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)
459 func = cmd_to_func(nfit_mem, cmd, call_pkg, &family);
464 struct acpi_device *adev = nfit_mem->adev;
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;
477 struct acpi_device *adev = to_acpi_dev(acpi_desc);
479 cmd_name = nvdimm_bus_cmd_name(cmd);
480 cmd_mask = nd_desc->cmd_mask;
481 if (cmd == ND_CMD_CALL && call_pkg->nd_family) {
482 family = call_pkg->nd_family;
483 if (family > NVDIMM_BUS_FAMILY_MAX ||
484 !test_bit(family, &nd_desc->bus_family_mask))
486 family = array_index_nospec(family,
487 NVDIMM_BUS_FAMILY_MAX + 1);
488 dsm_mask = acpi_desc->family_dsm_mask[family];
489 guid = to_nfit_bus_uuid(family);
491 dsm_mask = acpi_desc->bus_dsm_mask;
492 guid = to_nfit_uuid(NFIT_DEV_BUS);
494 desc = nd_cmd_bus_desc(cmd);
495 handle = adev->handle;
499 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
503 * Check for a valid command. For ND_CMD_CALL, we also have to
504 * make sure that the DSM function is supported.
506 if (cmd == ND_CMD_CALL &&
507 (func > NVDIMM_CMD_MAX || !test_bit(func, &dsm_mask)))
509 else if (!test_bit(cmd, &cmd_mask))
512 in_obj.type = ACPI_TYPE_PACKAGE;
513 in_obj.package.count = 1;
514 in_obj.package.elements = &in_buf;
515 in_buf.type = ACPI_TYPE_BUFFER;
516 in_buf.buffer.pointer = buf;
517 in_buf.buffer.length = 0;
519 /* libnvdimm has already validated the input envelope */
520 for (i = 0; i < desc->in_num; i++)
521 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
525 /* skip over package wrapper */
526 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
527 in_buf.buffer.length = call_pkg->nd_size_in;
530 dev_dbg(dev, "%s cmd: %d: family: %d func: %d input length: %d\n",
531 dimm_name, cmd, family, func, in_buf.buffer.length);
532 if (payload_dumpable(nvdimm, func))
533 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
534 in_buf.buffer.pointer,
535 min_t(u32, 256, in_buf.buffer.length), true);
537 /* call the BIOS, prefer the named methods over _DSM if available */
538 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE
539 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags))
540 out_obj = acpi_label_info(handle);
541 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA
542 && test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
543 struct nd_cmd_get_config_data_hdr *p = buf;
545 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
546 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
547 && test_bit(NFIT_MEM_LSW, &nfit_mem->flags)) {
548 struct nd_cmd_set_config_hdr *p = buf;
550 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
556 revid = nfit_dsm_revid(nfit_mem->family, func);
559 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
563 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
567 if (out_obj->type != ACPI_TYPE_BUFFER) {
568 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
569 dimm_name, cmd_name, out_obj->type);
574 dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
575 cmd_name, out_obj->buffer.length);
576 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
577 out_obj->buffer.pointer,
578 min_t(u32, 128, out_obj->buffer.length), true);
581 call_pkg->nd_fw_size = out_obj->buffer.length;
582 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
583 out_obj->buffer.pointer,
584 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
588 * Need to support FW function w/o known size in advance.
589 * Caller can determine required size based upon nd_fw_size.
590 * If we return an error (like elsewhere) then caller wouldn't
591 * be able to rely upon data returned to make calculation.
598 for (i = 0, offset = 0; i < desc->out_num; i++) {
599 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
600 (u32 *) out_obj->buffer.pointer,
601 out_obj->buffer.length - offset);
603 if (offset + out_size > out_obj->buffer.length) {
604 dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
605 dimm_name, cmd_name, i);
609 if (in_buf.buffer.length + offset + out_size > buf_len) {
610 dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
611 dimm_name, cmd_name, i);
615 memcpy(buf + in_buf.buffer.length + offset,
616 out_obj->buffer.pointer + offset, out_size);
621 * Set fw_status for all the commands with a known format to be
622 * later interpreted by xlat_status().
624 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
625 && cmd <= ND_CMD_CLEAR_ERROR)
626 || (nvdimm && cmd >= ND_CMD_SMART
627 && cmd <= ND_CMD_VENDOR)))
628 fw_status = *(u32 *) out_obj->buffer.pointer;
630 if (offset + in_buf.buffer.length < buf_len) {
633 * status valid, return the number of bytes left
634 * unfilled in the output buffer
636 rc = buf_len - offset - in_buf.buffer.length;
638 *cmd_rc = xlat_status(nvdimm, buf, cmd,
641 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
642 __func__, dimm_name, cmd_name, buf_len,
649 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
657 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
659 static const char *spa_type_name(u16 type)
661 static const char *to_name[] = {
662 [NFIT_SPA_VOLATILE] = "volatile",
663 [NFIT_SPA_PM] = "pmem",
664 [NFIT_SPA_DCR] = "dimm-control-region",
665 [NFIT_SPA_BDW] = "block-data-window",
666 [NFIT_SPA_VDISK] = "volatile-disk",
667 [NFIT_SPA_VCD] = "volatile-cd",
668 [NFIT_SPA_PDISK] = "persistent-disk",
669 [NFIT_SPA_PCD] = "persistent-cd",
673 if (type > NFIT_SPA_PCD)
676 return to_name[type];
679 int nfit_spa_type(struct acpi_nfit_system_address *spa)
684 import_guid(&guid, spa->range_guid);
685 for (i = 0; i < NFIT_UUID_MAX; i++)
686 if (guid_equal(to_nfit_uuid(i), &guid))
691 static size_t sizeof_spa(struct acpi_nfit_system_address *spa)
693 if (spa->flags & ACPI_NFIT_LOCATION_COOKIE_VALID)
695 return sizeof(*spa) - 8;
698 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
699 struct nfit_table_prev *prev,
700 struct acpi_nfit_system_address *spa)
702 struct device *dev = acpi_desc->dev;
703 struct nfit_spa *nfit_spa;
705 if (spa->header.length != sizeof_spa(spa))
708 list_for_each_entry(nfit_spa, &prev->spas, list) {
709 if (memcmp(nfit_spa->spa, spa, sizeof_spa(spa)) == 0) {
710 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
715 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof_spa(spa),
719 INIT_LIST_HEAD(&nfit_spa->list);
720 memcpy(nfit_spa->spa, spa, sizeof_spa(spa));
721 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
722 dev_dbg(dev, "spa index: %d type: %s\n",
724 spa_type_name(nfit_spa_type(spa)));
728 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
729 struct nfit_table_prev *prev,
730 struct acpi_nfit_memory_map *memdev)
732 struct device *dev = acpi_desc->dev;
733 struct nfit_memdev *nfit_memdev;
735 if (memdev->header.length != sizeof(*memdev))
738 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
739 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
740 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
744 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
748 INIT_LIST_HEAD(&nfit_memdev->list);
749 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
750 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
751 dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
752 memdev->device_handle, memdev->range_index,
753 memdev->region_index, memdev->flags);
757 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
759 struct acpi_nfit_memory_map *memdev;
760 struct acpi_nfit_desc *acpi_desc;
761 struct nfit_mem *nfit_mem;
764 mutex_lock(&acpi_desc_lock);
765 list_for_each_entry(acpi_desc, &acpi_descs, list) {
766 mutex_lock(&acpi_desc->init_mutex);
767 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
768 memdev = __to_nfit_memdev(nfit_mem);
769 if (memdev->device_handle == device_handle) {
770 *flags = memdev->flags;
771 physical_id = memdev->physical_id;
772 mutex_unlock(&acpi_desc->init_mutex);
773 mutex_unlock(&acpi_desc_lock);
777 mutex_unlock(&acpi_desc->init_mutex);
779 mutex_unlock(&acpi_desc_lock);
783 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
786 * An implementation may provide a truncated control region if no block windows
789 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
791 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
796 return offsetof(struct acpi_nfit_control_region, window_size);
799 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
800 struct nfit_table_prev *prev,
801 struct acpi_nfit_control_region *dcr)
803 struct device *dev = acpi_desc->dev;
804 struct nfit_dcr *nfit_dcr;
806 if (!sizeof_dcr(dcr))
809 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
810 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
811 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
815 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
819 INIT_LIST_HEAD(&nfit_dcr->list);
820 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
821 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
822 dev_dbg(dev, "dcr index: %d windows: %d\n",
823 dcr->region_index, dcr->windows);
827 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
828 struct nfit_table_prev *prev,
829 struct acpi_nfit_data_region *bdw)
831 struct device *dev = acpi_desc->dev;
832 struct nfit_bdw *nfit_bdw;
834 if (bdw->header.length != sizeof(*bdw))
836 list_for_each_entry(nfit_bdw, &prev->bdws, list)
837 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
838 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
842 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
846 INIT_LIST_HEAD(&nfit_bdw->list);
847 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
848 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
849 dev_dbg(dev, "bdw dcr: %d windows: %d\n",
850 bdw->region_index, bdw->windows);
854 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
856 if (idt->header.length < sizeof(*idt))
858 return sizeof(*idt) + sizeof(u32) * idt->line_count;
861 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
862 struct nfit_table_prev *prev,
863 struct acpi_nfit_interleave *idt)
865 struct device *dev = acpi_desc->dev;
866 struct nfit_idt *nfit_idt;
868 if (!sizeof_idt(idt))
871 list_for_each_entry(nfit_idt, &prev->idts, list) {
872 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
875 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
876 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
881 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
885 INIT_LIST_HEAD(&nfit_idt->list);
886 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
887 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
888 dev_dbg(dev, "idt index: %d num_lines: %d\n",
889 idt->interleave_index, idt->line_count);
893 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
895 if (flush->header.length < sizeof(*flush))
897 return struct_size(flush, hint_address, flush->hint_count);
900 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
901 struct nfit_table_prev *prev,
902 struct acpi_nfit_flush_address *flush)
904 struct device *dev = acpi_desc->dev;
905 struct nfit_flush *nfit_flush;
907 if (!sizeof_flush(flush))
910 list_for_each_entry(nfit_flush, &prev->flushes, list) {
911 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
914 if (memcmp(nfit_flush->flush, flush,
915 sizeof_flush(flush)) == 0) {
916 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
921 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
922 + sizeof_flush(flush), GFP_KERNEL);
925 INIT_LIST_HEAD(&nfit_flush->list);
926 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
927 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
928 dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
929 flush->device_handle, flush->hint_count);
933 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
934 struct acpi_nfit_capabilities *pcap)
936 struct device *dev = acpi_desc->dev;
939 mask = (1 << (pcap->highest_capability + 1)) - 1;
940 acpi_desc->platform_cap = pcap->capabilities & mask;
941 dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
945 static void *add_table(struct acpi_nfit_desc *acpi_desc,
946 struct nfit_table_prev *prev, void *table, const void *end)
948 struct device *dev = acpi_desc->dev;
949 struct acpi_nfit_header *hdr;
950 void *err = ERR_PTR(-ENOMEM);
957 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
963 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
964 if (!add_spa(acpi_desc, prev, table))
967 case ACPI_NFIT_TYPE_MEMORY_MAP:
968 if (!add_memdev(acpi_desc, prev, table))
971 case ACPI_NFIT_TYPE_CONTROL_REGION:
972 if (!add_dcr(acpi_desc, prev, table))
975 case ACPI_NFIT_TYPE_DATA_REGION:
976 if (!add_bdw(acpi_desc, prev, table))
979 case ACPI_NFIT_TYPE_INTERLEAVE:
980 if (!add_idt(acpi_desc, prev, table))
983 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
984 if (!add_flush(acpi_desc, prev, table))
987 case ACPI_NFIT_TYPE_SMBIOS:
988 dev_dbg(dev, "smbios\n");
990 case ACPI_NFIT_TYPE_CAPABILITIES:
991 if (!add_platform_cap(acpi_desc, table))
995 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
999 return table + hdr->length;
1002 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1003 struct acpi_nfit_system_address *spa)
1005 struct nfit_mem *nfit_mem, *found;
1006 struct nfit_memdev *nfit_memdev;
1007 int type = spa ? nfit_spa_type(spa) : 0;
1019 * This loop runs in two modes, when a dimm is mapped the loop
1020 * adds memdev associations to an existing dimm, or creates a
1021 * dimm. In the unmapped dimm case this loop sweeps for memdev
1022 * instances with an invalid / zero range_index and adds those
1023 * dimms without spa associations.
1025 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1026 struct nfit_flush *nfit_flush;
1027 struct nfit_dcr *nfit_dcr;
1031 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1033 if (!spa && nfit_memdev->memdev->range_index)
1036 dcr = nfit_memdev->memdev->region_index;
1037 device_handle = nfit_memdev->memdev->device_handle;
1038 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1039 if (__to_nfit_memdev(nfit_mem)->device_handle
1048 nfit_mem = devm_kzalloc(acpi_desc->dev,
1049 sizeof(*nfit_mem), GFP_KERNEL);
1052 INIT_LIST_HEAD(&nfit_mem->list);
1053 nfit_mem->acpi_desc = acpi_desc;
1054 list_add(&nfit_mem->list, &acpi_desc->dimms);
1057 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1058 if (nfit_dcr->dcr->region_index != dcr)
1061 * Record the control region for the dimm. For
1062 * the ACPI 6.1 case, where there are separate
1063 * control regions for the pmem vs blk
1064 * interfaces, be sure to record the extended
1068 nfit_mem->dcr = nfit_dcr->dcr;
1069 else if (nfit_mem->dcr->windows == 0
1070 && nfit_dcr->dcr->windows)
1071 nfit_mem->dcr = nfit_dcr->dcr;
1075 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1076 struct acpi_nfit_flush_address *flush;
1079 if (nfit_flush->flush->device_handle != device_handle)
1081 nfit_mem->nfit_flush = nfit_flush;
1082 flush = nfit_flush->flush;
1083 nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1085 sizeof(struct resource),
1087 if (!nfit_mem->flush_wpq)
1089 for (i = 0; i < flush->hint_count; i++) {
1090 struct resource *res = &nfit_mem->flush_wpq[i];
1092 res->start = flush->hint_address[i];
1093 res->end = res->start + 8 - 1;
1098 if (dcr && !nfit_mem->dcr) {
1099 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1100 spa->range_index, dcr);
1104 if (type == NFIT_SPA_DCR) {
1105 struct nfit_idt *nfit_idt;
1108 /* multiple dimms may share a SPA when interleaved */
1109 nfit_mem->spa_dcr = spa;
1110 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1111 idt_idx = nfit_memdev->memdev->interleave_index;
1112 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1113 if (nfit_idt->idt->interleave_index != idt_idx)
1115 nfit_mem->idt_dcr = nfit_idt->idt;
1118 } else if (type == NFIT_SPA_PM) {
1120 * A single dimm may belong to multiple SPA-PM
1121 * ranges, record at least one in addition to
1122 * any SPA-DCR range.
1124 nfit_mem->memdev_pmem = nfit_memdev->memdev;
1126 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1132 static int nfit_mem_cmp(void *priv, const struct list_head *_a,
1133 const struct list_head *_b)
1135 struct nfit_mem *a = container_of(_a, typeof(*a), list);
1136 struct nfit_mem *b = container_of(_b, typeof(*b), list);
1137 u32 handleA, handleB;
1139 handleA = __to_nfit_memdev(a)->device_handle;
1140 handleB = __to_nfit_memdev(b)->device_handle;
1141 if (handleA < handleB)
1143 else if (handleA > handleB)
1148 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1150 struct nfit_spa *nfit_spa;
1155 * For each SPA-DCR or SPA-PMEM address range find its
1156 * corresponding MEMDEV(s). From each MEMDEV find the
1157 * corresponding DCR. Then, if we're operating on a SPA-DCR,
1158 * try to find a SPA-BDW and a corresponding BDW that references
1159 * the DCR. Throw it all into an nfit_mem object. Note, that
1160 * BDWs are optional.
1162 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1163 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1169 * If a DIMM has failed to be mapped into SPA there will be no
1170 * SPA entries above. Find and register all the unmapped DIMMs
1171 * for reporting and recovery purposes.
1173 rc = __nfit_mem_init(acpi_desc, NULL);
1177 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1182 static ssize_t bus_dsm_mask_show(struct device *dev,
1183 struct device_attribute *attr, char *buf)
1185 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1186 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1187 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1189 return sprintf(buf, "%#lx\n", acpi_desc->bus_dsm_mask);
1191 static struct device_attribute dev_attr_bus_dsm_mask =
1192 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1194 static ssize_t revision_show(struct device *dev,
1195 struct device_attribute *attr, char *buf)
1197 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1198 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1199 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1201 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1203 static DEVICE_ATTR_RO(revision);
1205 static ssize_t hw_error_scrub_show(struct device *dev,
1206 struct device_attribute *attr, char *buf)
1208 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1209 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1210 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1212 return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1216 * The 'hw_error_scrub' attribute can have the following values written to it:
1217 * '0': Switch to the default mode where an exception will only insert
1218 * the address of the memory error into the poison and badblocks lists.
1219 * '1': Enable a full scrub to happen if an exception for a memory error is
1222 static ssize_t hw_error_scrub_store(struct device *dev,
1223 struct device_attribute *attr, const char *buf, size_t size)
1225 struct nvdimm_bus_descriptor *nd_desc;
1229 rc = kstrtol(buf, 0, &val);
1234 nd_desc = dev_get_drvdata(dev);
1236 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1239 case HW_ERROR_SCRUB_ON:
1240 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1242 case HW_ERROR_SCRUB_OFF:
1243 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1255 static DEVICE_ATTR_RW(hw_error_scrub);
1258 * This shows the number of full Address Range Scrubs that have been
1259 * completed since driver load time. Userspace can wait on this using
1260 * select/poll etc. A '+' at the end indicates an ARS is in progress
1262 static ssize_t scrub_show(struct device *dev,
1263 struct device_attribute *attr, char *buf)
1265 struct nvdimm_bus_descriptor *nd_desc;
1266 struct acpi_nfit_desc *acpi_desc;
1267 ssize_t rc = -ENXIO;
1271 nd_desc = dev_get_drvdata(dev);
1276 acpi_desc = to_acpi_desc(nd_desc);
1278 mutex_lock(&acpi_desc->init_mutex);
1279 busy = test_bit(ARS_BUSY, &acpi_desc->scrub_flags)
1280 && !test_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
1281 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count, busy ? "+\n" : "\n");
1282 /* Allow an admin to poll the busy state at a higher rate */
1283 if (busy && capable(CAP_SYS_RAWIO) && !test_and_set_bit(ARS_POLL,
1284 &acpi_desc->scrub_flags)) {
1285 acpi_desc->scrub_tmo = 1;
1286 mod_delayed_work(nfit_wq, &acpi_desc->dwork, HZ);
1289 mutex_unlock(&acpi_desc->init_mutex);
1294 static ssize_t scrub_store(struct device *dev,
1295 struct device_attribute *attr, const char *buf, size_t size)
1297 struct nvdimm_bus_descriptor *nd_desc;
1301 rc = kstrtol(buf, 0, &val);
1308 nd_desc = dev_get_drvdata(dev);
1310 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1312 rc = acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
1319 static DEVICE_ATTR_RW(scrub);
1321 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1323 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1324 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1325 | 1 << ND_CMD_ARS_STATUS;
1327 return (nd_desc->cmd_mask & mask) == mask;
1330 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1332 struct device *dev = kobj_to_dev(kobj);
1333 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1335 if (a == &dev_attr_scrub.attr)
1336 return ars_supported(nvdimm_bus) ? a->mode : 0;
1338 if (a == &dev_attr_firmware_activate_noidle.attr)
1339 return intel_fwa_supported(nvdimm_bus) ? a->mode : 0;
1344 static struct attribute *acpi_nfit_attributes[] = {
1345 &dev_attr_revision.attr,
1346 &dev_attr_scrub.attr,
1347 &dev_attr_hw_error_scrub.attr,
1348 &dev_attr_bus_dsm_mask.attr,
1349 &dev_attr_firmware_activate_noidle.attr,
1353 static const struct attribute_group acpi_nfit_attribute_group = {
1355 .attrs = acpi_nfit_attributes,
1356 .is_visible = nfit_visible,
1359 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1360 &acpi_nfit_attribute_group,
1364 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1366 struct nvdimm *nvdimm = to_nvdimm(dev);
1367 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1369 return __to_nfit_memdev(nfit_mem);
1372 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1374 struct nvdimm *nvdimm = to_nvdimm(dev);
1375 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1377 return nfit_mem->dcr;
1380 static ssize_t handle_show(struct device *dev,
1381 struct device_attribute *attr, char *buf)
1383 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1385 return sprintf(buf, "%#x\n", memdev->device_handle);
1387 static DEVICE_ATTR_RO(handle);
1389 static ssize_t phys_id_show(struct device *dev,
1390 struct device_attribute *attr, char *buf)
1392 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1394 return sprintf(buf, "%#x\n", memdev->physical_id);
1396 static DEVICE_ATTR_RO(phys_id);
1398 static ssize_t vendor_show(struct device *dev,
1399 struct device_attribute *attr, char *buf)
1401 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1403 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1405 static DEVICE_ATTR_RO(vendor);
1407 static ssize_t rev_id_show(struct device *dev,
1408 struct device_attribute *attr, char *buf)
1410 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1412 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1414 static DEVICE_ATTR_RO(rev_id);
1416 static ssize_t device_show(struct device *dev,
1417 struct device_attribute *attr, char *buf)
1419 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1421 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1423 static DEVICE_ATTR_RO(device);
1425 static ssize_t subsystem_vendor_show(struct device *dev,
1426 struct device_attribute *attr, char *buf)
1428 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1430 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1432 static DEVICE_ATTR_RO(subsystem_vendor);
1434 static ssize_t subsystem_rev_id_show(struct device *dev,
1435 struct device_attribute *attr, char *buf)
1437 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1439 return sprintf(buf, "0x%04x\n",
1440 be16_to_cpu(dcr->subsystem_revision_id));
1442 static DEVICE_ATTR_RO(subsystem_rev_id);
1444 static ssize_t subsystem_device_show(struct device *dev,
1445 struct device_attribute *attr, char *buf)
1447 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1449 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1451 static DEVICE_ATTR_RO(subsystem_device);
1453 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1455 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1458 if (nfit_mem->memdev_pmem)
1463 static ssize_t format_show(struct device *dev,
1464 struct device_attribute *attr, char *buf)
1466 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1468 return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1470 static DEVICE_ATTR_RO(format);
1472 static ssize_t format1_show(struct device *dev,
1473 struct device_attribute *attr, char *buf)
1476 ssize_t rc = -ENXIO;
1477 struct nfit_mem *nfit_mem;
1478 struct nfit_memdev *nfit_memdev;
1479 struct acpi_nfit_desc *acpi_desc;
1480 struct nvdimm *nvdimm = to_nvdimm(dev);
1481 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1483 nfit_mem = nvdimm_provider_data(nvdimm);
1484 acpi_desc = nfit_mem->acpi_desc;
1485 handle = to_nfit_memdev(dev)->device_handle;
1487 /* assumes DIMMs have at most 2 published interface codes */
1488 mutex_lock(&acpi_desc->init_mutex);
1489 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1490 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1491 struct nfit_dcr *nfit_dcr;
1493 if (memdev->device_handle != handle)
1496 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1497 if (nfit_dcr->dcr->region_index != memdev->region_index)
1499 if (nfit_dcr->dcr->code == dcr->code)
1501 rc = sprintf(buf, "0x%04x\n",
1502 le16_to_cpu(nfit_dcr->dcr->code));
1508 mutex_unlock(&acpi_desc->init_mutex);
1511 static DEVICE_ATTR_RO(format1);
1513 static ssize_t formats_show(struct device *dev,
1514 struct device_attribute *attr, char *buf)
1516 struct nvdimm *nvdimm = to_nvdimm(dev);
1518 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1520 static DEVICE_ATTR_RO(formats);
1522 static ssize_t serial_show(struct device *dev,
1523 struct device_attribute *attr, char *buf)
1525 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1527 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1529 static DEVICE_ATTR_RO(serial);
1531 static ssize_t family_show(struct device *dev,
1532 struct device_attribute *attr, char *buf)
1534 struct nvdimm *nvdimm = to_nvdimm(dev);
1535 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1537 if (nfit_mem->family < 0)
1539 return sprintf(buf, "%d\n", nfit_mem->family);
1541 static DEVICE_ATTR_RO(family);
1543 static ssize_t dsm_mask_show(struct device *dev,
1544 struct device_attribute *attr, char *buf)
1546 struct nvdimm *nvdimm = to_nvdimm(dev);
1547 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1549 if (nfit_mem->family < 0)
1551 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1553 static DEVICE_ATTR_RO(dsm_mask);
1555 static ssize_t flags_show(struct device *dev,
1556 struct device_attribute *attr, char *buf)
1558 struct nvdimm *nvdimm = to_nvdimm(dev);
1559 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1560 u16 flags = __to_nfit_memdev(nfit_mem)->flags;
1562 if (test_bit(NFIT_MEM_DIRTY, &nfit_mem->flags))
1563 flags |= ACPI_NFIT_MEM_FLUSH_FAILED;
1565 return sprintf(buf, "%s%s%s%s%s%s%s\n",
1566 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1567 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1568 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1569 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1570 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1571 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1572 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1574 static DEVICE_ATTR_RO(flags);
1576 static ssize_t id_show(struct device *dev,
1577 struct device_attribute *attr, char *buf)
1579 struct nvdimm *nvdimm = to_nvdimm(dev);
1580 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1582 return sprintf(buf, "%s\n", nfit_mem->id);
1584 static DEVICE_ATTR_RO(id);
1586 static ssize_t dirty_shutdown_show(struct device *dev,
1587 struct device_attribute *attr, char *buf)
1589 struct nvdimm *nvdimm = to_nvdimm(dev);
1590 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1592 return sprintf(buf, "%d\n", nfit_mem->dirty_shutdown);
1594 static DEVICE_ATTR_RO(dirty_shutdown);
1596 static struct attribute *acpi_nfit_dimm_attributes[] = {
1597 &dev_attr_handle.attr,
1598 &dev_attr_phys_id.attr,
1599 &dev_attr_vendor.attr,
1600 &dev_attr_device.attr,
1601 &dev_attr_rev_id.attr,
1602 &dev_attr_subsystem_vendor.attr,
1603 &dev_attr_subsystem_device.attr,
1604 &dev_attr_subsystem_rev_id.attr,
1605 &dev_attr_format.attr,
1606 &dev_attr_formats.attr,
1607 &dev_attr_format1.attr,
1608 &dev_attr_serial.attr,
1609 &dev_attr_flags.attr,
1611 &dev_attr_family.attr,
1612 &dev_attr_dsm_mask.attr,
1613 &dev_attr_dirty_shutdown.attr,
1617 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1618 struct attribute *a, int n)
1620 struct device *dev = kobj_to_dev(kobj);
1621 struct nvdimm *nvdimm = to_nvdimm(dev);
1622 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1624 if (!to_nfit_dcr(dev)) {
1625 /* Without a dcr only the memdev attributes can be surfaced */
1626 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1627 || a == &dev_attr_flags.attr
1628 || a == &dev_attr_family.attr
1629 || a == &dev_attr_dsm_mask.attr)
1634 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1637 if (!test_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags)
1638 && a == &dev_attr_dirty_shutdown.attr)
1644 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1646 .attrs = acpi_nfit_dimm_attributes,
1647 .is_visible = acpi_nfit_dimm_attr_visible,
1650 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1651 &acpi_nfit_dimm_attribute_group,
1655 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1658 struct nfit_mem *nfit_mem;
1660 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1661 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1662 return nfit_mem->nvdimm;
1667 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1669 struct nfit_mem *nfit_mem;
1670 struct acpi_nfit_desc *acpi_desc;
1672 dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1675 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1676 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1681 acpi_desc = dev_get_drvdata(dev->parent);
1686 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1689 nfit_mem = dev_get_drvdata(dev);
1690 if (nfit_mem && nfit_mem->flags_attr)
1691 sysfs_notify_dirent(nfit_mem->flags_attr);
1693 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1695 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1697 struct acpi_device *adev = data;
1698 struct device *dev = &adev->dev;
1700 device_lock(dev->parent);
1701 __acpi_nvdimm_notify(dev, event);
1702 device_unlock(dev->parent);
1705 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1710 status = acpi_get_handle(adev->handle, method, &handle);
1712 if (ACPI_SUCCESS(status))
1717 __weak void nfit_intel_shutdown_status(struct nfit_mem *nfit_mem)
1719 struct device *dev = &nfit_mem->adev->dev;
1720 struct nd_intel_smart smart = { 0 };
1721 union acpi_object in_buf = {
1722 .buffer.type = ACPI_TYPE_BUFFER,
1725 union acpi_object in_obj = {
1726 .package.type = ACPI_TYPE_PACKAGE,
1728 .package.elements = &in_buf,
1730 const u8 func = ND_INTEL_SMART;
1731 const guid_t *guid = to_nfit_uuid(nfit_mem->family);
1732 u8 revid = nfit_dsm_revid(nfit_mem->family, func);
1733 struct acpi_device *adev = nfit_mem->adev;
1734 acpi_handle handle = adev->handle;
1735 union acpi_object *out_obj;
1737 if ((nfit_mem->dsm_mask & (1 << func)) == 0)
1740 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
1741 if (!out_obj || out_obj->type != ACPI_TYPE_BUFFER
1742 || out_obj->buffer.length < sizeof(smart)) {
1743 dev_dbg(dev->parent, "%s: failed to retrieve initial health\n",
1748 memcpy(&smart, out_obj->buffer.pointer, sizeof(smart));
1751 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_VALID) {
1752 if (smart.shutdown_state)
1753 set_bit(NFIT_MEM_DIRTY, &nfit_mem->flags);
1756 if (smart.flags & ND_INTEL_SMART_SHUTDOWN_COUNT_VALID) {
1757 set_bit(NFIT_MEM_DIRTY_COUNT, &nfit_mem->flags);
1758 nfit_mem->dirty_shutdown = smart.shutdown_count;
1762 static void populate_shutdown_status(struct nfit_mem *nfit_mem)
1765 * For DIMMs that provide a dynamic facility to retrieve a
1766 * dirty-shutdown status and/or a dirty-shutdown count, cache
1767 * these values in nfit_mem.
1769 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1770 nfit_intel_shutdown_status(nfit_mem);
1773 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1774 struct nfit_mem *nfit_mem, u32 device_handle)
1776 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1777 struct acpi_device *adev, *adev_dimm;
1778 struct device *dev = acpi_desc->dev;
1779 unsigned long dsm_mask, label_mask;
1783 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1785 /* nfit test assumes 1:1 relationship between commands and dsms */
1786 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1787 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1788 set_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1790 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1791 sprintf(nfit_mem->id, "%04x-%02x-%04x-%08x",
1792 be16_to_cpu(dcr->vendor_id),
1793 dcr->manufacturing_location,
1794 be16_to_cpu(dcr->manufacturing_date),
1795 be32_to_cpu(dcr->serial_number));
1797 sprintf(nfit_mem->id, "%04x-%08x",
1798 be16_to_cpu(dcr->vendor_id),
1799 be32_to_cpu(dcr->serial_number));
1801 adev = to_acpi_dev(acpi_desc);
1803 /* unit test case */
1804 populate_shutdown_status(nfit_mem);
1808 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1809 nfit_mem->adev = adev_dimm;
1811 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1813 return force_enable_dimms ? 0 : -ENODEV;
1816 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1817 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1818 dev_err(dev, "%s: notification registration failed\n",
1819 dev_name(&adev_dimm->dev));
1823 * Record nfit_mem for the notification path to track back to
1824 * the nfit sysfs attributes for this dimm device object.
1826 dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1829 * There are 4 "legacy" NVDIMM command sets
1830 * (NVDIMM_FAMILY_{INTEL,MSFT,HPE1,HPE2}) that were created before
1831 * an EFI working group was established to constrain this
1832 * proliferation. The nfit driver probes for the supported command
1833 * set by GUID. Note, if you're a platform developer looking to add
1834 * a new command set to this probe, consider using an existing set,
1835 * or otherwise seek approval to publish the command set at
1836 * http://www.uefi.org/RFIC_LIST.
1838 * Note, that checking for function0 (bit0) tells us if any commands
1839 * are reachable through this GUID.
1841 clear_bit(NVDIMM_FAMILY_INTEL, &nd_desc->dimm_family_mask);
1842 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1843 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1)) {
1844 set_bit(i, &nd_desc->dimm_family_mask);
1845 if (family < 0 || i == default_dsm_family)
1849 /* limit the supported commands to those that are publicly documented */
1850 nfit_mem->family = family;
1851 if (override_dsm_mask && !disable_vendor_specific)
1852 dsm_mask = override_dsm_mask;
1853 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1854 dsm_mask = NVDIMM_INTEL_CMDMASK;
1855 if (disable_vendor_specific)
1856 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1857 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1858 dsm_mask = 0x1c3c76;
1859 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1861 if (disable_vendor_specific)
1862 dsm_mask &= ~(1 << 8);
1863 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1864 dsm_mask = 0xffffffff;
1865 } else if (nfit_mem->family == NVDIMM_FAMILY_HYPERV) {
1868 dev_dbg(dev, "unknown dimm command family\n");
1869 nfit_mem->family = -1;
1870 /* DSMs are optional, continue loading the driver... */
1875 * Function 0 is the command interrogation function, don't
1876 * export it to potential userspace use, and enable it to be
1877 * used as an error value in acpi_nfit_ctl().
1881 guid = to_nfit_uuid(nfit_mem->family);
1882 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1883 if (acpi_check_dsm(adev_dimm->handle, guid,
1884 nfit_dsm_revid(nfit_mem->family, i),
1886 set_bit(i, &nfit_mem->dsm_mask);
1889 * Prefer the NVDIMM_FAMILY_INTEL label read commands if present
1890 * due to their better semantics handling locked capacity.
1892 label_mask = 1 << ND_CMD_GET_CONFIG_SIZE | 1 << ND_CMD_GET_CONFIG_DATA
1893 | 1 << ND_CMD_SET_CONFIG_DATA;
1894 if (family == NVDIMM_FAMILY_INTEL
1895 && (dsm_mask & label_mask) == label_mask)
1896 /* skip _LS{I,R,W} enabling */;
1898 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1899 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1900 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1901 set_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1904 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)
1905 && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1906 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1907 set_bit(NFIT_MEM_LSW, &nfit_mem->flags);
1911 * Quirk read-only label configurations to preserve
1912 * access to label-less namespaces by default.
1914 if (!test_bit(NFIT_MEM_LSW, &nfit_mem->flags)
1916 dev_dbg(dev, "%s: No _LSW, disable labels\n",
1917 dev_name(&adev_dimm->dev));
1918 clear_bit(NFIT_MEM_LSR, &nfit_mem->flags);
1920 dev_dbg(dev, "%s: Force enable labels\n",
1921 dev_name(&adev_dimm->dev));
1924 populate_shutdown_status(nfit_mem);
1929 static void shutdown_dimm_notify(void *data)
1931 struct acpi_nfit_desc *acpi_desc = data;
1932 struct nfit_mem *nfit_mem;
1934 mutex_lock(&acpi_desc->init_mutex);
1936 * Clear out the nfit_mem->flags_attr and shut down dimm event
1939 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1940 struct acpi_device *adev_dimm = nfit_mem->adev;
1942 if (nfit_mem->flags_attr) {
1943 sysfs_put(nfit_mem->flags_attr);
1944 nfit_mem->flags_attr = NULL;
1947 acpi_remove_notify_handler(adev_dimm->handle,
1948 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1949 dev_set_drvdata(&adev_dimm->dev, NULL);
1952 mutex_unlock(&acpi_desc->init_mutex);
1955 static const struct nvdimm_security_ops *acpi_nfit_get_security_ops(int family)
1958 case NVDIMM_FAMILY_INTEL:
1959 return intel_security_ops;
1965 static const struct nvdimm_fw_ops *acpi_nfit_get_fw_ops(
1966 struct nfit_mem *nfit_mem)
1969 struct acpi_nfit_desc *acpi_desc = nfit_mem->acpi_desc;
1970 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1972 if (!nd_desc->fw_ops)
1975 if (nfit_mem->family != NVDIMM_FAMILY_INTEL)
1978 mask = nfit_mem->dsm_mask & NVDIMM_INTEL_FW_ACTIVATE_CMDMASK;
1979 if (mask != NVDIMM_INTEL_FW_ACTIVATE_CMDMASK)
1982 return intel_fw_ops;
1985 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1987 struct nfit_mem *nfit_mem;
1988 int dimm_count = 0, rc;
1989 struct nvdimm *nvdimm;
1991 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1992 struct acpi_nfit_flush_address *flush;
1993 unsigned long flags = 0, cmd_mask;
1994 struct nfit_memdev *nfit_memdev;
1998 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1999 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
2005 /* collate flags across all memdevs for this dimm */
2006 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2007 struct acpi_nfit_memory_map *dimm_memdev;
2009 dimm_memdev = __to_nfit_memdev(nfit_mem);
2010 if (dimm_memdev->device_handle
2011 != nfit_memdev->memdev->device_handle)
2013 dimm_memdev->flags |= nfit_memdev->memdev->flags;
2016 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
2017 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
2018 set_bit(NDD_UNARMED, &flags);
2020 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
2025 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
2026 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
2027 * userspace interface.
2029 cmd_mask = 1UL << ND_CMD_CALL;
2030 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
2032 * These commands have a 1:1 correspondence
2033 * between DSM payload and libnvdimm ioctl
2036 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
2039 if (test_bit(NFIT_MEM_LSR, &nfit_mem->flags)) {
2040 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
2041 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
2043 if (test_bit(NFIT_MEM_LSW, &nfit_mem->flags))
2044 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
2046 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
2048 nvdimm = __nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
2049 acpi_nfit_dimm_attribute_groups,
2050 flags, cmd_mask, flush ? flush->hint_count : 0,
2051 nfit_mem->flush_wpq, &nfit_mem->id[0],
2052 acpi_nfit_get_security_ops(nfit_mem->family),
2053 acpi_nfit_get_fw_ops(nfit_mem));
2057 nfit_mem->nvdimm = nvdimm;
2060 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
2063 dev_err(acpi_desc->dev, "Error found in NVDIMM %s flags:%s%s%s%s%s\n",
2064 nvdimm_name(nvdimm),
2065 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
2066 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
2067 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
2068 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
2069 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
2073 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
2078 * Now that dimms are successfully registered, and async registration
2079 * is flushed, attempt to enable event notification.
2081 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
2082 struct kernfs_node *nfit_kernfs;
2084 nvdimm = nfit_mem->nvdimm;
2088 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
2090 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
2092 sysfs_put(nfit_kernfs);
2093 if (!nfit_mem->flags_attr)
2094 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
2095 nvdimm_name(nvdimm));
2098 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
2103 * These constants are private because there are no kernel consumers of
2106 enum nfit_aux_cmds {
2107 NFIT_CMD_TRANSLATE_SPA = 5,
2108 NFIT_CMD_ARS_INJECT_SET = 7,
2109 NFIT_CMD_ARS_INJECT_CLEAR = 8,
2110 NFIT_CMD_ARS_INJECT_GET = 9,
2113 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
2115 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2116 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
2117 unsigned long dsm_mask, *mask;
2118 struct acpi_device *adev;
2121 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
2122 set_bit(NVDIMM_BUS_FAMILY_NFIT, &nd_desc->bus_family_mask);
2124 /* enable nfit_test to inject bus command emulation */
2125 if (acpi_desc->bus_cmd_force_en) {
2126 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
2127 mask = &nd_desc->bus_family_mask;
2128 if (acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL]) {
2129 set_bit(NVDIMM_BUS_FAMILY_INTEL, mask);
2130 nd_desc->fw_ops = intel_bus_fw_ops;
2134 adev = to_acpi_dev(acpi_desc);
2138 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
2139 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2140 set_bit(i, &nd_desc->cmd_mask);
2143 (1 << ND_CMD_ARS_CAP) |
2144 (1 << ND_CMD_ARS_START) |
2145 (1 << ND_CMD_ARS_STATUS) |
2146 (1 << ND_CMD_CLEAR_ERROR) |
2147 (1 << NFIT_CMD_TRANSLATE_SPA) |
2148 (1 << NFIT_CMD_ARS_INJECT_SET) |
2149 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
2150 (1 << NFIT_CMD_ARS_INJECT_GET);
2151 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2152 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2153 set_bit(i, &acpi_desc->bus_dsm_mask);
2155 /* Enumerate allowed NVDIMM_BUS_FAMILY_INTEL commands */
2156 dsm_mask = NVDIMM_BUS_INTEL_FW_ACTIVATE_CMDMASK;
2157 guid = to_nfit_bus_uuid(NVDIMM_BUS_FAMILY_INTEL);
2158 mask = &acpi_desc->family_dsm_mask[NVDIMM_BUS_FAMILY_INTEL];
2159 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
2160 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
2163 if (*mask == dsm_mask) {
2164 set_bit(NVDIMM_BUS_FAMILY_INTEL, &nd_desc->bus_family_mask);
2165 nd_desc->fw_ops = intel_bus_fw_ops;
2169 static ssize_t range_index_show(struct device *dev,
2170 struct device_attribute *attr, char *buf)
2172 struct nd_region *nd_region = to_nd_region(dev);
2173 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
2175 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
2177 static DEVICE_ATTR_RO(range_index);
2179 static struct attribute *acpi_nfit_region_attributes[] = {
2180 &dev_attr_range_index.attr,
2184 static const struct attribute_group acpi_nfit_region_attribute_group = {
2186 .attrs = acpi_nfit_region_attributes,
2189 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2190 &acpi_nfit_region_attribute_group,
2194 /* enough info to uniquely specify an interleave set */
2195 struct nfit_set_info {
2201 struct nfit_set_info2 {
2205 u16 manufacturing_date;
2206 u8 manufacturing_location;
2210 static int cmp_map_compat(const void *m0, const void *m1)
2212 const struct nfit_set_info *map0 = m0;
2213 const struct nfit_set_info *map1 = m1;
2215 return memcmp(&map0->region_offset, &map1->region_offset,
2219 static int cmp_map(const void *m0, const void *m1)
2221 const struct nfit_set_info *map0 = m0;
2222 const struct nfit_set_info *map1 = m1;
2224 if (map0->region_offset < map1->region_offset)
2226 else if (map0->region_offset > map1->region_offset)
2231 static int cmp_map2(const void *m0, const void *m1)
2233 const struct nfit_set_info2 *map0 = m0;
2234 const struct nfit_set_info2 *map1 = m1;
2236 if (map0->region_offset < map1->region_offset)
2238 else if (map0->region_offset > map1->region_offset)
2243 /* Retrieve the nth entry referencing this spa */
2244 static struct acpi_nfit_memory_map *memdev_from_spa(
2245 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2247 struct nfit_memdev *nfit_memdev;
2249 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2250 if (nfit_memdev->memdev->range_index == range_index)
2252 return nfit_memdev->memdev;
2256 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2257 struct nd_region_desc *ndr_desc,
2258 struct acpi_nfit_system_address *spa)
2260 struct device *dev = acpi_desc->dev;
2261 struct nd_interleave_set *nd_set;
2262 u16 nr = ndr_desc->num_mappings;
2263 struct nfit_set_info2 *info2;
2264 struct nfit_set_info *info;
2267 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2270 import_guid(&nd_set->type_guid, spa->range_guid);
2272 info = devm_kcalloc(dev, nr, sizeof(*info), GFP_KERNEL);
2276 info2 = devm_kcalloc(dev, nr, sizeof(*info2), GFP_KERNEL);
2280 for (i = 0; i < nr; i++) {
2281 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2282 struct nvdimm *nvdimm = mapping->nvdimm;
2283 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2284 struct nfit_set_info *map = &info[i];
2285 struct nfit_set_info2 *map2 = &info2[i];
2286 struct acpi_nfit_memory_map *memdev =
2287 memdev_from_spa(acpi_desc, spa->range_index, i);
2288 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2290 if (!memdev || !nfit_mem->dcr) {
2291 dev_err(dev, "%s: failed to find DCR\n", __func__);
2295 map->region_offset = memdev->region_offset;
2296 map->serial_number = dcr->serial_number;
2298 map2->region_offset = memdev->region_offset;
2299 map2->serial_number = dcr->serial_number;
2300 map2->vendor_id = dcr->vendor_id;
2301 map2->manufacturing_date = dcr->manufacturing_date;
2302 map2->manufacturing_location = dcr->manufacturing_location;
2305 /* v1.1 namespaces */
2306 sort(info, nr, sizeof(*info), cmp_map, NULL);
2307 nd_set->cookie1 = nd_fletcher64(info, sizeof(*info) * nr, 0);
2309 /* v1.2 namespaces */
2310 sort(info2, nr, sizeof(*info2), cmp_map2, NULL);
2311 nd_set->cookie2 = nd_fletcher64(info2, sizeof(*info2) * nr, 0);
2313 /* support v1.1 namespaces created with the wrong sort order */
2314 sort(info, nr, sizeof(*info), cmp_map_compat, NULL);
2315 nd_set->altcookie = nd_fletcher64(info, sizeof(*info) * nr, 0);
2317 /* record the result of the sort for the mapping position */
2318 for (i = 0; i < nr; i++) {
2319 struct nfit_set_info2 *map2 = &info2[i];
2322 for (j = 0; j < nr; j++) {
2323 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2324 struct nvdimm *nvdimm = mapping->nvdimm;
2325 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2326 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2328 if (map2->serial_number == dcr->serial_number &&
2329 map2->vendor_id == dcr->vendor_id &&
2330 map2->manufacturing_date == dcr->manufacturing_date &&
2331 map2->manufacturing_location
2332 == dcr->manufacturing_location) {
2333 mapping->position = i;
2339 ndr_desc->nd_set = nd_set;
2340 devm_kfree(dev, info);
2341 devm_kfree(dev, info2);
2346 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2347 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2349 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2350 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2353 cmd->address = spa->address;
2354 cmd->length = spa->length;
2355 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2356 sizeof(*cmd), &cmd_rc);
2362 static int ars_start(struct acpi_nfit_desc *acpi_desc,
2363 struct nfit_spa *nfit_spa, enum nfit_ars_state req_type)
2367 struct nd_cmd_ars_start ars_start;
2368 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2369 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2371 memset(&ars_start, 0, sizeof(ars_start));
2372 ars_start.address = spa->address;
2373 ars_start.length = spa->length;
2374 if (req_type == ARS_REQ_SHORT)
2375 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2376 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2377 ars_start.type = ND_ARS_PERSISTENT;
2378 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2379 ars_start.type = ND_ARS_VOLATILE;
2383 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2384 sizeof(ars_start), &cmd_rc);
2390 set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2394 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2397 struct nd_cmd_ars_start ars_start;
2398 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2399 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2401 ars_start = (struct nd_cmd_ars_start) {
2402 .address = ars_status->restart_address,
2403 .length = ars_status->restart_length,
2404 .type = ars_status->type,
2406 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2407 sizeof(ars_start), &cmd_rc);
2413 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2415 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2416 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2419 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2420 acpi_desc->max_ars, &cmd_rc);
2426 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2427 struct nfit_spa *nfit_spa)
2429 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2430 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2431 struct nd_region *nd_region = nfit_spa->nd_region;
2434 lockdep_assert_held(&acpi_desc->init_mutex);
2436 * Only advance the ARS state for ARS runs initiated by the
2437 * kernel, ignore ARS results from BIOS initiated runs for scrub
2438 * completion tracking.
2440 if (acpi_desc->scrub_spa != nfit_spa)
2443 if ((ars_status->address >= spa->address && ars_status->address
2444 < spa->address + spa->length)
2445 || (ars_status->address < spa->address)) {
2447 * Assume that if a scrub starts at an offset from the
2448 * start of nfit_spa that we are in the continuation
2451 * Otherwise, if the scrub covers the spa range, mark
2452 * any pending request complete.
2454 if (ars_status->address + ars_status->length
2455 >= spa->address + spa->length)
2462 acpi_desc->scrub_spa = NULL;
2464 dev = nd_region_dev(nd_region);
2465 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2467 dev = acpi_desc->dev;
2468 dev_dbg(dev, "ARS: range %d complete\n", spa->range_index);
2471 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2473 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2474 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2479 * First record starts at 44 byte offset from the start of the
2482 if (ars_status->out_length < 44)
2486 * Ignore potentially stale results that are only refreshed
2487 * after a start-ARS event.
2489 if (!test_and_clear_bit(ARS_VALID, &acpi_desc->scrub_flags)) {
2490 dev_dbg(acpi_desc->dev, "skip %d stale records\n",
2491 ars_status->num_records);
2495 for (i = 0; i < ars_status->num_records; i++) {
2496 /* only process full records */
2497 if (ars_status->out_length
2498 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2500 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2501 ars_status->records[i].err_address,
2502 ars_status->records[i].length);
2506 if (i < ars_status->num_records)
2507 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2512 static void acpi_nfit_remove_resource(void *data)
2514 struct resource *res = data;
2516 remove_resource(res);
2519 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2520 struct nd_region_desc *ndr_desc)
2522 struct resource *res, *nd_res = ndr_desc->res;
2525 /* No operation if the region is already registered as PMEM */
2526 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2527 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2528 if (is_pmem == REGION_INTERSECTS)
2531 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2535 res->name = "Persistent Memory";
2536 res->start = nd_res->start;
2537 res->end = nd_res->end;
2538 res->flags = IORESOURCE_MEM;
2539 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2541 ret = insert_resource(&iomem_resource, res);
2545 ret = devm_add_action_or_reset(acpi_desc->dev,
2546 acpi_nfit_remove_resource,
2554 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2555 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2556 struct acpi_nfit_memory_map *memdev,
2557 struct nfit_spa *nfit_spa)
2559 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2560 memdev->device_handle);
2561 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2564 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2565 spa->range_index, memdev->device_handle);
2569 mapping->nvdimm = nvdimm;
2570 switch (nfit_spa_type(spa)) {
2572 case NFIT_SPA_VOLATILE:
2573 mapping->start = memdev->address;
2574 mapping->size = memdev->region_size;
2581 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2583 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2584 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2585 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2586 nfit_spa_type(spa) == NFIT_SPA_PCD);
2589 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2591 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2592 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2593 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2596 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2597 struct nfit_spa *nfit_spa)
2599 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2600 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2601 struct nd_region_desc *ndr_desc, _ndr_desc;
2602 struct nfit_memdev *nfit_memdev;
2603 struct nvdimm_bus *nvdimm_bus;
2604 struct resource res;
2607 if (nfit_spa->nd_region)
2610 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2611 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2615 memset(&res, 0, sizeof(res));
2616 memset(&mappings, 0, sizeof(mappings));
2617 memset(&_ndr_desc, 0, sizeof(_ndr_desc));
2618 res.start = spa->address;
2619 res.end = res.start + spa->length - 1;
2620 ndr_desc = &_ndr_desc;
2621 ndr_desc->res = &res;
2622 ndr_desc->provider_data = nfit_spa;
2623 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2624 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID) {
2625 ndr_desc->numa_node = pxm_to_online_node(spa->proximity_domain);
2626 ndr_desc->target_node = pxm_to_node(spa->proximity_domain);
2628 ndr_desc->numa_node = NUMA_NO_NODE;
2629 ndr_desc->target_node = NUMA_NO_NODE;
2632 /* Fallback to address based numa information if node lookup failed */
2633 if (ndr_desc->numa_node == NUMA_NO_NODE) {
2634 ndr_desc->numa_node = memory_add_physaddr_to_nid(spa->address);
2635 dev_info(acpi_desc->dev, "changing numa node from %d to %d for nfit region [%pa-%pa]",
2636 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
2638 if (ndr_desc->target_node == NUMA_NO_NODE) {
2639 ndr_desc->target_node = phys_to_target_node(spa->address);
2640 dev_info(acpi_desc->dev, "changing target node from %d to %d for nfit region [%pa-%pa]",
2641 NUMA_NO_NODE, ndr_desc->numa_node, &res.start, &res.end);
2645 * Persistence domain bits are hierarchical, if
2646 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2647 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2649 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2650 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2651 else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2652 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2654 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2655 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2656 struct nd_mapping_desc *mapping;
2658 /* range index 0 == unmapped in SPA or invalid-SPA */
2659 if (memdev->range_index == 0 || spa->range_index == 0)
2661 if (memdev->range_index != spa->range_index)
2663 if (count >= ND_MAX_MAPPINGS) {
2664 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2665 spa->range_index, ND_MAX_MAPPINGS);
2668 mapping = &mappings[count++];
2669 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2675 ndr_desc->mapping = mappings;
2676 ndr_desc->num_mappings = count;
2677 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2681 nvdimm_bus = acpi_desc->nvdimm_bus;
2682 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2683 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2685 dev_warn(acpi_desc->dev,
2686 "failed to insert pmem resource to iomem: %d\n",
2691 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2693 if (!nfit_spa->nd_region)
2695 } else if (nfit_spa_is_volatile(spa)) {
2696 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2698 if (!nfit_spa->nd_region)
2700 } else if (nfit_spa_is_virtual(spa)) {
2701 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2703 if (!nfit_spa->nd_region)
2709 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2710 nfit_spa->spa->range_index);
2714 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
2716 struct device *dev = acpi_desc->dev;
2717 struct nd_cmd_ars_status *ars_status;
2719 if (acpi_desc->ars_status) {
2720 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2724 ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
2727 acpi_desc->ars_status = ars_status;
2731 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
2735 if (ars_status_alloc(acpi_desc))
2738 rc = ars_get_status(acpi_desc);
2740 if (rc < 0 && rc != -ENOSPC)
2743 if (ars_status_process_records(acpi_desc))
2744 dev_err(acpi_desc->dev, "Failed to process ARS records\n");
2749 static int ars_register(struct acpi_nfit_desc *acpi_desc,
2750 struct nfit_spa *nfit_spa)
2754 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2755 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2757 set_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
2759 set_bit(ARS_REQ_LONG, &nfit_spa->ars_state);
2761 switch (acpi_nfit_query_poison(acpi_desc)) {
2765 rc = ars_start(acpi_desc, nfit_spa, ARS_REQ_SHORT);
2766 /* shouldn't happen, try again later */
2770 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2773 clear_bit(ARS_REQ_SHORT, &nfit_spa->ars_state);
2774 rc = acpi_nfit_query_poison(acpi_desc);
2777 acpi_desc->scrub_spa = nfit_spa;
2778 ars_complete(acpi_desc, nfit_spa);
2780 * If ars_complete() says we didn't complete the
2781 * short scrub, we'll try again with a long
2784 acpi_desc->scrub_spa = NULL;
2789 * BIOS was using ARS, wait for it to complete (or
2790 * resources to become available) and then perform our
2795 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2799 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2802 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
2804 struct nfit_spa *nfit_spa;
2806 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2807 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2809 ars_complete(acpi_desc, nfit_spa);
2813 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
2816 unsigned int tmo = acpi_desc->scrub_tmo;
2817 struct device *dev = acpi_desc->dev;
2818 struct nfit_spa *nfit_spa;
2820 lockdep_assert_held(&acpi_desc->init_mutex);
2822 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags))
2825 if (query_rc == -EBUSY) {
2826 dev_dbg(dev, "ARS: ARS busy\n");
2827 return min(30U * 60U, tmo * 2);
2829 if (query_rc == -ENOSPC) {
2830 dev_dbg(dev, "ARS: ARS continue\n");
2831 ars_continue(acpi_desc);
2834 if (query_rc && query_rc != -EAGAIN) {
2835 unsigned long long addr, end;
2837 addr = acpi_desc->ars_status->address;
2838 end = addr + acpi_desc->ars_status->length;
2839 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
2843 ars_complete_all(acpi_desc);
2844 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2845 enum nfit_ars_state req_type;
2848 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2851 /* prefer short ARS requests first */
2852 if (test_bit(ARS_REQ_SHORT, &nfit_spa->ars_state))
2853 req_type = ARS_REQ_SHORT;
2854 else if (test_bit(ARS_REQ_LONG, &nfit_spa->ars_state))
2855 req_type = ARS_REQ_LONG;
2858 rc = ars_start(acpi_desc, nfit_spa, req_type);
2860 dev = nd_region_dev(nfit_spa->nd_region);
2861 dev_dbg(dev, "ARS: range %d ARS start %s (%d)\n",
2862 nfit_spa->spa->range_index,
2863 req_type == ARS_REQ_SHORT ? "short" : "long",
2866 * Hmm, we raced someone else starting ARS? Try again in
2872 dev_WARN_ONCE(dev, acpi_desc->scrub_spa,
2873 "scrub start while range %d active\n",
2874 acpi_desc->scrub_spa->spa->range_index);
2875 clear_bit(req_type, &nfit_spa->ars_state);
2876 acpi_desc->scrub_spa = nfit_spa;
2878 * Consider this spa last for future scrub
2881 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
2885 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
2886 nfit_spa->spa->range_index, rc);
2887 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2892 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
2894 lockdep_assert_held(&acpi_desc->init_mutex);
2896 set_bit(ARS_BUSY, &acpi_desc->scrub_flags);
2897 /* note this should only be set from within the workqueue */
2899 acpi_desc->scrub_tmo = tmo;
2900 queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
2903 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
2905 __sched_ars(acpi_desc, 0);
2908 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
2910 lockdep_assert_held(&acpi_desc->init_mutex);
2912 clear_bit(ARS_BUSY, &acpi_desc->scrub_flags);
2913 acpi_desc->scrub_count++;
2914 if (acpi_desc->scrub_count_state)
2915 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2918 static void acpi_nfit_scrub(struct work_struct *work)
2920 struct acpi_nfit_desc *acpi_desc;
2924 acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
2925 mutex_lock(&acpi_desc->init_mutex);
2926 query_rc = acpi_nfit_query_poison(acpi_desc);
2927 tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
2929 __sched_ars(acpi_desc, tmo);
2931 notify_ars_done(acpi_desc);
2932 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2933 clear_bit(ARS_POLL, &acpi_desc->scrub_flags);
2934 mutex_unlock(&acpi_desc->init_mutex);
2937 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
2938 struct nfit_spa *nfit_spa)
2940 int type = nfit_spa_type(nfit_spa->spa);
2941 struct nd_cmd_ars_cap ars_cap;
2944 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2945 memset(&ars_cap, 0, sizeof(ars_cap));
2946 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2949 /* check that the supported scrub types match the spa type */
2950 if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
2951 & ND_ARS_VOLATILE) == 0)
2953 if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
2954 & ND_ARS_PERSISTENT) == 0)
2957 nfit_spa->max_ars = ars_cap.max_ars_out;
2958 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2959 acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
2960 clear_bit(ARS_FAILED, &nfit_spa->ars_state);
2963 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2965 struct nfit_spa *nfit_spa;
2966 int rc, do_sched_ars = 0;
2968 set_bit(ARS_VALID, &acpi_desc->scrub_flags);
2969 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2970 switch (nfit_spa_type(nfit_spa->spa)) {
2971 case NFIT_SPA_VOLATILE:
2973 acpi_nfit_init_ars(acpi_desc, nfit_spa);
2978 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2979 switch (nfit_spa_type(nfit_spa->spa)) {
2980 case NFIT_SPA_VOLATILE:
2982 /* register regions and kick off initial ARS run */
2983 rc = ars_register(acpi_desc, nfit_spa);
2988 * Kick off background ARS if at least one
2989 * region successfully registered ARS
2991 if (!test_bit(ARS_FAILED, &nfit_spa->ars_state))
2995 /* nothing to register */
2998 case NFIT_SPA_VDISK:
3000 case NFIT_SPA_PDISK:
3002 /* register known regions that don't support ARS */
3003 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3008 /* don't register unknown regions */
3014 sched_ars(acpi_desc);
3018 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3019 struct nfit_table_prev *prev)
3021 struct device *dev = acpi_desc->dev;
3023 if (!list_empty(&prev->spas) ||
3024 !list_empty(&prev->memdevs) ||
3025 !list_empty(&prev->dcrs) ||
3026 !list_empty(&prev->bdws) ||
3027 !list_empty(&prev->idts) ||
3028 !list_empty(&prev->flushes)) {
3029 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3035 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3037 struct device *dev = acpi_desc->dev;
3038 struct kernfs_node *nfit;
3039 struct device *bus_dev;
3041 if (!ars_supported(acpi_desc->nvdimm_bus))
3044 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3045 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3047 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3050 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3052 if (!acpi_desc->scrub_count_state) {
3053 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3060 static void acpi_nfit_unregister(void *data)
3062 struct acpi_nfit_desc *acpi_desc = data;
3064 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3067 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3069 struct device *dev = acpi_desc->dev;
3070 struct nfit_table_prev prev;
3074 if (!acpi_desc->nvdimm_bus) {
3075 acpi_nfit_init_dsms(acpi_desc);
3077 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3078 &acpi_desc->nd_desc);
3079 if (!acpi_desc->nvdimm_bus)
3082 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3087 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3091 /* register this acpi_desc for mce notifications */
3092 mutex_lock(&acpi_desc_lock);
3093 list_add_tail(&acpi_desc->list, &acpi_descs);
3094 mutex_unlock(&acpi_desc_lock);
3097 mutex_lock(&acpi_desc->init_mutex);
3099 INIT_LIST_HEAD(&prev.spas);
3100 INIT_LIST_HEAD(&prev.memdevs);
3101 INIT_LIST_HEAD(&prev.dcrs);
3102 INIT_LIST_HEAD(&prev.bdws);
3103 INIT_LIST_HEAD(&prev.idts);
3104 INIT_LIST_HEAD(&prev.flushes);
3106 list_cut_position(&prev.spas, &acpi_desc->spas,
3107 acpi_desc->spas.prev);
3108 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3109 acpi_desc->memdevs.prev);
3110 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3111 acpi_desc->dcrs.prev);
3112 list_cut_position(&prev.bdws, &acpi_desc->bdws,
3113 acpi_desc->bdws.prev);
3114 list_cut_position(&prev.idts, &acpi_desc->idts,
3115 acpi_desc->idts.prev);
3116 list_cut_position(&prev.flushes, &acpi_desc->flushes,
3117 acpi_desc->flushes.prev);
3120 while (!IS_ERR_OR_NULL(data))
3121 data = add_table(acpi_desc, &prev, data, end);
3124 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3129 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3133 rc = nfit_mem_init(acpi_desc);
3137 rc = acpi_nfit_register_dimms(acpi_desc);
3141 rc = acpi_nfit_register_regions(acpi_desc);
3144 mutex_unlock(&acpi_desc->init_mutex);
3147 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3149 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3151 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3152 struct device *dev = acpi_desc->dev;
3154 /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3158 /* Bounce the init_mutex to complete initial registration */
3159 mutex_lock(&acpi_desc->init_mutex);
3160 mutex_unlock(&acpi_desc->init_mutex);
3165 static int __acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3166 struct nvdimm *nvdimm, unsigned int cmd)
3168 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
3172 if (cmd != ND_CMD_ARS_START)
3176 * The kernel and userspace may race to initiate a scrub, but
3177 * the scrub thread is prepared to lose that initial race. It
3178 * just needs guarantees that any ARS it initiates are not
3179 * interrupted by any intervening start requests from userspace.
3181 if (work_busy(&acpi_desc->dwork.work))
3188 * Prevent security and firmware activate commands from being issued via
3191 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3192 struct nvdimm *nvdimm, unsigned int cmd, void *buf)
3194 struct nd_cmd_pkg *call_pkg = buf;
3197 if (nvdimm && cmd == ND_CMD_CALL &&
3198 call_pkg->nd_family == NVDIMM_FAMILY_INTEL) {
3199 func = call_pkg->nd_command;
3200 if (func > NVDIMM_CMD_MAX ||
3201 (1 << func) & NVDIMM_INTEL_DENY_CMDMASK)
3205 /* block all non-nfit bus commands */
3206 if (!nvdimm && cmd == ND_CMD_CALL &&
3207 call_pkg->nd_family != NVDIMM_BUS_FAMILY_NFIT)
3210 return __acpi_nfit_clear_to_send(nd_desc, nvdimm, cmd);
3213 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc,
3214 enum nfit_ars_state req_type)
3216 struct device *dev = acpi_desc->dev;
3217 int scheduled = 0, busy = 0;
3218 struct nfit_spa *nfit_spa;
3220 mutex_lock(&acpi_desc->init_mutex);
3221 if (test_bit(ARS_CANCEL, &acpi_desc->scrub_flags)) {
3222 mutex_unlock(&acpi_desc->init_mutex);
3226 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3227 int type = nfit_spa_type(nfit_spa->spa);
3229 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3231 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3234 if (test_and_set_bit(req_type, &nfit_spa->ars_state))
3240 sched_ars(acpi_desc);
3241 dev_dbg(dev, "ars_scan triggered\n");
3243 mutex_unlock(&acpi_desc->init_mutex);
3252 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3254 struct nvdimm_bus_descriptor *nd_desc;
3256 dev_set_drvdata(dev, acpi_desc);
3257 acpi_desc->dev = dev;
3258 nd_desc = &acpi_desc->nd_desc;
3259 nd_desc->provider_name = "ACPI.NFIT";
3260 nd_desc->module = THIS_MODULE;
3261 nd_desc->ndctl = acpi_nfit_ctl;
3262 nd_desc->flush_probe = acpi_nfit_flush_probe;
3263 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3264 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3266 INIT_LIST_HEAD(&acpi_desc->spas);
3267 INIT_LIST_HEAD(&acpi_desc->dcrs);
3268 INIT_LIST_HEAD(&acpi_desc->bdws);
3269 INIT_LIST_HEAD(&acpi_desc->idts);
3270 INIT_LIST_HEAD(&acpi_desc->flushes);
3271 INIT_LIST_HEAD(&acpi_desc->memdevs);
3272 INIT_LIST_HEAD(&acpi_desc->dimms);
3273 INIT_LIST_HEAD(&acpi_desc->list);
3274 mutex_init(&acpi_desc->init_mutex);
3275 acpi_desc->scrub_tmo = 1;
3276 INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3278 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3280 static void acpi_nfit_put_table(void *table)
3282 acpi_put_table(table);
3285 static void acpi_nfit_notify(acpi_handle handle, u32 event, void *data)
3287 struct acpi_device *adev = data;
3289 device_lock(&adev->dev);
3290 __acpi_nfit_notify(&adev->dev, handle, event);
3291 device_unlock(&adev->dev);
3294 static void acpi_nfit_remove_notify_handler(void *data)
3296 struct acpi_device *adev = data;
3298 acpi_dev_remove_notify_handler(adev, ACPI_DEVICE_NOTIFY,
3302 void acpi_nfit_shutdown(void *data)
3304 struct acpi_nfit_desc *acpi_desc = data;
3305 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3308 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3311 mutex_lock(&acpi_desc_lock);
3312 list_del(&acpi_desc->list);
3313 mutex_unlock(&acpi_desc_lock);
3315 mutex_lock(&acpi_desc->init_mutex);
3316 set_bit(ARS_CANCEL, &acpi_desc->scrub_flags);
3317 mutex_unlock(&acpi_desc->init_mutex);
3318 cancel_delayed_work_sync(&acpi_desc->dwork);
3321 * Bounce the nvdimm bus lock to make sure any in-flight
3322 * acpi_nfit_ars_rescan() submissions have had a chance to
3323 * either submit or see ->cancel set.
3325 device_lock(bus_dev);
3326 device_unlock(bus_dev);
3328 flush_workqueue(nfit_wq);
3330 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3332 static int acpi_nfit_add(struct acpi_device *adev)
3334 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3335 struct acpi_nfit_desc *acpi_desc;
3336 struct device *dev = &adev->dev;
3337 struct acpi_table_header *tbl;
3338 acpi_status status = AE_OK;
3342 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3343 if (ACPI_FAILURE(status)) {
3344 /* The NVDIMM root device allows OS to trigger enumeration of
3345 * NVDIMMs through NFIT at boot time and re-enumeration at
3346 * root level via the _FIT method during runtime.
3347 * This is ok to return 0 here, we could have an nvdimm
3348 * hotplugged later and evaluate _FIT method which returns
3349 * data in the format of a series of NFIT Structures.
3351 dev_dbg(dev, "failed to find NFIT at startup\n");
3355 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3360 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3363 acpi_nfit_desc_init(acpi_desc, &adev->dev);
3365 /* Save the acpi header for exporting the revision via sysfs */
3366 acpi_desc->acpi_header = *tbl;
3368 /* Evaluate _FIT and override with that if present */
3369 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3370 if (ACPI_SUCCESS(status) && buf.length > 0) {
3371 union acpi_object *obj = buf.pointer;
3373 if (obj->type == ACPI_TYPE_BUFFER)
3374 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3375 obj->buffer.length);
3377 dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3381 /* skip over the lead-in header table */
3382 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3383 + sizeof(struct acpi_table_nfit),
3384 sz - sizeof(struct acpi_table_nfit));
3389 rc = devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3393 rc = acpi_dev_install_notify_handler(adev, ACPI_DEVICE_NOTIFY,
3398 return devm_add_action_or_reset(dev, acpi_nfit_remove_notify_handler,
3402 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3404 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3405 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3406 union acpi_object *obj;
3411 /* dev->driver may be null if we're being removed */
3412 dev_dbg(dev, "no driver found for dev\n");
3417 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3420 acpi_nfit_desc_init(acpi_desc, dev);
3423 * Finish previous registration before considering new
3426 flush_workqueue(nfit_wq);
3430 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3431 if (ACPI_FAILURE(status)) {
3432 dev_err(dev, "failed to evaluate _FIT\n");
3437 if (obj->type == ACPI_TYPE_BUFFER) {
3438 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3439 obj->buffer.length);
3441 dev_err(dev, "failed to merge updated NFIT\n");
3443 dev_err(dev, "Invalid _FIT\n");
3447 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3449 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3451 if (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON)
3452 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_LONG);
3454 acpi_nfit_ars_rescan(acpi_desc, ARS_REQ_SHORT);
3457 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3459 dev_dbg(dev, "event: 0x%x\n", event);
3462 case NFIT_NOTIFY_UPDATE:
3463 return acpi_nfit_update_notify(dev, handle);
3464 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3465 return acpi_nfit_uc_error_notify(dev, handle);
3470 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3472 static const struct acpi_device_id acpi_nfit_ids[] = {
3476 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3478 static struct acpi_driver acpi_nfit_driver = {
3479 .name = KBUILD_MODNAME,
3480 .ids = acpi_nfit_ids,
3482 .add = acpi_nfit_add,
3486 static __init int nfit_init(void)
3490 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3491 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 64);
3492 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3493 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 16);
3494 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 8);
3495 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3496 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3497 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3499 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3500 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3501 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3502 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3503 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3504 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3505 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3506 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3507 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3508 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3509 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3510 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3511 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3512 guid_parse(UUID_NFIT_DIMM_N_HYPERV, &nfit_uuid[NFIT_DEV_DIMM_N_HYPERV]);
3513 guid_parse(UUID_INTEL_BUS, &nfit_uuid[NFIT_BUS_INTEL]);
3515 nfit_wq = create_singlethread_workqueue("nfit");
3519 nfit_mce_register();
3520 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3522 nfit_mce_unregister();
3523 destroy_workqueue(nfit_wq);
3530 static __exit void nfit_exit(void)
3532 nfit_mce_unregister();
3533 acpi_bus_unregister_driver(&acpi_nfit_driver);
3534 destroy_workqueue(nfit_wq);
3535 WARN_ON(!list_empty(&acpi_descs));
3538 module_init(nfit_init);
3539 module_exit(nfit_exit);
3540 MODULE_LICENSE("GPL v2");
3541 MODULE_AUTHOR("Intel Corporation");