nfit test assumes 1:1 relationship between commands and dsms */
1790 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1791 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1792 set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1794 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1795 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1796 be16_to_cpu(dcr->vendor_id),
1797 dcr->manufacturing_location,
1798 be16_to_cpu(dcr->manufacturing_date),
1799 be32_to_cpu(dcr->serial_number));
1801 sprintf(nfit_mem->id, "%04x-%08x",
1802 be16_to_cpu(dcr->vendor_id),
1803 be32_to_cpu(dcr->serial_number));
1805 adev = to_acpi_dev(acpi_desc);
1807 /* unit test case */
1808 populate_shutdown_status(nfit_mem);
1812 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1813 nfit_mem->adev = adev_dimm;
1815 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1817 return force_enable_dimms ? 0 : -ENODEV;
1820 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1821 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1822 dev_err(dev, "%s: notification registration failed\n",
1823 dev_name(&adev_dimm->dev));
1827 * Record nfit_mem for the notification path to track back to
1828 * the nfit sysfs attributes for this dimm device object.
1830 dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1833 * There are 4 "legacy" NVDIMM command sets
1834 * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1835 * an EFI working group was established to constrain this
1836 * proliferation. The nfit driver probes for the supported command
1837 * set by GUID. Note, if you're a platform developer looking to add
1838 * a new command set to this probe, consider using an existing set,
1839 * or otherwise seek approval to publish the command set at
1840 * http://www.uefi.org/RFIC_LIST.
1842 * Note, that checking for function0 (bit0) tells us if any commands
1843 * are reachable through this GUID.
1845 clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1846 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1847 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
1848 set_bit(i, &nd_desc->dimm_family_mask);
1849 if (family < 0 || i == default_dsm_family)
1853 /* limit the supported commands to those that are publicly documented */
1854 nfit_mem->family = family;
1855 if (override_dsm_mask && !disable_vendor_specific)
1856 dsm_mask = override_dsm_mask;
1857 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1858 dsm_mask = NVDIMM_INTEL_CMDMASK;
1859 if (disable_vendor_specific)
1860 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1861 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1862 dsm_mask = 0x1c3c76;
1863 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1865 if (disable_vendor_specific)
1866 dsm_mask &= ~(1 << 8);
1867 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1868 dsm_mask = 0xffffffff;
1869 } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1872 dev_dbg(dev, "unknown dimm command family\n");
1873 nfit_mem->family = -1;
1874 /* DSMs are optional, continue loading the driver... */
1879 * Function 0 is the command interrogation function, don't
1880 * export it to potential userspace use, and enable it to be
1881 * used as an error value in acpi_nfit_ctl().
1885 guid = to_nfit_uuid(nfit_mem->family);
1886 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1887 if (acpi_check_dsm(adev_dimm->handle, guid,
1888 nfit_dsm_revid(nfit_mem->family, i),
1890 set_bit(i, &nfit_mem->dsm_mask);
1893 * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1894 * due to their better semantics handling locked capacity.
1896 label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1897 | 1 << ND_CMD_SET_CONFIG_DATA;
1898 if (family == NVDIMM_FAMILY_INTEL
1899 && (dsm_mask & label_mask) == label_mask)
1900 /* skip _LS{I,R,W} enabling */;
1902 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1903 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1904 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1905 set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1908 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1909 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1910 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1911 set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1915 * Quirk read-only label configurations to preserve
1916 * access to label-less namespaces by default.
1918 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1920 dev_dbg(dev, "%s: No _LSW, disable labels\n",
1921 dev_name(&adev_dimm->dev));
1922 clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1924 dev_dbg(dev, "%s: Force enable labels\n",
1925 dev_name(&adev_dimm->dev));
1928 populate_shutdown_status(nfit_mem);
1933 static void shutdown_dimm_notify(void *data)
1935 struct acpi_nfit_desc *acpi_desc = data;
1936 struct nfit_mem *nfit_mem;
1938 mutex_lock(&acpi_desc->init_mutex);
1940 * Clear out the nfit_mem->flags_attr and shut down dimm event
1943 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1944 struct acpi_device *adev_dimm = nfit_mem->adev;
1946 if (nfit_mem->flags_attr) {
1947 sysfs_put(nfit_mem->flags_attr);
1948 nfit_mem->flags_attr = NULL;
1951 acpi_remove_notify_handler(adev_dimm->handle,
1952 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1953 dev_set_drvdata(&adev_dimm->dev, NULL);
1956 mutex_unlock(&acpi_desc->init_mutex);
1959 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
1962 case NVDIMM_FAMILY_INTEL:
1963 return intel_security_ops;
1969 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops(
1970 struct nfit_mem *nfit_mem)
1973 struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc;
1974 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1976 if (!nd_desc->fw_ops)
1979 if (nfit_mem->family != NVDIMM_FAMILY_INTEL)
1982 mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK;
1983 if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK)
1986 return intel_fw_ops;
1989 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1991 struct nfit_mem *nfit_mem;
1992 int dimm_count = 0, rc;
1993 struct nvdimm *nvdimm;
1995 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1996 struct acpi_nfit_flush_address *flush;
1997 unsigned long flags = 0, cmd_mask;
1998 struct nfit_memdev *nfit_memdev;
2002 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
2003 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2009 /* collate flags across all memdevs for this dimm */
2010 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2011 struct acpi_nfit_memory_map *dimm_memdev;
2013 dimm_memdev = __to_nfit_memdev(nfit_mem);
2014 if (dimm_memdev->device_handle
2015 != nfit_memdev->memdev->device_handle)
2017 dimm_memdev->flags |= nfit_memdev->memdev->flags;
2020 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2021 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2022 set_bit(NDD_UNARMED, &flags);
2024 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2029 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2030 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2031 * userspace interface.
2033 cmd_mask = 1UL << ND_CMD_CALL;
2034 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2036 * These commands have a 1:1 correspondence
2037 * between DSM payload and libnvdimm ioctl
2040 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2043 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2044 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2045 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2047 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2048 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2050 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2052 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2053 acpi_nfit_dimm_attribute_groups,
2054 flags, cmd_mask, flush ? flush->hint_count : 0,
2055 nfit_mem->flush_wpq, &nfit_mem->id[0],
2056 acpi_nfit_get_security_ops(nfit_mem->family),
2057 acpi_nfit_get_fw_ops(nfit_mem));
2061 nfit_mem->nvdimm = nvdimm;
2064 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2067 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2068 nvdimm_name(nvdimm),
2069 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2070 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2071 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2072 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2073 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2077 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2082 * Now that dimms are successfully registered, and async registration
2083 * is flushed, attempt to enable event notification.
2085 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2086 struct kernfs_node *nfit_kernfs;
2088 nvdimm = nfit_mem->nvdimm;
2092 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2094 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2096 sysfs_put(nfit_kernfs);
2097 if (!nfit_mem->flags_attr)
2098 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2099 nvdimm_name(nvdimm));
2102 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2107 * These constants are private because there are no kernel consumers of
2110 enum nfit_aux_cmds {
2111 NFIT_CMD_TRANSLATE_SPA = 5,
2112 NFIT_CMD_ARS_INJECT_SET = 7,
2113 NFIT_CMD_ARS_INJECT_CLEAR = 8,
2114 NFIT_CMD_ARS_INJECT_GET = 9,
2117 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2119 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2120 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2121 unsigned long dsm_mask, *mask;
2122 struct acpi_device *adev;
2125 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2126 set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask);
2128 /* enable nfit_test to inject bus command emulation */
2129 if (acpi_desc->bus_cmd_force_en) {
2130 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2131 mask = &nd_desc->bus_family_mask;
2132 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) {
2133 set_bit(NVDIMM_BUS_FAMILY_INTEL, mask);
2134 nd_desc->fw_ops = intel_bus_fw_ops;
2138 adev = to_acpi_dev(acpi_desc);
2142 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2143 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2144 set_bit(i, &nd_desc->cmd_mask);
2147 (1 << ND_CMD_ARS_CAP) |
2148 (1 << ND_CMD_ARS_START) |
2149 (1 << ND_CMD_ARS_STATUS) |
2150 (1 << ND_CMD_CLEAR_ERROR) |
2151 (1 << NFIT_CMD_TRANSLATE_SPA) |
2152 (1 << NFIT_CMD_ARS_INJECT_SET) |
2153 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2154 (1 << NFIT_CMD_ARS_INJECT_GET);
2155 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2156 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2157 set_bit(i, &acpi_desc->bus_dsm_mask);
2159 /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */
2160 dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK;
2161 guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL);
2162 mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL];
2163 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2164 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2167 if (*mask == dsm_mask) {
2168 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask);
2169 nd_desc->fw_ops = intel_bus_fw_ops;
2173 static ssize_t range_index_show(struct device *dev,
2174 struct device_attribute *attr, char *buf)
2176 struct nd_region *nd_region = to_nd_region(dev);
2177 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2179 return sysfs_emit(buf, "%d\n", nfit_spa->spa->range_index);
2181 static DEVICE_ATTR_RO(range_index);
2183 static struct attribute *acpi_nfit_region_attributes[] = {
2184 &dev_attr_range_index.attr,
2188 static const struct attribute_group acpi_nfit_region_attribute_group = {
2190 .attrs = acpi_nfit_region_attributes,
2193 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2194 &acpi_nfit_region_attribute_group,
2198 /* enough info to uniquely specify an interleave set */
2199 struct nfit_set_info {
2205 struct nfit_set_info2 {
2209 u16 manufacturing_date;
2210 u8 manufacturing_location;
2214 static int cmp_map_compat(const void *m0, const void *m1)
2216 const struct nfit_set_info *map0 = m0;
2217 const struct nfit_set_info *map1 = m1;
2219 return memcmp(&map0->region_offset, &map1->region_offset,
2223 static int cmp_map(const void *m0, const void *m1)
2225 const struct nfit_set_info *map0 = m0;
2226 const struct nfit_set_info *map1 = m1;
2228 if (map0->region_offset < map1->region_offset)
2230 else if (map0->region_offset > map1->region_offset)
2235 static int cmp_map2(const void *m0, const void *m1)
2237 const struct nfit_set_info2 *map0 = m0;
2238 const struct nfit_set_info2 *map1 = m1;
2240 if (map0->region_offset < map1->region_offset)
2242 else if (map0->region_offset > map1->region_offset)
2247 /* Retrieve the nth entry referencing this spa */
2248 static struct acpi_nfit_memory_map *memdev_from_spa(
2249 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2251 struct nfit_memdev *nfit_memdev;
2253 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2254 if (nfit_memdev->memdev->range_index == range_index)
2256 return nfit_memdev->memdev;
2260 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2261 struct nd_region_desc *ndr_desc,
2262 struct acpi_nfit_system_address *spa)
2264 u16 nr = ndr_desc->num_mappings;
2265 struct nfit_set_info2 *info2 __free(kfree) =
2266 kcalloc(nr, sizeof(*info2), GFP_KERNEL);
2267 struct nfit_set_info *info __free(kfree) =
2268 kcalloc(nr, sizeof(*info), GFP_KERNEL);
2269 struct device *dev = acpi_desc->dev;
2270 struct nd_interleave_set *nd_set;
2273 if (!info || !info2)
2276 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2279 import_guid(&nd_set->type_guid, spa->range_guid);
2281 for (i = 0; i < nr; i++) {
2282 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2283 struct nvdimm *nvdimm = mapping->nvdimm;
2284 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2285 struct nfit_set_info *map = &info[i];
2286 struct nfit_set_info2 *map2 = &info2[i];
2287 struct acpi_nfit_memory_map *memdev =
2288 memdev_from_spa(acpi_desc, spa->range_index, i);
2289 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2291 if (!memdev || !nfit_mem->dcr) {
2292 dev_err(dev, "%s: failed to find DCR\n", __func__);
2296 map->region_offset = memdev->region_offset;
2297 map->serial_number = dcr->serial_number;
2299 map2->region_offset = memdev->region_offset;
2300 map2->serial_number = dcr->serial_number;
2301 map2->vendor_id = dcr->vendor_id;
2302 map2->manufacturing_date = dcr->manufacturing_date;
2303 map2->manufacturing_location = dcr->manufacturing_location;
2306 /* v1.1 namespaces */
2307 sort(info, nr, sizeof(*info), cmp_map, NULL);
2308 nd_set->cookie1 = nd_fletcher64(info, sizeof(*info) * nr, 0);
2310 /* v1.2 namespaces */
2311 sort(info2, nr, sizeof(*info2), cmp_map2, NULL);
2312 nd_set->cookie2 = nd_fletcher64(info2, sizeof(*info2) * nr, 0);
2314 /* support v1.1 namespaces created with the wrong sort order */
2315 sort(info, nr, sizeof(*info), cmp_map_compat, NULL);
2316 nd_set->altcookie = nd_fletcher64(info, sizeof(*info) * nr, 0);
2318 /* record the result of the sort for the mapping position */
2319 for (i = 0; i < nr; i++) {
2320 struct nfit_set_info2 *map2 = &info2[i];
2323 for (j = 0; j < nr; j++) {
2324 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2325 struct nvdimm *nvdimm = mapping->nvdimm;
2326 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2327 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2329 if (map2->serial_number == dcr->serial_number &&
2330 map2->vendor_id == dcr->vendor_id &&
2331 map2->manufacturing_date == dcr->manufacturing_date &&
2332 map2->manufacturing_location
2333 == dcr->manufacturing_location) {
2334 mapping->position = i;
2340 ndr_desc->nd_set = nd_set;
2345 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2346 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2348 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2349 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2352 cmd->address = spa->address;
2353 cmd->length = spa->length;
2354 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2355 sizeof(*cmd), &cmd_rc);
2361 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2362 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2366 struct nd_cmd_ars_start ars_start;
2367 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2368 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2370 memset(&ars_start, 0, sizeof(ars_start));
2371 ars_start.address = spa->address;
2372 ars_start.length = spa->length;
2373 if (req_type == ARS_REQ_SHORT)
2374 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2375 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2376 ars_start.type = ND_ARS_PERSISTENT;
2377 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2378 ars_start.type = ND_ARS_VOLATILE;
2382 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2383 sizeof(ars_start), &cmd_rc);
2389 set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2393 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2396 struct nd_cmd_ars_start ars_start;
2397 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2398 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2400 ars_start = (struct nd_cmd_ars_start) {
2401 .address = ars_status->restart_address,
2402 .length = ars_status->restart_length,
2403 .type = ars_status->type,
2405 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2406 sizeof(ars_start), &cmd_rc);
2412 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2414 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2415 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2418 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2419 acpi_desc->max_ars, &cmd_rc);
2425 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2426 struct nfit_spa *nfit_spa)
2428 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2429 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2430 struct nd_region *nd_region = nfit_spa->nd_region;
2433 lockdep_assert_held(&acpi_desc->init_mutex);
2435 * Only advance the ARS state for ARS runs initiated by the
2436 * kernel, ignore ARS results from BIOS initiated runs for scrub
2437 * completion tracking.
2439 if (acpi_desc->scrub_spa != nfit_spa)
2442 if ((ars_status->address >= spa->address && ars_status->address
2443 < spa->address + spa->length)
2444 || (ars_status->address < spa->address)) {
2446 * Assume that if a scrub starts at an offset from the
2447 * start of nfit_spa that we are in the continuation
2450 * Otherwise, if the scrub covers the spa range, mark
2451 * any pending request complete.
2453 if (ars_status->address + ars_status->length
2454 >= spa->address + spa->length)
2461 acpi_desc->scrub_spa = NULL;
2463 dev = nd_region_dev(nd_region);
2464 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2466 dev = acpi_desc->dev;
2467 dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2470 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2472 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2473 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2478 * First record starts at 44 byte offset from the start of the
2481 if (ars_status->out_length < 44)
2485 * Ignore potentially stale results that are only refreshed
2486 * after a start-ARS event.
2488 if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2489 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2490 ars_status->num_records);
2494 for (i = 0; i < ars_status->num_records; i++) {
2495 /* only process full records */
2496 if (ars_status->out_length
2497 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2499 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2500 ars_status->records[i].err_address,
2501 ars_status->records[i].length);
2505 if (i < ars_status->num_records)
2506 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2511 static void acpi_nfit_remove_resource(void *data)
2513 struct resource *res = data;
2515 remove_resource(res);
2518 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2519 struct nd_region_desc *ndr_desc)
2521 struct resource *res, *nd_res = ndr_desc->res;
2524 /* No operation if the region is already registered as PMEM */
2525 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2526 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2527 if (is_pmem == REGION_INTERSECTS)
2530 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2534 res->name = "Persistent Memory";
2535 res->start = nd_res->start;
2536 res->end = nd_res->end;
2537 res->flags = IORESOURCE_MEM;
2538 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2540 ret = insert_resource(&iomem_resource, res);
2544 ret = devm_add_action_or_reset(acpi_desc->dev,
2545 acpi_nfit_remove_resource,
2553 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2554 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2555 struct acpi_nfit_memory_map *memdev,
2556 struct nfit_spa *nfit_spa)
2558 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2559 memdev->device_handle);
2560 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2563 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2564 spa->range_index, memdev->device_handle);
2568 mapping->nvdimm = nvdimm;
2569 switch (nfit_spa_type(spa)) {
2571 case NFIT_SPA_VOLATILE:
2572 mapping->start = memdev->address;
2573 mapping->size = memdev->region_size;
2580 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2582 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2583 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2584 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2585 nfit_spa_type(spa) == NFIT_SPA_PCD);
2588 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2590 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2591 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2592 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2595 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2596 struct nfit_spa *nfit_spa)
2598 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2599 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2600 struct nd_region_desc *ndr_desc, _ndr_desc;
2601 struct nfit_memdev *nfit_memdev;
2602 struct nvdimm_bus *nvdimm_bus;
2603 struct resource res;
2606 if (nfit_spa->nd_region)
2609 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2610 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2614 memset(&res, 0, sizeof(res));
2615 memset(&mappings, 0, sizeof(mappings));
2616 memset(&_ndr_desc, 0, sizeof(_ndr_desc));
2617 res.start = spa->address;
2618 res.end = res.start + spa->length - 1;
2619 ndr_desc = &_ndr_desc;
2620 ndr_desc->res = &res;
2621 ndr_desc->provider_data = nfit_spa;
2622 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2623 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2624 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain);
2625 ndr_desc->target_node = pxm_to_node(spa->proximity_domain);
2627 ndr_desc->numa_node = NUMA_NO_NODE;
2628 ndr_desc->target_node = NUMA_NO_NODE;
2631 /* Fallback to address based numa information if node lookup failed */
2632 if (ndr_desc->numa_node == NUMA_NO_NODE) {
2633 ndr_desc->numa_node = memory_add_physaddr_to_nid(spa->address);
2634 dev_info(acpi_desc->dev, "changing numa node from %d to %d for nfit region [%pa-%pa]",
2635 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
2637 if (ndr_desc->target_node == NUMA_NO_NODE) {
2638 ndr_desc->target_node = phys_to_target_node(spa->address);
2639 dev_info(acpi_desc->dev, "changing target node from %d to %d for nfit region [%pa-%pa]",
2640 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
2644 * Persistence domain bits are hierarchical, if
2645 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2646 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2648 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2649 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2650 else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2651 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2653 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2654 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2655 struct nd_mapping_desc *mapping;
2657 /* range index 0 == unmapped in SPA or invalid-SPA */
2658 if (memdev->range_index == 0 || spa->range_index == 0)
2660 if (memdev->range_index != spa->range_index)
2662 if (count >= ND_MAX_MAPPINGS) {
2663 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2664 spa->range_index, ND_MAX_MAPPINGS);
2667 mapping = &mappings[count++];
2668 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2674 ndr_desc->mapping = mappings;
2675 ndr_desc->num_mappings = count;
2676 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2680 nvdimm_bus = acpi_desc->nvdimm_bus;
2681 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2682 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2684 dev_warn(acpi_desc->dev,
2685 "failed to insert pmem resource to iomem: %d\n",
2690 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2692 if (!nfit_spa->nd_region)
2694 } else if (nfit_spa_is_volatile(spa)) {
2695 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2697 if (!nfit_spa->nd_region)
2699 } else if (nfit_spa_is_virtual(spa)) {
2700 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2702 if (!nfit_spa->nd_region)
2708 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2709 nfit_spa->spa->range_index);
2713 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
2715 struct device *dev = acpi_desc->dev;
2716 struct nd_cmd_ars_status *ars_status;
2718 if (acpi_desc->ars_status) {
2719 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2723 ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
2726 acpi_desc->ars_status = ars_status;
2730 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
2734 if (ars_status_alloc(acpi_desc))
2737 rc = ars_get_status(acpi_desc);
2739 if (rc < 0 && rc != -ENOSPC)
2742 if (ars_status_process_records(acpi_desc))
2743 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
2748 static int ars_register(struct acpi_nfit_desc *acpi_desc,
2749 struct nfit_spa *nfit_spa)
2753 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2754 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2756 set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
2758 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
2760 switch (acpi_nfit_query_poison(acpi_desc)) {
2764 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
2765 /* shouldn't happen, try again later */
2769 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2772 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
2773 rc = acpi_nfit_query_poison(acpi_desc);
2776 acpi_desc->scrub_spa = nfit_spa;
2777 ars_complete(acpi_desc, nfit_spa);
2779 * If ars_complete() says we didn't complete the
2780 * short scrub, we'll try again with a long
2783 acpi_desc->scrub_spa = NULL;
2788 * BIOS was using ARS, wait for it to complete (or
2789 * resources to become available) and then perform our
2794 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2798 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2801 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
2803 struct nfit_spa *nfit_spa;
2805 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2806 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2808 ars_complete(acpi_desc, nfit_spa);
2812 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
2815 unsigned int tmo = acpi_desc->scrub_tmo;
2816 struct device *dev = acpi_desc->dev;
2817 struct nfit_spa *nfit_spa;
2819 lockdep_assert_held(&acpi_desc->init_mutex);
2821 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
2824 if (query_rc == -EBUSY) {
2825 dev_dbg(dev, "ARS: ARS busy\n");
2826 return min(30U * 60U, tmo * 2);
2828 if (query_rc == -ENOSPC) {
2829 dev_dbg(dev, "ARS: ARS continue\n");
2830 ars_continue(acpi_desc);
2833 if (query_rc && query_rc != -EAGAIN) {
2834 unsigned long long addr, end;
2836 addr = acpi_desc->ars_status->address;
2837 end = addr + acpi_desc->ars_status->length;
2838 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
2842 ars_complete_all(acpi_desc);
2843 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2844 enum nfit_ars_state req_type;
2847 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2850 /* prefer short ARS requests first */
2851 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
2852 req_type = ARS_REQ_SHORT;
2853 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
2854 req_type = ARS_REQ_LONG;
2857 rc = ars_start(acpi_desc, nfit_spa, req_type);
2859 dev = nd_region_dev(nfit_spa->nd_region);
2860 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
2861 nfit_spa->spa->range_index,
2862 req_type == ARS_REQ_SHORT ? "short" : "long",
2865 * Hmm, we raced someone else starting ARS? Try again in
2871 dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
2872 "scrub start while range %d active\n",
2873 acpi_desc->scrub_spa->spa->range_index);
2874 clear_bit(req_type, &nfit_spa->ars_state);
2875 acpi_desc->scrub_spa = nfit_spa;
2877 * Consider this spa last for future scrub
2880 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
2884 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
2885 nfit_spa->spa->range_index, rc);
2886 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2891 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
2893 lockdep_assert_held(&acpi_desc->init_mutex);
2895 set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
2896 /* note this should only be set from within the workqueue */
2898 acpi_desc->scrub_tmo = tmo;
2899 queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
2902 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
2904 __sched_ars(acpi_desc, 0);
2907 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
2909 lockdep_assert_held(&acpi_desc->init_mutex);
2911 clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
2912 acpi_desc->scrub_count++;
2913 if (acpi_desc->scrub_count_state)
2914 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2917 static void acpi_nfit_scrub(struct work_struct *work)
2919 struct acpi_nfit_desc *acpi_desc;
2923 acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
2924 mutex_lock(&acpi_desc->init_mutex);
2925 query_rc = acpi_nfit_query_poison(acpi_desc);
2926 tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
2928 __sched_ars(acpi_desc, tmo);
2930 notify_ars_done(acpi_desc);
2931 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2932 clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
2933 mutex_unlock(&acpi_desc->init_mutex);
2936 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
2937 struct nfit_spa *nfit_spa)
2939 int type = nfit_spa_type(nfit_spa->spa);
2940 struct nd_cmd_ars_cap ars_cap;
2943 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2944 memset(&ars_cap, 0, sizeof(ars_cap));
2945 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2948 /* check that the supported scrub types match the spa type */
2949 if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
2950 & ND_ARS_VOLATILE) == 0)
2952 if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
2953 & ND_ARS_PERSISTENT) == 0)
2956 nfit_spa->max_ars = ars_cap.max_ars_out;
2957 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2958 acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
2959 clear_bit(ARS_FAILED, &nfit_spa->ars_state);
2962 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2964 struct nfit_spa *nfit_spa;
2965 int rc, do_sched_ars = 0;
2967 set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2968 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2969 switch (nfit_spa_type(nfit_spa->spa)) {
2970 case NFIT_SPA_VOLATILE:
2972 acpi_nfit_init_ars(acpi_desc, nfit_spa);
2977 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2978 switch (nfit_spa_type(nfit_spa->spa)) {
2979 case NFIT_SPA_VOLATILE:
2981 /* register regions and kick off initial ARS run */
2982 rc = ars_register(acpi_desc, nfit_spa);
2987 * Kick off background ARS if at least one
2988 * region successfully registered ARS
2990 if (!test_bit(ARS_FAILED, &nfit_spa->ars_state))
2994 /* nothing to register */
2997 case NFIT_SPA_VDISK:
2999 case NFIT_SPA_PDISK:
3001 /* register known regions that don't support ARS */
3002 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3007 /* don't register unknown regions */
3013 sched_ars(acpi_desc);
3017 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3018 struct nfit_table_prev *prev)
3020 struct device *dev = acpi_desc->dev;
3022 if (!list_empty(&prev->spas) ||
3023 !list_empty(&prev->memdevs) ||
3024 !list_empty(&prev->dcrs) ||
3025 !list_empty(&prev->bdws) ||
3026 !list_empty(&prev->idts) ||
3027 !list_empty(&prev->flushes)) {
3028 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3034 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3036 struct device *dev = acpi_desc->dev;
3037 struct kernfs_node *nfit;
3038 struct device *bus_dev;
3040 if (!ars_supported(acpi_desc->nvdimm_bus))
3043 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3044 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3046 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3049 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3051 if (!acpi_desc->scrub_count_state) {
3052 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3059 static void acpi_nfit_unregister(void *data)
3061 struct acpi_nfit_desc *acpi_desc = data;
3063 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3066 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3068 struct device *dev = acpi_desc->dev;
3069 struct nfit_table_prev prev;
3073 if (!acpi_desc->nvdimm_bus) {
3074 acpi_nfit_init_dsms(acpi_desc);
3076 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3077 &acpi_desc->nd_desc);
3078 if (!acpi_desc->nvdimm_bus)
3081 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3086 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3090 /* register this acpi_desc for mce notifications */
3091 mutex_lock(&acpi_desc_lock);
3092 list_add_tail(&acpi_desc->list, &acpi_descs);
3093 mutex_unlock(&acpi_desc_lock);
3096 mutex_lock(&acpi_desc->init_mutex);
3098 INIT_LIST_HEAD(&prev.spas);
3099 INIT_LIST_HEAD(&prev.memdevs);
3100 INIT_LIST_HEAD(&prev.dcrs);
3101 INIT_LIST_HEAD(&prev.bdws);
3102 INIT_LIST_HEAD(&prev.idts);
3103 INIT_LIST_HEAD(&prev.flushes);
3105 list_cut_position(&prev.spas, &acpi_desc->spas,
3106 acpi_desc->spas.prev);
3107 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3108 acpi_desc->memdevs.prev);
3109 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3110 acpi_desc->dcrs.prev);
3111 list_cut_position(&prev.bdws, &acpi_desc->bdws,
3112 acpi_desc->bdws.prev);
3113 list_cut_position(&prev.idts, &acpi_desc->idts,
3114 acpi_desc->idts.prev);
3115 list_cut_position(&prev.flushes, &acpi_desc->flushes,
3116 acpi_desc->flushes.prev);
3119 while (!IS_ERR_OR_NULL(data))
3120 data = add_table(acpi_desc, &prev, data, end);
3123 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3128 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3132 rc = nfit_mem_init(acpi_desc);
3136 rc = acpi_nfit_register_dimms(acpi_desc);
3140 rc = acpi_nfit_register_regions(acpi_desc);
3143 mutex_unlock(&acpi_desc->init_mutex);
3146 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3148 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3150 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3151 struct device *dev = acpi_desc->dev;
3153 /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3157 /* Bounce the init_mutex to complete initial registration */
3158 mutex_lock(&acpi_desc->init_mutex);
3159 mutex_unlock(&acpi_desc->init_mutex);
3164 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3165 struct nvdimm *nvdimm, unsigned int cmd)
3167 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3171 if (cmd != ND_CMD_ARS_START)
3175 * The kernel and userspace may race to initiate a scrub, but
3176 * the scrub thread is prepared to lose that initial race. It
3177 * just needs guarantees that any ARS it initiates are not
3178 * interrupted by any intervening start requests from userspace.
3180 if (work_busy(&acpi_desc->dwork.work))
3187 * Prevent security and firmware activate commands from being issued via
3190 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3191 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3193 struct nd_cmd_pkg *call_pkg = buf;
3196 if (nvdimm && cmd == ND_CMD_CALL &&
3197 call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3198 func = call_pkg->nd_command;
3199 if (func > NVDIMM_CMD_MAX ||
3200 (1 << func) & NVDIMM_INTEL_DENY_CMDMASK)
3204 /* block all non-nfit bus commands */
3205 if (!nvdimm && cmd == ND_CMD_CALL &&
3206 call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT)
3209 return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3212 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3213 enum nfit_ars_state req_type)
3215 struct device *dev = acpi_desc->dev;
3216 int scheduled = 0, busy = 0;
3217 struct nfit_spa *nfit_spa;
3219 mutex_lock(&acpi_desc->init_mutex);
3220 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3221 mutex_unlock(&acpi_desc->init_mutex);
3225 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3226 int type = nfit_spa_type(nfit_spa->spa);
3228 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3230 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3233 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3239 sched_ars(acpi_desc);
3240 dev_dbg(dev, "ars_scan triggered\n");
3242 mutex_unlock(&acpi_desc->init_mutex);
3251 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3253 struct nvdimm_bus_descriptor *nd_desc;
3255 dev_set_drvdata(dev, acpi_desc);
3256 acpi_desc->dev = dev;
3257 nd_desc = &acpi_desc->nd_desc;
3258 nd_desc->provider_name = "ACPI.NFIT";
3259 nd_desc->module = THIS_MODULE;
3260 nd_desc->ndctl = acpi_nfit_ctl;
3261 nd_desc->flush_probe = acpi_nfit_flush_probe;
3262 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3263 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3265 INIT_LIST_HEAD(&acpi_desc->spas);
3266 INIT_LIST_HEAD(&acpi_desc->dcrs);
3267 INIT_LIST_HEAD(&acpi_desc->bdws);
3268 INIT_LIST_HEAD(&acpi_desc->idts);
3269 INIT_LIST_HEAD(&acpi_desc->flushes);
3270 INIT_LIST_HEAD(&acpi_desc->memdevs);
3271 INIT_LIST_HEAD(&acpi_desc->dimms);
3272 INIT_LIST_HEAD(&acpi_desc->list);
3273 mutex_init(&acpi_desc->init_mutex);
3274 acpi_desc->scrub_tmo = 1;
3275 INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3277 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3279 static void acpi_nfit_put_table(void *table)
3281 acpi_put_table(table);
3284 static void acpi_nfit_notify(acpi_handle handle, u32 event, void *data)
3286 struct acpi_device *adev = data;
3288 device_lock(&adev->dev);
3289 __acpi_nfit_notify(&adev->dev, handle, event);
3290 device_unlock(&adev->dev);
3293 static void acpi_nfit_remove_notify_handler(void *data)
3295 struct acpi_device *adev = data;
3297 acpi_dev_remove_notify_handler(adev, ACPI_DEVICE_NOTIFY,
3301 void acpi_nfit_shutdown(void *data)
3303 struct acpi_nfit_desc *acpi_desc = data;
3304 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3307 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3310 mutex_lock(&acpi_desc_lock);
3311 list_del(&acpi_desc->list);
3312 mutex_unlock(&acpi_desc_lock);
3314 mutex_lock(&acpi_desc->init_mutex);
3315 set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3316 mutex_unlock(&acpi_desc->init_mutex);
3317 cancel_delayed_work_sync(&acpi_desc->dwork);
3320 * Bounce the nvdimm bus lock to make sure any in-flight
3321 * acpi_nfit_ars_rescan() submissions have had a chance to
3322 * either submit or see ->cancel set.
3324 device_lock(bus_dev);
3325 device_unlock(bus_dev);
3327 flush_workqueue(nfit_wq);
3329 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3331 static int acpi_nfit_add(struct acpi_device *adev)
3333 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3334 struct acpi_nfit_desc *acpi_desc;
3335 struct device *dev = &adev->dev;
3336 struct acpi_table_header *tbl;
3337 acpi_status status = AE_OK;
3341 rc = acpi_dev_install_notify_handler(adev, ACPI_DEVICE_NOTIFY,
3342 acpi_nfit_notify, adev);
3346 rc = devm_add_action_or_reset(dev, acpi_nfit_remove_notify_handler,
3351 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3352 if (ACPI_FAILURE(status)) {
3353 /* The NVDIMM root device allows OS to trigger enumeration of
3354 * NVDIMMs through NFIT at boot time and re-enumeration at
3355 * root level via the _FIT method during runtime.
3356 * This is ok to return 0 here, we could have an nvdimm
3357 * hotplugged later and evaluate _FIT method which returns
3358 * data in the format of a series of NFIT Structures.
3360 dev_dbg(dev, "failed to find NFIT at startup\n");
3364 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3369 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3372 acpi_nfit_desc_init(acpi_desc, &adev->dev);
3374 /* Save the acpi header for exporting the revision via sysfs */
3375 acpi_desc->acpi_header = *tbl;
3377 /* Evaluate _FIT and override with that if present */
3378 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3379 if (ACPI_SUCCESS(status) && buf.length > 0) {
3380 union acpi_object *obj = buf.pointer;
3382 if (obj->type == ACPI_TYPE_BUFFER)
3383 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3384 obj->buffer.length);
3386 dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3390 /* skip over the lead-in header table */
3391 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3392 + sizeof(struct acpi_table_nfit),
3393 sz - sizeof(struct acpi_table_nfit));
3398 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3401 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3403 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3404 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3405 union acpi_object *obj;
3410 /* dev->driver may be null if we're being removed */
3411 dev_dbg(dev, "no driver found for dev\n");
3416 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3419 acpi_nfit_desc_init(acpi_desc, dev);
3422 * Finish previous registration before considering new
3425 flush_workqueue(nfit_wq);
3429 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3430 if (ACPI_FAILURE(status)) {
3431 dev_err(dev, "failed to evaluate _FIT\n");
3436 if (obj->type == ACPI_TYPE_BUFFER) {
3437 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3438 obj->buffer.length);
3440 dev_err(dev, "failed to merge updated NFIT\n");
3442 dev_err(dev, "Invalid _FIT\n");
3446 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3448 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3450 if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3451 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3453 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3456 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3458 dev_dbg(dev, "event: 0x%x\n", event);
3461 case NFIT_NOTIFY_UPDATE:
3462 return acpi_nfit_update_notify(dev, handle);
3463 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3464 return acpi_nfit_uc_error_notify(dev, handle);
3469 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3471 static const struct acpi_device_id acpi_nfit_ids[] = {
3475 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3477 static struct acpi_driver acpi_nfit_driver = {
3478 .name = KBUILD_MODNAME,
3479 .ids = acpi_nfit_ids,
3481 .add = acpi_nfit_add,
3485 static __init int nfit_init(void)
3489 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3490 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 64);
3491 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3492 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 16);
3493 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 8);
3494 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3495 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3496 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3498 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3499 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3500 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3501 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3502 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3503 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3504 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3505 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3506 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3507 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3508 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3509 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3510 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3511 guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3512 guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]);
3514 nfit_wq = create_singlethread_workqueue("nfit");
3518 nfit_mce_register();
3519 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3521 nfit_mce_unregister();
3522 destroy_workqueue(nfit_wq);
3529 static __exit void nfit_exit(void)
3531 nfit_mce_unregister();
3532 acpi_bus_unregister_driver(&acpi_nfit_driver);
3533 destroy_workqueue(nfit_wq);
3534 WARN_ON(!list_empty(&acpi_descs));
3537 module_init(nfit_init);
3538 module_exit(nfit_exit);
3539 MODULE_DESCRIPTION("ACPI NVDIMM Firmware Interface Table (NFIT) driver");
3540 MODULE_LICENSE("GPL v2");
3541 MODULE_AUTHOR("Intel Corporation");