2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/list_sort.h>
14 #include <linux/libnvdimm.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/ndctl.h>
18 #include <linux/sysfs.h>
19 #include <linux/delay.h>
20 #include <linux/list.h>
21 #include <linux/acpi.h>
22 #include <linux/sort.h>
25 #include <asm/cacheflush.h>
26 #include <acpi/nfit.h>
30 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
33 #include <linux/io-64-nonatomic-hi-lo.h>
35 static bool force_enable_dimms;
36 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
37 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
39 static bool disable_vendor_specific;
40 module_param(disable_vendor_specific, bool, S_IRUGO);
41 MODULE_PARM_DESC(disable_vendor_specific,
42 "Limit commands to the publicly specified set");
44 static unsigned long override_dsm_mask;
45 module_param(override_dsm_mask, ulong, S_IRUGO);
46 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
48 static int default_dsm_family = -1;
49 module_param(default_dsm_family, int, S_IRUGO);
50 MODULE_PARM_DESC(default_dsm_family,
51 "Try this DSM type first when identifying NVDIMM family");
53 static bool no_init_ars;
54 module_param(no_init_ars, bool, 0644);
55 MODULE_PARM_DESC(no_init_ars, "Skip ARS run at nfit init time");
57 LIST_HEAD(acpi_descs);
58 DEFINE_MUTEX(acpi_desc_lock);
60 static struct workqueue_struct *nfit_wq;
62 struct nfit_table_prev {
63 struct list_head spas;
64 struct list_head memdevs;
65 struct list_head dcrs;
66 struct list_head bdws;
67 struct list_head idts;
68 struct list_head flushes;
71 static guid_t nfit_uuid[NFIT_UUID_MAX];
73 const guid_t *to_nfit_uuid(enum nfit_uuids id)
75 return &nfit_uuid[id];
77 EXPORT_SYMBOL(to_nfit_uuid);
79 static struct acpi_nfit_desc *to_acpi_nfit_desc(
80 struct nvdimm_bus_descriptor *nd_desc)
82 return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
85 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
87 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
90 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
93 if (!nd_desc->provider_name
94 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
97 return to_acpi_device(acpi_desc->dev);
100 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
102 struct nd_cmd_clear_error *clear_err;
103 struct nd_cmd_ars_status *ars_status;
108 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
115 /* No supported scan types for this range */
116 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
117 if ((status >> 16 & flags) == 0)
120 case ND_CMD_ARS_START:
121 /* ARS is in progress */
122 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
129 case ND_CMD_ARS_STATUS:
134 /* Check extended status (Upper two bytes) */
135 if (status == NFIT_ARS_STATUS_DONE)
138 /* ARS is in progress */
139 if (status == NFIT_ARS_STATUS_BUSY)
142 /* No ARS performed for the current boot */
143 if (status == NFIT_ARS_STATUS_NONE)
147 * ARS interrupted, either we overflowed or some other
148 * agent wants the scan to stop. If we didn't overflow
149 * then just continue with the returned results.
151 if (status == NFIT_ARS_STATUS_INTR) {
152 if (ars_status->out_length >= 40 && (ars_status->flags
153 & NFIT_ARS_F_OVERFLOW))
162 case ND_CMD_CLEAR_ERROR:
166 if (!clear_err->cleared)
168 if (clear_err->length > clear_err->cleared)
169 return clear_err->cleared;
175 /* all other non-zero status results in an error */
181 #define ACPI_LABELS_LOCKED 3
183 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
186 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
189 case ND_CMD_GET_CONFIG_SIZE:
191 * In the _LSI, _LSR, _LSW case the locked status is
192 * communicated via the read/write commands
194 if (nfit_mem->has_lsr)
197 if (status >> 16 & ND_CONFIG_LOCKED)
200 case ND_CMD_GET_CONFIG_DATA:
201 if (nfit_mem->has_lsr && status == ACPI_LABELS_LOCKED)
204 case ND_CMD_SET_CONFIG_DATA:
205 if (nfit_mem->has_lsw && status == ACPI_LABELS_LOCKED)
212 /* all other non-zero status results in an error */
218 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
222 return xlat_bus_status(buf, cmd, status);
223 return xlat_nvdimm_status(nvdimm, buf, cmd, status);
226 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
227 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
232 union acpi_object *buf = NULL;
234 if (pkg->type != ACPI_TYPE_PACKAGE) {
235 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
240 for (i = 0; i < pkg->package.count; i++) {
241 union acpi_object *obj = &pkg->package.elements[i];
243 if (obj->type == ACPI_TYPE_INTEGER)
245 else if (obj->type == ACPI_TYPE_BUFFER)
246 size += obj->buffer.length;
248 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
254 buf = ACPI_ALLOCATE(sizeof(*buf) + size);
259 buf->type = ACPI_TYPE_BUFFER;
260 buf->buffer.length = size;
261 buf->buffer.pointer = dst;
262 for (i = 0; i < pkg->package.count; i++) {
263 union acpi_object *obj = &pkg->package.elements[i];
265 if (obj->type == ACPI_TYPE_INTEGER) {
266 memcpy(dst, &obj->integer.value, 4);
268 } else if (obj->type == ACPI_TYPE_BUFFER) {
269 memcpy(dst, obj->buffer.pointer, obj->buffer.length);
270 dst += obj->buffer.length;
278 static union acpi_object *int_to_buf(union acpi_object *integer)
280 union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
286 if (integer->type != ACPI_TYPE_INTEGER) {
287 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
293 buf->type = ACPI_TYPE_BUFFER;
294 buf->buffer.length = 4;
295 buf->buffer.pointer = dst;
296 memcpy(dst, &integer->integer.value, 4);
302 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
306 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
307 struct acpi_object_list input = {
309 .pointer = (union acpi_object []) {
311 .integer.type = ACPI_TYPE_INTEGER,
312 .integer.value = offset,
315 .integer.type = ACPI_TYPE_INTEGER,
316 .integer.value = len,
319 .buffer.type = ACPI_TYPE_BUFFER,
320 .buffer.pointer = data,
321 .buffer.length = len,
326 rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
327 if (ACPI_FAILURE(rc))
329 return int_to_buf(buf.pointer);
332 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
336 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
337 struct acpi_object_list input = {
339 .pointer = (union acpi_object []) {
341 .integer.type = ACPI_TYPE_INTEGER,
342 .integer.value = offset,
345 .integer.type = ACPI_TYPE_INTEGER,
346 .integer.value = len,
351 rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
352 if (ACPI_FAILURE(rc))
354 return pkg_to_buf(buf.pointer);
357 static union acpi_object *acpi_label_info(acpi_handle handle)
360 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
362 rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
363 if (ACPI_FAILURE(rc))
365 return pkg_to_buf(buf.pointer);
368 static u8 nfit_dsm_revid(unsigned family, unsigned func)
370 static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
371 [NVDIMM_FAMILY_INTEL] = {
372 [NVDIMM_INTEL_GET_MODES] = 2,
373 [NVDIMM_INTEL_GET_FWINFO] = 2,
374 [NVDIMM_INTEL_START_FWUPDATE] = 2,
375 [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
376 [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
377 [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
378 [NVDIMM_INTEL_SET_THRESHOLD] = 2,
379 [NVDIMM_INTEL_INJECT_ERROR] = 2,
384 if (family > NVDIMM_FAMILY_MAX)
388 id = revid_table[family][func];
390 return 1; /* default */
394 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
395 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
397 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
398 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
399 union acpi_object in_obj, in_buf, *out_obj;
400 const struct nd_cmd_desc *desc = NULL;
401 struct device *dev = acpi_desc->dev;
402 struct nd_cmd_pkg *call_pkg = NULL;
403 const char *cmd_name, *dimm_name;
404 unsigned long cmd_mask, dsm_mask;
405 u32 offset, fw_status = 0;
413 if (cmd == ND_CMD_CALL) {
415 func = call_pkg->nd_command;
417 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
418 if (call_pkg->nd_reserved2[i])
423 struct acpi_device *adev = nfit_mem->adev;
427 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
430 dimm_name = nvdimm_name(nvdimm);
431 cmd_name = nvdimm_cmd_name(cmd);
432 cmd_mask = nvdimm_cmd_mask(nvdimm);
433 dsm_mask = nfit_mem->dsm_mask;
434 desc = nd_cmd_dimm_desc(cmd);
435 guid = to_nfit_uuid(nfit_mem->family);
436 handle = adev->handle;
438 struct acpi_device *adev = to_acpi_dev(acpi_desc);
440 cmd_name = nvdimm_bus_cmd_name(cmd);
441 cmd_mask = nd_desc->cmd_mask;
443 if (cmd == ND_CMD_CALL)
444 dsm_mask = nd_desc->bus_dsm_mask;
445 desc = nd_cmd_bus_desc(cmd);
446 guid = to_nfit_uuid(NFIT_DEV_BUS);
447 handle = adev->handle;
451 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
454 if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
457 in_obj.type = ACPI_TYPE_PACKAGE;
458 in_obj.package.count = 1;
459 in_obj.package.elements = &in_buf;
460 in_buf.type = ACPI_TYPE_BUFFER;
461 in_buf.buffer.pointer = buf;
462 in_buf.buffer.length = 0;
464 /* libnvdimm has already validated the input envelope */
465 for (i = 0; i < desc->in_num; i++)
466 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
470 /* skip over package wrapper */
471 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
472 in_buf.buffer.length = call_pkg->nd_size_in;
475 dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
476 dimm_name, cmd, func, in_buf.buffer.length);
477 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
478 in_buf.buffer.pointer,
479 min_t(u32, 256, in_buf.buffer.length), true);
481 /* call the BIOS, prefer the named methods over _DSM if available */
482 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE && nfit_mem->has_lsr)
483 out_obj = acpi_label_info(handle);
484 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && nfit_mem->has_lsr) {
485 struct nd_cmd_get_config_data_hdr *p = buf;
487 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
488 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
489 && nfit_mem->has_lsw) {
490 struct nd_cmd_set_config_hdr *p = buf;
492 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
498 revid = nfit_dsm_revid(nfit_mem->family, func);
501 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
505 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
510 call_pkg->nd_fw_size = out_obj->buffer.length;
511 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
512 out_obj->buffer.pointer,
513 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
517 * Need to support FW function w/o known size in advance.
518 * Caller can determine required size based upon nd_fw_size.
519 * If we return an error (like elsewhere) then caller wouldn't
520 * be able to rely upon data returned to make calculation.
526 if (out_obj->package.type != ACPI_TYPE_BUFFER) {
527 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
528 dimm_name, cmd_name, out_obj->type);
533 dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
534 cmd_name, out_obj->buffer.length);
535 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
536 out_obj->buffer.pointer,
537 min_t(u32, 128, out_obj->buffer.length), true);
539 for (i = 0, offset = 0; i < desc->out_num; i++) {
540 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
541 (u32 *) out_obj->buffer.pointer,
542 out_obj->buffer.length - offset);
544 if (offset + out_size > out_obj->buffer.length) {
545 dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
546 dimm_name, cmd_name, i);
550 if (in_buf.buffer.length + offset + out_size > buf_len) {
551 dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
552 dimm_name, cmd_name, i);
556 memcpy(buf + in_buf.buffer.length + offset,
557 out_obj->buffer.pointer + offset, out_size);
562 * Set fw_status for all the commands with a known format to be
563 * later interpreted by xlat_status().
565 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
566 && cmd <= ND_CMD_CLEAR_ERROR)
567 || (nvdimm && cmd >= ND_CMD_SMART
568 && cmd <= ND_CMD_VENDOR)))
569 fw_status = *(u32 *) out_obj->buffer.pointer;
571 if (offset + in_buf.buffer.length < buf_len) {
574 * status valid, return the number of bytes left
575 * unfilled in the output buffer
577 rc = buf_len - offset - in_buf.buffer.length;
579 *cmd_rc = xlat_status(nvdimm, buf, cmd,
582 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
583 __func__, dimm_name, cmd_name, buf_len,
590 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
598 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
600 static const char *spa_type_name(u16 type)
602 static const char *to_name[] = {
603 [NFIT_SPA_VOLATILE] = "volatile",
604 [NFIT_SPA_PM] = "pmem",
605 [NFIT_SPA_DCR] = "dimm-control-region",
606 [NFIT_SPA_BDW] = "block-data-window",
607 [NFIT_SPA_VDISK] = "volatile-disk",
608 [NFIT_SPA_VCD] = "volatile-cd",
609 [NFIT_SPA_PDISK] = "persistent-disk",
610 [NFIT_SPA_PCD] = "persistent-cd",
614 if (type > NFIT_SPA_PCD)
617 return to_name[type];
620 int nfit_spa_type(struct acpi_nfit_system_address *spa)
624 for (i = 0; i < NFIT_UUID_MAX; i++)
625 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
630 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
631 struct nfit_table_prev *prev,
632 struct acpi_nfit_system_address *spa)
634 struct device *dev = acpi_desc->dev;
635 struct nfit_spa *nfit_spa;
637 if (spa->header.length != sizeof(*spa))
640 list_for_each_entry(nfit_spa, &prev->spas, list) {
641 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
642 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
647 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
651 INIT_LIST_HEAD(&nfit_spa->list);
652 memcpy(nfit_spa->spa, spa, sizeof(*spa));
653 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
654 dev_dbg(dev, "spa index: %d type: %s\n",
656 spa_type_name(nfit_spa_type(spa)));
660 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
661 struct nfit_table_prev *prev,
662 struct acpi_nfit_memory_map *memdev)
664 struct device *dev = acpi_desc->dev;
665 struct nfit_memdev *nfit_memdev;
667 if (memdev->header.length != sizeof(*memdev))
670 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
671 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
672 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
676 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
680 INIT_LIST_HEAD(&nfit_memdev->list);
681 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
682 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
683 dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
684 memdev->device_handle, memdev->range_index,
685 memdev->region_index, memdev->flags);
689 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
691 struct acpi_nfit_memory_map *memdev;
692 struct acpi_nfit_desc *acpi_desc;
693 struct nfit_mem *nfit_mem;
695 mutex_lock(&acpi_desc_lock);
696 list_for_each_entry(acpi_desc, &acpi_descs, list) {
697 mutex_lock(&acpi_desc->init_mutex);
698 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
699 memdev = __to_nfit_memdev(nfit_mem);
700 if (memdev->device_handle == device_handle) {
701 mutex_unlock(&acpi_desc->init_mutex);
702 mutex_unlock(&acpi_desc_lock);
703 *flags = memdev->flags;
704 return memdev->physical_id;
707 mutex_unlock(&acpi_desc->init_mutex);
709 mutex_unlock(&acpi_desc_lock);
713 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
716 * An implementation may provide a truncated control region if no block windows
719 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
721 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
726 return offsetof(struct acpi_nfit_control_region, window_size);
729 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
730 struct nfit_table_prev *prev,
731 struct acpi_nfit_control_region *dcr)
733 struct device *dev = acpi_desc->dev;
734 struct nfit_dcr *nfit_dcr;
736 if (!sizeof_dcr(dcr))
739 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
740 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
741 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
745 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
749 INIT_LIST_HEAD(&nfit_dcr->list);
750 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
751 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
752 dev_dbg(dev, "dcr index: %d windows: %d\n",
753 dcr->region_index, dcr->windows);
757 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
758 struct nfit_table_prev *prev,
759 struct acpi_nfit_data_region *bdw)
761 struct device *dev = acpi_desc->dev;
762 struct nfit_bdw *nfit_bdw;
764 if (bdw->header.length != sizeof(*bdw))
766 list_for_each_entry(nfit_bdw, &prev->bdws, list)
767 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
768 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
772 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
776 INIT_LIST_HEAD(&nfit_bdw->list);
777 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
778 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
779 dev_dbg(dev, "bdw dcr: %d windows: %d\n",
780 bdw->region_index, bdw->windows);
784 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
786 if (idt->header.length < sizeof(*idt))
788 return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
791 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
792 struct nfit_table_prev *prev,
793 struct acpi_nfit_interleave *idt)
795 struct device *dev = acpi_desc->dev;
796 struct nfit_idt *nfit_idt;
798 if (!sizeof_idt(idt))
801 list_for_each_entry(nfit_idt, &prev->idts, list) {
802 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
805 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
806 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
811 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
815 INIT_LIST_HEAD(&nfit_idt->list);
816 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
817 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
818 dev_dbg(dev, "idt index: %d num_lines: %d\n",
819 idt->interleave_index, idt->line_count);
823 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
825 if (flush->header.length < sizeof(*flush))
827 return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
830 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
831 struct nfit_table_prev *prev,
832 struct acpi_nfit_flush_address *flush)
834 struct device *dev = acpi_desc->dev;
835 struct nfit_flush *nfit_flush;
837 if (!sizeof_flush(flush))
840 list_for_each_entry(nfit_flush, &prev->flushes, list) {
841 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
844 if (memcmp(nfit_flush->flush, flush,
845 sizeof_flush(flush)) == 0) {
846 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
851 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
852 + sizeof_flush(flush), GFP_KERNEL);
855 INIT_LIST_HEAD(&nfit_flush->list);
856 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
857 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
858 dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
859 flush->device_handle, flush->hint_count);
863 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
864 struct acpi_nfit_capabilities *pcap)
866 struct device *dev = acpi_desc->dev;
869 mask = (1 << (pcap->highest_capability + 1)) - 1;
870 acpi_desc->platform_cap = pcap->capabilities & mask;
871 dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
875 static void *add_table(struct acpi_nfit_desc *acpi_desc,
876 struct nfit_table_prev *prev, void *table, const void *end)
878 struct device *dev = acpi_desc->dev;
879 struct acpi_nfit_header *hdr;
880 void *err = ERR_PTR(-ENOMEM);
887 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
893 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
894 if (!add_spa(acpi_desc, prev, table))
897 case ACPI_NFIT_TYPE_MEMORY_MAP:
898 if (!add_memdev(acpi_desc, prev, table))
901 case ACPI_NFIT_TYPE_CONTROL_REGION:
902 if (!add_dcr(acpi_desc, prev, table))
905 case ACPI_NFIT_TYPE_DATA_REGION:
906 if (!add_bdw(acpi_desc, prev, table))
909 case ACPI_NFIT_TYPE_INTERLEAVE:
910 if (!add_idt(acpi_desc, prev, table))
913 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
914 if (!add_flush(acpi_desc, prev, table))
917 case ACPI_NFIT_TYPE_SMBIOS:
918 dev_dbg(dev, "smbios\n");
920 case ACPI_NFIT_TYPE_CAPABILITIES:
921 if (!add_platform_cap(acpi_desc, table))
925 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
929 return table + hdr->length;
932 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
933 struct nfit_mem *nfit_mem)
935 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
936 u16 dcr = nfit_mem->dcr->region_index;
937 struct nfit_spa *nfit_spa;
939 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
940 u16 range_index = nfit_spa->spa->range_index;
941 int type = nfit_spa_type(nfit_spa->spa);
942 struct nfit_memdev *nfit_memdev;
944 if (type != NFIT_SPA_BDW)
947 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
948 if (nfit_memdev->memdev->range_index != range_index)
950 if (nfit_memdev->memdev->device_handle != device_handle)
952 if (nfit_memdev->memdev->region_index != dcr)
955 nfit_mem->spa_bdw = nfit_spa->spa;
960 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
961 nfit_mem->spa_dcr->range_index);
962 nfit_mem->bdw = NULL;
965 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
966 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
968 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
969 struct nfit_memdev *nfit_memdev;
970 struct nfit_bdw *nfit_bdw;
971 struct nfit_idt *nfit_idt;
972 u16 idt_idx, range_index;
974 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
975 if (nfit_bdw->bdw->region_index != dcr)
977 nfit_mem->bdw = nfit_bdw->bdw;
984 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
986 if (!nfit_mem->spa_bdw)
989 range_index = nfit_mem->spa_bdw->range_index;
990 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
991 if (nfit_memdev->memdev->range_index != range_index ||
992 nfit_memdev->memdev->region_index != dcr)
994 nfit_mem->memdev_bdw = nfit_memdev->memdev;
995 idt_idx = nfit_memdev->memdev->interleave_index;
996 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
997 if (nfit_idt->idt->interleave_index != idt_idx)
999 nfit_mem->idt_bdw = nfit_idt->idt;
1006 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1007 struct acpi_nfit_system_address *spa)
1009 struct nfit_mem *nfit_mem, *found;
1010 struct nfit_memdev *nfit_memdev;
1011 int type = spa ? nfit_spa_type(spa) : 0;
1023 * This loop runs in two modes, when a dimm is mapped the loop
1024 * adds memdev associations to an existing dimm, or creates a
1025 * dimm. In the unmapped dimm case this loop sweeps for memdev
1026 * instances with an invalid / zero range_index and adds those
1027 * dimms without spa associations.
1029 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1030 struct nfit_flush *nfit_flush;
1031 struct nfit_dcr *nfit_dcr;
1035 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1037 if (!spa && nfit_memdev->memdev->range_index)
1040 dcr = nfit_memdev->memdev->region_index;
1041 device_handle = nfit_memdev->memdev->device_handle;
1042 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1043 if (__to_nfit_memdev(nfit_mem)->device_handle
1052 nfit_mem = devm_kzalloc(acpi_desc->dev,
1053 sizeof(*nfit_mem), GFP_KERNEL);
1056 INIT_LIST_HEAD(&nfit_mem->list);
1057 nfit_mem->acpi_desc = acpi_desc;
1058 list_add(&nfit_mem->list, &acpi_desc->dimms);
1061 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1062 if (nfit_dcr->dcr->region_index != dcr)
1065 * Record the control region for the dimm. For
1066 * the ACPI 6.1 case, where there are separate
1067 * control regions for the pmem vs blk
1068 * interfaces, be sure to record the extended
1072 nfit_mem->dcr = nfit_dcr->dcr;
1073 else if (nfit_mem->dcr->windows == 0
1074 && nfit_dcr->dcr->windows)
1075 nfit_mem->dcr = nfit_dcr->dcr;
1079 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1080 struct acpi_nfit_flush_address *flush;
1083 if (nfit_flush->flush->device_handle != device_handle)
1085 nfit_mem->nfit_flush = nfit_flush;
1086 flush = nfit_flush->flush;
1087 nfit_mem->flush_wpq = devm_kcalloc(acpi_desc->dev,
1089 sizeof(struct resource),
1091 if (!nfit_mem->flush_wpq)
1093 for (i = 0; i < flush->hint_count; i++) {
1094 struct resource *res = &nfit_mem->flush_wpq[i];
1096 res->start = flush->hint_address[i];
1097 res->end = res->start + 8 - 1;
1102 if (dcr && !nfit_mem->dcr) {
1103 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1104 spa->range_index, dcr);
1108 if (type == NFIT_SPA_DCR) {
1109 struct nfit_idt *nfit_idt;
1112 /* multiple dimms may share a SPA when interleaved */
1113 nfit_mem->spa_dcr = spa;
1114 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1115 idt_idx = nfit_memdev->memdev->interleave_index;
1116 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1117 if (nfit_idt->idt->interleave_index != idt_idx)
1119 nfit_mem->idt_dcr = nfit_idt->idt;
1122 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
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, struct list_head *_a, struct list_head *_b)
1139 struct nfit_mem *a = container_of(_a, typeof(*a), list);
1140 struct nfit_mem *b = container_of(_b, typeof(*b), list);
1141 u32 handleA, handleB;
1143 handleA = __to_nfit_memdev(a)->device_handle;
1144 handleB = __to_nfit_memdev(b)->device_handle;
1145 if (handleA < handleB)
1147 else if (handleA > handleB)
1152 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1154 struct nfit_spa *nfit_spa;
1159 * For each SPA-DCR or SPA-PMEM address range find its
1160 * corresponding MEMDEV(s). From each MEMDEV find the
1161 * corresponding DCR. Then, if we're operating on a SPA-DCR,
1162 * try to find a SPA-BDW and a corresponding BDW that references
1163 * the DCR. Throw it all into an nfit_mem object. Note, that
1164 * BDWs are optional.
1166 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1167 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1173 * If a DIMM has failed to be mapped into SPA there will be no
1174 * SPA entries above. Find and register all the unmapped DIMMs
1175 * for reporting and recovery purposes.
1177 rc = __nfit_mem_init(acpi_desc, NULL);
1181 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1186 static ssize_t bus_dsm_mask_show(struct device *dev,
1187 struct device_attribute *attr, char *buf)
1189 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1190 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1192 return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1194 static struct device_attribute dev_attr_bus_dsm_mask =
1195 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1197 static ssize_t revision_show(struct device *dev,
1198 struct device_attribute *attr, char *buf)
1200 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1201 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1202 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1204 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1206 static DEVICE_ATTR_RO(revision);
1208 static ssize_t hw_error_scrub_show(struct device *dev,
1209 struct device_attribute *attr, char *buf)
1211 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1212 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1213 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1215 return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1219 * The 'hw_error_scrub' attribute can have the following values written to it:
1220 * '0': Switch to the default mode where an exception will only insert
1221 * the address of the memory error into the poison and badblocks lists.
1222 * '1': Enable a full scrub to happen if an exception for a memory error is
1225 static ssize_t hw_error_scrub_store(struct device *dev,
1226 struct device_attribute *attr, const char *buf, size_t size)
1228 struct nvdimm_bus_descriptor *nd_desc;
1232 rc = kstrtol(buf, 0, &val);
1237 nd_desc = dev_get_drvdata(dev);
1239 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1242 case HW_ERROR_SCRUB_ON:
1243 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1245 case HW_ERROR_SCRUB_OFF:
1246 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1258 static DEVICE_ATTR_RW(hw_error_scrub);
1261 * This shows the number of full Address Range Scrubs that have been
1262 * completed since driver load time. Userspace can wait on this using
1263 * select/poll etc. A '+' at the end indicates an ARS is in progress
1265 static ssize_t scrub_show(struct device *dev,
1266 struct device_attribute *attr, char *buf)
1268 struct nvdimm_bus_descriptor *nd_desc;
1269 ssize_t rc = -ENXIO;
1272 nd_desc = dev_get_drvdata(dev);
1274 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1276 mutex_lock(&acpi_desc->init_mutex);
1277 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1278 acpi_desc->scrub_busy
1279 && !acpi_desc->cancel ? "+\n" : "\n");
1280 mutex_unlock(&acpi_desc->init_mutex);
1286 static ssize_t scrub_store(struct device *dev,
1287 struct device_attribute *attr, const char *buf, size_t size)
1289 struct nvdimm_bus_descriptor *nd_desc;
1293 rc = kstrtol(buf, 0, &val);
1300 nd_desc = dev_get_drvdata(dev);
1302 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1304 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1311 static DEVICE_ATTR_RW(scrub);
1313 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1315 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1316 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1317 | 1 << ND_CMD_ARS_STATUS;
1319 return (nd_desc->cmd_mask & mask) == mask;
1322 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1324 struct device *dev = container_of(kobj, struct device, kobj);
1325 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1327 if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1332 static struct attribute *acpi_nfit_attributes[] = {
1333 &dev_attr_revision.attr,
1334 &dev_attr_scrub.attr,
1335 &dev_attr_hw_error_scrub.attr,
1336 &dev_attr_bus_dsm_mask.attr,
1340 static const struct attribute_group acpi_nfit_attribute_group = {
1342 .attrs = acpi_nfit_attributes,
1343 .is_visible = nfit_visible,
1346 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1347 &nvdimm_bus_attribute_group,
1348 &acpi_nfit_attribute_group,
1352 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1354 struct nvdimm *nvdimm = to_nvdimm(dev);
1355 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1357 return __to_nfit_memdev(nfit_mem);
1360 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1362 struct nvdimm *nvdimm = to_nvdimm(dev);
1363 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1365 return nfit_mem->dcr;
1368 static ssize_t handle_show(struct device *dev,
1369 struct device_attribute *attr, char *buf)
1371 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1373 return sprintf(buf, "%#x\n", memdev->device_handle);
1375 static DEVICE_ATTR_RO(handle);
1377 static ssize_t phys_id_show(struct device *dev,
1378 struct device_attribute *attr, char *buf)
1380 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1382 return sprintf(buf, "%#x\n", memdev->physical_id);
1384 static DEVICE_ATTR_RO(phys_id);
1386 static ssize_t vendor_show(struct device *dev,
1387 struct device_attribute *attr, char *buf)
1389 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1391 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1393 static DEVICE_ATTR_RO(vendor);
1395 static ssize_t rev_id_show(struct device *dev,
1396 struct device_attribute *attr, char *buf)
1398 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1400 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1402 static DEVICE_ATTR_RO(rev_id);
1404 static ssize_t device_show(struct device *dev,
1405 struct device_attribute *attr, char *buf)
1407 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1409 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1411 static DEVICE_ATTR_RO(device);
1413 static ssize_t subsystem_vendor_show(struct device *dev,
1414 struct device_attribute *attr, char *buf)
1416 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1418 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1420 static DEVICE_ATTR_RO(subsystem_vendor);
1422 static ssize_t subsystem_rev_id_show(struct device *dev,
1423 struct device_attribute *attr, char *buf)
1425 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1427 return sprintf(buf, "0x%04x\n",
1428 be16_to_cpu(dcr->subsystem_revision_id));
1430 static DEVICE_ATTR_RO(subsystem_rev_id);
1432 static ssize_t subsystem_device_show(struct device *dev,
1433 struct device_attribute *attr, char *buf)
1435 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1437 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1439 static DEVICE_ATTR_RO(subsystem_device);
1441 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1443 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1446 if (nfit_mem->memdev_pmem)
1448 if (nfit_mem->memdev_bdw)
1453 static ssize_t format_show(struct device *dev,
1454 struct device_attribute *attr, char *buf)
1456 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1458 return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1460 static DEVICE_ATTR_RO(format);
1462 static ssize_t format1_show(struct device *dev,
1463 struct device_attribute *attr, char *buf)
1466 ssize_t rc = -ENXIO;
1467 struct nfit_mem *nfit_mem;
1468 struct nfit_memdev *nfit_memdev;
1469 struct acpi_nfit_desc *acpi_desc;
1470 struct nvdimm *nvdimm = to_nvdimm(dev);
1471 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1473 nfit_mem = nvdimm_provider_data(nvdimm);
1474 acpi_desc = nfit_mem->acpi_desc;
1475 handle = to_nfit_memdev(dev)->device_handle;
1477 /* assumes DIMMs have at most 2 published interface codes */
1478 mutex_lock(&acpi_desc->init_mutex);
1479 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1480 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1481 struct nfit_dcr *nfit_dcr;
1483 if (memdev->device_handle != handle)
1486 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1487 if (nfit_dcr->dcr->region_index != memdev->region_index)
1489 if (nfit_dcr->dcr->code == dcr->code)
1491 rc = sprintf(buf, "0x%04x\n",
1492 le16_to_cpu(nfit_dcr->dcr->code));
1498 mutex_unlock(&acpi_desc->init_mutex);
1501 static DEVICE_ATTR_RO(format1);
1503 static ssize_t formats_show(struct device *dev,
1504 struct device_attribute *attr, char *buf)
1506 struct nvdimm *nvdimm = to_nvdimm(dev);
1508 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1510 static DEVICE_ATTR_RO(formats);
1512 static ssize_t serial_show(struct device *dev,
1513 struct device_attribute *attr, char *buf)
1515 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1517 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1519 static DEVICE_ATTR_RO(serial);
1521 static ssize_t family_show(struct device *dev,
1522 struct device_attribute *attr, char *buf)
1524 struct nvdimm *nvdimm = to_nvdimm(dev);
1525 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1527 if (nfit_mem->family < 0)
1529 return sprintf(buf, "%d\n", nfit_mem->family);
1531 static DEVICE_ATTR_RO(family);
1533 static ssize_t dsm_mask_show(struct device *dev,
1534 struct device_attribute *attr, char *buf)
1536 struct nvdimm *nvdimm = to_nvdimm(dev);
1537 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1539 if (nfit_mem->family < 0)
1541 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1543 static DEVICE_ATTR_RO(dsm_mask);
1545 static ssize_t flags_show(struct device *dev,
1546 struct device_attribute *attr, char *buf)
1548 u16 flags = to_nfit_memdev(dev)->flags;
1550 return sprintf(buf, "%s%s%s%s%s%s%s\n",
1551 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1552 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1553 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1554 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1555 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1556 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1557 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1559 static DEVICE_ATTR_RO(flags);
1561 static ssize_t id_show(struct device *dev,
1562 struct device_attribute *attr, char *buf)
1564 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1566 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1567 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1568 be16_to_cpu(dcr->vendor_id),
1569 dcr->manufacturing_location,
1570 be16_to_cpu(dcr->manufacturing_date),
1571 be32_to_cpu(dcr->serial_number));
1573 return sprintf(buf, "%04x-%08x\n",
1574 be16_to_cpu(dcr->vendor_id),
1575 be32_to_cpu(dcr->serial_number));
1577 static DEVICE_ATTR_RO(id);
1579 static struct attribute *acpi_nfit_dimm_attributes[] = {
1580 &dev_attr_handle.attr,
1581 &dev_attr_phys_id.attr,
1582 &dev_attr_vendor.attr,
1583 &dev_attr_device.attr,
1584 &dev_attr_rev_id.attr,
1585 &dev_attr_subsystem_vendor.attr,
1586 &dev_attr_subsystem_device.attr,
1587 &dev_attr_subsystem_rev_id.attr,
1588 &dev_attr_format.attr,
1589 &dev_attr_formats.attr,
1590 &dev_attr_format1.attr,
1591 &dev_attr_serial.attr,
1592 &dev_attr_flags.attr,
1594 &dev_attr_family.attr,
1595 &dev_attr_dsm_mask.attr,
1599 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1600 struct attribute *a, int n)
1602 struct device *dev = container_of(kobj, struct device, kobj);
1603 struct nvdimm *nvdimm = to_nvdimm(dev);
1605 if (!to_nfit_dcr(dev)) {
1606 /* Without a dcr only the memdev attributes can be surfaced */
1607 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1608 || a == &dev_attr_flags.attr
1609 || a == &dev_attr_family.attr
1610 || a == &dev_attr_dsm_mask.attr)
1615 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1620 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1622 .attrs = acpi_nfit_dimm_attributes,
1623 .is_visible = acpi_nfit_dimm_attr_visible,
1626 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1627 &nvdimm_attribute_group,
1628 &nd_device_attribute_group,
1629 &acpi_nfit_dimm_attribute_group,
1633 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1636 struct nfit_mem *nfit_mem;
1638 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1639 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1640 return nfit_mem->nvdimm;
1645 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1647 struct nfit_mem *nfit_mem;
1648 struct acpi_nfit_desc *acpi_desc;
1650 dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1653 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1654 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1659 acpi_desc = dev_get_drvdata(dev->parent);
1664 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1667 nfit_mem = dev_get_drvdata(dev);
1668 if (nfit_mem && nfit_mem->flags_attr)
1669 sysfs_notify_dirent(nfit_mem->flags_attr);
1671 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1673 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1675 struct acpi_device *adev = data;
1676 struct device *dev = &adev->dev;
1678 device_lock(dev->parent);
1679 __acpi_nvdimm_notify(dev, event);
1680 device_unlock(dev->parent);
1683 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1688 status = acpi_get_handle(adev->handle, method, &handle);
1690 if (ACPI_SUCCESS(status))
1695 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1696 struct nfit_mem *nfit_mem, u32 device_handle)
1698 struct acpi_device *adev, *adev_dimm;
1699 struct device *dev = acpi_desc->dev;
1700 unsigned long dsm_mask;
1705 /* nfit test assumes 1:1 relationship between commands and dsms */
1706 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1707 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1708 adev = to_acpi_dev(acpi_desc);
1712 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1713 nfit_mem->adev = adev_dimm;
1715 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1717 return force_enable_dimms ? 0 : -ENODEV;
1720 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1721 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1722 dev_err(dev, "%s: notification registration failed\n",
1723 dev_name(&adev_dimm->dev));
1727 * Record nfit_mem for the notification path to track back to
1728 * the nfit sysfs attributes for this dimm device object.
1730 dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1733 * Until standardization materializes we need to consider 4
1734 * different command sets. Note, that checking for function0 (bit0)
1735 * tells us if any commands are reachable through this GUID.
1737 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1738 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1739 if (family < 0 || i == default_dsm_family)
1742 /* limit the supported commands to those that are publicly documented */
1743 nfit_mem->family = family;
1744 if (override_dsm_mask && !disable_vendor_specific)
1745 dsm_mask = override_dsm_mask;
1746 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1747 dsm_mask = NVDIMM_INTEL_CMDMASK;
1748 if (disable_vendor_specific)
1749 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1750 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1751 dsm_mask = 0x1c3c76;
1752 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1754 if (disable_vendor_specific)
1755 dsm_mask &= ~(1 << 8);
1756 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1757 dsm_mask = 0xffffffff;
1759 dev_dbg(dev, "unknown dimm command family\n");
1760 nfit_mem->family = -1;
1761 /* DSMs are optional, continue loading the driver... */
1765 guid = to_nfit_uuid(nfit_mem->family);
1766 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1767 if (acpi_check_dsm(adev_dimm->handle, guid,
1768 nfit_dsm_revid(nfit_mem->family, i),
1770 set_bit(i, &nfit_mem->dsm_mask);
1772 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1773 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1774 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1775 nfit_mem->has_lsr = true;
1778 if (nfit_mem->has_lsr && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1779 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1780 nfit_mem->has_lsw = true;
1786 static void shutdown_dimm_notify(void *data)
1788 struct acpi_nfit_desc *acpi_desc = data;
1789 struct nfit_mem *nfit_mem;
1791 mutex_lock(&acpi_desc->init_mutex);
1793 * Clear out the nfit_mem->flags_attr and shut down dimm event
1796 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1797 struct acpi_device *adev_dimm = nfit_mem->adev;
1799 if (nfit_mem->flags_attr) {
1800 sysfs_put(nfit_mem->flags_attr);
1801 nfit_mem->flags_attr = NULL;
1804 acpi_remove_notify_handler(adev_dimm->handle,
1805 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1806 dev_set_drvdata(&adev_dimm->dev, NULL);
1809 mutex_unlock(&acpi_desc->init_mutex);
1812 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1814 struct nfit_mem *nfit_mem;
1815 int dimm_count = 0, rc;
1816 struct nvdimm *nvdimm;
1818 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1819 struct acpi_nfit_flush_address *flush;
1820 unsigned long flags = 0, cmd_mask;
1821 struct nfit_memdev *nfit_memdev;
1825 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1826 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1832 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1833 set_bit(NDD_ALIASING, &flags);
1835 /* collate flags across all memdevs for this dimm */
1836 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1837 struct acpi_nfit_memory_map *dimm_memdev;
1839 dimm_memdev = __to_nfit_memdev(nfit_mem);
1840 if (dimm_memdev->device_handle
1841 != nfit_memdev->memdev->device_handle)
1843 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1846 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1847 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1848 set_bit(NDD_UNARMED, &flags);
1850 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1855 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1856 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1857 * userspace interface.
1859 cmd_mask = 1UL << ND_CMD_CALL;
1860 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1862 * These commands have a 1:1 correspondence
1863 * between DSM payload and libnvdimm ioctl
1866 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1869 if (nfit_mem->has_lsr) {
1870 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1871 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1873 if (nfit_mem->has_lsw)
1874 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1876 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1878 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1879 acpi_nfit_dimm_attribute_groups,
1880 flags, cmd_mask, flush ? flush->hint_count : 0,
1881 nfit_mem->flush_wpq);
1885 nfit_mem->nvdimm = nvdimm;
1888 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1891 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1892 nvdimm_name(nvdimm),
1893 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1894 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1895 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1896 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1897 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1901 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1906 * Now that dimms are successfully registered, and async registration
1907 * is flushed, attempt to enable event notification.
1909 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1910 struct kernfs_node *nfit_kernfs;
1912 nvdimm = nfit_mem->nvdimm;
1916 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1918 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1920 sysfs_put(nfit_kernfs);
1921 if (!nfit_mem->flags_attr)
1922 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1923 nvdimm_name(nvdimm));
1926 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1931 * These constants are private because there are no kernel consumers of
1934 enum nfit_aux_cmds {
1935 NFIT_CMD_TRANSLATE_SPA = 5,
1936 NFIT_CMD_ARS_INJECT_SET = 7,
1937 NFIT_CMD_ARS_INJECT_CLEAR = 8,
1938 NFIT_CMD_ARS_INJECT_GET = 9,
1941 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1943 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1944 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1945 struct acpi_device *adev;
1946 unsigned long dsm_mask;
1949 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1950 nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1951 adev = to_acpi_dev(acpi_desc);
1955 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1956 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1957 set_bit(i, &nd_desc->cmd_mask);
1958 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1961 (1 << ND_CMD_ARS_CAP) |
1962 (1 << ND_CMD_ARS_START) |
1963 (1 << ND_CMD_ARS_STATUS) |
1964 (1 << ND_CMD_CLEAR_ERROR) |
1965 (1 << NFIT_CMD_TRANSLATE_SPA) |
1966 (1 << NFIT_CMD_ARS_INJECT_SET) |
1967 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1968 (1 << NFIT_CMD_ARS_INJECT_GET);
1969 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1970 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1971 set_bit(i, &nd_desc->bus_dsm_mask);
1974 static ssize_t range_index_show(struct device *dev,
1975 struct device_attribute *attr, char *buf)
1977 struct nd_region *nd_region = to_nd_region(dev);
1978 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1980 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1982 static DEVICE_ATTR_RO(range_index);
1984 static struct attribute *acpi_nfit_region_attributes[] = {
1985 &dev_attr_range_index.attr,
1989 static const struct attribute_group acpi_nfit_region_attribute_group = {
1991 .attrs = acpi_nfit_region_attributes,
1994 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1995 &nd_region_attribute_group,
1996 &nd_mapping_attribute_group,
1997 &nd_device_attribute_group,
1998 &nd_numa_attribute_group,
1999 &acpi_nfit_region_attribute_group,
2003 /* enough info to uniquely specify an interleave set */
2004 struct nfit_set_info {
2005 struct nfit_set_info_map {
2012 struct nfit_set_info2 {
2013 struct nfit_set_info_map2 {
2017 u16 manufacturing_date;
2018 u8 manufacturing_location;
2023 static size_t sizeof_nfit_set_info(int num_mappings)
2025 return sizeof(struct nfit_set_info)
2026 + num_mappings * sizeof(struct nfit_set_info_map);
2029 static size_t sizeof_nfit_set_info2(int num_mappings)
2031 return sizeof(struct nfit_set_info2)
2032 + num_mappings * sizeof(struct nfit_set_info_map2);
2035 static int cmp_map_compat(const void *m0, const void *m1)
2037 const struct nfit_set_info_map *map0 = m0;
2038 const struct nfit_set_info_map *map1 = m1;
2040 return memcmp(&map0->region_offset, &map1->region_offset,
2044 static int cmp_map(const void *m0, const void *m1)
2046 const struct nfit_set_info_map *map0 = m0;
2047 const struct nfit_set_info_map *map1 = m1;
2049 if (map0->region_offset < map1->region_offset)
2051 else if (map0->region_offset > map1->region_offset)
2056 static int cmp_map2(const void *m0, const void *m1)
2058 const struct nfit_set_info_map2 *map0 = m0;
2059 const struct nfit_set_info_map2 *map1 = m1;
2061 if (map0->region_offset < map1->region_offset)
2063 else if (map0->region_offset > map1->region_offset)
2068 /* Retrieve the nth entry referencing this spa */
2069 static struct acpi_nfit_memory_map *memdev_from_spa(
2070 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2072 struct nfit_memdev *nfit_memdev;
2074 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2075 if (nfit_memdev->memdev->range_index == range_index)
2077 return nfit_memdev->memdev;
2081 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2082 struct nd_region_desc *ndr_desc,
2083 struct acpi_nfit_system_address *spa)
2085 struct device *dev = acpi_desc->dev;
2086 struct nd_interleave_set *nd_set;
2087 u16 nr = ndr_desc->num_mappings;
2088 struct nfit_set_info2 *info2;
2089 struct nfit_set_info *info;
2092 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2095 ndr_desc->nd_set = nd_set;
2096 guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2098 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2102 info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2106 for (i = 0; i < nr; i++) {
2107 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2108 struct nfit_set_info_map *map = &info->mapping[i];
2109 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2110 struct nvdimm *nvdimm = mapping->nvdimm;
2111 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2112 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2113 spa->range_index, i);
2114 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2116 if (!memdev || !nfit_mem->dcr) {
2117 dev_err(dev, "%s: failed to find DCR\n", __func__);
2121 map->region_offset = memdev->region_offset;
2122 map->serial_number = dcr->serial_number;
2124 map2->region_offset = memdev->region_offset;
2125 map2->serial_number = dcr->serial_number;
2126 map2->vendor_id = dcr->vendor_id;
2127 map2->manufacturing_date = dcr->manufacturing_date;
2128 map2->manufacturing_location = dcr->manufacturing_location;
2131 /* v1.1 namespaces */
2132 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2134 nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2136 /* v1.2 namespaces */
2137 sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2139 nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2141 /* support v1.1 namespaces created with the wrong sort order */
2142 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2143 cmp_map_compat, NULL);
2144 nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2146 /* record the result of the sort for the mapping position */
2147 for (i = 0; i < nr; i++) {
2148 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2151 for (j = 0; j < nr; j++) {
2152 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2153 struct nvdimm *nvdimm = mapping->nvdimm;
2154 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2155 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2157 if (map2->serial_number == dcr->serial_number &&
2158 map2->vendor_id == dcr->vendor_id &&
2159 map2->manufacturing_date == dcr->manufacturing_date &&
2160 map2->manufacturing_location
2161 == dcr->manufacturing_location) {
2162 mapping->position = i;
2168 ndr_desc->nd_set = nd_set;
2169 devm_kfree(dev, info);
2170 devm_kfree(dev, info2);
2175 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2177 struct acpi_nfit_interleave *idt = mmio->idt;
2178 u32 sub_line_offset, line_index, line_offset;
2179 u64 line_no, table_skip_count, table_offset;
2181 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2182 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2183 line_offset = idt->line_offset[line_index]
2185 table_offset = table_skip_count * mmio->table_size;
2187 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2190 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2192 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2193 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2194 const u32 STATUS_MASK = 0x80000037;
2196 if (mmio->num_lines)
2197 offset = to_interleave_offset(offset, mmio);
2199 return readl(mmio->addr.base + offset) & STATUS_MASK;
2202 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2203 resource_size_t dpa, unsigned int len, unsigned int write)
2206 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2209 BCW_OFFSET_MASK = (1ULL << 48)-1,
2211 BCW_LEN_MASK = (1ULL << 8) - 1,
2215 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2216 len = len >> L1_CACHE_SHIFT;
2217 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2218 cmd |= ((u64) write) << BCW_CMD_SHIFT;
2220 offset = nfit_blk->cmd_offset + mmio->size * bw;
2221 if (mmio->num_lines)
2222 offset = to_interleave_offset(offset, mmio);
2224 writeq(cmd, mmio->addr.base + offset);
2225 nvdimm_flush(nfit_blk->nd_region);
2227 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2228 readq(mmio->addr.base + offset);
2231 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2232 resource_size_t dpa, void *iobuf, size_t len, int rw,
2235 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2236 unsigned int copied = 0;
2240 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2241 + lane * mmio->size;
2242 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2247 if (mmio->num_lines) {
2250 offset = to_interleave_offset(base_offset + copied,
2252 div_u64_rem(offset, mmio->line_size, &line_offset);
2253 c = min_t(size_t, len, mmio->line_size - line_offset);
2255 offset = base_offset + nfit_blk->bdw_offset;
2260 memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2262 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2263 arch_invalidate_pmem((void __force *)
2264 mmio->addr.aperture + offset, c);
2266 memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2274 nvdimm_flush(nfit_blk->nd_region);
2276 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2280 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2281 resource_size_t dpa, void *iobuf, u64 len, int rw)
2283 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2284 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2285 struct nd_region *nd_region = nfit_blk->nd_region;
2286 unsigned int lane, copied = 0;
2289 lane = nd_region_acquire_lane(nd_region);
2291 u64 c = min(len, mmio->size);
2293 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2294 iobuf + copied, c, rw, lane);
2301 nd_region_release_lane(nd_region, lane);
2306 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2307 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2310 mmio->num_lines = idt->line_count;
2311 mmio->line_size = idt->line_size;
2312 if (interleave_ways == 0)
2314 mmio->table_size = mmio->num_lines * interleave_ways
2321 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2322 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2324 struct nd_cmd_dimm_flags flags;
2327 memset(&flags, 0, sizeof(flags));
2328 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2329 sizeof(flags), NULL);
2331 if (rc >= 0 && flags.status == 0)
2332 nfit_blk->dimm_flags = flags.flags;
2333 else if (rc == -ENOTTY) {
2334 /* fall back to a conservative default */
2335 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2343 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2346 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2347 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2348 struct nfit_blk_mmio *mmio;
2349 struct nfit_blk *nfit_blk;
2350 struct nfit_mem *nfit_mem;
2351 struct nvdimm *nvdimm;
2354 nvdimm = nd_blk_region_to_dimm(ndbr);
2355 nfit_mem = nvdimm_provider_data(nvdimm);
2356 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2357 dev_dbg(dev, "missing%s%s%s\n",
2358 nfit_mem ? "" : " nfit_mem",
2359 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2360 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2364 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2367 nd_blk_region_set_provider_data(ndbr, nfit_blk);
2368 nfit_blk->nd_region = to_nd_region(dev);
2370 /* map block aperture memory */
2371 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2372 mmio = &nfit_blk->mmio[BDW];
2373 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2374 nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2375 if (!mmio->addr.base) {
2376 dev_dbg(dev, "%s failed to map bdw\n",
2377 nvdimm_name(nvdimm));
2380 mmio->size = nfit_mem->bdw->size;
2381 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2382 mmio->idt = nfit_mem->idt_bdw;
2383 mmio->spa = nfit_mem->spa_bdw;
2384 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2385 nfit_mem->memdev_bdw->interleave_ways);
2387 dev_dbg(dev, "%s failed to init bdw interleave\n",
2388 nvdimm_name(nvdimm));
2392 /* map block control memory */
2393 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2394 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2395 mmio = &nfit_blk->mmio[DCR];
2396 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2397 nfit_mem->spa_dcr->length);
2398 if (!mmio->addr.base) {
2399 dev_dbg(dev, "%s failed to map dcr\n",
2400 nvdimm_name(nvdimm));
2403 mmio->size = nfit_mem->dcr->window_size;
2404 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2405 mmio->idt = nfit_mem->idt_dcr;
2406 mmio->spa = nfit_mem->spa_dcr;
2407 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2408 nfit_mem->memdev_dcr->interleave_ways);
2410 dev_dbg(dev, "%s failed to init dcr interleave\n",
2411 nvdimm_name(nvdimm));
2415 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2417 dev_dbg(dev, "%s failed get DIMM flags\n",
2418 nvdimm_name(nvdimm));
2422 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2423 dev_warn(dev, "unable to guarantee persistence of writes\n");
2425 if (mmio->line_size == 0)
2428 if ((u32) nfit_blk->cmd_offset % mmio->line_size
2429 + 8 > mmio->line_size) {
2430 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2432 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2433 + 8 > mmio->line_size) {
2434 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2441 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2442 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2444 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2445 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2448 cmd->address = spa->address;
2449 cmd->length = spa->length;
2450 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2451 sizeof(*cmd), &cmd_rc);
2457 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2461 struct nd_cmd_ars_start ars_start;
2462 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2463 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2465 memset(&ars_start, 0, sizeof(ars_start));
2466 ars_start.address = spa->address;
2467 ars_start.length = spa->length;
2468 if (test_bit(ARS_SHORT, &nfit_spa->ars_state))
2469 ars_start.flags = ND_ARS_RETURN_PREV_DATA;
2470 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2471 ars_start.type = ND_ARS_PERSISTENT;
2472 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2473 ars_start.type = ND_ARS_VOLATILE;
2477 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2478 sizeof(ars_start), &cmd_rc);
2485 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2488 struct nd_cmd_ars_start ars_start;
2489 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2490 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2492 memset(&ars_start, 0, sizeof(ars_start));
2493 ars_start.address = ars_status->restart_address;
2494 ars_start.length = ars_status->restart_length;
2495 ars_start.type = ars_status->type;
2496 ars_start.flags = acpi_desc->ars_start_flags;
2497 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2498 sizeof(ars_start), &cmd_rc);
2504 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2506 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2507 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2510 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2511 acpi_desc->max_ars, &cmd_rc);
2517 static void ars_complete(struct acpi_nfit_desc *acpi_desc,
2518 struct nfit_spa *nfit_spa)
2520 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2521 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2522 struct nd_region *nd_region = nfit_spa->nd_region;
2525 if ((ars_status->address >= spa->address && ars_status->address
2526 < spa->address + spa->length)
2527 || (ars_status->address < spa->address)) {
2529 * Assume that if a scrub starts at an offset from the
2530 * start of nfit_spa that we are in the continuation
2533 * Otherwise, if the scrub covers the spa range, mark
2534 * any pending request complete.
2536 if (ars_status->address + ars_status->length
2537 >= spa->address + spa->length)
2544 if (test_bit(ARS_DONE, &nfit_spa->ars_state))
2547 if (!test_and_clear_bit(ARS_REQ, &nfit_spa->ars_state))
2551 dev = nd_region_dev(nd_region);
2552 nvdimm_region_notify(nd_region, NVDIMM_REVALIDATE_POISON);
2554 dev = acpi_desc->dev;
2556 dev_dbg(dev, "ARS: range %d %s complete\n", spa->range_index,
2557 test_bit(ARS_SHORT, &nfit_spa->ars_state)
2558 ? "short" : "long");
2559 clear_bit(ARS_SHORT, &nfit_spa->ars_state);
2560 set_bit(ARS_DONE, &nfit_spa->ars_state);
2563 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc)
2565 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2566 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2571 * First record starts at 44 byte offset from the start of the
2574 if (ars_status->out_length < 44)
2576 for (i = 0; i < ars_status->num_records; i++) {
2577 /* only process full records */
2578 if (ars_status->out_length
2579 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2581 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2582 ars_status->records[i].err_address,
2583 ars_status->records[i].length);
2587 if (i < ars_status->num_records)
2588 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2593 static void acpi_nfit_remove_resource(void *data)
2595 struct resource *res = data;
2597 remove_resource(res);
2600 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2601 struct nd_region_desc *ndr_desc)
2603 struct resource *res, *nd_res = ndr_desc->res;
2606 /* No operation if the region is already registered as PMEM */
2607 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2608 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2609 if (is_pmem == REGION_INTERSECTS)
2612 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2616 res->name = "Persistent Memory";
2617 res->start = nd_res->start;
2618 res->end = nd_res->end;
2619 res->flags = IORESOURCE_MEM;
2620 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2622 ret = insert_resource(&iomem_resource, res);
2626 ret = devm_add_action_or_reset(acpi_desc->dev,
2627 acpi_nfit_remove_resource,
2635 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2636 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2637 struct acpi_nfit_memory_map *memdev,
2638 struct nfit_spa *nfit_spa)
2640 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2641 memdev->device_handle);
2642 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2643 struct nd_blk_region_desc *ndbr_desc;
2644 struct nfit_mem *nfit_mem;
2648 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2649 spa->range_index, memdev->device_handle);
2653 mapping->nvdimm = nvdimm;
2654 switch (nfit_spa_type(spa)) {
2656 case NFIT_SPA_VOLATILE:
2657 mapping->start = memdev->address;
2658 mapping->size = memdev->region_size;
2661 nfit_mem = nvdimm_provider_data(nvdimm);
2662 if (!nfit_mem || !nfit_mem->bdw) {
2663 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2664 spa->range_index, nvdimm_name(nvdimm));
2668 mapping->size = nfit_mem->bdw->capacity;
2669 mapping->start = nfit_mem->bdw->start_address;
2670 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2671 ndr_desc->mapping = mapping;
2672 ndr_desc->num_mappings = 1;
2673 ndbr_desc = to_blk_region_desc(ndr_desc);
2674 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2675 ndbr_desc->do_io = acpi_desc->blk_do_io;
2676 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2679 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2681 if (!nfit_spa->nd_region)
2689 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2691 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2692 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2693 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2694 nfit_spa_type(spa) == NFIT_SPA_PCD);
2697 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2699 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2700 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2701 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2704 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2705 struct nfit_spa *nfit_spa)
2707 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2708 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2709 struct nd_blk_region_desc ndbr_desc;
2710 struct nd_region_desc *ndr_desc;
2711 struct nfit_memdev *nfit_memdev;
2712 struct nvdimm_bus *nvdimm_bus;
2713 struct resource res;
2716 if (nfit_spa->nd_region)
2719 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2720 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2724 memset(&res, 0, sizeof(res));
2725 memset(&mappings, 0, sizeof(mappings));
2726 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2727 res.start = spa->address;
2728 res.end = res.start + spa->length - 1;
2729 ndr_desc = &ndbr_desc.ndr_desc;
2730 ndr_desc->res = &res;
2731 ndr_desc->provider_data = nfit_spa;
2732 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2733 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2734 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2735 spa->proximity_domain);
2737 ndr_desc->numa_node = NUMA_NO_NODE;
2740 * Persistence domain bits are hierarchical, if
2741 * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2742 * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2744 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2745 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2746 else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2747 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2749 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2750 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2751 struct nd_mapping_desc *mapping;
2753 if (memdev->range_index != spa->range_index)
2755 if (count >= ND_MAX_MAPPINGS) {
2756 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2757 spa->range_index, ND_MAX_MAPPINGS);
2760 mapping = &mappings[count++];
2761 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2767 ndr_desc->mapping = mappings;
2768 ndr_desc->num_mappings = count;
2769 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2773 nvdimm_bus = acpi_desc->nvdimm_bus;
2774 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2775 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2777 dev_warn(acpi_desc->dev,
2778 "failed to insert pmem resource to iomem: %d\n",
2783 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2785 if (!nfit_spa->nd_region)
2787 } else if (nfit_spa_is_volatile(spa)) {
2788 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2790 if (!nfit_spa->nd_region)
2792 } else if (nfit_spa_is_virtual(spa)) {
2793 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2795 if (!nfit_spa->nd_region)
2801 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2802 nfit_spa->spa->range_index);
2806 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc)
2808 struct device *dev = acpi_desc->dev;
2809 struct nd_cmd_ars_status *ars_status;
2811 if (acpi_desc->ars_status) {
2812 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2816 ars_status = devm_kzalloc(dev, acpi_desc->max_ars, GFP_KERNEL);
2819 acpi_desc->ars_status = ars_status;
2823 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc)
2827 if (ars_status_alloc(acpi_desc))
2830 rc = ars_get_status(acpi_desc);
2832 if (rc < 0 && rc != -ENOSPC)
2835 if (ars_status_process_records(acpi_desc))
2841 static int ars_register(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa,
2847 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2849 set_bit(ARS_REQ, &nfit_spa->ars_state);
2850 set_bit(ARS_SHORT, &nfit_spa->ars_state);
2855 rc = ars_start(acpi_desc, nfit_spa);
2859 } else if (rc == 0) {
2860 rc = acpi_nfit_query_poison(acpi_desc);
2862 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2866 clear_bit(ARS_SHORT, &nfit_spa->ars_state);
2868 ars_complete(acpi_desc, nfit_spa);
2874 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2878 if (test_and_clear_bit(ARS_DONE, &nfit_spa->ars_state))
2879 set_bit(ARS_REQ, &nfit_spa->ars_state);
2881 return acpi_nfit_register_region(acpi_desc, nfit_spa);
2884 static void ars_complete_all(struct acpi_nfit_desc *acpi_desc)
2886 struct nfit_spa *nfit_spa;
2888 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2889 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2891 ars_complete(acpi_desc, nfit_spa);
2895 static unsigned int __acpi_nfit_scrub(struct acpi_nfit_desc *acpi_desc,
2898 unsigned int tmo = acpi_desc->scrub_tmo;
2899 struct device *dev = acpi_desc->dev;
2900 struct nfit_spa *nfit_spa;
2902 if (acpi_desc->cancel)
2905 if (query_rc == -EBUSY) {
2906 dev_dbg(dev, "ARS: ARS busy\n");
2907 return min(30U * 60U, tmo * 2);
2909 if (query_rc == -ENOSPC) {
2910 dev_dbg(dev, "ARS: ARS continue\n");
2911 ars_continue(acpi_desc);
2914 if (query_rc && query_rc != -EAGAIN) {
2915 unsigned long long addr, end;
2917 addr = acpi_desc->ars_status->address;
2918 end = addr + acpi_desc->ars_status->length;
2919 dev_dbg(dev, "ARS: %llx-%llx failed (%d)\n", addr, end,
2923 ars_complete_all(acpi_desc);
2924 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2925 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
2927 if (test_bit(ARS_REQ, &nfit_spa->ars_state)) {
2928 int rc = ars_start(acpi_desc, nfit_spa);
2930 clear_bit(ARS_DONE, &nfit_spa->ars_state);
2931 dev = nd_region_dev(nfit_spa->nd_region);
2932 dev_dbg(dev, "ARS: range %d ARS start (%d)\n",
2933 nfit_spa->spa->range_index, rc);
2934 if (rc == 0 || rc == -EBUSY)
2936 dev_err(dev, "ARS: range %d ARS failed (%d)\n",
2937 nfit_spa->spa->range_index, rc);
2938 set_bit(ARS_FAILED, &nfit_spa->ars_state);
2944 static void __sched_ars(struct acpi_nfit_desc *acpi_desc, unsigned int tmo)
2946 lockdep_assert_held(&acpi_desc->init_mutex);
2948 acpi_desc->scrub_busy = 1;
2949 /* note this should only be set from within the workqueue */
2951 acpi_desc->scrub_tmo = tmo;
2952 queue_delayed_work(nfit_wq, &acpi_desc->dwork, tmo * HZ);
2955 static void sched_ars(struct acpi_nfit_desc *acpi_desc)
2957 __sched_ars(acpi_desc, 0);
2960 static void notify_ars_done(struct acpi_nfit_desc *acpi_desc)
2962 lockdep_assert_held(&acpi_desc->init_mutex);
2964 acpi_desc->scrub_busy = 0;
2965 acpi_desc->scrub_count++;
2966 if (acpi_desc->scrub_count_state)
2967 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2970 static void acpi_nfit_scrub(struct work_struct *work)
2972 struct acpi_nfit_desc *acpi_desc;
2976 acpi_desc = container_of(work, typeof(*acpi_desc), dwork.work);
2977 mutex_lock(&acpi_desc->init_mutex);
2978 query_rc = acpi_nfit_query_poison(acpi_desc);
2979 tmo = __acpi_nfit_scrub(acpi_desc, query_rc);
2981 __sched_ars(acpi_desc, tmo);
2983 notify_ars_done(acpi_desc);
2984 memset(acpi_desc->ars_status, 0, acpi_desc->max_ars);
2985 mutex_unlock(&acpi_desc->init_mutex);
2988 static void acpi_nfit_init_ars(struct acpi_nfit_desc *acpi_desc,
2989 struct nfit_spa *nfit_spa)
2991 int type = nfit_spa_type(nfit_spa->spa);
2992 struct nd_cmd_ars_cap ars_cap;
2995 memset(&ars_cap, 0, sizeof(ars_cap));
2996 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2999 /* check that the supported scrub types match the spa type */
3000 if (type == NFIT_SPA_VOLATILE && ((ars_cap.status >> 16)
3001 & ND_ARS_VOLATILE) == 0)
3003 if (type == NFIT_SPA_PM && ((ars_cap.status >> 16)
3004 & ND_ARS_PERSISTENT) == 0)
3007 nfit_spa->max_ars = ars_cap.max_ars_out;
3008 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
3009 acpi_desc->max_ars = max(nfit_spa->max_ars, acpi_desc->max_ars);
3010 clear_bit(ARS_FAILED, &nfit_spa->ars_state);
3011 set_bit(ARS_REQ, &nfit_spa->ars_state);
3014 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3016 struct nfit_spa *nfit_spa;
3019 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3020 set_bit(ARS_FAILED, &nfit_spa->ars_state);
3021 switch (nfit_spa_type(nfit_spa->spa)) {
3022 case NFIT_SPA_VOLATILE:
3024 acpi_nfit_init_ars(acpi_desc, nfit_spa);
3030 * Reap any results that might be pending before starting new
3033 query_rc = acpi_nfit_query_poison(acpi_desc);
3035 ars_complete_all(acpi_desc);
3037 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3038 switch (nfit_spa_type(nfit_spa->spa)) {
3039 case NFIT_SPA_VOLATILE:
3041 /* register regions and kick off initial ARS run */
3042 rc = ars_register(acpi_desc, nfit_spa, &query_rc);
3047 /* nothing to register */
3050 case NFIT_SPA_VDISK:
3052 case NFIT_SPA_PDISK:
3054 /* register known regions that don't support ARS */
3055 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3060 /* don't register unknown regions */
3064 sched_ars(acpi_desc);
3068 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3069 struct nfit_table_prev *prev)
3071 struct device *dev = acpi_desc->dev;
3073 if (!list_empty(&prev->spas) ||
3074 !list_empty(&prev->memdevs) ||
3075 !list_empty(&prev->dcrs) ||
3076 !list_empty(&prev->bdws) ||
3077 !list_empty(&prev->idts) ||
3078 !list_empty(&prev->flushes)) {
3079 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3085 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3087 struct device *dev = acpi_desc->dev;
3088 struct kernfs_node *nfit;
3089 struct device *bus_dev;
3091 if (!ars_supported(acpi_desc->nvdimm_bus))
3094 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3095 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3097 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3100 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3102 if (!acpi_desc->scrub_count_state) {
3103 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3110 static void acpi_nfit_unregister(void *data)
3112 struct acpi_nfit_desc *acpi_desc = data;
3114 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3117 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3119 struct device *dev = acpi_desc->dev;
3120 struct nfit_table_prev prev;
3124 if (!acpi_desc->nvdimm_bus) {
3125 acpi_nfit_init_dsms(acpi_desc);
3127 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3128 &acpi_desc->nd_desc);
3129 if (!acpi_desc->nvdimm_bus)
3132 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3137 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3141 /* register this acpi_desc for mce notifications */
3142 mutex_lock(&acpi_desc_lock);
3143 list_add_tail(&acpi_desc->list, &acpi_descs);
3144 mutex_unlock(&acpi_desc_lock);
3147 mutex_lock(&acpi_desc->init_mutex);
3149 INIT_LIST_HEAD(&prev.spas);
3150 INIT_LIST_HEAD(&prev.memdevs);
3151 INIT_LIST_HEAD(&prev.dcrs);
3152 INIT_LIST_HEAD(&prev.bdws);
3153 INIT_LIST_HEAD(&prev.idts);
3154 INIT_LIST_HEAD(&prev.flushes);
3156 list_cut_position(&prev.spas, &acpi_desc->spas,
3157 acpi_desc->spas.prev);
3158 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3159 acpi_desc->memdevs.prev);
3160 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3161 acpi_desc->dcrs.prev);
3162 list_cut_position(&prev.bdws, &acpi_desc->bdws,
3163 acpi_desc->bdws.prev);
3164 list_cut_position(&prev.idts, &acpi_desc->idts,
3165 acpi_desc->idts.prev);
3166 list_cut_position(&prev.flushes, &acpi_desc->flushes,
3167 acpi_desc->flushes.prev);
3170 while (!IS_ERR_OR_NULL(data))
3171 data = add_table(acpi_desc, &prev, data, end);
3174 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3179 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3183 rc = nfit_mem_init(acpi_desc);
3187 rc = acpi_nfit_register_dimms(acpi_desc);
3191 rc = acpi_nfit_register_regions(acpi_desc);
3194 mutex_unlock(&acpi_desc->init_mutex);
3197 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3199 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3201 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3202 struct device *dev = acpi_desc->dev;
3204 /* Bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3208 /* Bounce the init_mutex to complete initial registration */
3209 mutex_lock(&acpi_desc->init_mutex);
3210 mutex_unlock(&acpi_desc->init_mutex);
3215 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3216 struct nvdimm *nvdimm, unsigned int cmd)
3218 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3222 if (cmd != ND_CMD_ARS_START)
3226 * The kernel and userspace may race to initiate a scrub, but
3227 * the scrub thread is prepared to lose that initial race. It
3228 * just needs guarantees that any ars it initiates are not
3229 * interrupted by any intervening start reqeusts from userspace.
3231 if (work_busy(&acpi_desc->dwork.work))
3237 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, unsigned long flags)
3239 struct device *dev = acpi_desc->dev;
3240 int scheduled = 0, busy = 0;
3241 struct nfit_spa *nfit_spa;
3243 mutex_lock(&acpi_desc->init_mutex);
3244 if (acpi_desc->cancel) {
3245 mutex_unlock(&acpi_desc->init_mutex);
3249 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3250 int type = nfit_spa_type(nfit_spa->spa);
3252 if (type != NFIT_SPA_PM && type != NFIT_SPA_VOLATILE)
3254 if (test_bit(ARS_FAILED, &nfit_spa->ars_state))
3257 if (test_and_set_bit(ARS_REQ, &nfit_spa->ars_state))
3260 if (test_bit(ARS_SHORT, &flags))
3261 set_bit(ARS_SHORT, &nfit_spa->ars_state);
3266 sched_ars(acpi_desc);
3267 dev_dbg(dev, "ars_scan triggered\n");
3269 mutex_unlock(&acpi_desc->init_mutex);
3278 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3280 struct nvdimm_bus_descriptor *nd_desc;
3282 dev_set_drvdata(dev, acpi_desc);
3283 acpi_desc->dev = dev;
3284 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3285 nd_desc = &acpi_desc->nd_desc;
3286 nd_desc->provider_name = "ACPI.NFIT";
3287 nd_desc->module = THIS_MODULE;
3288 nd_desc->ndctl = acpi_nfit_ctl;
3289 nd_desc->flush_probe = acpi_nfit_flush_probe;
3290 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3291 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3293 INIT_LIST_HEAD(&acpi_desc->spas);
3294 INIT_LIST_HEAD(&acpi_desc->dcrs);
3295 INIT_LIST_HEAD(&acpi_desc->bdws);
3296 INIT_LIST_HEAD(&acpi_desc->idts);
3297 INIT_LIST_HEAD(&acpi_desc->flushes);
3298 INIT_LIST_HEAD(&acpi_desc->memdevs);
3299 INIT_LIST_HEAD(&acpi_desc->dimms);
3300 INIT_LIST_HEAD(&acpi_desc->list);
3301 mutex_init(&acpi_desc->init_mutex);
3302 acpi_desc->scrub_tmo = 1;
3303 INIT_DELAYED_WORK(&acpi_desc->dwork, acpi_nfit_scrub);
3305 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3307 static void acpi_nfit_put_table(void *table)
3309 acpi_put_table(table);
3312 void acpi_nfit_shutdown(void *data)
3314 struct acpi_nfit_desc *acpi_desc = data;
3315 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3318 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3321 mutex_lock(&acpi_desc_lock);
3322 list_del(&acpi_desc->list);
3323 mutex_unlock(&acpi_desc_lock);
3325 mutex_lock(&acpi_desc->init_mutex);
3326 acpi_desc->cancel = 1;
3327 cancel_delayed_work_sync(&acpi_desc->dwork);
3328 mutex_unlock(&acpi_desc->init_mutex);
3331 * Bounce the nvdimm bus lock to make sure any in-flight
3332 * acpi_nfit_ars_rescan() submissions have had a chance to
3333 * either submit or see ->cancel set.
3335 device_lock(bus_dev);
3336 device_unlock(bus_dev);
3338 flush_workqueue(nfit_wq);
3340 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3342 static int acpi_nfit_add(struct acpi_device *adev)
3344 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3345 struct acpi_nfit_desc *acpi_desc;
3346 struct device *dev = &adev->dev;
3347 struct acpi_table_header *tbl;
3348 acpi_status status = AE_OK;
3352 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3353 if (ACPI_FAILURE(status)) {
3354 /* This is ok, we could have an nvdimm hotplugged later */
3355 dev_dbg(dev, "failed to find NFIT at startup\n");
3359 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3364 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3367 acpi_nfit_desc_init(acpi_desc, &adev->dev);
3369 /* Save the acpi header for exporting the revision via sysfs */
3370 acpi_desc->acpi_header = *tbl;
3372 /* Evaluate _FIT and override with that if present */
3373 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3374 if (ACPI_SUCCESS(status) && buf.length > 0) {
3375 union acpi_object *obj = buf.pointer;
3377 if (obj->type == ACPI_TYPE_BUFFER)
3378 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3379 obj->buffer.length);
3381 dev_dbg(dev, "invalid type %d, ignoring _FIT\n",
3385 /* skip over the lead-in header table */
3386 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3387 + sizeof(struct acpi_table_nfit),
3388 sz - sizeof(struct acpi_table_nfit));
3392 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3395 static int acpi_nfit_remove(struct acpi_device *adev)
3397 /* see acpi_nfit_unregister */
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);
3449 unsigned long flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3452 acpi_nfit_ars_rescan(acpi_desc, flags);
3455 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3457 dev_dbg(dev, "event: 0x%x\n", event);
3460 case NFIT_NOTIFY_UPDATE:
3461 return acpi_nfit_update_notify(dev, handle);
3462 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3463 return acpi_nfit_uc_error_notify(dev, handle);
3468 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3470 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3472 device_lock(&adev->dev);
3473 __acpi_nfit_notify(&adev->dev, adev->handle, event);
3474 device_unlock(&adev->dev);
3477 static const struct acpi_device_id acpi_nfit_ids[] = {
3481 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3483 static struct acpi_driver acpi_nfit_driver = {
3484 .name = KBUILD_MODNAME,
3485 .ids = acpi_nfit_ids,
3487 .add = acpi_nfit_add,
3488 .remove = acpi_nfit_remove,
3489 .notify = acpi_nfit_notify,
3493 static __init int nfit_init(void)
3497 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3498 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3499 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3500 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3501 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3502 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3503 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3504 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3506 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3507 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3508 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3509 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3510 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3511 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3512 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3513 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3514 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3515 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3516 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3517 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3518 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3520 nfit_wq = create_singlethread_workqueue("nfit");
3524 nfit_mce_register();
3525 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3527 nfit_mce_unregister();
3528 destroy_workqueue(nfit_wq);
3535 static __exit void nfit_exit(void)
3537 nfit_mce_unregister();
3538 acpi_bus_unregister_driver(&acpi_nfit_driver);
3539 destroy_workqueue(nfit_wq);
3540 WARN_ON(!list_empty(&acpi_descs));
3543 module_init(nfit_init);
3544 module_exit(nfit_exit);
3545 MODULE_LICENSE("GPL v2");
3546 MODULE_AUTHOR("Intel Corporation");