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>
29 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
32 #include <linux/io-64-nonatomic-hi-lo.h>
34 static bool force_enable_dimms;
35 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
36 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
38 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
39 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
40 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
42 /* after three payloads of overflow, it's dead jim */
43 static unsigned int scrub_overflow_abort = 3;
44 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
45 MODULE_PARM_DESC(scrub_overflow_abort,
46 "Number of times we overflow ARS results before abort");
48 static bool disable_vendor_specific;
49 module_param(disable_vendor_specific, bool, S_IRUGO);
50 MODULE_PARM_DESC(disable_vendor_specific,
51 "Limit commands to the publicly specified set");
53 static unsigned long override_dsm_mask;
54 module_param(override_dsm_mask, ulong, S_IRUGO);
55 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
57 static int default_dsm_family = -1;
58 module_param(default_dsm_family, int, S_IRUGO);
59 MODULE_PARM_DESC(default_dsm_family,
60 "Try this DSM type first when identifying NVDIMM family");
62 LIST_HEAD(acpi_descs);
63 DEFINE_MUTEX(acpi_desc_lock);
65 static struct workqueue_struct *nfit_wq;
67 struct nfit_table_prev {
68 struct list_head spas;
69 struct list_head memdevs;
70 struct list_head dcrs;
71 struct list_head bdws;
72 struct list_head idts;
73 struct list_head flushes;
76 static guid_t nfit_uuid[NFIT_UUID_MAX];
78 const guid_t *to_nfit_uuid(enum nfit_uuids id)
80 return &nfit_uuid[id];
82 EXPORT_SYMBOL(to_nfit_uuid);
84 static struct acpi_nfit_desc *to_acpi_nfit_desc(
85 struct nvdimm_bus_descriptor *nd_desc)
87 return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
90 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
92 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
95 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
98 if (!nd_desc->provider_name
99 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
102 return to_acpi_device(acpi_desc->dev);
105 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
107 struct nd_cmd_clear_error *clear_err;
108 struct nd_cmd_ars_status *ars_status;
113 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
120 /* No supported scan types for this range */
121 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
122 if ((status >> 16 & flags) == 0)
125 case ND_CMD_ARS_START:
126 /* ARS is in progress */
127 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
134 case ND_CMD_ARS_STATUS:
139 /* Check extended status (Upper two bytes) */
140 if (status == NFIT_ARS_STATUS_DONE)
143 /* ARS is in progress */
144 if (status == NFIT_ARS_STATUS_BUSY)
147 /* No ARS performed for the current boot */
148 if (status == NFIT_ARS_STATUS_NONE)
152 * ARS interrupted, either we overflowed or some other
153 * agent wants the scan to stop. If we didn't overflow
154 * then just continue with the returned results.
156 if (status == NFIT_ARS_STATUS_INTR) {
157 if (ars_status->out_length >= 40 && (ars_status->flags
158 & NFIT_ARS_F_OVERFLOW))
167 case ND_CMD_CLEAR_ERROR:
171 if (!clear_err->cleared)
173 if (clear_err->length > clear_err->cleared)
174 return clear_err->cleared;
180 /* all other non-zero status results in an error */
186 #define ACPI_LABELS_LOCKED 3
188 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
191 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
194 case ND_CMD_GET_CONFIG_SIZE:
196 * In the _LSI, _LSR, _LSW case the locked status is
197 * communicated via the read/write commands
199 if (nfit_mem->has_lsr)
202 if (status >> 16 & ND_CONFIG_LOCKED)
205 case ND_CMD_GET_CONFIG_DATA:
206 if (nfit_mem->has_lsr && status == ACPI_LABELS_LOCKED)
209 case ND_CMD_SET_CONFIG_DATA:
210 if (nfit_mem->has_lsw && status == ACPI_LABELS_LOCKED)
217 /* all other non-zero status results in an error */
223 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
227 return xlat_bus_status(buf, cmd, status);
228 return xlat_nvdimm_status(nvdimm, buf, cmd, status);
231 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
232 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
237 union acpi_object *buf = NULL;
239 if (pkg->type != ACPI_TYPE_PACKAGE) {
240 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
245 for (i = 0; i < pkg->package.count; i++) {
246 union acpi_object *obj = &pkg->package.elements[i];
248 if (obj->type == ACPI_TYPE_INTEGER)
250 else if (obj->type == ACPI_TYPE_BUFFER)
251 size += obj->buffer.length;
253 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
259 buf = ACPI_ALLOCATE(sizeof(*buf) + size);
264 buf->type = ACPI_TYPE_BUFFER;
265 buf->buffer.length = size;
266 buf->buffer.pointer = dst;
267 for (i = 0; i < pkg->package.count; i++) {
268 union acpi_object *obj = &pkg->package.elements[i];
270 if (obj->type == ACPI_TYPE_INTEGER) {
271 memcpy(dst, &obj->integer.value, 4);
273 } else if (obj->type == ACPI_TYPE_BUFFER) {
274 memcpy(dst, obj->buffer.pointer, obj->buffer.length);
275 dst += obj->buffer.length;
283 static union acpi_object *int_to_buf(union acpi_object *integer)
285 union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
291 if (integer->type != ACPI_TYPE_INTEGER) {
292 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
298 buf->type = ACPI_TYPE_BUFFER;
299 buf->buffer.length = 4;
300 buf->buffer.pointer = dst;
301 memcpy(dst, &integer->integer.value, 4);
307 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
311 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
312 struct acpi_object_list input = {
314 .pointer = (union acpi_object []) {
316 .integer.type = ACPI_TYPE_INTEGER,
317 .integer.value = offset,
320 .integer.type = ACPI_TYPE_INTEGER,
321 .integer.value = len,
324 .buffer.type = ACPI_TYPE_BUFFER,
325 .buffer.pointer = data,
326 .buffer.length = len,
331 rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
332 if (ACPI_FAILURE(rc))
334 return int_to_buf(buf.pointer);
337 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
341 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
342 struct acpi_object_list input = {
344 .pointer = (union acpi_object []) {
346 .integer.type = ACPI_TYPE_INTEGER,
347 .integer.value = offset,
350 .integer.type = ACPI_TYPE_INTEGER,
351 .integer.value = len,
356 rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
357 if (ACPI_FAILURE(rc))
359 return pkg_to_buf(buf.pointer);
362 static union acpi_object *acpi_label_info(acpi_handle handle)
365 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
367 rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
368 if (ACPI_FAILURE(rc))
370 return pkg_to_buf(buf.pointer);
373 static u8 nfit_dsm_revid(unsigned family, unsigned func)
375 static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
376 [NVDIMM_FAMILY_INTEL] = {
377 [NVDIMM_INTEL_GET_MODES] = 2,
378 [NVDIMM_INTEL_GET_FWINFO] = 2,
379 [NVDIMM_INTEL_START_FWUPDATE] = 2,
380 [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
381 [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
382 [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
383 [NVDIMM_INTEL_SET_THRESHOLD] = 2,
384 [NVDIMM_INTEL_INJECT_ERROR] = 2,
389 if (family > NVDIMM_FAMILY_MAX)
393 id = revid_table[family][func];
395 return 1; /* default */
399 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
400 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
402 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
403 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
404 union acpi_object in_obj, in_buf, *out_obj;
405 const struct nd_cmd_desc *desc = NULL;
406 struct device *dev = acpi_desc->dev;
407 struct nd_cmd_pkg *call_pkg = NULL;
408 const char *cmd_name, *dimm_name;
409 unsigned long cmd_mask, dsm_mask;
410 u32 offset, fw_status = 0;
417 if (cmd == ND_CMD_CALL) {
419 func = call_pkg->nd_command;
421 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
422 if (call_pkg->nd_reserved2[i])
427 struct acpi_device *adev = nfit_mem->adev;
431 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
434 dimm_name = nvdimm_name(nvdimm);
435 cmd_name = nvdimm_cmd_name(cmd);
436 cmd_mask = nvdimm_cmd_mask(nvdimm);
437 dsm_mask = nfit_mem->dsm_mask;
438 desc = nd_cmd_dimm_desc(cmd);
439 guid = to_nfit_uuid(nfit_mem->family);
440 handle = adev->handle;
442 struct acpi_device *adev = to_acpi_dev(acpi_desc);
444 cmd_name = nvdimm_bus_cmd_name(cmd);
445 cmd_mask = nd_desc->cmd_mask;
447 if (cmd == ND_CMD_CALL)
448 dsm_mask = nd_desc->bus_dsm_mask;
449 desc = nd_cmd_bus_desc(cmd);
450 guid = to_nfit_uuid(NFIT_DEV_BUS);
451 handle = adev->handle;
455 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
458 if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
461 in_obj.type = ACPI_TYPE_PACKAGE;
462 in_obj.package.count = 1;
463 in_obj.package.elements = &in_buf;
464 in_buf.type = ACPI_TYPE_BUFFER;
465 in_buf.buffer.pointer = buf;
466 in_buf.buffer.length = 0;
468 /* libnvdimm has already validated the input envelope */
469 for (i = 0; i < desc->in_num; i++)
470 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
474 /* skip over package wrapper */
475 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
476 in_buf.buffer.length = call_pkg->nd_size_in;
479 dev_dbg(dev, "%s cmd: %d: func: %d input length: %d\n",
480 dimm_name, cmd, func, in_buf.buffer.length);
481 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
482 in_buf.buffer.pointer,
483 min_t(u32, 256, in_buf.buffer.length), true);
485 /* call the BIOS, prefer the named methods over _DSM if available */
486 if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE && nfit_mem->has_lsr)
487 out_obj = acpi_label_info(handle);
488 else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && nfit_mem->has_lsr) {
489 struct nd_cmd_get_config_data_hdr *p = buf;
491 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
492 } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
493 && nfit_mem->has_lsw) {
494 struct nd_cmd_set_config_hdr *p = buf;
496 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
502 revid = nfit_dsm_revid(nfit_mem->family, func);
505 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
509 dev_dbg(dev, "%s _DSM failed cmd: %s\n", dimm_name, cmd_name);
514 call_pkg->nd_fw_size = out_obj->buffer.length;
515 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
516 out_obj->buffer.pointer,
517 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
521 * Need to support FW function w/o known size in advance.
522 * Caller can determine required size based upon nd_fw_size.
523 * If we return an error (like elsewhere) then caller wouldn't
524 * be able to rely upon data returned to make calculation.
529 if (out_obj->package.type != ACPI_TYPE_BUFFER) {
530 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
531 dimm_name, cmd_name, out_obj->type);
536 dev_dbg(dev, "%s cmd: %s output length: %d\n", dimm_name,
537 cmd_name, out_obj->buffer.length);
538 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
539 out_obj->buffer.pointer,
540 min_t(u32, 128, out_obj->buffer.length), true);
542 for (i = 0, offset = 0; i < desc->out_num; i++) {
543 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
544 (u32 *) out_obj->buffer.pointer,
545 out_obj->buffer.length - offset);
547 if (offset + out_size > out_obj->buffer.length) {
548 dev_dbg(dev, "%s output object underflow cmd: %s field: %d\n",
549 dimm_name, cmd_name, i);
553 if (in_buf.buffer.length + offset + out_size > buf_len) {
554 dev_dbg(dev, "%s output overrun cmd: %s field: %d\n",
555 dimm_name, cmd_name, i);
559 memcpy(buf + in_buf.buffer.length + offset,
560 out_obj->buffer.pointer + offset, out_size);
565 * Set fw_status for all the commands with a known format to be
566 * later interpreted by xlat_status().
568 if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
569 && cmd <= ND_CMD_CLEAR_ERROR)
570 || (nvdimm && cmd >= ND_CMD_SMART
571 && cmd <= ND_CMD_VENDOR)))
572 fw_status = *(u32 *) out_obj->buffer.pointer;
574 if (offset + in_buf.buffer.length < buf_len) {
577 * status valid, return the number of bytes left
578 * unfilled in the output buffer
580 rc = buf_len - offset - in_buf.buffer.length;
582 *cmd_rc = xlat_status(nvdimm, buf, cmd,
585 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
586 __func__, dimm_name, cmd_name, buf_len,
593 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
601 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
603 static const char *spa_type_name(u16 type)
605 static const char *to_name[] = {
606 [NFIT_SPA_VOLATILE] = "volatile",
607 [NFIT_SPA_PM] = "pmem",
608 [NFIT_SPA_DCR] = "dimm-control-region",
609 [NFIT_SPA_BDW] = "block-data-window",
610 [NFIT_SPA_VDISK] = "volatile-disk",
611 [NFIT_SPA_VCD] = "volatile-cd",
612 [NFIT_SPA_PDISK] = "persistent-disk",
613 [NFIT_SPA_PCD] = "persistent-cd",
617 if (type > NFIT_SPA_PCD)
620 return to_name[type];
623 int nfit_spa_type(struct acpi_nfit_system_address *spa)
627 for (i = 0; i < NFIT_UUID_MAX; i++)
628 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
633 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
634 struct nfit_table_prev *prev,
635 struct acpi_nfit_system_address *spa)
637 struct device *dev = acpi_desc->dev;
638 struct nfit_spa *nfit_spa;
640 if (spa->header.length != sizeof(*spa))
643 list_for_each_entry(nfit_spa, &prev->spas, list) {
644 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
645 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
650 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
654 INIT_LIST_HEAD(&nfit_spa->list);
655 memcpy(nfit_spa->spa, spa, sizeof(*spa));
656 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
657 dev_dbg(dev, "spa index: %d type: %s\n",
659 spa_type_name(nfit_spa_type(spa)));
663 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
664 struct nfit_table_prev *prev,
665 struct acpi_nfit_memory_map *memdev)
667 struct device *dev = acpi_desc->dev;
668 struct nfit_memdev *nfit_memdev;
670 if (memdev->header.length != sizeof(*memdev))
673 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
674 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
675 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
679 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
683 INIT_LIST_HEAD(&nfit_memdev->list);
684 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
685 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
686 dev_dbg(dev, "memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
687 memdev->device_handle, memdev->range_index,
688 memdev->region_index, memdev->flags);
693 * An implementation may provide a truncated control region if no block windows
696 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
698 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
703 return offsetof(struct acpi_nfit_control_region, window_size);
706 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
707 struct nfit_table_prev *prev,
708 struct acpi_nfit_control_region *dcr)
710 struct device *dev = acpi_desc->dev;
711 struct nfit_dcr *nfit_dcr;
713 if (!sizeof_dcr(dcr))
716 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
717 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
718 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
722 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
726 INIT_LIST_HEAD(&nfit_dcr->list);
727 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
728 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
729 dev_dbg(dev, "dcr index: %d windows: %d\n",
730 dcr->region_index, dcr->windows);
734 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
735 struct nfit_table_prev *prev,
736 struct acpi_nfit_data_region *bdw)
738 struct device *dev = acpi_desc->dev;
739 struct nfit_bdw *nfit_bdw;
741 if (bdw->header.length != sizeof(*bdw))
743 list_for_each_entry(nfit_bdw, &prev->bdws, list)
744 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
745 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
749 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
753 INIT_LIST_HEAD(&nfit_bdw->list);
754 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
755 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
756 dev_dbg(dev, "bdw dcr: %d windows: %d\n",
757 bdw->region_index, bdw->windows);
761 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
763 if (idt->header.length < sizeof(*idt))
765 return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
768 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
769 struct nfit_table_prev *prev,
770 struct acpi_nfit_interleave *idt)
772 struct device *dev = acpi_desc->dev;
773 struct nfit_idt *nfit_idt;
775 if (!sizeof_idt(idt))
778 list_for_each_entry(nfit_idt, &prev->idts, list) {
779 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
782 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
783 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
788 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
792 INIT_LIST_HEAD(&nfit_idt->list);
793 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
794 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
795 dev_dbg(dev, "idt index: %d num_lines: %d\n",
796 idt->interleave_index, idt->line_count);
800 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
802 if (flush->header.length < sizeof(*flush))
804 return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
807 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
808 struct nfit_table_prev *prev,
809 struct acpi_nfit_flush_address *flush)
811 struct device *dev = acpi_desc->dev;
812 struct nfit_flush *nfit_flush;
814 if (!sizeof_flush(flush))
817 list_for_each_entry(nfit_flush, &prev->flushes, list) {
818 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
821 if (memcmp(nfit_flush->flush, flush,
822 sizeof_flush(flush)) == 0) {
823 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
828 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
829 + sizeof_flush(flush), GFP_KERNEL);
832 INIT_LIST_HEAD(&nfit_flush->list);
833 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
834 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
835 dev_dbg(dev, "nfit_flush handle: %d hint_count: %d\n",
836 flush->device_handle, flush->hint_count);
840 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
841 struct acpi_nfit_capabilities *pcap)
843 struct device *dev = acpi_desc->dev;
846 mask = (1 << (pcap->highest_capability + 1)) - 1;
847 acpi_desc->platform_cap = pcap->capabilities & mask;
848 dev_dbg(dev, "cap: %#x\n", acpi_desc->platform_cap);
852 static void *add_table(struct acpi_nfit_desc *acpi_desc,
853 struct nfit_table_prev *prev, void *table, const void *end)
855 struct device *dev = acpi_desc->dev;
856 struct acpi_nfit_header *hdr;
857 void *err = ERR_PTR(-ENOMEM);
864 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
870 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
871 if (!add_spa(acpi_desc, prev, table))
874 case ACPI_NFIT_TYPE_MEMORY_MAP:
875 if (!add_memdev(acpi_desc, prev, table))
878 case ACPI_NFIT_TYPE_CONTROL_REGION:
879 if (!add_dcr(acpi_desc, prev, table))
882 case ACPI_NFIT_TYPE_DATA_REGION:
883 if (!add_bdw(acpi_desc, prev, table))
886 case ACPI_NFIT_TYPE_INTERLEAVE:
887 if (!add_idt(acpi_desc, prev, table))
890 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
891 if (!add_flush(acpi_desc, prev, table))
894 case ACPI_NFIT_TYPE_SMBIOS:
895 dev_dbg(dev, "smbios\n");
897 case ACPI_NFIT_TYPE_CAPABILITIES:
898 if (!add_platform_cap(acpi_desc, table))
902 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
906 return table + hdr->length;
909 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
910 struct nfit_mem *nfit_mem)
912 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
913 u16 dcr = nfit_mem->dcr->region_index;
914 struct nfit_spa *nfit_spa;
916 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
917 u16 range_index = nfit_spa->spa->range_index;
918 int type = nfit_spa_type(nfit_spa->spa);
919 struct nfit_memdev *nfit_memdev;
921 if (type != NFIT_SPA_BDW)
924 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
925 if (nfit_memdev->memdev->range_index != range_index)
927 if (nfit_memdev->memdev->device_handle != device_handle)
929 if (nfit_memdev->memdev->region_index != dcr)
932 nfit_mem->spa_bdw = nfit_spa->spa;
937 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
938 nfit_mem->spa_dcr->range_index);
939 nfit_mem->bdw = NULL;
942 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
943 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
945 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
946 struct nfit_memdev *nfit_memdev;
947 struct nfit_bdw *nfit_bdw;
948 struct nfit_idt *nfit_idt;
949 u16 idt_idx, range_index;
951 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
952 if (nfit_bdw->bdw->region_index != dcr)
954 nfit_mem->bdw = nfit_bdw->bdw;
961 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
963 if (!nfit_mem->spa_bdw)
966 range_index = nfit_mem->spa_bdw->range_index;
967 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
968 if (nfit_memdev->memdev->range_index != range_index ||
969 nfit_memdev->memdev->region_index != dcr)
971 nfit_mem->memdev_bdw = nfit_memdev->memdev;
972 idt_idx = nfit_memdev->memdev->interleave_index;
973 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
974 if (nfit_idt->idt->interleave_index != idt_idx)
976 nfit_mem->idt_bdw = nfit_idt->idt;
983 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
984 struct acpi_nfit_system_address *spa)
986 struct nfit_mem *nfit_mem, *found;
987 struct nfit_memdev *nfit_memdev;
988 int type = spa ? nfit_spa_type(spa) : 0;
1000 * This loop runs in two modes, when a dimm is mapped the loop
1001 * adds memdev associations to an existing dimm, or creates a
1002 * dimm. In the unmapped dimm case this loop sweeps for memdev
1003 * instances with an invalid / zero range_index and adds those
1004 * dimms without spa associations.
1006 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1007 struct nfit_flush *nfit_flush;
1008 struct nfit_dcr *nfit_dcr;
1012 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1014 if (!spa && nfit_memdev->memdev->range_index)
1017 dcr = nfit_memdev->memdev->region_index;
1018 device_handle = nfit_memdev->memdev->device_handle;
1019 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1020 if (__to_nfit_memdev(nfit_mem)->device_handle
1029 nfit_mem = devm_kzalloc(acpi_desc->dev,
1030 sizeof(*nfit_mem), GFP_KERNEL);
1033 INIT_LIST_HEAD(&nfit_mem->list);
1034 nfit_mem->acpi_desc = acpi_desc;
1035 list_add(&nfit_mem->list, &acpi_desc->dimms);
1038 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1039 if (nfit_dcr->dcr->region_index != dcr)
1042 * Record the control region for the dimm. For
1043 * the ACPI 6.1 case, where there are separate
1044 * control regions for the pmem vs blk
1045 * interfaces, be sure to record the extended
1049 nfit_mem->dcr = nfit_dcr->dcr;
1050 else if (nfit_mem->dcr->windows == 0
1051 && nfit_dcr->dcr->windows)
1052 nfit_mem->dcr = nfit_dcr->dcr;
1056 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1057 struct acpi_nfit_flush_address *flush;
1060 if (nfit_flush->flush->device_handle != device_handle)
1062 nfit_mem->nfit_flush = nfit_flush;
1063 flush = nfit_flush->flush;
1064 nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
1066 * sizeof(struct resource), GFP_KERNEL);
1067 if (!nfit_mem->flush_wpq)
1069 for (i = 0; i < flush->hint_count; i++) {
1070 struct resource *res = &nfit_mem->flush_wpq[i];
1072 res->start = flush->hint_address[i];
1073 res->end = res->start + 8 - 1;
1078 if (dcr && !nfit_mem->dcr) {
1079 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1080 spa->range_index, dcr);
1084 if (type == NFIT_SPA_DCR) {
1085 struct nfit_idt *nfit_idt;
1088 /* multiple dimms may share a SPA when interleaved */
1089 nfit_mem->spa_dcr = spa;
1090 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1091 idt_idx = nfit_memdev->memdev->interleave_index;
1092 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1093 if (nfit_idt->idt->interleave_index != idt_idx)
1095 nfit_mem->idt_dcr = nfit_idt->idt;
1098 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1099 } else if (type == NFIT_SPA_PM) {
1101 * A single dimm may belong to multiple SPA-PM
1102 * ranges, record at least one in addition to
1103 * any SPA-DCR range.
1105 nfit_mem->memdev_pmem = nfit_memdev->memdev;
1107 nfit_mem->memdev_dcr = nfit_memdev->memdev;
1113 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1115 struct nfit_mem *a = container_of(_a, typeof(*a), list);
1116 struct nfit_mem *b = container_of(_b, typeof(*b), list);
1117 u32 handleA, handleB;
1119 handleA = __to_nfit_memdev(a)->device_handle;
1120 handleB = __to_nfit_memdev(b)->device_handle;
1121 if (handleA < handleB)
1123 else if (handleA > handleB)
1128 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1130 struct nfit_spa *nfit_spa;
1135 * For each SPA-DCR or SPA-PMEM address range find its
1136 * corresponding MEMDEV(s). From each MEMDEV find the
1137 * corresponding DCR. Then, if we're operating on a SPA-DCR,
1138 * try to find a SPA-BDW and a corresponding BDW that references
1139 * the DCR. Throw it all into an nfit_mem object. Note, that
1140 * BDWs are optional.
1142 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1143 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1149 * If a DIMM has failed to be mapped into SPA there will be no
1150 * SPA entries above. Find and register all the unmapped DIMMs
1151 * for reporting and recovery purposes.
1153 rc = __nfit_mem_init(acpi_desc, NULL);
1157 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1162 static ssize_t bus_dsm_mask_show(struct device *dev,
1163 struct device_attribute *attr, char *buf)
1165 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1166 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1168 return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1170 static struct device_attribute dev_attr_bus_dsm_mask =
1171 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1173 static ssize_t revision_show(struct device *dev,
1174 struct device_attribute *attr, char *buf)
1176 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1177 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1178 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1180 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1182 static DEVICE_ATTR_RO(revision);
1184 static ssize_t hw_error_scrub_show(struct device *dev,
1185 struct device_attribute *attr, char *buf)
1187 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1188 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1189 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1191 return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1195 * The 'hw_error_scrub' attribute can have the following values written to it:
1196 * '0': Switch to the default mode where an exception will only insert
1197 * the address of the memory error into the poison and badblocks lists.
1198 * '1': Enable a full scrub to happen if an exception for a memory error is
1201 static ssize_t hw_error_scrub_store(struct device *dev,
1202 struct device_attribute *attr, const char *buf, size_t size)
1204 struct nvdimm_bus_descriptor *nd_desc;
1208 rc = kstrtol(buf, 0, &val);
1213 nd_desc = dev_get_drvdata(dev);
1215 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1218 case HW_ERROR_SCRUB_ON:
1219 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1221 case HW_ERROR_SCRUB_OFF:
1222 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1234 static DEVICE_ATTR_RW(hw_error_scrub);
1237 * This shows the number of full Address Range Scrubs that have been
1238 * completed since driver load time. Userspace can wait on this using
1239 * select/poll etc. A '+' at the end indicates an ARS is in progress
1241 static ssize_t scrub_show(struct device *dev,
1242 struct device_attribute *attr, char *buf)
1244 struct nvdimm_bus_descriptor *nd_desc;
1245 ssize_t rc = -ENXIO;
1248 nd_desc = dev_get_drvdata(dev);
1250 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1252 mutex_lock(&acpi_desc->init_mutex);
1253 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1254 work_busy(&acpi_desc->work)
1255 && !acpi_desc->cancel ? "+\n" : "\n");
1256 mutex_unlock(&acpi_desc->init_mutex);
1262 static ssize_t scrub_store(struct device *dev,
1263 struct device_attribute *attr, const char *buf, size_t size)
1265 struct nvdimm_bus_descriptor *nd_desc;
1269 rc = kstrtol(buf, 0, &val);
1276 nd_desc = dev_get_drvdata(dev);
1278 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1280 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1287 static DEVICE_ATTR_RW(scrub);
1289 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1291 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1292 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1293 | 1 << ND_CMD_ARS_STATUS;
1295 return (nd_desc->cmd_mask & mask) == mask;
1298 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1300 struct device *dev = container_of(kobj, struct device, kobj);
1301 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1303 if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1308 static struct attribute *acpi_nfit_attributes[] = {
1309 &dev_attr_revision.attr,
1310 &dev_attr_scrub.attr,
1311 &dev_attr_hw_error_scrub.attr,
1312 &dev_attr_bus_dsm_mask.attr,
1316 static const struct attribute_group acpi_nfit_attribute_group = {
1318 .attrs = acpi_nfit_attributes,
1319 .is_visible = nfit_visible,
1322 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1323 &nvdimm_bus_attribute_group,
1324 &acpi_nfit_attribute_group,
1328 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1330 struct nvdimm *nvdimm = to_nvdimm(dev);
1331 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1333 return __to_nfit_memdev(nfit_mem);
1336 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1338 struct nvdimm *nvdimm = to_nvdimm(dev);
1339 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1341 return nfit_mem->dcr;
1344 static ssize_t handle_show(struct device *dev,
1345 struct device_attribute *attr, char *buf)
1347 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1349 return sprintf(buf, "%#x\n", memdev->device_handle);
1351 static DEVICE_ATTR_RO(handle);
1353 static ssize_t phys_id_show(struct device *dev,
1354 struct device_attribute *attr, char *buf)
1356 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1358 return sprintf(buf, "%#x\n", memdev->physical_id);
1360 static DEVICE_ATTR_RO(phys_id);
1362 static ssize_t vendor_show(struct device *dev,
1363 struct device_attribute *attr, char *buf)
1365 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1367 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1369 static DEVICE_ATTR_RO(vendor);
1371 static ssize_t rev_id_show(struct device *dev,
1372 struct device_attribute *attr, char *buf)
1374 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1376 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1378 static DEVICE_ATTR_RO(rev_id);
1380 static ssize_t device_show(struct device *dev,
1381 struct device_attribute *attr, char *buf)
1383 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1385 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1387 static DEVICE_ATTR_RO(device);
1389 static ssize_t subsystem_vendor_show(struct device *dev,
1390 struct device_attribute *attr, char *buf)
1392 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1394 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1396 static DEVICE_ATTR_RO(subsystem_vendor);
1398 static ssize_t subsystem_rev_id_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",
1404 be16_to_cpu(dcr->subsystem_revision_id));
1406 static DEVICE_ATTR_RO(subsystem_rev_id);
1408 static ssize_t subsystem_device_show(struct device *dev,
1409 struct device_attribute *attr, char *buf)
1411 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1413 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1415 static DEVICE_ATTR_RO(subsystem_device);
1417 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1419 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1422 if (nfit_mem->memdev_pmem)
1424 if (nfit_mem->memdev_bdw)
1429 static ssize_t format_show(struct device *dev,
1430 struct device_attribute *attr, char *buf)
1432 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1434 return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1436 static DEVICE_ATTR_RO(format);
1438 static ssize_t format1_show(struct device *dev,
1439 struct device_attribute *attr, char *buf)
1442 ssize_t rc = -ENXIO;
1443 struct nfit_mem *nfit_mem;
1444 struct nfit_memdev *nfit_memdev;
1445 struct acpi_nfit_desc *acpi_desc;
1446 struct nvdimm *nvdimm = to_nvdimm(dev);
1447 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1449 nfit_mem = nvdimm_provider_data(nvdimm);
1450 acpi_desc = nfit_mem->acpi_desc;
1451 handle = to_nfit_memdev(dev)->device_handle;
1453 /* assumes DIMMs have at most 2 published interface codes */
1454 mutex_lock(&acpi_desc->init_mutex);
1455 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1456 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1457 struct nfit_dcr *nfit_dcr;
1459 if (memdev->device_handle != handle)
1462 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1463 if (nfit_dcr->dcr->region_index != memdev->region_index)
1465 if (nfit_dcr->dcr->code == dcr->code)
1467 rc = sprintf(buf, "0x%04x\n",
1468 le16_to_cpu(nfit_dcr->dcr->code));
1474 mutex_unlock(&acpi_desc->init_mutex);
1477 static DEVICE_ATTR_RO(format1);
1479 static ssize_t formats_show(struct device *dev,
1480 struct device_attribute *attr, char *buf)
1482 struct nvdimm *nvdimm = to_nvdimm(dev);
1484 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1486 static DEVICE_ATTR_RO(formats);
1488 static ssize_t serial_show(struct device *dev,
1489 struct device_attribute *attr, char *buf)
1491 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1493 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1495 static DEVICE_ATTR_RO(serial);
1497 static ssize_t family_show(struct device *dev,
1498 struct device_attribute *attr, char *buf)
1500 struct nvdimm *nvdimm = to_nvdimm(dev);
1501 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1503 if (nfit_mem->family < 0)
1505 return sprintf(buf, "%d\n", nfit_mem->family);
1507 static DEVICE_ATTR_RO(family);
1509 static ssize_t dsm_mask_show(struct device *dev,
1510 struct device_attribute *attr, char *buf)
1512 struct nvdimm *nvdimm = to_nvdimm(dev);
1513 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1515 if (nfit_mem->family < 0)
1517 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1519 static DEVICE_ATTR_RO(dsm_mask);
1521 static ssize_t flags_show(struct device *dev,
1522 struct device_attribute *attr, char *buf)
1524 u16 flags = to_nfit_memdev(dev)->flags;
1526 return sprintf(buf, "%s%s%s%s%s%s%s\n",
1527 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1528 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1529 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1530 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1531 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1532 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1533 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1535 static DEVICE_ATTR_RO(flags);
1537 static ssize_t id_show(struct device *dev,
1538 struct device_attribute *attr, char *buf)
1540 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1542 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1543 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1544 be16_to_cpu(dcr->vendor_id),
1545 dcr->manufacturing_location,
1546 be16_to_cpu(dcr->manufacturing_date),
1547 be32_to_cpu(dcr->serial_number));
1549 return sprintf(buf, "%04x-%08x\n",
1550 be16_to_cpu(dcr->vendor_id),
1551 be32_to_cpu(dcr->serial_number));
1553 static DEVICE_ATTR_RO(id);
1555 static struct attribute *acpi_nfit_dimm_attributes[] = {
1556 &dev_attr_handle.attr,
1557 &dev_attr_phys_id.attr,
1558 &dev_attr_vendor.attr,
1559 &dev_attr_device.attr,
1560 &dev_attr_rev_id.attr,
1561 &dev_attr_subsystem_vendor.attr,
1562 &dev_attr_subsystem_device.attr,
1563 &dev_attr_subsystem_rev_id.attr,
1564 &dev_attr_format.attr,
1565 &dev_attr_formats.attr,
1566 &dev_attr_format1.attr,
1567 &dev_attr_serial.attr,
1568 &dev_attr_flags.attr,
1570 &dev_attr_family.attr,
1571 &dev_attr_dsm_mask.attr,
1575 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1576 struct attribute *a, int n)
1578 struct device *dev = container_of(kobj, struct device, kobj);
1579 struct nvdimm *nvdimm = to_nvdimm(dev);
1581 if (!to_nfit_dcr(dev)) {
1582 /* Without a dcr only the memdev attributes can be surfaced */
1583 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1584 || a == &dev_attr_flags.attr
1585 || a == &dev_attr_family.attr
1586 || a == &dev_attr_dsm_mask.attr)
1591 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1596 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1598 .attrs = acpi_nfit_dimm_attributes,
1599 .is_visible = acpi_nfit_dimm_attr_visible,
1602 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1603 &nvdimm_attribute_group,
1604 &nd_device_attribute_group,
1605 &acpi_nfit_dimm_attribute_group,
1609 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1612 struct nfit_mem *nfit_mem;
1614 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1615 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1616 return nfit_mem->nvdimm;
1621 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1623 struct nfit_mem *nfit_mem;
1624 struct acpi_nfit_desc *acpi_desc;
1626 dev_dbg(dev->parent, "%s: event: %d\n", dev_name(dev),
1629 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1630 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1635 acpi_desc = dev_get_drvdata(dev->parent);
1640 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1643 nfit_mem = dev_get_drvdata(dev);
1644 if (nfit_mem && nfit_mem->flags_attr)
1645 sysfs_notify_dirent(nfit_mem->flags_attr);
1647 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1649 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1651 struct acpi_device *adev = data;
1652 struct device *dev = &adev->dev;
1654 device_lock(dev->parent);
1655 __acpi_nvdimm_notify(dev, event);
1656 device_unlock(dev->parent);
1659 static bool acpi_nvdimm_has_method(struct acpi_device *adev, char *method)
1664 status = acpi_get_handle(adev->handle, method, &handle);
1666 if (ACPI_SUCCESS(status))
1671 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1672 struct nfit_mem *nfit_mem, u32 device_handle)
1674 struct acpi_device *adev, *adev_dimm;
1675 struct device *dev = acpi_desc->dev;
1676 unsigned long dsm_mask;
1681 /* nfit test assumes 1:1 relationship between commands and dsms */
1682 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1683 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1684 adev = to_acpi_dev(acpi_desc);
1688 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1689 nfit_mem->adev = adev_dimm;
1691 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1693 return force_enable_dimms ? 0 : -ENODEV;
1696 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1697 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1698 dev_err(dev, "%s: notification registration failed\n",
1699 dev_name(&adev_dimm->dev));
1703 * Record nfit_mem for the notification path to track back to
1704 * the nfit sysfs attributes for this dimm device object.
1706 dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1709 * Until standardization materializes we need to consider 4
1710 * different command sets. Note, that checking for function0 (bit0)
1711 * tells us if any commands are reachable through this GUID.
1713 for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1714 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1715 if (family < 0 || i == default_dsm_family)
1718 /* limit the supported commands to those that are publicly documented */
1719 nfit_mem->family = family;
1720 if (override_dsm_mask && !disable_vendor_specific)
1721 dsm_mask = override_dsm_mask;
1722 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1723 dsm_mask = NVDIMM_INTEL_CMDMASK;
1724 if (disable_vendor_specific)
1725 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1726 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1727 dsm_mask = 0x1c3c76;
1728 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1730 if (disable_vendor_specific)
1731 dsm_mask &= ~(1 << 8);
1732 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1733 dsm_mask = 0xffffffff;
1735 dev_dbg(dev, "unknown dimm command family\n");
1736 nfit_mem->family = -1;
1737 /* DSMs are optional, continue loading the driver... */
1741 guid = to_nfit_uuid(nfit_mem->family);
1742 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1743 if (acpi_check_dsm(adev_dimm->handle, guid,
1744 nfit_dsm_revid(nfit_mem->family, i),
1746 set_bit(i, &nfit_mem->dsm_mask);
1748 if (acpi_nvdimm_has_method(adev_dimm, "_LSI")
1749 && acpi_nvdimm_has_method(adev_dimm, "_LSR")) {
1750 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1751 nfit_mem->has_lsr = true;
1754 if (nfit_mem->has_lsr && acpi_nvdimm_has_method(adev_dimm, "_LSW")) {
1755 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1756 nfit_mem->has_lsw = true;
1762 static void shutdown_dimm_notify(void *data)
1764 struct acpi_nfit_desc *acpi_desc = data;
1765 struct nfit_mem *nfit_mem;
1767 mutex_lock(&acpi_desc->init_mutex);
1769 * Clear out the nfit_mem->flags_attr and shut down dimm event
1772 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1773 struct acpi_device *adev_dimm = nfit_mem->adev;
1775 if (nfit_mem->flags_attr) {
1776 sysfs_put(nfit_mem->flags_attr);
1777 nfit_mem->flags_attr = NULL;
1780 acpi_remove_notify_handler(adev_dimm->handle,
1781 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1782 dev_set_drvdata(&adev_dimm->dev, NULL);
1785 mutex_unlock(&acpi_desc->init_mutex);
1788 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1790 struct nfit_mem *nfit_mem;
1791 int dimm_count = 0, rc;
1792 struct nvdimm *nvdimm;
1794 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1795 struct acpi_nfit_flush_address *flush;
1796 unsigned long flags = 0, cmd_mask;
1797 struct nfit_memdev *nfit_memdev;
1801 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1802 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1808 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1809 set_bit(NDD_ALIASING, &flags);
1811 /* collate flags across all memdevs for this dimm */
1812 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1813 struct acpi_nfit_memory_map *dimm_memdev;
1815 dimm_memdev = __to_nfit_memdev(nfit_mem);
1816 if (dimm_memdev->device_handle
1817 != nfit_memdev->memdev->device_handle)
1819 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1822 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1823 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1824 set_bit(NDD_UNARMED, &flags);
1826 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1831 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1832 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1833 * userspace interface.
1835 cmd_mask = 1UL << ND_CMD_CALL;
1836 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1838 * These commands have a 1:1 correspondence
1839 * between DSM payload and libnvdimm ioctl
1842 cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1845 if (nfit_mem->has_lsr) {
1846 set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1847 set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1849 if (nfit_mem->has_lsw)
1850 set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1852 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1854 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1855 acpi_nfit_dimm_attribute_groups,
1856 flags, cmd_mask, flush ? flush->hint_count : 0,
1857 nfit_mem->flush_wpq);
1861 nfit_mem->nvdimm = nvdimm;
1864 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1867 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1868 nvdimm_name(nvdimm),
1869 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1870 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1871 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1872 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1873 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1877 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1882 * Now that dimms are successfully registered, and async registration
1883 * is flushed, attempt to enable event notification.
1885 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1886 struct kernfs_node *nfit_kernfs;
1888 nvdimm = nfit_mem->nvdimm;
1892 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1894 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1896 sysfs_put(nfit_kernfs);
1897 if (!nfit_mem->flags_attr)
1898 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1899 nvdimm_name(nvdimm));
1902 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1907 * These constants are private because there are no kernel consumers of
1910 enum nfit_aux_cmds {
1911 NFIT_CMD_TRANSLATE_SPA = 5,
1912 NFIT_CMD_ARS_INJECT_SET = 7,
1913 NFIT_CMD_ARS_INJECT_CLEAR = 8,
1914 NFIT_CMD_ARS_INJECT_GET = 9,
1917 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1919 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1920 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1921 struct acpi_device *adev;
1922 unsigned long dsm_mask;
1925 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1926 nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1927 adev = to_acpi_dev(acpi_desc);
1931 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1932 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1933 set_bit(i, &nd_desc->cmd_mask);
1934 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1937 (1 << ND_CMD_ARS_CAP) |
1938 (1 << ND_CMD_ARS_START) |
1939 (1 << ND_CMD_ARS_STATUS) |
1940 (1 << ND_CMD_CLEAR_ERROR) |
1941 (1 << NFIT_CMD_TRANSLATE_SPA) |
1942 (1 << NFIT_CMD_ARS_INJECT_SET) |
1943 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1944 (1 << NFIT_CMD_ARS_INJECT_GET);
1945 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1946 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1947 set_bit(i, &nd_desc->bus_dsm_mask);
1950 static ssize_t range_index_show(struct device *dev,
1951 struct device_attribute *attr, char *buf)
1953 struct nd_region *nd_region = to_nd_region(dev);
1954 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1956 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1958 static DEVICE_ATTR_RO(range_index);
1960 static ssize_t ecc_unit_size_show(struct device *dev,
1961 struct device_attribute *attr, char *buf)
1963 struct nd_region *nd_region = to_nd_region(dev);
1964 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1966 return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1968 static DEVICE_ATTR_RO(ecc_unit_size);
1970 static struct attribute *acpi_nfit_region_attributes[] = {
1971 &dev_attr_range_index.attr,
1972 &dev_attr_ecc_unit_size.attr,
1976 static const struct attribute_group acpi_nfit_region_attribute_group = {
1978 .attrs = acpi_nfit_region_attributes,
1981 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1982 &nd_region_attribute_group,
1983 &nd_mapping_attribute_group,
1984 &nd_device_attribute_group,
1985 &nd_numa_attribute_group,
1986 &acpi_nfit_region_attribute_group,
1990 /* enough info to uniquely specify an interleave set */
1991 struct nfit_set_info {
1992 struct nfit_set_info_map {
1999 struct nfit_set_info2 {
2000 struct nfit_set_info_map2 {
2004 u16 manufacturing_date;
2005 u8 manufacturing_location;
2010 static size_t sizeof_nfit_set_info(int num_mappings)
2012 return sizeof(struct nfit_set_info)
2013 + num_mappings * sizeof(struct nfit_set_info_map);
2016 static size_t sizeof_nfit_set_info2(int num_mappings)
2018 return sizeof(struct nfit_set_info2)
2019 + num_mappings * sizeof(struct nfit_set_info_map2);
2022 static int cmp_map_compat(const void *m0, const void *m1)
2024 const struct nfit_set_info_map *map0 = m0;
2025 const struct nfit_set_info_map *map1 = m1;
2027 return memcmp(&map0->region_offset, &map1->region_offset,
2031 static int cmp_map(const void *m0, const void *m1)
2033 const struct nfit_set_info_map *map0 = m0;
2034 const struct nfit_set_info_map *map1 = m1;
2036 if (map0->region_offset < map1->region_offset)
2038 else if (map0->region_offset > map1->region_offset)
2043 static int cmp_map2(const void *m0, const void *m1)
2045 const struct nfit_set_info_map2 *map0 = m0;
2046 const struct nfit_set_info_map2 *map1 = m1;
2048 if (map0->region_offset < map1->region_offset)
2050 else if (map0->region_offset > map1->region_offset)
2055 /* Retrieve the nth entry referencing this spa */
2056 static struct acpi_nfit_memory_map *memdev_from_spa(
2057 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2059 struct nfit_memdev *nfit_memdev;
2061 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2062 if (nfit_memdev->memdev->range_index == range_index)
2064 return nfit_memdev->memdev;
2068 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2069 struct nd_region_desc *ndr_desc,
2070 struct acpi_nfit_system_address *spa)
2072 struct device *dev = acpi_desc->dev;
2073 struct nd_interleave_set *nd_set;
2074 u16 nr = ndr_desc->num_mappings;
2075 struct nfit_set_info2 *info2;
2076 struct nfit_set_info *info;
2079 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2082 ndr_desc->nd_set = nd_set;
2083 guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2085 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2089 info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2093 for (i = 0; i < nr; i++) {
2094 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2095 struct nfit_set_info_map *map = &info->mapping[i];
2096 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2097 struct nvdimm *nvdimm = mapping->nvdimm;
2098 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2099 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2100 spa->range_index, i);
2101 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2103 if (!memdev || !nfit_mem->dcr) {
2104 dev_err(dev, "%s: failed to find DCR\n", __func__);
2108 map->region_offset = memdev->region_offset;
2109 map->serial_number = dcr->serial_number;
2111 map2->region_offset = memdev->region_offset;
2112 map2->serial_number = dcr->serial_number;
2113 map2->vendor_id = dcr->vendor_id;
2114 map2->manufacturing_date = dcr->manufacturing_date;
2115 map2->manufacturing_location = dcr->manufacturing_location;
2118 /* v1.1 namespaces */
2119 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2121 nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2123 /* v1.2 namespaces */
2124 sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2126 nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2128 /* support v1.1 namespaces created with the wrong sort order */
2129 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2130 cmp_map_compat, NULL);
2131 nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2133 /* record the result of the sort for the mapping position */
2134 for (i = 0; i < nr; i++) {
2135 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2138 for (j = 0; j < nr; j++) {
2139 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2140 struct nvdimm *nvdimm = mapping->nvdimm;
2141 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2142 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2144 if (map2->serial_number == dcr->serial_number &&
2145 map2->vendor_id == dcr->vendor_id &&
2146 map2->manufacturing_date == dcr->manufacturing_date &&
2147 map2->manufacturing_location
2148 == dcr->manufacturing_location) {
2149 mapping->position = i;
2155 ndr_desc->nd_set = nd_set;
2156 devm_kfree(dev, info);
2157 devm_kfree(dev, info2);
2162 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2164 struct acpi_nfit_interleave *idt = mmio->idt;
2165 u32 sub_line_offset, line_index, line_offset;
2166 u64 line_no, table_skip_count, table_offset;
2168 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2169 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2170 line_offset = idt->line_offset[line_index]
2172 table_offset = table_skip_count * mmio->table_size;
2174 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2177 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2179 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2180 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2181 const u32 STATUS_MASK = 0x80000037;
2183 if (mmio->num_lines)
2184 offset = to_interleave_offset(offset, mmio);
2186 return readl(mmio->addr.base + offset) & STATUS_MASK;
2189 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2190 resource_size_t dpa, unsigned int len, unsigned int write)
2193 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2196 BCW_OFFSET_MASK = (1ULL << 48)-1,
2198 BCW_LEN_MASK = (1ULL << 8) - 1,
2202 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2203 len = len >> L1_CACHE_SHIFT;
2204 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2205 cmd |= ((u64) write) << BCW_CMD_SHIFT;
2207 offset = nfit_blk->cmd_offset + mmio->size * bw;
2208 if (mmio->num_lines)
2209 offset = to_interleave_offset(offset, mmio);
2211 writeq(cmd, mmio->addr.base + offset);
2212 nvdimm_flush(nfit_blk->nd_region);
2214 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2215 readq(mmio->addr.base + offset);
2218 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2219 resource_size_t dpa, void *iobuf, size_t len, int rw,
2222 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2223 unsigned int copied = 0;
2227 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2228 + lane * mmio->size;
2229 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2234 if (mmio->num_lines) {
2237 offset = to_interleave_offset(base_offset + copied,
2239 div_u64_rem(offset, mmio->line_size, &line_offset);
2240 c = min_t(size_t, len, mmio->line_size - line_offset);
2242 offset = base_offset + nfit_blk->bdw_offset;
2247 memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2249 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2250 arch_invalidate_pmem((void __force *)
2251 mmio->addr.aperture + offset, c);
2253 memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2261 nvdimm_flush(nfit_blk->nd_region);
2263 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2267 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2268 resource_size_t dpa, void *iobuf, u64 len, int rw)
2270 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2271 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2272 struct nd_region *nd_region = nfit_blk->nd_region;
2273 unsigned int lane, copied = 0;
2276 lane = nd_region_acquire_lane(nd_region);
2278 u64 c = min(len, mmio->size);
2280 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2281 iobuf + copied, c, rw, lane);
2288 nd_region_release_lane(nd_region, lane);
2293 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2294 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2297 mmio->num_lines = idt->line_count;
2298 mmio->line_size = idt->line_size;
2299 if (interleave_ways == 0)
2301 mmio->table_size = mmio->num_lines * interleave_ways
2308 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2309 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2311 struct nd_cmd_dimm_flags flags;
2314 memset(&flags, 0, sizeof(flags));
2315 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2316 sizeof(flags), NULL);
2318 if (rc >= 0 && flags.status == 0)
2319 nfit_blk->dimm_flags = flags.flags;
2320 else if (rc == -ENOTTY) {
2321 /* fall back to a conservative default */
2322 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2330 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2333 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2334 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2335 struct nfit_blk_mmio *mmio;
2336 struct nfit_blk *nfit_blk;
2337 struct nfit_mem *nfit_mem;
2338 struct nvdimm *nvdimm;
2341 nvdimm = nd_blk_region_to_dimm(ndbr);
2342 nfit_mem = nvdimm_provider_data(nvdimm);
2343 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2344 dev_dbg(dev, "missing%s%s%s\n",
2345 nfit_mem ? "" : " nfit_mem",
2346 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2347 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2351 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2354 nd_blk_region_set_provider_data(ndbr, nfit_blk);
2355 nfit_blk->nd_region = to_nd_region(dev);
2357 /* map block aperture memory */
2358 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2359 mmio = &nfit_blk->mmio[BDW];
2360 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2361 nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2362 if (!mmio->addr.base) {
2363 dev_dbg(dev, "%s failed to map bdw\n",
2364 nvdimm_name(nvdimm));
2367 mmio->size = nfit_mem->bdw->size;
2368 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2369 mmio->idt = nfit_mem->idt_bdw;
2370 mmio->spa = nfit_mem->spa_bdw;
2371 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2372 nfit_mem->memdev_bdw->interleave_ways);
2374 dev_dbg(dev, "%s failed to init bdw interleave\n",
2375 nvdimm_name(nvdimm));
2379 /* map block control memory */
2380 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2381 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2382 mmio = &nfit_blk->mmio[DCR];
2383 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2384 nfit_mem->spa_dcr->length);
2385 if (!mmio->addr.base) {
2386 dev_dbg(dev, "%s failed to map dcr\n",
2387 nvdimm_name(nvdimm));
2390 mmio->size = nfit_mem->dcr->window_size;
2391 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2392 mmio->idt = nfit_mem->idt_dcr;
2393 mmio->spa = nfit_mem->spa_dcr;
2394 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2395 nfit_mem->memdev_dcr->interleave_ways);
2397 dev_dbg(dev, "%s failed to init dcr interleave\n",
2398 nvdimm_name(nvdimm));
2402 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2404 dev_dbg(dev, "%s failed get DIMM flags\n",
2405 nvdimm_name(nvdimm));
2409 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2410 dev_warn(dev, "unable to guarantee persistence of writes\n");
2412 if (mmio->line_size == 0)
2415 if ((u32) nfit_blk->cmd_offset % mmio->line_size
2416 + 8 > mmio->line_size) {
2417 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2419 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2420 + 8 > mmio->line_size) {
2421 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2428 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2429 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2431 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2432 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2435 cmd->address = spa->address;
2436 cmd->length = spa->length;
2437 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2438 sizeof(*cmd), &cmd_rc);
2444 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2448 struct nd_cmd_ars_start ars_start;
2449 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2450 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2452 memset(&ars_start, 0, sizeof(ars_start));
2453 ars_start.address = spa->address;
2454 ars_start.length = spa->length;
2455 ars_start.flags = acpi_desc->ars_start_flags;
2456 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2457 ars_start.type = ND_ARS_PERSISTENT;
2458 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2459 ars_start.type = ND_ARS_VOLATILE;
2463 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2464 sizeof(ars_start), &cmd_rc);
2471 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2474 struct nd_cmd_ars_start ars_start;
2475 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2476 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2478 memset(&ars_start, 0, sizeof(ars_start));
2479 ars_start.address = ars_status->restart_address;
2480 ars_start.length = ars_status->restart_length;
2481 ars_start.type = ars_status->type;
2482 ars_start.flags = acpi_desc->ars_start_flags;
2483 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2484 sizeof(ars_start), &cmd_rc);
2490 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2492 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2493 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2496 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2497 acpi_desc->ars_status_size, &cmd_rc);
2503 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2504 struct nd_cmd_ars_status *ars_status)
2506 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2511 * First record starts at 44 byte offset from the start of the
2514 if (ars_status->out_length < 44)
2516 for (i = 0; i < ars_status->num_records; i++) {
2517 /* only process full records */
2518 if (ars_status->out_length
2519 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2521 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2522 ars_status->records[i].err_address,
2523 ars_status->records[i].length);
2527 if (i < ars_status->num_records)
2528 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2533 static void acpi_nfit_remove_resource(void *data)
2535 struct resource *res = data;
2537 remove_resource(res);
2540 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2541 struct nd_region_desc *ndr_desc)
2543 struct resource *res, *nd_res = ndr_desc->res;
2546 /* No operation if the region is already registered as PMEM */
2547 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2548 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2549 if (is_pmem == REGION_INTERSECTS)
2552 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2556 res->name = "Persistent Memory";
2557 res->start = nd_res->start;
2558 res->end = nd_res->end;
2559 res->flags = IORESOURCE_MEM;
2560 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2562 ret = insert_resource(&iomem_resource, res);
2566 ret = devm_add_action_or_reset(acpi_desc->dev,
2567 acpi_nfit_remove_resource,
2575 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2576 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2577 struct acpi_nfit_memory_map *memdev,
2578 struct nfit_spa *nfit_spa)
2580 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2581 memdev->device_handle);
2582 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2583 struct nd_blk_region_desc *ndbr_desc;
2584 struct nfit_mem *nfit_mem;
2588 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2589 spa->range_index, memdev->device_handle);
2593 mapping->nvdimm = nvdimm;
2594 switch (nfit_spa_type(spa)) {
2596 case NFIT_SPA_VOLATILE:
2597 mapping->start = memdev->address;
2598 mapping->size = memdev->region_size;
2601 nfit_mem = nvdimm_provider_data(nvdimm);
2602 if (!nfit_mem || !nfit_mem->bdw) {
2603 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2604 spa->range_index, nvdimm_name(nvdimm));
2608 mapping->size = nfit_mem->bdw->capacity;
2609 mapping->start = nfit_mem->bdw->start_address;
2610 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2611 ndr_desc->mapping = mapping;
2612 ndr_desc->num_mappings = 1;
2613 ndbr_desc = to_blk_region_desc(ndr_desc);
2614 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2615 ndbr_desc->do_io = acpi_desc->blk_do_io;
2616 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2619 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2621 if (!nfit_spa->nd_region)
2629 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2631 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2632 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2633 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2634 nfit_spa_type(spa) == NFIT_SPA_PCD);
2637 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2639 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2640 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2641 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2644 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2645 struct nfit_spa *nfit_spa)
2647 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2648 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2649 struct nd_blk_region_desc ndbr_desc;
2650 struct nd_region_desc *ndr_desc;
2651 struct nfit_memdev *nfit_memdev;
2652 struct nvdimm_bus *nvdimm_bus;
2653 struct resource res;
2656 if (nfit_spa->nd_region)
2659 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2660 dev_dbg(acpi_desc->dev, "detected invalid spa index\n");
2664 memset(&res, 0, sizeof(res));
2665 memset(&mappings, 0, sizeof(mappings));
2666 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2667 res.start = spa->address;
2668 res.end = res.start + spa->length - 1;
2669 ndr_desc = &ndbr_desc.ndr_desc;
2670 ndr_desc->res = &res;
2671 ndr_desc->provider_data = nfit_spa;
2672 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2673 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2674 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2675 spa->proximity_domain);
2677 ndr_desc->numa_node = NUMA_NO_NODE;
2679 if(acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2680 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2682 if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2683 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2685 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2686 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2687 struct nd_mapping_desc *mapping;
2689 if (memdev->range_index != spa->range_index)
2691 if (count >= ND_MAX_MAPPINGS) {
2692 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2693 spa->range_index, ND_MAX_MAPPINGS);
2696 mapping = &mappings[count++];
2697 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2703 ndr_desc->mapping = mappings;
2704 ndr_desc->num_mappings = count;
2705 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2709 nvdimm_bus = acpi_desc->nvdimm_bus;
2710 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2711 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2713 dev_warn(acpi_desc->dev,
2714 "failed to insert pmem resource to iomem: %d\n",
2719 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2721 if (!nfit_spa->nd_region)
2723 } else if (nfit_spa_is_volatile(spa)) {
2724 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2726 if (!nfit_spa->nd_region)
2728 } else if (nfit_spa_is_virtual(spa)) {
2729 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2731 if (!nfit_spa->nd_region)
2737 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2738 nfit_spa->spa->range_index);
2742 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2745 struct device *dev = acpi_desc->dev;
2746 struct nd_cmd_ars_status *ars_status;
2748 if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2749 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2753 if (acpi_desc->ars_status)
2754 devm_kfree(dev, acpi_desc->ars_status);
2755 acpi_desc->ars_status = NULL;
2756 ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2759 acpi_desc->ars_status = ars_status;
2760 acpi_desc->ars_status_size = max_ars;
2764 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2765 struct nfit_spa *nfit_spa)
2767 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2770 if (!nfit_spa->max_ars) {
2771 struct nd_cmd_ars_cap ars_cap;
2773 memset(&ars_cap, 0, sizeof(ars_cap));
2774 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2777 nfit_spa->max_ars = ars_cap.max_ars_out;
2778 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2779 /* check that the supported scrub types match the spa type */
2780 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2781 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2783 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2784 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2788 if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2791 rc = ars_get_status(acpi_desc);
2792 if (rc < 0 && rc != -ENOSPC)
2795 if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2801 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2802 struct nfit_spa *nfit_spa)
2804 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2805 unsigned int overflow_retry = scrub_overflow_abort;
2806 u64 init_ars_start = 0, init_ars_len = 0;
2807 struct device *dev = acpi_desc->dev;
2808 unsigned int tmo = scrub_timeout;
2811 if (!test_bit(ARS_REQ, &nfit_spa->ars_state) || !nfit_spa->nd_region)
2814 rc = ars_start(acpi_desc, nfit_spa);
2816 * If we timed out the initial scan we'll still be busy here,
2817 * and will wait another timeout before giving up permanently.
2819 if (rc < 0 && rc != -EBUSY)
2823 u64 ars_start, ars_len;
2825 if (acpi_desc->cancel)
2827 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2830 if (rc == -EBUSY && !tmo) {
2831 dev_warn(dev, "range %d ars timeout, aborting\n",
2838 * Note, entries may be appended to the list
2839 * while the lock is dropped, but the workqueue
2840 * being active prevents entries being deleted /
2843 mutex_unlock(&acpi_desc->init_mutex);
2846 mutex_lock(&acpi_desc->init_mutex);
2850 /* we got some results, but there are more pending... */
2851 if (rc == -ENOSPC && overflow_retry--) {
2852 if (!init_ars_len) {
2853 init_ars_len = acpi_desc->ars_status->length;
2854 init_ars_start = acpi_desc->ars_status->address;
2856 rc = ars_continue(acpi_desc);
2860 dev_warn(dev, "range %d ars continuation failed\n",
2866 ars_start = init_ars_start;
2867 ars_len = init_ars_len;
2869 ars_start = acpi_desc->ars_status->address;
2870 ars_len = acpi_desc->ars_status->length;
2872 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2873 spa->range_index, ars_start, ars_len);
2874 /* notify the region about new poison entries */
2875 nvdimm_region_notify(nfit_spa->nd_region,
2876 NVDIMM_REVALIDATE_POISON);
2881 static void acpi_nfit_scrub(struct work_struct *work)
2884 u64 init_scrub_length = 0;
2885 struct nfit_spa *nfit_spa;
2886 u64 init_scrub_address = 0;
2887 bool init_ars_done = false;
2888 struct acpi_nfit_desc *acpi_desc;
2889 unsigned int tmo = scrub_timeout;
2890 unsigned int overflow_retry = scrub_overflow_abort;
2892 acpi_desc = container_of(work, typeof(*acpi_desc), work);
2893 dev = acpi_desc->dev;
2896 * We scrub in 2 phases. The first phase waits for any platform
2897 * firmware initiated scrubs to complete and then we go search for the
2898 * affected spa regions to mark them scanned. In the second phase we
2899 * initiate a directed scrub for every range that was not scrubbed in
2900 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2901 * the first phase, but really only care about running phase 2, where
2902 * regions can be notified of new poison.
2905 /* process platform firmware initiated scrubs */
2907 mutex_lock(&acpi_desc->init_mutex);
2908 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2909 struct nd_cmd_ars_status *ars_status;
2910 struct acpi_nfit_system_address *spa;
2911 u64 ars_start, ars_len;
2914 if (acpi_desc->cancel)
2917 if (nfit_spa->nd_region)
2920 if (init_ars_done) {
2922 * No need to re-query, we're now just
2923 * reconciling all the ranges covered by the
2928 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2930 if (rc == -ENOTTY) {
2931 /* no ars capability, just register spa and move on */
2932 acpi_nfit_register_region(acpi_desc, nfit_spa);
2936 if (rc == -EBUSY && !tmo) {
2937 /* fallthrough to directed scrub in phase 2 */
2938 dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2940 } else if (rc == -EBUSY) {
2941 mutex_unlock(&acpi_desc->init_mutex);
2947 /* we got some results, but there are more pending... */
2948 if (rc == -ENOSPC && overflow_retry--) {
2949 ars_status = acpi_desc->ars_status;
2951 * Record the original scrub range, so that we
2952 * can recall all the ranges impacted by the
2955 if (!init_scrub_length) {
2956 init_scrub_length = ars_status->length;
2957 init_scrub_address = ars_status->address;
2959 rc = ars_continue(acpi_desc);
2961 mutex_unlock(&acpi_desc->init_mutex);
2968 * Initial scrub failed, we'll give it one more
2974 /* We got some final results, record completed ranges */
2975 ars_status = acpi_desc->ars_status;
2976 if (init_scrub_length) {
2977 ars_start = init_scrub_address;
2978 ars_len = ars_start + init_scrub_length;
2980 ars_start = ars_status->address;
2981 ars_len = ars_status->length;
2983 spa = nfit_spa->spa;
2985 if (!init_ars_done) {
2986 init_ars_done = true;
2987 dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2988 ars_start, ars_len);
2990 if (ars_start <= spa->address && ars_start + ars_len
2991 >= spa->address + spa->length)
2992 acpi_nfit_register_region(acpi_desc, nfit_spa);
2996 * For all the ranges not covered by an initial scrub we still
2997 * want to see if there are errors, but it's ok to discover them
3000 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3002 * Flag all the ranges that still need scrubbing, but
3003 * register them now to make data available.
3005 if (!nfit_spa->nd_region) {
3006 set_bit(ARS_REQ, &nfit_spa->ars_state);
3007 acpi_nfit_register_region(acpi_desc, nfit_spa);
3010 acpi_desc->init_complete = 1;
3012 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3013 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
3014 acpi_desc->scrub_count++;
3015 acpi_desc->ars_start_flags = 0;
3016 if (acpi_desc->scrub_count_state)
3017 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3018 mutex_unlock(&acpi_desc->init_mutex);
3021 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3023 struct nfit_spa *nfit_spa;
3025 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3026 int rc, type = nfit_spa_type(nfit_spa->spa);
3028 /* PMEM and VMEM will be registered by the ARS workqueue */
3029 if (type == NFIT_SPA_PM || type == NFIT_SPA_VOLATILE)
3031 /* BLK apertures belong to BLK region registration below */
3032 if (type == NFIT_SPA_BDW)
3034 /* BLK regions don't need to wait for ARS results */
3035 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3040 acpi_desc->ars_start_flags = 0;
3041 if (!acpi_desc->cancel)
3042 queue_work(nfit_wq, &acpi_desc->work);
3046 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3047 struct nfit_table_prev *prev)
3049 struct device *dev = acpi_desc->dev;
3051 if (!list_empty(&prev->spas) ||
3052 !list_empty(&prev->memdevs) ||
3053 !list_empty(&prev->dcrs) ||
3054 !list_empty(&prev->bdws) ||
3055 !list_empty(&prev->idts) ||
3056 !list_empty(&prev->flushes)) {
3057 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3063 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3065 struct device *dev = acpi_desc->dev;
3066 struct kernfs_node *nfit;
3067 struct device *bus_dev;
3069 if (!ars_supported(acpi_desc->nvdimm_bus))
3072 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3073 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3075 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3078 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3080 if (!acpi_desc->scrub_count_state) {
3081 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3088 static void acpi_nfit_unregister(void *data)
3090 struct acpi_nfit_desc *acpi_desc = data;
3092 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3095 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3097 struct device *dev = acpi_desc->dev;
3098 struct nfit_table_prev prev;
3102 if (!acpi_desc->nvdimm_bus) {
3103 acpi_nfit_init_dsms(acpi_desc);
3105 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3106 &acpi_desc->nd_desc);
3107 if (!acpi_desc->nvdimm_bus)
3110 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3115 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3119 /* register this acpi_desc for mce notifications */
3120 mutex_lock(&acpi_desc_lock);
3121 list_add_tail(&acpi_desc->list, &acpi_descs);
3122 mutex_unlock(&acpi_desc_lock);
3125 mutex_lock(&acpi_desc->init_mutex);
3127 INIT_LIST_HEAD(&prev.spas);
3128 INIT_LIST_HEAD(&prev.memdevs);
3129 INIT_LIST_HEAD(&prev.dcrs);
3130 INIT_LIST_HEAD(&prev.bdws);
3131 INIT_LIST_HEAD(&prev.idts);
3132 INIT_LIST_HEAD(&prev.flushes);
3134 list_cut_position(&prev.spas, &acpi_desc->spas,
3135 acpi_desc->spas.prev);
3136 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3137 acpi_desc->memdevs.prev);
3138 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3139 acpi_desc->dcrs.prev);
3140 list_cut_position(&prev.bdws, &acpi_desc->bdws,
3141 acpi_desc->bdws.prev);
3142 list_cut_position(&prev.idts, &acpi_desc->idts,
3143 acpi_desc->idts.prev);
3144 list_cut_position(&prev.flushes, &acpi_desc->flushes,
3145 acpi_desc->flushes.prev);
3148 while (!IS_ERR_OR_NULL(data))
3149 data = add_table(acpi_desc, &prev, data, end);
3152 dev_dbg(dev, "nfit table parsing error: %ld\n", PTR_ERR(data));
3157 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3161 rc = nfit_mem_init(acpi_desc);
3165 rc = acpi_nfit_register_dimms(acpi_desc);
3169 rc = acpi_nfit_register_regions(acpi_desc);
3172 mutex_unlock(&acpi_desc->init_mutex);
3175 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3177 struct acpi_nfit_flush_work {
3178 struct work_struct work;
3179 struct completion cmp;
3182 static void flush_probe(struct work_struct *work)
3184 struct acpi_nfit_flush_work *flush;
3186 flush = container_of(work, typeof(*flush), work);
3187 complete(&flush->cmp);
3190 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3192 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3193 struct device *dev = acpi_desc->dev;
3194 struct acpi_nfit_flush_work flush;
3197 /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3201 /* bounce the init_mutex to make init_complete valid */
3202 mutex_lock(&acpi_desc->init_mutex);
3203 if (acpi_desc->cancel || acpi_desc->init_complete) {
3204 mutex_unlock(&acpi_desc->init_mutex);
3209 * Scrub work could take 10s of seconds, userspace may give up so we
3210 * need to be interruptible while waiting.
3212 INIT_WORK_ONSTACK(&flush.work, flush_probe);
3213 init_completion(&flush.cmp);
3214 queue_work(nfit_wq, &flush.work);
3215 mutex_unlock(&acpi_desc->init_mutex);
3217 rc = wait_for_completion_interruptible(&flush.cmp);
3218 cancel_work_sync(&flush.work);
3222 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3223 struct nvdimm *nvdimm, unsigned int cmd)
3225 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3229 if (cmd != ND_CMD_ARS_START)
3233 * The kernel and userspace may race to initiate a scrub, but
3234 * the scrub thread is prepared to lose that initial race. It
3235 * just needs guarantees that any ars it initiates are not
3236 * interrupted by any intervening start reqeusts from userspace.
3238 if (work_busy(&acpi_desc->work))
3244 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, unsigned long flags)
3246 struct device *dev = acpi_desc->dev;
3247 struct nfit_spa *nfit_spa;
3249 if (work_busy(&acpi_desc->work))
3252 mutex_lock(&acpi_desc->init_mutex);
3253 if (acpi_desc->cancel) {
3254 mutex_unlock(&acpi_desc->init_mutex);
3258 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3259 struct acpi_nfit_system_address *spa = nfit_spa->spa;
3261 if (nfit_spa_type(spa) != NFIT_SPA_PM)
3264 set_bit(ARS_REQ, &nfit_spa->ars_state);
3266 acpi_desc->ars_start_flags = 0;
3267 if (test_bit(ARS_SHORT, &flags))
3268 acpi_desc->ars_start_flags |= ND_ARS_RETURN_PREV_DATA;
3269 queue_work(nfit_wq, &acpi_desc->work);
3270 dev_dbg(dev, "ars_scan triggered\n");
3271 mutex_unlock(&acpi_desc->init_mutex);
3276 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3278 struct nvdimm_bus_descriptor *nd_desc;
3280 dev_set_drvdata(dev, acpi_desc);
3281 acpi_desc->dev = dev;
3282 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3283 nd_desc = &acpi_desc->nd_desc;
3284 nd_desc->provider_name = "ACPI.NFIT";
3285 nd_desc->module = THIS_MODULE;
3286 nd_desc->ndctl = acpi_nfit_ctl;
3287 nd_desc->flush_probe = acpi_nfit_flush_probe;
3288 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3289 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3291 INIT_LIST_HEAD(&acpi_desc->spas);
3292 INIT_LIST_HEAD(&acpi_desc->dcrs);
3293 INIT_LIST_HEAD(&acpi_desc->bdws);
3294 INIT_LIST_HEAD(&acpi_desc->idts);
3295 INIT_LIST_HEAD(&acpi_desc->flushes);
3296 INIT_LIST_HEAD(&acpi_desc->memdevs);
3297 INIT_LIST_HEAD(&acpi_desc->dimms);
3298 INIT_LIST_HEAD(&acpi_desc->list);
3299 mutex_init(&acpi_desc->init_mutex);
3300 INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3302 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3304 static void acpi_nfit_put_table(void *table)
3306 acpi_put_table(table);
3309 void acpi_nfit_shutdown(void *data)
3311 struct acpi_nfit_desc *acpi_desc = data;
3312 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3315 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3318 mutex_lock(&acpi_desc_lock);
3319 list_del(&acpi_desc->list);
3320 mutex_unlock(&acpi_desc_lock);
3322 mutex_lock(&acpi_desc->init_mutex);
3323 acpi_desc->cancel = 1;
3324 mutex_unlock(&acpi_desc->init_mutex);
3327 * Bounce the nvdimm bus lock to make sure any in-flight
3328 * acpi_nfit_ars_rescan() submissions have had a chance to
3329 * either submit or see ->cancel set.
3331 device_lock(bus_dev);
3332 device_unlock(bus_dev);
3334 flush_workqueue(nfit_wq);
3336 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3338 static int acpi_nfit_add(struct acpi_device *adev)
3340 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3341 struct acpi_nfit_desc *acpi_desc;
3342 struct device *dev = &adev->dev;
3343 struct acpi_table_header *tbl;
3344 acpi_status status = AE_OK;
3348 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3349 if (ACPI_FAILURE(status)) {
3350 /* This is ok, we could have an nvdimm hotplugged later */
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));
3388 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3391 static int acpi_nfit_remove(struct acpi_device *adev)
3393 /* see acpi_nfit_unregister */
3397 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3399 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3400 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3401 union acpi_object *obj;
3406 /* dev->driver may be null if we're being removed */
3407 dev_dbg(dev, "no driver found for dev\n");
3412 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3415 acpi_nfit_desc_init(acpi_desc, dev);
3418 * Finish previous registration before considering new
3421 flush_workqueue(nfit_wq);
3425 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3426 if (ACPI_FAILURE(status)) {
3427 dev_err(dev, "failed to evaluate _FIT\n");
3432 if (obj->type == ACPI_TYPE_BUFFER) {
3433 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3434 obj->buffer.length);
3436 dev_err(dev, "failed to merge updated NFIT\n");
3438 dev_err(dev, "Invalid _FIT\n");
3442 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3444 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3445 unsigned long flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3448 acpi_nfit_ars_rescan(acpi_desc, flags);
3451 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3453 dev_dbg(dev, "event: 0x%x\n", event);
3456 case NFIT_NOTIFY_UPDATE:
3457 return acpi_nfit_update_notify(dev, handle);
3458 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3459 return acpi_nfit_uc_error_notify(dev, handle);
3464 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3466 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3468 device_lock(&adev->dev);
3469 __acpi_nfit_notify(&adev->dev, adev->handle, event);
3470 device_unlock(&adev->dev);
3473 static const struct acpi_device_id acpi_nfit_ids[] = {
3477 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3479 static struct acpi_driver acpi_nfit_driver = {
3480 .name = KBUILD_MODNAME,
3481 .ids = acpi_nfit_ids,
3483 .add = acpi_nfit_add,
3484 .remove = acpi_nfit_remove,
3485 .notify = acpi_nfit_notify,
3489 static __init int nfit_init(void)
3493 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3494 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3495 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3496 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3497 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3498 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3499 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3500 BUILD_BUG_ON(sizeof(struct acpi_nfit_capabilities) != 16);
3502 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3503 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3504 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3505 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3506 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3507 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3508 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3509 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3510 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3511 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3512 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3513 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3514 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3516 nfit_wq = create_singlethread_workqueue("nfit");
3520 nfit_mce_register();
3521 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3523 nfit_mce_unregister();
3524 destroy_workqueue(nfit_wq);
3531 static __exit void nfit_exit(void)
3533 nfit_mce_unregister();
3534 acpi_bus_unregister_driver(&acpi_nfit_driver);
3535 destroy_workqueue(nfit_wq);
3536 WARN_ON(!list_empty(&acpi_descs));
3539 module_init(nfit_init);
3540 module_exit(nfit_exit);
3541 MODULE_LICENSE("GPL v2");
3542 MODULE_AUTHOR("Intel Corporation");