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>
23 #include <linux/pmem.h>
26 #include <asm/cacheflush.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 unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
40 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
41 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
43 /* after three payloads of overflow, it's dead jim */
44 static unsigned int scrub_overflow_abort = 3;
45 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
46 MODULE_PARM_DESC(scrub_overflow_abort,
47 "Number of times we overflow ARS results before abort");
49 static bool disable_vendor_specific;
50 module_param(disable_vendor_specific, bool, S_IRUGO);
51 MODULE_PARM_DESC(disable_vendor_specific,
52 "Limit commands to the publicly specified set\n");
54 static struct workqueue_struct *nfit_wq;
56 struct nfit_table_prev {
57 struct list_head spas;
58 struct list_head memdevs;
59 struct list_head dcrs;
60 struct list_head bdws;
61 struct list_head idts;
62 struct list_head flushes;
65 static u8 nfit_uuid[NFIT_UUID_MAX][16];
67 const u8 *to_nfit_uuid(enum nfit_uuids id)
71 EXPORT_SYMBOL(to_nfit_uuid);
73 static struct acpi_nfit_desc *to_acpi_nfit_desc(
74 struct nvdimm_bus_descriptor *nd_desc)
76 return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
79 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
81 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
84 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
87 if (!nd_desc->provider_name
88 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
91 return to_acpi_device(acpi_desc->dev);
94 static int xlat_status(void *buf, unsigned int cmd)
96 struct nd_cmd_clear_error *clear_err;
97 struct nd_cmd_ars_status *ars_status;
98 struct nd_cmd_ars_start *ars_start;
99 struct nd_cmd_ars_cap *ars_cap;
105 if ((ars_cap->status & 0xffff) == NFIT_ARS_CAP_NONE)
109 if (ars_cap->status & 0xffff)
112 /* No supported scan types for this range */
113 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
114 if ((ars_cap->status >> 16 & flags) == 0)
117 case ND_CMD_ARS_START:
119 /* ARS is in progress */
120 if ((ars_start->status & 0xffff) == NFIT_ARS_START_BUSY)
124 if (ars_start->status & 0xffff)
127 case ND_CMD_ARS_STATUS:
130 if (ars_status->status & 0xffff)
132 /* Check extended status (Upper two bytes) */
133 if (ars_status->status == NFIT_ARS_STATUS_DONE)
136 /* ARS is in progress */
137 if (ars_status->status == NFIT_ARS_STATUS_BUSY)
140 /* No ARS performed for the current boot */
141 if (ars_status->status == NFIT_ARS_STATUS_NONE)
145 * ARS interrupted, either we overflowed or some other
146 * agent wants the scan to stop. If we didn't overflow
147 * then just continue with the returned results.
149 if (ars_status->status == NFIT_ARS_STATUS_INTR) {
150 if (ars_status->flags & NFIT_ARS_F_OVERFLOW)
156 if (ars_status->status >> 16)
159 case ND_CMD_CLEAR_ERROR:
161 if (clear_err->status & 0xffff)
163 if (!clear_err->cleared)
165 if (clear_err->length > clear_err->cleared)
166 return clear_err->cleared;
175 static int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc,
176 struct nvdimm *nvdimm, unsigned int cmd, void *buf,
177 unsigned int buf_len, int *cmd_rc)
179 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
180 union acpi_object in_obj, in_buf, *out_obj;
181 const struct nd_cmd_desc *desc = NULL;
182 struct device *dev = acpi_desc->dev;
183 struct nd_cmd_pkg *call_pkg = NULL;
184 const char *cmd_name, *dimm_name;
185 unsigned long cmd_mask, dsm_mask;
193 if (cmd == ND_CMD_CALL) {
195 func = call_pkg->nd_command;
199 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
200 struct acpi_device *adev = nfit_mem->adev;
204 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
207 dimm_name = nvdimm_name(nvdimm);
208 cmd_name = nvdimm_cmd_name(cmd);
209 cmd_mask = nvdimm_cmd_mask(nvdimm);
210 dsm_mask = nfit_mem->dsm_mask;
211 desc = nd_cmd_dimm_desc(cmd);
212 uuid = to_nfit_uuid(nfit_mem->family);
213 handle = adev->handle;
215 struct acpi_device *adev = to_acpi_dev(acpi_desc);
217 cmd_name = nvdimm_bus_cmd_name(cmd);
218 cmd_mask = nd_desc->cmd_mask;
220 desc = nd_cmd_bus_desc(cmd);
221 uuid = to_nfit_uuid(NFIT_DEV_BUS);
222 handle = adev->handle;
226 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
229 if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
232 in_obj.type = ACPI_TYPE_PACKAGE;
233 in_obj.package.count = 1;
234 in_obj.package.elements = &in_buf;
235 in_buf.type = ACPI_TYPE_BUFFER;
236 in_buf.buffer.pointer = buf;
237 in_buf.buffer.length = 0;
239 /* libnvdimm has already validated the input envelope */
240 for (i = 0; i < desc->in_num; i++)
241 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
245 /* skip over package wrapper */
246 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
247 in_buf.buffer.length = call_pkg->nd_size_in;
250 if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
251 dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
252 __func__, dimm_name, cmd, func,
253 in_buf.buffer.length);
254 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
255 in_buf.buffer.pointer,
256 min_t(u32, 256, in_buf.buffer.length), true);
259 out_obj = acpi_evaluate_dsm(handle, uuid, 1, func, &in_obj);
261 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
267 call_pkg->nd_fw_size = out_obj->buffer.length;
268 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
269 out_obj->buffer.pointer,
270 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
274 * Need to support FW function w/o known size in advance.
275 * Caller can determine required size based upon nd_fw_size.
276 * If we return an error (like elsewhere) then caller wouldn't
277 * be able to rely upon data returned to make calculation.
282 if (out_obj->package.type != ACPI_TYPE_BUFFER) {
283 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
284 __func__, dimm_name, cmd_name, out_obj->type);
289 if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
290 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__,
291 dimm_name, cmd_name, out_obj->buffer.length);
292 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
293 4, out_obj->buffer.pointer, min_t(u32, 128,
294 out_obj->buffer.length), true);
297 for (i = 0, offset = 0; i < desc->out_num; i++) {
298 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
299 (u32 *) out_obj->buffer.pointer);
301 if (offset + out_size > out_obj->buffer.length) {
302 dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
303 __func__, dimm_name, cmd_name, i);
307 if (in_buf.buffer.length + offset + out_size > buf_len) {
308 dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
309 __func__, dimm_name, cmd_name, i);
313 memcpy(buf + in_buf.buffer.length + offset,
314 out_obj->buffer.pointer + offset, out_size);
317 if (offset + in_buf.buffer.length < buf_len) {
320 * status valid, return the number of bytes left
321 * unfilled in the output buffer
323 rc = buf_len - offset - in_buf.buffer.length;
325 *cmd_rc = xlat_status(buf, cmd);
327 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
328 __func__, dimm_name, cmd_name, buf_len,
335 *cmd_rc = xlat_status(buf, cmd);
344 static const char *spa_type_name(u16 type)
346 static const char *to_name[] = {
347 [NFIT_SPA_VOLATILE] = "volatile",
348 [NFIT_SPA_PM] = "pmem",
349 [NFIT_SPA_DCR] = "dimm-control-region",
350 [NFIT_SPA_BDW] = "block-data-window",
351 [NFIT_SPA_VDISK] = "volatile-disk",
352 [NFIT_SPA_VCD] = "volatile-cd",
353 [NFIT_SPA_PDISK] = "persistent-disk",
354 [NFIT_SPA_PCD] = "persistent-cd",
358 if (type > NFIT_SPA_PCD)
361 return to_name[type];
364 static int nfit_spa_type(struct acpi_nfit_system_address *spa)
368 for (i = 0; i < NFIT_UUID_MAX; i++)
369 if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
374 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
375 struct nfit_table_prev *prev,
376 struct acpi_nfit_system_address *spa)
378 struct device *dev = acpi_desc->dev;
379 struct nfit_spa *nfit_spa;
381 if (spa->header.length != sizeof(*spa))
384 list_for_each_entry(nfit_spa, &prev->spas, list) {
385 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
386 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
391 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
395 INIT_LIST_HEAD(&nfit_spa->list);
396 memcpy(nfit_spa->spa, spa, sizeof(*spa));
397 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
398 dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
400 spa_type_name(nfit_spa_type(spa)));
404 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
405 struct nfit_table_prev *prev,
406 struct acpi_nfit_memory_map *memdev)
408 struct device *dev = acpi_desc->dev;
409 struct nfit_memdev *nfit_memdev;
411 if (memdev->header.length != sizeof(*memdev))
414 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
415 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
416 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
420 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
424 INIT_LIST_HEAD(&nfit_memdev->list);
425 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
426 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
427 dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
428 __func__, memdev->device_handle, memdev->range_index,
429 memdev->region_index);
434 * An implementation may provide a truncated control region if no block windows
437 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
439 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
444 return offsetof(struct acpi_nfit_control_region, window_size);
447 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
448 struct nfit_table_prev *prev,
449 struct acpi_nfit_control_region *dcr)
451 struct device *dev = acpi_desc->dev;
452 struct nfit_dcr *nfit_dcr;
454 if (!sizeof_dcr(dcr))
457 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
458 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
459 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
463 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
467 INIT_LIST_HEAD(&nfit_dcr->list);
468 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
469 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
470 dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
471 dcr->region_index, dcr->windows);
475 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
476 struct nfit_table_prev *prev,
477 struct acpi_nfit_data_region *bdw)
479 struct device *dev = acpi_desc->dev;
480 struct nfit_bdw *nfit_bdw;
482 if (bdw->header.length != sizeof(*bdw))
484 list_for_each_entry(nfit_bdw, &prev->bdws, list)
485 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
486 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
490 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
494 INIT_LIST_HEAD(&nfit_bdw->list);
495 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
496 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
497 dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
498 bdw->region_index, bdw->windows);
502 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
504 if (idt->header.length < sizeof(*idt))
506 return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
509 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
510 struct nfit_table_prev *prev,
511 struct acpi_nfit_interleave *idt)
513 struct device *dev = acpi_desc->dev;
514 struct nfit_idt *nfit_idt;
516 if (!sizeof_idt(idt))
519 list_for_each_entry(nfit_idt, &prev->idts, list) {
520 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
523 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
524 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
529 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
533 INIT_LIST_HEAD(&nfit_idt->list);
534 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
535 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
536 dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
537 idt->interleave_index, idt->line_count);
541 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
543 if (flush->header.length < sizeof(*flush))
545 return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
548 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
549 struct nfit_table_prev *prev,
550 struct acpi_nfit_flush_address *flush)
552 struct device *dev = acpi_desc->dev;
553 struct nfit_flush *nfit_flush;
555 if (!sizeof_flush(flush))
558 list_for_each_entry(nfit_flush, &prev->flushes, list) {
559 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
562 if (memcmp(nfit_flush->flush, flush,
563 sizeof_flush(flush)) == 0) {
564 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
569 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
570 + sizeof_flush(flush), GFP_KERNEL);
573 INIT_LIST_HEAD(&nfit_flush->list);
574 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
575 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
576 dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
577 flush->device_handle, flush->hint_count);
581 static void *add_table(struct acpi_nfit_desc *acpi_desc,
582 struct nfit_table_prev *prev, void *table, const void *end)
584 struct device *dev = acpi_desc->dev;
585 struct acpi_nfit_header *hdr;
586 void *err = ERR_PTR(-ENOMEM);
593 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
599 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
600 if (!add_spa(acpi_desc, prev, table))
603 case ACPI_NFIT_TYPE_MEMORY_MAP:
604 if (!add_memdev(acpi_desc, prev, table))
607 case ACPI_NFIT_TYPE_CONTROL_REGION:
608 if (!add_dcr(acpi_desc, prev, table))
611 case ACPI_NFIT_TYPE_DATA_REGION:
612 if (!add_bdw(acpi_desc, prev, table))
615 case ACPI_NFIT_TYPE_INTERLEAVE:
616 if (!add_idt(acpi_desc, prev, table))
619 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
620 if (!add_flush(acpi_desc, prev, table))
623 case ACPI_NFIT_TYPE_SMBIOS:
624 dev_dbg(dev, "%s: smbios\n", __func__);
627 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
631 return table + hdr->length;
634 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
635 struct nfit_mem *nfit_mem)
637 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
638 u16 dcr = nfit_mem->dcr->region_index;
639 struct nfit_spa *nfit_spa;
641 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
642 u16 range_index = nfit_spa->spa->range_index;
643 int type = nfit_spa_type(nfit_spa->spa);
644 struct nfit_memdev *nfit_memdev;
646 if (type != NFIT_SPA_BDW)
649 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
650 if (nfit_memdev->memdev->range_index != range_index)
652 if (nfit_memdev->memdev->device_handle != device_handle)
654 if (nfit_memdev->memdev->region_index != dcr)
657 nfit_mem->spa_bdw = nfit_spa->spa;
662 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
663 nfit_mem->spa_dcr->range_index);
664 nfit_mem->bdw = NULL;
667 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
668 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
670 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
671 struct nfit_memdev *nfit_memdev;
672 struct nfit_bdw *nfit_bdw;
673 struct nfit_idt *nfit_idt;
674 u16 idt_idx, range_index;
676 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
677 if (nfit_bdw->bdw->region_index != dcr)
679 nfit_mem->bdw = nfit_bdw->bdw;
686 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
688 if (!nfit_mem->spa_bdw)
691 range_index = nfit_mem->spa_bdw->range_index;
692 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
693 if (nfit_memdev->memdev->range_index != range_index ||
694 nfit_memdev->memdev->region_index != dcr)
696 nfit_mem->memdev_bdw = nfit_memdev->memdev;
697 idt_idx = nfit_memdev->memdev->interleave_index;
698 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
699 if (nfit_idt->idt->interleave_index != idt_idx)
701 nfit_mem->idt_bdw = nfit_idt->idt;
708 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
709 struct acpi_nfit_system_address *spa)
711 struct nfit_mem *nfit_mem, *found;
712 struct nfit_memdev *nfit_memdev;
713 int type = nfit_spa_type(spa);
723 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
724 struct nfit_flush *nfit_flush;
725 struct nfit_dcr *nfit_dcr;
729 if (nfit_memdev->memdev->range_index != spa->range_index)
732 dcr = nfit_memdev->memdev->region_index;
733 device_handle = nfit_memdev->memdev->device_handle;
734 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
735 if (__to_nfit_memdev(nfit_mem)->device_handle
744 nfit_mem = devm_kzalloc(acpi_desc->dev,
745 sizeof(*nfit_mem), GFP_KERNEL);
748 INIT_LIST_HEAD(&nfit_mem->list);
749 nfit_mem->acpi_desc = acpi_desc;
750 list_add(&nfit_mem->list, &acpi_desc->dimms);
753 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
754 if (nfit_dcr->dcr->region_index != dcr)
757 * Record the control region for the dimm. For
758 * the ACPI 6.1 case, where there are separate
759 * control regions for the pmem vs blk
760 * interfaces, be sure to record the extended
764 nfit_mem->dcr = nfit_dcr->dcr;
765 else if (nfit_mem->dcr->windows == 0
766 && nfit_dcr->dcr->windows)
767 nfit_mem->dcr = nfit_dcr->dcr;
771 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
772 struct acpi_nfit_flush_address *flush;
775 if (nfit_flush->flush->device_handle != device_handle)
777 nfit_mem->nfit_flush = nfit_flush;
778 flush = nfit_flush->flush;
779 nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
781 * sizeof(struct resource), GFP_KERNEL);
782 if (!nfit_mem->flush_wpq)
784 for (i = 0; i < flush->hint_count; i++) {
785 struct resource *res = &nfit_mem->flush_wpq[i];
787 res->start = flush->hint_address[i];
788 res->end = res->start + 8 - 1;
793 if (dcr && !nfit_mem->dcr) {
794 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
795 spa->range_index, dcr);
799 if (type == NFIT_SPA_DCR) {
800 struct nfit_idt *nfit_idt;
803 /* multiple dimms may share a SPA when interleaved */
804 nfit_mem->spa_dcr = spa;
805 nfit_mem->memdev_dcr = nfit_memdev->memdev;
806 idt_idx = nfit_memdev->memdev->interleave_index;
807 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
808 if (nfit_idt->idt->interleave_index != idt_idx)
810 nfit_mem->idt_dcr = nfit_idt->idt;
813 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
816 * A single dimm may belong to multiple SPA-PM
817 * ranges, record at least one in addition to
820 nfit_mem->memdev_pmem = nfit_memdev->memdev;
827 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
829 struct nfit_mem *a = container_of(_a, typeof(*a), list);
830 struct nfit_mem *b = container_of(_b, typeof(*b), list);
831 u32 handleA, handleB;
833 handleA = __to_nfit_memdev(a)->device_handle;
834 handleB = __to_nfit_memdev(b)->device_handle;
835 if (handleA < handleB)
837 else if (handleA > handleB)
842 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
844 struct nfit_spa *nfit_spa;
847 * For each SPA-DCR or SPA-PMEM address range find its
848 * corresponding MEMDEV(s). From each MEMDEV find the
849 * corresponding DCR. Then, if we're operating on a SPA-DCR,
850 * try to find a SPA-BDW and a corresponding BDW that references
851 * the DCR. Throw it all into an nfit_mem object. Note, that
854 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
857 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
862 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
867 static ssize_t revision_show(struct device *dev,
868 struct device_attribute *attr, char *buf)
870 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
871 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
872 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
874 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
876 static DEVICE_ATTR_RO(revision);
879 * This shows the number of full Address Range Scrubs that have been
880 * completed since driver load time. Userspace can wait on this using
881 * select/poll etc. A '+' at the end indicates an ARS is in progress
883 static ssize_t scrub_show(struct device *dev,
884 struct device_attribute *attr, char *buf)
886 struct nvdimm_bus_descriptor *nd_desc;
890 nd_desc = dev_get_drvdata(dev);
892 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
894 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
895 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
901 static int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc);
903 static ssize_t scrub_store(struct device *dev,
904 struct device_attribute *attr, const char *buf, size_t size)
906 struct nvdimm_bus_descriptor *nd_desc;
910 rc = kstrtol(buf, 0, &val);
917 nd_desc = dev_get_drvdata(dev);
919 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
921 rc = acpi_nfit_ars_rescan(acpi_desc);
928 static DEVICE_ATTR_RW(scrub);
930 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
932 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
933 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
934 | 1 << ND_CMD_ARS_STATUS;
936 return (nd_desc->cmd_mask & mask) == mask;
939 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
941 struct device *dev = container_of(kobj, struct device, kobj);
942 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
944 if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
949 static struct attribute *acpi_nfit_attributes[] = {
950 &dev_attr_revision.attr,
951 &dev_attr_scrub.attr,
955 static struct attribute_group acpi_nfit_attribute_group = {
957 .attrs = acpi_nfit_attributes,
958 .is_visible = nfit_visible,
961 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
962 &nvdimm_bus_attribute_group,
963 &acpi_nfit_attribute_group,
967 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
969 struct nvdimm *nvdimm = to_nvdimm(dev);
970 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
972 return __to_nfit_memdev(nfit_mem);
975 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
977 struct nvdimm *nvdimm = to_nvdimm(dev);
978 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
980 return nfit_mem->dcr;
983 static ssize_t handle_show(struct device *dev,
984 struct device_attribute *attr, char *buf)
986 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
988 return sprintf(buf, "%#x\n", memdev->device_handle);
990 static DEVICE_ATTR_RO(handle);
992 static ssize_t phys_id_show(struct device *dev,
993 struct device_attribute *attr, char *buf)
995 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
997 return sprintf(buf, "%#x\n", memdev->physical_id);
999 static DEVICE_ATTR_RO(phys_id);
1001 static ssize_t vendor_show(struct device *dev,
1002 struct device_attribute *attr, char *buf)
1004 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1006 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1008 static DEVICE_ATTR_RO(vendor);
1010 static ssize_t rev_id_show(struct device *dev,
1011 struct device_attribute *attr, char *buf)
1013 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1015 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1017 static DEVICE_ATTR_RO(rev_id);
1019 static ssize_t device_show(struct device *dev,
1020 struct device_attribute *attr, char *buf)
1022 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1024 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1026 static DEVICE_ATTR_RO(device);
1028 static ssize_t subsystem_vendor_show(struct device *dev,
1029 struct device_attribute *attr, char *buf)
1031 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1033 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1035 static DEVICE_ATTR_RO(subsystem_vendor);
1037 static ssize_t subsystem_rev_id_show(struct device *dev,
1038 struct device_attribute *attr, char *buf)
1040 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1042 return sprintf(buf, "0x%04x\n",
1043 be16_to_cpu(dcr->subsystem_revision_id));
1045 static DEVICE_ATTR_RO(subsystem_rev_id);
1047 static ssize_t subsystem_device_show(struct device *dev,
1048 struct device_attribute *attr, char *buf)
1050 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1052 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1054 static DEVICE_ATTR_RO(subsystem_device);
1056 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1058 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1061 if (nfit_mem->memdev_pmem)
1063 if (nfit_mem->memdev_bdw)
1068 static ssize_t format_show(struct device *dev,
1069 struct device_attribute *attr, char *buf)
1071 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1073 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->code));
1075 static DEVICE_ATTR_RO(format);
1077 static ssize_t format1_show(struct device *dev,
1078 struct device_attribute *attr, char *buf)
1081 ssize_t rc = -ENXIO;
1082 struct nfit_mem *nfit_mem;
1083 struct nfit_memdev *nfit_memdev;
1084 struct acpi_nfit_desc *acpi_desc;
1085 struct nvdimm *nvdimm = to_nvdimm(dev);
1086 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1088 nfit_mem = nvdimm_provider_data(nvdimm);
1089 acpi_desc = nfit_mem->acpi_desc;
1090 handle = to_nfit_memdev(dev)->device_handle;
1092 /* assumes DIMMs have at most 2 published interface codes */
1093 mutex_lock(&acpi_desc->init_mutex);
1094 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1095 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1096 struct nfit_dcr *nfit_dcr;
1098 if (memdev->device_handle != handle)
1101 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1102 if (nfit_dcr->dcr->region_index != memdev->region_index)
1104 if (nfit_dcr->dcr->code == dcr->code)
1106 rc = sprintf(buf, "%#x\n",
1107 be16_to_cpu(nfit_dcr->dcr->code));
1113 mutex_unlock(&acpi_desc->init_mutex);
1116 static DEVICE_ATTR_RO(format1);
1118 static ssize_t formats_show(struct device *dev,
1119 struct device_attribute *attr, char *buf)
1121 struct nvdimm *nvdimm = to_nvdimm(dev);
1123 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1125 static DEVICE_ATTR_RO(formats);
1127 static ssize_t serial_show(struct device *dev,
1128 struct device_attribute *attr, char *buf)
1130 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1132 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1134 static DEVICE_ATTR_RO(serial);
1136 static ssize_t family_show(struct device *dev,
1137 struct device_attribute *attr, char *buf)
1139 struct nvdimm *nvdimm = to_nvdimm(dev);
1140 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1142 if (nfit_mem->family < 0)
1144 return sprintf(buf, "%d\n", nfit_mem->family);
1146 static DEVICE_ATTR_RO(family);
1148 static ssize_t dsm_mask_show(struct device *dev,
1149 struct device_attribute *attr, char *buf)
1151 struct nvdimm *nvdimm = to_nvdimm(dev);
1152 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1154 if (nfit_mem->family < 0)
1156 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1158 static DEVICE_ATTR_RO(dsm_mask);
1160 static ssize_t flags_show(struct device *dev,
1161 struct device_attribute *attr, char *buf)
1163 u16 flags = to_nfit_memdev(dev)->flags;
1165 return sprintf(buf, "%s%s%s%s%s\n",
1166 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1167 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1168 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1169 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1170 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
1172 static DEVICE_ATTR_RO(flags);
1174 static ssize_t id_show(struct device *dev,
1175 struct device_attribute *attr, char *buf)
1177 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1179 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1180 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1181 be16_to_cpu(dcr->vendor_id),
1182 dcr->manufacturing_location,
1183 be16_to_cpu(dcr->manufacturing_date),
1184 be32_to_cpu(dcr->serial_number));
1186 return sprintf(buf, "%04x-%08x\n",
1187 be16_to_cpu(dcr->vendor_id),
1188 be32_to_cpu(dcr->serial_number));
1190 static DEVICE_ATTR_RO(id);
1192 static struct attribute *acpi_nfit_dimm_attributes[] = {
1193 &dev_attr_handle.attr,
1194 &dev_attr_phys_id.attr,
1195 &dev_attr_vendor.attr,
1196 &dev_attr_device.attr,
1197 &dev_attr_rev_id.attr,
1198 &dev_attr_subsystem_vendor.attr,
1199 &dev_attr_subsystem_device.attr,
1200 &dev_attr_subsystem_rev_id.attr,
1201 &dev_attr_format.attr,
1202 &dev_attr_formats.attr,
1203 &dev_attr_format1.attr,
1204 &dev_attr_serial.attr,
1205 &dev_attr_flags.attr,
1207 &dev_attr_family.attr,
1208 &dev_attr_dsm_mask.attr,
1212 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1213 struct attribute *a, int n)
1215 struct device *dev = container_of(kobj, struct device, kobj);
1216 struct nvdimm *nvdimm = to_nvdimm(dev);
1218 if (!to_nfit_dcr(dev))
1220 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1225 static struct attribute_group acpi_nfit_dimm_attribute_group = {
1227 .attrs = acpi_nfit_dimm_attributes,
1228 .is_visible = acpi_nfit_dimm_attr_visible,
1231 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1232 &nvdimm_attribute_group,
1233 &nd_device_attribute_group,
1234 &acpi_nfit_dimm_attribute_group,
1238 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1241 struct nfit_mem *nfit_mem;
1243 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1244 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1245 return nfit_mem->nvdimm;
1250 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1251 struct nfit_mem *nfit_mem, u32 device_handle)
1253 struct acpi_device *adev, *adev_dimm;
1254 struct device *dev = acpi_desc->dev;
1255 unsigned long dsm_mask;
1259 /* nfit test assumes 1:1 relationship between commands and dsms */
1260 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1261 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1262 adev = to_acpi_dev(acpi_desc);
1266 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1267 nfit_mem->adev = adev_dimm;
1269 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1271 return force_enable_dimms ? 0 : -ENODEV;
1275 * Until standardization materializes we need to consider 4
1276 * different command sets. Note, that checking for function0 (bit0)
1277 * tells us if any commands are reachable through this uuid.
1279 for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1280 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1283 /* limit the supported commands to those that are publicly documented */
1284 nfit_mem->family = i;
1285 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1287 if (disable_vendor_specific)
1288 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1289 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1290 dsm_mask = 0x1c3c76;
1291 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1293 if (disable_vendor_specific)
1294 dsm_mask &= ~(1 << 8);
1295 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1296 dsm_mask = 0xffffffff;
1298 dev_err(dev, "unknown dimm command family\n");
1299 nfit_mem->family = -1;
1300 return force_enable_dimms ? 0 : -ENODEV;
1303 uuid = to_nfit_uuid(nfit_mem->family);
1304 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1305 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1306 set_bit(i, &nfit_mem->dsm_mask);
1311 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1313 struct nfit_mem *nfit_mem;
1316 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1317 struct acpi_nfit_flush_address *flush;
1318 unsigned long flags = 0, cmd_mask;
1319 struct nvdimm *nvdimm;
1324 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1325 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1331 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1332 flags |= NDD_ALIASING;
1334 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1335 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1336 flags |= NDD_UNARMED;
1338 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1343 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1344 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1345 * userspace interface.
1347 cmd_mask = 1UL << ND_CMD_CALL;
1348 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1349 cmd_mask |= nfit_mem->dsm_mask;
1351 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1353 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1354 acpi_nfit_dimm_attribute_groups,
1355 flags, cmd_mask, flush ? flush->hint_count : 0,
1356 nfit_mem->flush_wpq);
1360 nfit_mem->nvdimm = nvdimm;
1363 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1366 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1367 nvdimm_name(nvdimm),
1368 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1369 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1370 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1371 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1375 return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1378 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1380 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1381 const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1382 struct acpi_device *adev;
1385 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1386 adev = to_acpi_dev(acpi_desc);
1390 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1391 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1392 set_bit(i, &nd_desc->cmd_mask);
1395 static ssize_t range_index_show(struct device *dev,
1396 struct device_attribute *attr, char *buf)
1398 struct nd_region *nd_region = to_nd_region(dev);
1399 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1401 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1403 static DEVICE_ATTR_RO(range_index);
1405 static struct attribute *acpi_nfit_region_attributes[] = {
1406 &dev_attr_range_index.attr,
1410 static struct attribute_group acpi_nfit_region_attribute_group = {
1412 .attrs = acpi_nfit_region_attributes,
1415 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1416 &nd_region_attribute_group,
1417 &nd_mapping_attribute_group,
1418 &nd_device_attribute_group,
1419 &nd_numa_attribute_group,
1420 &acpi_nfit_region_attribute_group,
1424 /* enough info to uniquely specify an interleave set */
1425 struct nfit_set_info {
1426 struct nfit_set_info_map {
1433 static size_t sizeof_nfit_set_info(int num_mappings)
1435 return sizeof(struct nfit_set_info)
1436 + num_mappings * sizeof(struct nfit_set_info_map);
1439 static int cmp_map(const void *m0, const void *m1)
1441 const struct nfit_set_info_map *map0 = m0;
1442 const struct nfit_set_info_map *map1 = m1;
1444 return memcmp(&map0->region_offset, &map1->region_offset,
1448 /* Retrieve the nth entry referencing this spa */
1449 static struct acpi_nfit_memory_map *memdev_from_spa(
1450 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1452 struct nfit_memdev *nfit_memdev;
1454 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1455 if (nfit_memdev->memdev->range_index == range_index)
1457 return nfit_memdev->memdev;
1461 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1462 struct nd_region_desc *ndr_desc,
1463 struct acpi_nfit_system_address *spa)
1465 int i, spa_type = nfit_spa_type(spa);
1466 struct device *dev = acpi_desc->dev;
1467 struct nd_interleave_set *nd_set;
1468 u16 nr = ndr_desc->num_mappings;
1469 struct nfit_set_info *info;
1471 if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1476 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1480 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1483 for (i = 0; i < nr; i++) {
1484 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1485 struct nfit_set_info_map *map = &info->mapping[i];
1486 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1487 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1488 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1489 spa->range_index, i);
1491 if (!memdev || !nfit_mem->dcr) {
1492 dev_err(dev, "%s: failed to find DCR\n", __func__);
1496 map->region_offset = memdev->region_offset;
1497 map->serial_number = nfit_mem->dcr->serial_number;
1500 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1502 nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1503 ndr_desc->nd_set = nd_set;
1504 devm_kfree(dev, info);
1509 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1511 struct acpi_nfit_interleave *idt = mmio->idt;
1512 u32 sub_line_offset, line_index, line_offset;
1513 u64 line_no, table_skip_count, table_offset;
1515 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1516 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1517 line_offset = idt->line_offset[line_index]
1519 table_offset = table_skip_count * mmio->table_size;
1521 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1524 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1526 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1527 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1529 if (mmio->num_lines)
1530 offset = to_interleave_offset(offset, mmio);
1532 return readl(mmio->addr.base + offset);
1535 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1536 resource_size_t dpa, unsigned int len, unsigned int write)
1539 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1542 BCW_OFFSET_MASK = (1ULL << 48)-1,
1544 BCW_LEN_MASK = (1ULL << 8) - 1,
1548 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1549 len = len >> L1_CACHE_SHIFT;
1550 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1551 cmd |= ((u64) write) << BCW_CMD_SHIFT;
1553 offset = nfit_blk->cmd_offset + mmio->size * bw;
1554 if (mmio->num_lines)
1555 offset = to_interleave_offset(offset, mmio);
1557 writeq(cmd, mmio->addr.base + offset);
1558 nvdimm_flush(nfit_blk->nd_region);
1560 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1561 readq(mmio->addr.base + offset);
1564 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1565 resource_size_t dpa, void *iobuf, size_t len, int rw,
1568 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1569 unsigned int copied = 0;
1573 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1574 + lane * mmio->size;
1575 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1580 if (mmio->num_lines) {
1583 offset = to_interleave_offset(base_offset + copied,
1585 div_u64_rem(offset, mmio->line_size, &line_offset);
1586 c = min_t(size_t, len, mmio->line_size - line_offset);
1588 offset = base_offset + nfit_blk->bdw_offset;
1593 memcpy_to_pmem(mmio->addr.aperture + offset,
1596 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1597 mmio_flush_range((void __force *)
1598 mmio->addr.aperture + offset, c);
1600 memcpy_from_pmem(iobuf + copied,
1601 mmio->addr.aperture + offset, c);
1609 nvdimm_flush(nfit_blk->nd_region);
1611 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1615 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1616 resource_size_t dpa, void *iobuf, u64 len, int rw)
1618 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1619 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1620 struct nd_region *nd_region = nfit_blk->nd_region;
1621 unsigned int lane, copied = 0;
1624 lane = nd_region_acquire_lane(nd_region);
1626 u64 c = min(len, mmio->size);
1628 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1629 iobuf + copied, c, rw, lane);
1636 nd_region_release_lane(nd_region, lane);
1641 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1642 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1645 mmio->num_lines = idt->line_count;
1646 mmio->line_size = idt->line_size;
1647 if (interleave_ways == 0)
1649 mmio->table_size = mmio->num_lines * interleave_ways
1656 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1657 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1659 struct nd_cmd_dimm_flags flags;
1662 memset(&flags, 0, sizeof(flags));
1663 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1664 sizeof(flags), NULL);
1666 if (rc >= 0 && flags.status == 0)
1667 nfit_blk->dimm_flags = flags.flags;
1668 else if (rc == -ENOTTY) {
1669 /* fall back to a conservative default */
1670 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1678 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1681 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1682 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1683 struct nfit_blk_mmio *mmio;
1684 struct nfit_blk *nfit_blk;
1685 struct nfit_mem *nfit_mem;
1686 struct nvdimm *nvdimm;
1689 nvdimm = nd_blk_region_to_dimm(ndbr);
1690 nfit_mem = nvdimm_provider_data(nvdimm);
1691 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1692 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1693 nfit_mem ? "" : " nfit_mem",
1694 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1695 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1699 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1702 nd_blk_region_set_provider_data(ndbr, nfit_blk);
1703 nfit_blk->nd_region = to_nd_region(dev);
1705 /* map block aperture memory */
1706 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1707 mmio = &nfit_blk->mmio[BDW];
1708 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
1709 nfit_mem->spa_bdw->length, ARCH_MEMREMAP_PMEM);
1710 if (!mmio->addr.base) {
1711 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1712 nvdimm_name(nvdimm));
1715 mmio->size = nfit_mem->bdw->size;
1716 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1717 mmio->idt = nfit_mem->idt_bdw;
1718 mmio->spa = nfit_mem->spa_bdw;
1719 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1720 nfit_mem->memdev_bdw->interleave_ways);
1722 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1723 __func__, nvdimm_name(nvdimm));
1727 /* map block control memory */
1728 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1729 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1730 mmio = &nfit_blk->mmio[DCR];
1731 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
1732 nfit_mem->spa_dcr->length);
1733 if (!mmio->addr.base) {
1734 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1735 nvdimm_name(nvdimm));
1738 mmio->size = nfit_mem->dcr->window_size;
1739 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1740 mmio->idt = nfit_mem->idt_dcr;
1741 mmio->spa = nfit_mem->spa_dcr;
1742 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1743 nfit_mem->memdev_dcr->interleave_ways);
1745 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1746 __func__, nvdimm_name(nvdimm));
1750 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1752 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1753 __func__, nvdimm_name(nvdimm));
1757 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
1758 dev_warn(dev, "unable to guarantee persistence of writes\n");
1760 if (mmio->line_size == 0)
1763 if ((u32) nfit_blk->cmd_offset % mmio->line_size
1764 + 8 > mmio->line_size) {
1765 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1767 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1768 + 8 > mmio->line_size) {
1769 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1776 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1777 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1779 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1780 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1783 cmd->address = spa->address;
1784 cmd->length = spa->length;
1785 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1786 sizeof(*cmd), &cmd_rc);
1792 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1796 struct nd_cmd_ars_start ars_start;
1797 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1798 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1800 memset(&ars_start, 0, sizeof(ars_start));
1801 ars_start.address = spa->address;
1802 ars_start.length = spa->length;
1803 if (nfit_spa_type(spa) == NFIT_SPA_PM)
1804 ars_start.type = ND_ARS_PERSISTENT;
1805 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1806 ars_start.type = ND_ARS_VOLATILE;
1810 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1811 sizeof(ars_start), &cmd_rc);
1818 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1821 struct nd_cmd_ars_start ars_start;
1822 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1823 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1825 memset(&ars_start, 0, sizeof(ars_start));
1826 ars_start.address = ars_status->restart_address;
1827 ars_start.length = ars_status->restart_length;
1828 ars_start.type = ars_status->type;
1829 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1830 sizeof(ars_start), &cmd_rc);
1836 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1838 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1839 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1842 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1843 acpi_desc->ars_status_size, &cmd_rc);
1849 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1850 struct nd_cmd_ars_status *ars_status)
1855 for (i = 0; i < ars_status->num_records; i++) {
1856 rc = nvdimm_bus_add_poison(nvdimm_bus,
1857 ars_status->records[i].err_address,
1858 ars_status->records[i].length);
1866 static void acpi_nfit_remove_resource(void *data)
1868 struct resource *res = data;
1870 remove_resource(res);
1873 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1874 struct nd_region_desc *ndr_desc)
1876 struct resource *res, *nd_res = ndr_desc->res;
1879 /* No operation if the region is already registered as PMEM */
1880 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1881 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1882 if (is_pmem == REGION_INTERSECTS)
1885 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1889 res->name = "Persistent Memory";
1890 res->start = nd_res->start;
1891 res->end = nd_res->end;
1892 res->flags = IORESOURCE_MEM;
1893 res->desc = IORES_DESC_PERSISTENT_MEMORY;
1895 ret = insert_resource(&iomem_resource, res);
1899 ret = devm_add_action_or_reset(acpi_desc->dev,
1900 acpi_nfit_remove_resource,
1908 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1909 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1910 struct acpi_nfit_memory_map *memdev,
1911 struct nfit_spa *nfit_spa)
1913 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1914 memdev->device_handle);
1915 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1916 struct nd_blk_region_desc *ndbr_desc;
1917 struct nfit_mem *nfit_mem;
1921 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1922 spa->range_index, memdev->device_handle);
1926 nd_mapping->nvdimm = nvdimm;
1927 switch (nfit_spa_type(spa)) {
1929 case NFIT_SPA_VOLATILE:
1930 nd_mapping->start = memdev->address;
1931 nd_mapping->size = memdev->region_size;
1934 nfit_mem = nvdimm_provider_data(nvdimm);
1935 if (!nfit_mem || !nfit_mem->bdw) {
1936 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1937 spa->range_index, nvdimm_name(nvdimm));
1939 nd_mapping->size = nfit_mem->bdw->capacity;
1940 nd_mapping->start = nfit_mem->bdw->start_address;
1941 ndr_desc->num_lanes = nfit_mem->bdw->windows;
1945 ndr_desc->nd_mapping = nd_mapping;
1946 ndr_desc->num_mappings = blk_valid;
1947 ndbr_desc = to_blk_region_desc(ndr_desc);
1948 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1949 ndbr_desc->do_io = acpi_desc->blk_do_io;
1950 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1952 if (!nfit_spa->nd_region)
1960 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
1962 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
1963 nfit_spa_type(spa) == NFIT_SPA_VCD ||
1964 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
1965 nfit_spa_type(spa) == NFIT_SPA_PCD);
1968 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1969 struct nfit_spa *nfit_spa)
1971 static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1972 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1973 struct nd_blk_region_desc ndbr_desc;
1974 struct nd_region_desc *ndr_desc;
1975 struct nfit_memdev *nfit_memdev;
1976 struct nvdimm_bus *nvdimm_bus;
1977 struct resource res;
1980 if (nfit_spa->nd_region)
1983 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
1984 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
1989 memset(&res, 0, sizeof(res));
1990 memset(&nd_mappings, 0, sizeof(nd_mappings));
1991 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
1992 res.start = spa->address;
1993 res.end = res.start + spa->length - 1;
1994 ndr_desc = &ndbr_desc.ndr_desc;
1995 ndr_desc->res = &res;
1996 ndr_desc->provider_data = nfit_spa;
1997 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
1998 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
1999 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2000 spa->proximity_domain);
2002 ndr_desc->numa_node = NUMA_NO_NODE;
2004 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2005 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2006 struct nd_mapping *nd_mapping;
2008 if (memdev->range_index != spa->range_index)
2010 if (count >= ND_MAX_MAPPINGS) {
2011 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2012 spa->range_index, ND_MAX_MAPPINGS);
2015 nd_mapping = &nd_mappings[count++];
2016 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
2022 ndr_desc->nd_mapping = nd_mappings;
2023 ndr_desc->num_mappings = count;
2024 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2028 nvdimm_bus = acpi_desc->nvdimm_bus;
2029 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2030 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2032 dev_warn(acpi_desc->dev,
2033 "failed to insert pmem resource to iomem: %d\n",
2038 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2040 if (!nfit_spa->nd_region)
2042 } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
2043 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2045 if (!nfit_spa->nd_region)
2047 } else if (nfit_spa_is_virtual(spa)) {
2048 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2050 if (!nfit_spa->nd_region)
2056 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2057 nfit_spa->spa->range_index);
2061 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2064 struct device *dev = acpi_desc->dev;
2065 struct nd_cmd_ars_status *ars_status;
2067 if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2068 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2072 if (acpi_desc->ars_status)
2073 devm_kfree(dev, acpi_desc->ars_status);
2074 acpi_desc->ars_status = NULL;
2075 ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2078 acpi_desc->ars_status = ars_status;
2079 acpi_desc->ars_status_size = max_ars;
2083 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2084 struct nfit_spa *nfit_spa)
2086 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2089 if (!nfit_spa->max_ars) {
2090 struct nd_cmd_ars_cap ars_cap;
2092 memset(&ars_cap, 0, sizeof(ars_cap));
2093 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2096 nfit_spa->max_ars = ars_cap.max_ars_out;
2097 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2098 /* check that the supported scrub types match the spa type */
2099 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2100 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2102 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2103 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2107 if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2110 rc = ars_get_status(acpi_desc);
2111 if (rc < 0 && rc != -ENOSPC)
2114 if (ars_status_process_records(acpi_desc->nvdimm_bus,
2115 acpi_desc->ars_status))
2121 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2122 struct nfit_spa *nfit_spa)
2124 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2125 unsigned int overflow_retry = scrub_overflow_abort;
2126 u64 init_ars_start = 0, init_ars_len = 0;
2127 struct device *dev = acpi_desc->dev;
2128 unsigned int tmo = scrub_timeout;
2131 if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2134 rc = ars_start(acpi_desc, nfit_spa);
2136 * If we timed out the initial scan we'll still be busy here,
2137 * and will wait another timeout before giving up permanently.
2139 if (rc < 0 && rc != -EBUSY)
2143 u64 ars_start, ars_len;
2145 if (acpi_desc->cancel)
2147 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2150 if (rc == -EBUSY && !tmo) {
2151 dev_warn(dev, "range %d ars timeout, aborting\n",
2158 * Note, entries may be appended to the list
2159 * while the lock is dropped, but the workqueue
2160 * being active prevents entries being deleted /
2163 mutex_unlock(&acpi_desc->init_mutex);
2166 mutex_lock(&acpi_desc->init_mutex);
2170 /* we got some results, but there are more pending... */
2171 if (rc == -ENOSPC && overflow_retry--) {
2172 if (!init_ars_len) {
2173 init_ars_len = acpi_desc->ars_status->length;
2174 init_ars_start = acpi_desc->ars_status->address;
2176 rc = ars_continue(acpi_desc);
2180 dev_warn(dev, "range %d ars continuation failed\n",
2186 ars_start = init_ars_start;
2187 ars_len = init_ars_len;
2189 ars_start = acpi_desc->ars_status->address;
2190 ars_len = acpi_desc->ars_status->length;
2192 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2193 spa->range_index, ars_start, ars_len);
2194 /* notify the region about new poison entries */
2195 nvdimm_region_notify(nfit_spa->nd_region,
2196 NVDIMM_REVALIDATE_POISON);
2201 static void acpi_nfit_scrub(struct work_struct *work)
2204 u64 init_scrub_length = 0;
2205 struct nfit_spa *nfit_spa;
2206 u64 init_scrub_address = 0;
2207 bool init_ars_done = false;
2208 struct acpi_nfit_desc *acpi_desc;
2209 unsigned int tmo = scrub_timeout;
2210 unsigned int overflow_retry = scrub_overflow_abort;
2212 acpi_desc = container_of(work, typeof(*acpi_desc), work);
2213 dev = acpi_desc->dev;
2216 * We scrub in 2 phases. The first phase waits for any platform
2217 * firmware initiated scrubs to complete and then we go search for the
2218 * affected spa regions to mark them scanned. In the second phase we
2219 * initiate a directed scrub for every range that was not scrubbed in
2220 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2221 * the first phase, but really only care about running phase 2, where
2222 * regions can be notified of new poison.
2225 /* process platform firmware initiated scrubs */
2227 mutex_lock(&acpi_desc->init_mutex);
2228 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2229 struct nd_cmd_ars_status *ars_status;
2230 struct acpi_nfit_system_address *spa;
2231 u64 ars_start, ars_len;
2234 if (acpi_desc->cancel)
2237 if (nfit_spa->nd_region)
2240 if (init_ars_done) {
2242 * No need to re-query, we're now just
2243 * reconciling all the ranges covered by the
2248 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2250 if (rc == -ENOTTY) {
2251 /* no ars capability, just register spa and move on */
2252 acpi_nfit_register_region(acpi_desc, nfit_spa);
2256 if (rc == -EBUSY && !tmo) {
2257 /* fallthrough to directed scrub in phase 2 */
2258 dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2260 } else if (rc == -EBUSY) {
2261 mutex_unlock(&acpi_desc->init_mutex);
2267 /* we got some results, but there are more pending... */
2268 if (rc == -ENOSPC && overflow_retry--) {
2269 ars_status = acpi_desc->ars_status;
2271 * Record the original scrub range, so that we
2272 * can recall all the ranges impacted by the
2275 if (!init_scrub_length) {
2276 init_scrub_length = ars_status->length;
2277 init_scrub_address = ars_status->address;
2279 rc = ars_continue(acpi_desc);
2281 mutex_unlock(&acpi_desc->init_mutex);
2288 * Initial scrub failed, we'll give it one more
2294 /* We got some final results, record completed ranges */
2295 ars_status = acpi_desc->ars_status;
2296 if (init_scrub_length) {
2297 ars_start = init_scrub_address;
2298 ars_len = ars_start + init_scrub_length;
2300 ars_start = ars_status->address;
2301 ars_len = ars_status->length;
2303 spa = nfit_spa->spa;
2305 if (!init_ars_done) {
2306 init_ars_done = true;
2307 dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2308 ars_start, ars_len);
2310 if (ars_start <= spa->address && ars_start + ars_len
2311 >= spa->address + spa->length)
2312 acpi_nfit_register_region(acpi_desc, nfit_spa);
2316 * For all the ranges not covered by an initial scrub we still
2317 * want to see if there are errors, but it's ok to discover them
2320 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2322 * Flag all the ranges that still need scrubbing, but
2323 * register them now to make data available.
2325 if (!nfit_spa->nd_region) {
2326 nfit_spa->ars_required = 1;
2327 acpi_nfit_register_region(acpi_desc, nfit_spa);
2331 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2332 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2333 acpi_desc->scrub_count++;
2334 if (acpi_desc->scrub_count_state)
2335 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2336 mutex_unlock(&acpi_desc->init_mutex);
2339 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2341 struct nfit_spa *nfit_spa;
2344 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2345 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2346 /* BLK regions don't need to wait for ars results */
2347 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2352 queue_work(nfit_wq, &acpi_desc->work);
2356 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2357 struct nfit_table_prev *prev)
2359 struct device *dev = acpi_desc->dev;
2361 if (!list_empty(&prev->spas) ||
2362 !list_empty(&prev->memdevs) ||
2363 !list_empty(&prev->dcrs) ||
2364 !list_empty(&prev->bdws) ||
2365 !list_empty(&prev->idts) ||
2366 !list_empty(&prev->flushes)) {
2367 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2373 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
2375 struct device *dev = acpi_desc->dev;
2376 struct kernfs_node *nfit;
2377 struct device *bus_dev;
2379 if (!ars_supported(acpi_desc->nvdimm_bus))
2382 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2383 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
2385 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
2388 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
2390 if (!acpi_desc->scrub_count_state) {
2391 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
2398 static void acpi_nfit_destruct(void *data)
2400 struct acpi_nfit_desc *acpi_desc = data;
2401 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2403 acpi_desc->cancel = 1;
2405 * Bounce the nvdimm bus lock to make sure any in-flight
2406 * acpi_nfit_ars_rescan() submissions have had a chance to
2407 * either submit or see ->cancel set.
2409 device_lock(bus_dev);
2410 device_unlock(bus_dev);
2412 flush_workqueue(nfit_wq);
2413 if (acpi_desc->scrub_count_state)
2414 sysfs_put(acpi_desc->scrub_count_state);
2415 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2416 acpi_desc->nvdimm_bus = NULL;
2419 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
2421 struct device *dev = acpi_desc->dev;
2422 struct nfit_table_prev prev;
2426 if (!acpi_desc->nvdimm_bus) {
2427 acpi_nfit_init_dsms(acpi_desc);
2429 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
2430 &acpi_desc->nd_desc);
2431 if (!acpi_desc->nvdimm_bus)
2434 rc = devm_add_action_or_reset(dev, acpi_nfit_destruct,
2439 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
2444 mutex_lock(&acpi_desc->init_mutex);
2446 INIT_LIST_HEAD(&prev.spas);
2447 INIT_LIST_HEAD(&prev.memdevs);
2448 INIT_LIST_HEAD(&prev.dcrs);
2449 INIT_LIST_HEAD(&prev.bdws);
2450 INIT_LIST_HEAD(&prev.idts);
2451 INIT_LIST_HEAD(&prev.flushes);
2453 list_cut_position(&prev.spas, &acpi_desc->spas,
2454 acpi_desc->spas.prev);
2455 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2456 acpi_desc->memdevs.prev);
2457 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2458 acpi_desc->dcrs.prev);
2459 list_cut_position(&prev.bdws, &acpi_desc->bdws,
2460 acpi_desc->bdws.prev);
2461 list_cut_position(&prev.idts, &acpi_desc->idts,
2462 acpi_desc->idts.prev);
2463 list_cut_position(&prev.flushes, &acpi_desc->flushes,
2464 acpi_desc->flushes.prev);
2467 while (!IS_ERR_OR_NULL(data))
2468 data = add_table(acpi_desc, &prev, data, end);
2471 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2477 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2481 rc = nfit_mem_init(acpi_desc);
2485 rc = acpi_nfit_register_dimms(acpi_desc);
2489 rc = acpi_nfit_register_regions(acpi_desc);
2492 mutex_unlock(&acpi_desc->init_mutex);
2495 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2497 struct acpi_nfit_flush_work {
2498 struct work_struct work;
2499 struct completion cmp;
2502 static void flush_probe(struct work_struct *work)
2504 struct acpi_nfit_flush_work *flush;
2506 flush = container_of(work, typeof(*flush), work);
2507 complete(&flush->cmp);
2510 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2512 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2513 struct device *dev = acpi_desc->dev;
2514 struct acpi_nfit_flush_work flush;
2516 /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2521 * Scrub work could take 10s of seconds, userspace may give up so we
2522 * need to be interruptible while waiting.
2524 INIT_WORK_ONSTACK(&flush.work, flush_probe);
2525 COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2526 queue_work(nfit_wq, &flush.work);
2527 return wait_for_completion_interruptible(&flush.cmp);
2530 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2531 struct nvdimm *nvdimm, unsigned int cmd)
2533 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2537 if (cmd != ND_CMD_ARS_START)
2541 * The kernel and userspace may race to initiate a scrub, but
2542 * the scrub thread is prepared to lose that initial race. It
2543 * just needs guarantees that any ars it initiates are not
2544 * interrupted by any intervening start reqeusts from userspace.
2546 if (work_busy(&acpi_desc->work))
2552 static int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc)
2554 struct device *dev = acpi_desc->dev;
2555 struct nfit_spa *nfit_spa;
2557 if (work_busy(&acpi_desc->work))
2560 if (acpi_desc->cancel)
2563 mutex_lock(&acpi_desc->init_mutex);
2564 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2565 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2567 if (nfit_spa_type(spa) != NFIT_SPA_PM)
2570 nfit_spa->ars_required = 1;
2572 queue_work(nfit_wq, &acpi_desc->work);
2573 dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
2574 mutex_unlock(&acpi_desc->init_mutex);
2579 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2581 struct nvdimm_bus_descriptor *nd_desc;
2583 dev_set_drvdata(dev, acpi_desc);
2584 acpi_desc->dev = dev;
2585 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2586 nd_desc = &acpi_desc->nd_desc;
2587 nd_desc->provider_name = "ACPI.NFIT";
2588 nd_desc->module = THIS_MODULE;
2589 nd_desc->ndctl = acpi_nfit_ctl;
2590 nd_desc->flush_probe = acpi_nfit_flush_probe;
2591 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2592 nd_desc->attr_groups = acpi_nfit_attribute_groups;
2594 INIT_LIST_HEAD(&acpi_desc->spas);
2595 INIT_LIST_HEAD(&acpi_desc->dcrs);
2596 INIT_LIST_HEAD(&acpi_desc->bdws);
2597 INIT_LIST_HEAD(&acpi_desc->idts);
2598 INIT_LIST_HEAD(&acpi_desc->flushes);
2599 INIT_LIST_HEAD(&acpi_desc->memdevs);
2600 INIT_LIST_HEAD(&acpi_desc->dimms);
2601 mutex_init(&acpi_desc->init_mutex);
2602 INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2604 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2606 static int acpi_nfit_add(struct acpi_device *adev)
2608 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2609 struct acpi_nfit_desc *acpi_desc;
2610 struct device *dev = &adev->dev;
2611 struct acpi_table_header *tbl;
2612 acpi_status status = AE_OK;
2616 status = acpi_get_table_with_size(ACPI_SIG_NFIT, 0, &tbl, &sz);
2617 if (ACPI_FAILURE(status)) {
2618 /* This is ok, we could have an nvdimm hotplugged later */
2619 dev_dbg(dev, "failed to find NFIT at startup\n");
2623 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2626 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2628 /* Save the acpi header for exporting the revision via sysfs */
2629 acpi_desc->acpi_header = *tbl;
2631 /* Evaluate _FIT and override with that if present */
2632 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2633 if (ACPI_SUCCESS(status) && buf.length > 0) {
2634 union acpi_object *obj = buf.pointer;
2636 if (obj->type == ACPI_TYPE_BUFFER)
2637 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
2638 obj->buffer.length);
2640 dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2641 __func__, (int) obj->type);
2644 /* skip over the lead-in header table */
2645 rc = acpi_nfit_init(acpi_desc, (void *) tbl
2646 + sizeof(struct acpi_table_nfit),
2647 sz - sizeof(struct acpi_table_nfit));
2651 static int acpi_nfit_remove(struct acpi_device *adev)
2653 /* see acpi_nfit_destruct */
2657 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2659 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2660 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2661 struct device *dev = &adev->dev;
2662 union acpi_object *obj;
2666 dev_dbg(dev, "%s: event: %d\n", __func__, event);
2670 /* dev->driver may be null if we're being removed */
2671 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2676 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2679 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2682 * Finish previous registration before considering new
2685 flush_workqueue(nfit_wq);
2689 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2690 if (ACPI_FAILURE(status)) {
2691 dev_err(dev, "failed to evaluate _FIT\n");
2696 if (obj->type == ACPI_TYPE_BUFFER) {
2697 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
2698 obj->buffer.length);
2700 dev_err(dev, "failed to merge updated NFIT\n");
2702 dev_err(dev, "Invalid _FIT\n");
2709 static const struct acpi_device_id acpi_nfit_ids[] = {
2713 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2715 static struct acpi_driver acpi_nfit_driver = {
2716 .name = KBUILD_MODNAME,
2717 .ids = acpi_nfit_ids,
2719 .add = acpi_nfit_add,
2720 .remove = acpi_nfit_remove,
2721 .notify = acpi_nfit_notify,
2725 static __init int nfit_init(void)
2727 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2728 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2729 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2730 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2731 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2732 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2733 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2735 acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2736 acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2737 acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2738 acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2739 acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2740 acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2741 acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2742 acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2743 acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2744 acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2745 acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2746 acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2747 acpi_str_to_uuid(UUID_NFIT_DIMM_N_MSFT, nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
2749 nfit_wq = create_singlethread_workqueue("nfit");
2753 return acpi_bus_register_driver(&acpi_nfit_driver);
2756 static __exit void nfit_exit(void)
2758 acpi_bus_unregister_driver(&acpi_nfit_driver);
2759 destroy_workqueue(nfit_wq);
2762 module_init(nfit_init);
2763 module_exit(nfit_exit);
2764 MODULE_LICENSE("GPL v2");
2765 MODULE_AUTHOR("Intel Corporation");