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/delay.h>
19 #include <linux/list.h>
20 #include <linux/acpi.h>
21 #include <linux/sort.h>
22 #include <linux/pmem.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\n");
53 static struct workqueue_struct *nfit_wq;
55 struct nfit_table_prev {
56 struct list_head spas;
57 struct list_head memdevs;
58 struct list_head dcrs;
59 struct list_head bdws;
60 struct list_head idts;
61 struct list_head flushes;
64 static u8 nfit_uuid[NFIT_UUID_MAX][16];
66 const u8 *to_nfit_uuid(enum nfit_uuids id)
70 EXPORT_SYMBOL(to_nfit_uuid);
72 static struct acpi_nfit_desc *to_acpi_nfit_desc(
73 struct nvdimm_bus_descriptor *nd_desc)
75 return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
78 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
80 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
83 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
86 if (!nd_desc->provider_name
87 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
90 return to_acpi_device(acpi_desc->dev);
93 static int xlat_status(void *buf, unsigned int cmd)
95 struct nd_cmd_clear_error *clear_err;
96 struct nd_cmd_ars_status *ars_status;
97 struct nd_cmd_ars_start *ars_start;
98 struct nd_cmd_ars_cap *ars_cap;
104 if ((ars_cap->status & 0xffff) == NFIT_ARS_CAP_NONE)
108 if (ars_cap->status & 0xffff)
111 /* No supported scan types for this range */
112 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
113 if ((ars_cap->status >> 16 & flags) == 0)
116 case ND_CMD_ARS_START:
118 /* ARS is in progress */
119 if ((ars_start->status & 0xffff) == NFIT_ARS_START_BUSY)
123 if (ars_start->status & 0xffff)
126 case ND_CMD_ARS_STATUS:
129 if (ars_status->status & 0xffff)
131 /* Check extended status (Upper two bytes) */
132 if (ars_status->status == NFIT_ARS_STATUS_DONE)
135 /* ARS is in progress */
136 if (ars_status->status == NFIT_ARS_STATUS_BUSY)
139 /* No ARS performed for the current boot */
140 if (ars_status->status == NFIT_ARS_STATUS_NONE)
144 * ARS interrupted, either we overflowed or some other
145 * agent wants the scan to stop. If we didn't overflow
146 * then just continue with the returned results.
148 if (ars_status->status == NFIT_ARS_STATUS_INTR) {
149 if (ars_status->flags & NFIT_ARS_F_OVERFLOW)
155 if (ars_status->status >> 16)
158 case ND_CMD_CLEAR_ERROR:
160 if (clear_err->status & 0xffff)
162 if (!clear_err->cleared)
164 if (clear_err->length > clear_err->cleared)
165 return clear_err->cleared;
174 static int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc,
175 struct nvdimm *nvdimm, unsigned int cmd, void *buf,
176 unsigned int buf_len, int *cmd_rc)
178 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
179 union acpi_object in_obj, in_buf, *out_obj;
180 const struct nd_cmd_desc *desc = NULL;
181 struct device *dev = acpi_desc->dev;
182 struct nd_cmd_pkg *call_pkg = NULL;
183 const char *cmd_name, *dimm_name;
184 unsigned long cmd_mask, dsm_mask;
192 if (cmd == ND_CMD_CALL) {
194 func = call_pkg->nd_command;
198 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
199 struct acpi_device *adev = nfit_mem->adev;
203 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
206 dimm_name = nvdimm_name(nvdimm);
207 cmd_name = nvdimm_cmd_name(cmd);
208 cmd_mask = nvdimm_cmd_mask(nvdimm);
209 dsm_mask = nfit_mem->dsm_mask;
210 desc = nd_cmd_dimm_desc(cmd);
211 uuid = to_nfit_uuid(nfit_mem->family);
212 handle = adev->handle;
214 struct acpi_device *adev = to_acpi_dev(acpi_desc);
216 cmd_name = nvdimm_bus_cmd_name(cmd);
217 cmd_mask = nd_desc->cmd_mask;
219 desc = nd_cmd_bus_desc(cmd);
220 uuid = to_nfit_uuid(NFIT_DEV_BUS);
221 handle = adev->handle;
225 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
228 if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
231 in_obj.type = ACPI_TYPE_PACKAGE;
232 in_obj.package.count = 1;
233 in_obj.package.elements = &in_buf;
234 in_buf.type = ACPI_TYPE_BUFFER;
235 in_buf.buffer.pointer = buf;
236 in_buf.buffer.length = 0;
238 /* libnvdimm has already validated the input envelope */
239 for (i = 0; i < desc->in_num; i++)
240 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
244 /* skip over package wrapper */
245 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
246 in_buf.buffer.length = call_pkg->nd_size_in;
249 if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
250 dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
251 __func__, dimm_name, cmd, func,
252 in_buf.buffer.length);
253 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
254 in_buf.buffer.pointer,
255 min_t(u32, 256, in_buf.buffer.length), true);
258 out_obj = acpi_evaluate_dsm(handle, uuid, 1, func, &in_obj);
260 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
266 call_pkg->nd_fw_size = out_obj->buffer.length;
267 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
268 out_obj->buffer.pointer,
269 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
273 * Need to support FW function w/o known size in advance.
274 * Caller can determine required size based upon nd_fw_size.
275 * If we return an error (like elsewhere) then caller wouldn't
276 * be able to rely upon data returned to make calculation.
281 if (out_obj->package.type != ACPI_TYPE_BUFFER) {
282 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
283 __func__, dimm_name, cmd_name, out_obj->type);
288 if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
289 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__,
290 dimm_name, cmd_name, out_obj->buffer.length);
291 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
292 4, out_obj->buffer.pointer, min_t(u32, 128,
293 out_obj->buffer.length), true);
296 for (i = 0, offset = 0; i < desc->out_num; i++) {
297 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
298 (u32 *) out_obj->buffer.pointer);
300 if (offset + out_size > out_obj->buffer.length) {
301 dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
302 __func__, dimm_name, cmd_name, i);
306 if (in_buf.buffer.length + offset + out_size > buf_len) {
307 dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
308 __func__, dimm_name, cmd_name, i);
312 memcpy(buf + in_buf.buffer.length + offset,
313 out_obj->buffer.pointer + offset, out_size);
316 if (offset + in_buf.buffer.length < buf_len) {
319 * status valid, return the number of bytes left
320 * unfilled in the output buffer
322 rc = buf_len - offset - in_buf.buffer.length;
324 *cmd_rc = xlat_status(buf, cmd);
326 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
327 __func__, dimm_name, cmd_name, buf_len,
334 *cmd_rc = xlat_status(buf, cmd);
343 static const char *spa_type_name(u16 type)
345 static const char *to_name[] = {
346 [NFIT_SPA_VOLATILE] = "volatile",
347 [NFIT_SPA_PM] = "pmem",
348 [NFIT_SPA_DCR] = "dimm-control-region",
349 [NFIT_SPA_BDW] = "block-data-window",
350 [NFIT_SPA_VDISK] = "volatile-disk",
351 [NFIT_SPA_VCD] = "volatile-cd",
352 [NFIT_SPA_PDISK] = "persistent-disk",
353 [NFIT_SPA_PCD] = "persistent-cd",
357 if (type > NFIT_SPA_PCD)
360 return to_name[type];
363 static int nfit_spa_type(struct acpi_nfit_system_address *spa)
367 for (i = 0; i < NFIT_UUID_MAX; i++)
368 if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
373 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
374 struct nfit_table_prev *prev,
375 struct acpi_nfit_system_address *spa)
377 size_t length = min_t(size_t, sizeof(*spa), spa->header.length);
378 struct device *dev = acpi_desc->dev;
379 struct nfit_spa *nfit_spa;
381 list_for_each_entry(nfit_spa, &prev->spas, list) {
382 if (memcmp(nfit_spa->spa, spa, length) == 0) {
383 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
388 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa), GFP_KERNEL);
391 INIT_LIST_HEAD(&nfit_spa->list);
393 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
394 dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
396 spa_type_name(nfit_spa_type(spa)));
400 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
401 struct nfit_table_prev *prev,
402 struct acpi_nfit_memory_map *memdev)
404 size_t length = min_t(size_t, sizeof(*memdev), memdev->header.length);
405 struct device *dev = acpi_desc->dev;
406 struct nfit_memdev *nfit_memdev;
408 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
409 if (memcmp(nfit_memdev->memdev, memdev, length) == 0) {
410 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
414 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev), GFP_KERNEL);
417 INIT_LIST_HEAD(&nfit_memdev->list);
418 nfit_memdev->memdev = memdev;
419 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
420 dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
421 __func__, memdev->device_handle, memdev->range_index,
422 memdev->region_index);
426 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
427 struct nfit_table_prev *prev,
428 struct acpi_nfit_control_region *dcr)
430 size_t length = min_t(size_t, sizeof(*dcr), dcr->header.length);
431 struct device *dev = acpi_desc->dev;
432 struct nfit_dcr *nfit_dcr;
434 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
435 if (memcmp(nfit_dcr->dcr, dcr, length) == 0) {
436 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
440 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr), GFP_KERNEL);
443 INIT_LIST_HEAD(&nfit_dcr->list);
445 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
446 dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
447 dcr->region_index, dcr->windows);
451 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
452 struct nfit_table_prev *prev,
453 struct acpi_nfit_data_region *bdw)
455 size_t length = min_t(size_t, sizeof(*bdw), bdw->header.length);
456 struct device *dev = acpi_desc->dev;
457 struct nfit_bdw *nfit_bdw;
459 list_for_each_entry(nfit_bdw, &prev->bdws, list)
460 if (memcmp(nfit_bdw->bdw, bdw, length) == 0) {
461 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
465 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw), GFP_KERNEL);
468 INIT_LIST_HEAD(&nfit_bdw->list);
470 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
471 dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
472 bdw->region_index, bdw->windows);
476 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
477 struct nfit_table_prev *prev,
478 struct acpi_nfit_interleave *idt)
480 size_t length = min_t(size_t, sizeof(*idt), idt->header.length);
481 struct device *dev = acpi_desc->dev;
482 struct nfit_idt *nfit_idt;
484 list_for_each_entry(nfit_idt, &prev->idts, list)
485 if (memcmp(nfit_idt->idt, idt, length) == 0) {
486 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
490 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt), GFP_KERNEL);
493 INIT_LIST_HEAD(&nfit_idt->list);
495 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
496 dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
497 idt->interleave_index, idt->line_count);
501 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
502 struct nfit_table_prev *prev,
503 struct acpi_nfit_flush_address *flush)
505 size_t length = min_t(size_t, sizeof(*flush), flush->header.length);
506 struct device *dev = acpi_desc->dev;
507 struct nfit_flush *nfit_flush;
509 list_for_each_entry(nfit_flush, &prev->flushes, list)
510 if (memcmp(nfit_flush->flush, flush, length) == 0) {
511 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
515 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush), GFP_KERNEL);
518 INIT_LIST_HEAD(&nfit_flush->list);
519 nfit_flush->flush = flush;
520 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
521 dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
522 flush->device_handle, flush->hint_count);
526 static void *add_table(struct acpi_nfit_desc *acpi_desc,
527 struct nfit_table_prev *prev, void *table, const void *end)
529 struct device *dev = acpi_desc->dev;
530 struct acpi_nfit_header *hdr;
531 void *err = ERR_PTR(-ENOMEM);
538 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
544 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
545 if (!add_spa(acpi_desc, prev, table))
548 case ACPI_NFIT_TYPE_MEMORY_MAP:
549 if (!add_memdev(acpi_desc, prev, table))
552 case ACPI_NFIT_TYPE_CONTROL_REGION:
553 if (!add_dcr(acpi_desc, prev, table))
556 case ACPI_NFIT_TYPE_DATA_REGION:
557 if (!add_bdw(acpi_desc, prev, table))
560 case ACPI_NFIT_TYPE_INTERLEAVE:
561 if (!add_idt(acpi_desc, prev, table))
564 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
565 if (!add_flush(acpi_desc, prev, table))
568 case ACPI_NFIT_TYPE_SMBIOS:
569 dev_dbg(dev, "%s: smbios\n", __func__);
572 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
576 return table + hdr->length;
579 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
580 struct nfit_mem *nfit_mem)
582 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
583 u16 dcr = nfit_mem->dcr->region_index;
584 struct nfit_spa *nfit_spa;
586 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
587 u16 range_index = nfit_spa->spa->range_index;
588 int type = nfit_spa_type(nfit_spa->spa);
589 struct nfit_memdev *nfit_memdev;
591 if (type != NFIT_SPA_BDW)
594 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
595 if (nfit_memdev->memdev->range_index != range_index)
597 if (nfit_memdev->memdev->device_handle != device_handle)
599 if (nfit_memdev->memdev->region_index != dcr)
602 nfit_mem->spa_bdw = nfit_spa->spa;
607 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
608 nfit_mem->spa_dcr->range_index);
609 nfit_mem->bdw = NULL;
612 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
613 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
615 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
616 struct nfit_memdev *nfit_memdev;
617 struct nfit_bdw *nfit_bdw;
618 struct nfit_idt *nfit_idt;
619 u16 idt_idx, range_index;
621 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
622 if (nfit_bdw->bdw->region_index != dcr)
624 nfit_mem->bdw = nfit_bdw->bdw;
631 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
633 if (!nfit_mem->spa_bdw)
636 range_index = nfit_mem->spa_bdw->range_index;
637 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
638 if (nfit_memdev->memdev->range_index != range_index ||
639 nfit_memdev->memdev->region_index != dcr)
641 nfit_mem->memdev_bdw = nfit_memdev->memdev;
642 idt_idx = nfit_memdev->memdev->interleave_index;
643 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
644 if (nfit_idt->idt->interleave_index != idt_idx)
646 nfit_mem->idt_bdw = nfit_idt->idt;
653 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
654 struct acpi_nfit_system_address *spa)
656 struct nfit_mem *nfit_mem, *found;
657 struct nfit_memdev *nfit_memdev;
658 int type = nfit_spa_type(spa);
668 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
669 struct nfit_flush *nfit_flush;
670 struct nfit_dcr *nfit_dcr;
674 if (nfit_memdev->memdev->range_index != spa->range_index)
677 dcr = nfit_memdev->memdev->region_index;
678 device_handle = nfit_memdev->memdev->device_handle;
679 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
680 if (__to_nfit_memdev(nfit_mem)->device_handle
689 nfit_mem = devm_kzalloc(acpi_desc->dev,
690 sizeof(*nfit_mem), GFP_KERNEL);
693 INIT_LIST_HEAD(&nfit_mem->list);
694 nfit_mem->acpi_desc = acpi_desc;
695 list_add(&nfit_mem->list, &acpi_desc->dimms);
698 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
699 if (nfit_dcr->dcr->region_index != dcr)
702 * Record the control region for the dimm. For
703 * the ACPI 6.1 case, where there are separate
704 * control regions for the pmem vs blk
705 * interfaces, be sure to record the extended
709 nfit_mem->dcr = nfit_dcr->dcr;
710 else if (nfit_mem->dcr->windows == 0
711 && nfit_dcr->dcr->windows)
712 nfit_mem->dcr = nfit_dcr->dcr;
716 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
717 struct acpi_nfit_flush_address *flush;
720 if (nfit_flush->flush->device_handle != device_handle)
722 nfit_mem->nfit_flush = nfit_flush;
723 flush = nfit_flush->flush;
724 nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
726 * sizeof(struct resource), GFP_KERNEL);
727 if (!nfit_mem->flush_wpq)
729 for (i = 0; i < flush->hint_count; i++) {
730 struct resource *res = &nfit_mem->flush_wpq[i];
732 res->start = flush->hint_address[i];
733 res->end = res->start + 8 - 1;
738 if (dcr && !nfit_mem->dcr) {
739 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
740 spa->range_index, dcr);
744 if (type == NFIT_SPA_DCR) {
745 struct nfit_idt *nfit_idt;
748 /* multiple dimms may share a SPA when interleaved */
749 nfit_mem->spa_dcr = spa;
750 nfit_mem->memdev_dcr = nfit_memdev->memdev;
751 idt_idx = nfit_memdev->memdev->interleave_index;
752 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
753 if (nfit_idt->idt->interleave_index != idt_idx)
755 nfit_mem->idt_dcr = nfit_idt->idt;
758 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
761 * A single dimm may belong to multiple SPA-PM
762 * ranges, record at least one in addition to
765 nfit_mem->memdev_pmem = nfit_memdev->memdev;
772 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
774 struct nfit_mem *a = container_of(_a, typeof(*a), list);
775 struct nfit_mem *b = container_of(_b, typeof(*b), list);
776 u32 handleA, handleB;
778 handleA = __to_nfit_memdev(a)->device_handle;
779 handleB = __to_nfit_memdev(b)->device_handle;
780 if (handleA < handleB)
782 else if (handleA > handleB)
787 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
789 struct nfit_spa *nfit_spa;
792 * For each SPA-DCR or SPA-PMEM address range find its
793 * corresponding MEMDEV(s). From each MEMDEV find the
794 * corresponding DCR. Then, if we're operating on a SPA-DCR,
795 * try to find a SPA-BDW and a corresponding BDW that references
796 * the DCR. Throw it all into an nfit_mem object. Note, that
799 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
802 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
807 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
812 static ssize_t revision_show(struct device *dev,
813 struct device_attribute *attr, char *buf)
815 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
816 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
817 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
819 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
821 static DEVICE_ATTR_RO(revision);
823 static struct attribute *acpi_nfit_attributes[] = {
824 &dev_attr_revision.attr,
828 static struct attribute_group acpi_nfit_attribute_group = {
830 .attrs = acpi_nfit_attributes,
833 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
834 &nvdimm_bus_attribute_group,
835 &acpi_nfit_attribute_group,
839 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
841 struct nvdimm *nvdimm = to_nvdimm(dev);
842 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
844 return __to_nfit_memdev(nfit_mem);
847 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
849 struct nvdimm *nvdimm = to_nvdimm(dev);
850 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
852 return nfit_mem->dcr;
855 static ssize_t handle_show(struct device *dev,
856 struct device_attribute *attr, char *buf)
858 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
860 return sprintf(buf, "%#x\n", memdev->device_handle);
862 static DEVICE_ATTR_RO(handle);
864 static ssize_t phys_id_show(struct device *dev,
865 struct device_attribute *attr, char *buf)
867 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
869 return sprintf(buf, "%#x\n", memdev->physical_id);
871 static DEVICE_ATTR_RO(phys_id);
873 static ssize_t vendor_show(struct device *dev,
874 struct device_attribute *attr, char *buf)
876 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
878 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
880 static DEVICE_ATTR_RO(vendor);
882 static ssize_t rev_id_show(struct device *dev,
883 struct device_attribute *attr, char *buf)
885 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
887 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
889 static DEVICE_ATTR_RO(rev_id);
891 static ssize_t device_show(struct device *dev,
892 struct device_attribute *attr, char *buf)
894 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
896 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
898 static DEVICE_ATTR_RO(device);
900 static ssize_t subsystem_vendor_show(struct device *dev,
901 struct device_attribute *attr, char *buf)
903 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
905 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
907 static DEVICE_ATTR_RO(subsystem_vendor);
909 static ssize_t subsystem_rev_id_show(struct device *dev,
910 struct device_attribute *attr, char *buf)
912 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
914 return sprintf(buf, "0x%04x\n",
915 be16_to_cpu(dcr->subsystem_revision_id));
917 static DEVICE_ATTR_RO(subsystem_rev_id);
919 static ssize_t subsystem_device_show(struct device *dev,
920 struct device_attribute *attr, char *buf)
922 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
924 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
926 static DEVICE_ATTR_RO(subsystem_device);
928 static int num_nvdimm_formats(struct nvdimm *nvdimm)
930 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
933 if (nfit_mem->memdev_pmem)
935 if (nfit_mem->memdev_bdw)
940 static ssize_t format_show(struct device *dev,
941 struct device_attribute *attr, char *buf)
943 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
945 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->code));
947 static DEVICE_ATTR_RO(format);
949 static ssize_t format1_show(struct device *dev,
950 struct device_attribute *attr, char *buf)
954 struct nfit_mem *nfit_mem;
955 struct nfit_memdev *nfit_memdev;
956 struct acpi_nfit_desc *acpi_desc;
957 struct nvdimm *nvdimm = to_nvdimm(dev);
958 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
960 nfit_mem = nvdimm_provider_data(nvdimm);
961 acpi_desc = nfit_mem->acpi_desc;
962 handle = to_nfit_memdev(dev)->device_handle;
964 /* assumes DIMMs have at most 2 published interface codes */
965 mutex_lock(&acpi_desc->init_mutex);
966 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
967 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
968 struct nfit_dcr *nfit_dcr;
970 if (memdev->device_handle != handle)
973 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
974 if (nfit_dcr->dcr->region_index != memdev->region_index)
976 if (nfit_dcr->dcr->code == dcr->code)
978 rc = sprintf(buf, "%#x\n",
979 be16_to_cpu(nfit_dcr->dcr->code));
985 mutex_unlock(&acpi_desc->init_mutex);
988 static DEVICE_ATTR_RO(format1);
990 static ssize_t formats_show(struct device *dev,
991 struct device_attribute *attr, char *buf)
993 struct nvdimm *nvdimm = to_nvdimm(dev);
995 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
997 static DEVICE_ATTR_RO(formats);
999 static ssize_t serial_show(struct device *dev,
1000 struct device_attribute *attr, char *buf)
1002 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1004 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1006 static DEVICE_ATTR_RO(serial);
1008 static ssize_t family_show(struct device *dev,
1009 struct device_attribute *attr, char *buf)
1011 struct nvdimm *nvdimm = to_nvdimm(dev);
1012 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1014 if (nfit_mem->family < 0)
1016 return sprintf(buf, "%d\n", nfit_mem->family);
1018 static DEVICE_ATTR_RO(family);
1020 static ssize_t dsm_mask_show(struct device *dev,
1021 struct device_attribute *attr, char *buf)
1023 struct nvdimm *nvdimm = to_nvdimm(dev);
1024 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1026 if (nfit_mem->family < 0)
1028 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1030 static DEVICE_ATTR_RO(dsm_mask);
1032 static ssize_t flags_show(struct device *dev,
1033 struct device_attribute *attr, char *buf)
1035 u16 flags = to_nfit_memdev(dev)->flags;
1037 return sprintf(buf, "%s%s%s%s%s\n",
1038 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1039 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1040 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1041 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1042 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
1044 static DEVICE_ATTR_RO(flags);
1046 static ssize_t id_show(struct device *dev,
1047 struct device_attribute *attr, char *buf)
1049 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1051 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1052 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1053 be16_to_cpu(dcr->vendor_id),
1054 dcr->manufacturing_location,
1055 be16_to_cpu(dcr->manufacturing_date),
1056 be32_to_cpu(dcr->serial_number));
1058 return sprintf(buf, "%04x-%08x\n",
1059 be16_to_cpu(dcr->vendor_id),
1060 be32_to_cpu(dcr->serial_number));
1062 static DEVICE_ATTR_RO(id);
1064 static struct attribute *acpi_nfit_dimm_attributes[] = {
1065 &dev_attr_handle.attr,
1066 &dev_attr_phys_id.attr,
1067 &dev_attr_vendor.attr,
1068 &dev_attr_device.attr,
1069 &dev_attr_rev_id.attr,
1070 &dev_attr_subsystem_vendor.attr,
1071 &dev_attr_subsystem_device.attr,
1072 &dev_attr_subsystem_rev_id.attr,
1073 &dev_attr_format.attr,
1074 &dev_attr_formats.attr,
1075 &dev_attr_format1.attr,
1076 &dev_attr_serial.attr,
1077 &dev_attr_flags.attr,
1079 &dev_attr_family.attr,
1080 &dev_attr_dsm_mask.attr,
1084 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1085 struct attribute *a, int n)
1087 struct device *dev = container_of(kobj, struct device, kobj);
1088 struct nvdimm *nvdimm = to_nvdimm(dev);
1090 if (!to_nfit_dcr(dev))
1092 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1097 static struct attribute_group acpi_nfit_dimm_attribute_group = {
1099 .attrs = acpi_nfit_dimm_attributes,
1100 .is_visible = acpi_nfit_dimm_attr_visible,
1103 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1104 &nvdimm_attribute_group,
1105 &nd_device_attribute_group,
1106 &acpi_nfit_dimm_attribute_group,
1110 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1113 struct nfit_mem *nfit_mem;
1115 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1116 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1117 return nfit_mem->nvdimm;
1122 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1123 struct nfit_mem *nfit_mem, u32 device_handle)
1125 struct acpi_device *adev, *adev_dimm;
1126 struct device *dev = acpi_desc->dev;
1127 unsigned long dsm_mask;
1131 /* nfit test assumes 1:1 relationship between commands and dsms */
1132 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1133 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1134 adev = to_acpi_dev(acpi_desc);
1138 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1139 nfit_mem->adev = adev_dimm;
1141 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1143 return force_enable_dimms ? 0 : -ENODEV;
1147 * Until standardization materializes we need to consider 4
1148 * different command sets. Note, that checking for function0 (bit0)
1149 * tells us if any commands are reachable through this uuid.
1151 for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1152 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1155 /* limit the supported commands to those that are publicly documented */
1156 nfit_mem->family = i;
1157 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1159 if (disable_vendor_specific)
1160 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1161 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1162 dsm_mask = 0x1c3c76;
1163 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1165 if (disable_vendor_specific)
1166 dsm_mask &= ~(1 << 8);
1167 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1168 dsm_mask = 0xffffffff;
1170 dev_err(dev, "unknown dimm command family\n");
1171 nfit_mem->family = -1;
1172 return force_enable_dimms ? 0 : -ENODEV;
1175 uuid = to_nfit_uuid(nfit_mem->family);
1176 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1177 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1178 set_bit(i, &nfit_mem->dsm_mask);
1183 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1185 struct nfit_mem *nfit_mem;
1188 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1189 struct acpi_nfit_flush_address *flush;
1190 unsigned long flags = 0, cmd_mask;
1191 struct nvdimm *nvdimm;
1196 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1197 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1203 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1204 flags |= NDD_ALIASING;
1206 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1207 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1208 flags |= NDD_UNARMED;
1210 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1215 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1216 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1217 * userspace interface.
1219 cmd_mask = 1UL << ND_CMD_CALL;
1220 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1221 cmd_mask |= nfit_mem->dsm_mask;
1223 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1225 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1226 acpi_nfit_dimm_attribute_groups,
1227 flags, cmd_mask, flush ? flush->hint_count : 0,
1228 nfit_mem->flush_wpq);
1232 nfit_mem->nvdimm = nvdimm;
1235 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1238 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1239 nvdimm_name(nvdimm),
1240 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1241 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1242 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1243 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1247 return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1250 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1252 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1253 const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1254 struct acpi_device *adev;
1257 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1258 adev = to_acpi_dev(acpi_desc);
1262 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1263 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1264 set_bit(i, &nd_desc->cmd_mask);
1267 static ssize_t range_index_show(struct device *dev,
1268 struct device_attribute *attr, char *buf)
1270 struct nd_region *nd_region = to_nd_region(dev);
1271 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1273 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1275 static DEVICE_ATTR_RO(range_index);
1277 static struct attribute *acpi_nfit_region_attributes[] = {
1278 &dev_attr_range_index.attr,
1282 static struct attribute_group acpi_nfit_region_attribute_group = {
1284 .attrs = acpi_nfit_region_attributes,
1287 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1288 &nd_region_attribute_group,
1289 &nd_mapping_attribute_group,
1290 &nd_device_attribute_group,
1291 &nd_numa_attribute_group,
1292 &acpi_nfit_region_attribute_group,
1296 /* enough info to uniquely specify an interleave set */
1297 struct nfit_set_info {
1298 struct nfit_set_info_map {
1305 static size_t sizeof_nfit_set_info(int num_mappings)
1307 return sizeof(struct nfit_set_info)
1308 + num_mappings * sizeof(struct nfit_set_info_map);
1311 static int cmp_map(const void *m0, const void *m1)
1313 const struct nfit_set_info_map *map0 = m0;
1314 const struct nfit_set_info_map *map1 = m1;
1316 return memcmp(&map0->region_offset, &map1->region_offset,
1320 /* Retrieve the nth entry referencing this spa */
1321 static struct acpi_nfit_memory_map *memdev_from_spa(
1322 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1324 struct nfit_memdev *nfit_memdev;
1326 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1327 if (nfit_memdev->memdev->range_index == range_index)
1329 return nfit_memdev->memdev;
1333 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1334 struct nd_region_desc *ndr_desc,
1335 struct acpi_nfit_system_address *spa)
1337 int i, spa_type = nfit_spa_type(spa);
1338 struct device *dev = acpi_desc->dev;
1339 struct nd_interleave_set *nd_set;
1340 u16 nr = ndr_desc->num_mappings;
1341 struct nfit_set_info *info;
1343 if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1348 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1352 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1355 for (i = 0; i < nr; i++) {
1356 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1357 struct nfit_set_info_map *map = &info->mapping[i];
1358 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1359 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1360 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1361 spa->range_index, i);
1363 if (!memdev || !nfit_mem->dcr) {
1364 dev_err(dev, "%s: failed to find DCR\n", __func__);
1368 map->region_offset = memdev->region_offset;
1369 map->serial_number = nfit_mem->dcr->serial_number;
1372 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1374 nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1375 ndr_desc->nd_set = nd_set;
1376 devm_kfree(dev, info);
1381 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1383 struct acpi_nfit_interleave *idt = mmio->idt;
1384 u32 sub_line_offset, line_index, line_offset;
1385 u64 line_no, table_skip_count, table_offset;
1387 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1388 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1389 line_offset = idt->line_offset[line_index]
1391 table_offset = table_skip_count * mmio->table_size;
1393 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1396 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1398 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1399 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1401 if (mmio->num_lines)
1402 offset = to_interleave_offset(offset, mmio);
1404 return readl(mmio->addr.base + offset);
1407 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1408 resource_size_t dpa, unsigned int len, unsigned int write)
1411 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1414 BCW_OFFSET_MASK = (1ULL << 48)-1,
1416 BCW_LEN_MASK = (1ULL << 8) - 1,
1420 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1421 len = len >> L1_CACHE_SHIFT;
1422 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1423 cmd |= ((u64) write) << BCW_CMD_SHIFT;
1425 offset = nfit_blk->cmd_offset + mmio->size * bw;
1426 if (mmio->num_lines)
1427 offset = to_interleave_offset(offset, mmio);
1429 writeq(cmd, mmio->addr.base + offset);
1430 nvdimm_flush(nfit_blk->nd_region);
1432 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1433 readq(mmio->addr.base + offset);
1436 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1437 resource_size_t dpa, void *iobuf, size_t len, int rw,
1440 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1441 unsigned int copied = 0;
1445 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1446 + lane * mmio->size;
1447 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1452 if (mmio->num_lines) {
1455 offset = to_interleave_offset(base_offset + copied,
1457 div_u64_rem(offset, mmio->line_size, &line_offset);
1458 c = min_t(size_t, len, mmio->line_size - line_offset);
1460 offset = base_offset + nfit_blk->bdw_offset;
1465 memcpy_to_pmem(mmio->addr.aperture + offset,
1468 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1469 mmio_flush_range((void __force *)
1470 mmio->addr.aperture + offset, c);
1472 memcpy_from_pmem(iobuf + copied,
1473 mmio->addr.aperture + offset, c);
1481 nvdimm_flush(nfit_blk->nd_region);
1483 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1487 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1488 resource_size_t dpa, void *iobuf, u64 len, int rw)
1490 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1491 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1492 struct nd_region *nd_region = nfit_blk->nd_region;
1493 unsigned int lane, copied = 0;
1496 lane = nd_region_acquire_lane(nd_region);
1498 u64 c = min(len, mmio->size);
1500 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1501 iobuf + copied, c, rw, lane);
1508 nd_region_release_lane(nd_region, lane);
1513 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1514 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1517 mmio->num_lines = idt->line_count;
1518 mmio->line_size = idt->line_size;
1519 if (interleave_ways == 0)
1521 mmio->table_size = mmio->num_lines * interleave_ways
1528 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1529 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1531 struct nd_cmd_dimm_flags flags;
1534 memset(&flags, 0, sizeof(flags));
1535 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1536 sizeof(flags), NULL);
1538 if (rc >= 0 && flags.status == 0)
1539 nfit_blk->dimm_flags = flags.flags;
1540 else if (rc == -ENOTTY) {
1541 /* fall back to a conservative default */
1542 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1550 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1553 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1554 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1555 struct nfit_blk_mmio *mmio;
1556 struct nfit_blk *nfit_blk;
1557 struct nfit_mem *nfit_mem;
1558 struct nvdimm *nvdimm;
1561 nvdimm = nd_blk_region_to_dimm(ndbr);
1562 nfit_mem = nvdimm_provider_data(nvdimm);
1563 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1564 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1565 nfit_mem ? "" : " nfit_mem",
1566 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1567 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1571 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1574 nd_blk_region_set_provider_data(ndbr, nfit_blk);
1575 nfit_blk->nd_region = to_nd_region(dev);
1577 /* map block aperture memory */
1578 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1579 mmio = &nfit_blk->mmio[BDW];
1580 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
1581 nfit_mem->spa_bdw->length, ARCH_MEMREMAP_PMEM);
1582 if (!mmio->addr.base) {
1583 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1584 nvdimm_name(nvdimm));
1587 mmio->size = nfit_mem->bdw->size;
1588 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1589 mmio->idt = nfit_mem->idt_bdw;
1590 mmio->spa = nfit_mem->spa_bdw;
1591 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1592 nfit_mem->memdev_bdw->interleave_ways);
1594 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1595 __func__, nvdimm_name(nvdimm));
1599 /* map block control memory */
1600 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1601 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1602 mmio = &nfit_blk->mmio[DCR];
1603 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
1604 nfit_mem->spa_dcr->length);
1605 if (!mmio->addr.base) {
1606 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1607 nvdimm_name(nvdimm));
1610 mmio->size = nfit_mem->dcr->window_size;
1611 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1612 mmio->idt = nfit_mem->idt_dcr;
1613 mmio->spa = nfit_mem->spa_dcr;
1614 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1615 nfit_mem->memdev_dcr->interleave_ways);
1617 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1618 __func__, nvdimm_name(nvdimm));
1622 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1624 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1625 __func__, nvdimm_name(nvdimm));
1629 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
1630 dev_warn(dev, "unable to guarantee persistence of writes\n");
1632 if (mmio->line_size == 0)
1635 if ((u32) nfit_blk->cmd_offset % mmio->line_size
1636 + 8 > mmio->line_size) {
1637 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1639 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1640 + 8 > mmio->line_size) {
1641 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1648 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1649 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1651 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1652 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1655 cmd->address = spa->address;
1656 cmd->length = spa->length;
1657 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1658 sizeof(*cmd), &cmd_rc);
1664 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1668 struct nd_cmd_ars_start ars_start;
1669 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1670 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1672 memset(&ars_start, 0, sizeof(ars_start));
1673 ars_start.address = spa->address;
1674 ars_start.length = spa->length;
1675 if (nfit_spa_type(spa) == NFIT_SPA_PM)
1676 ars_start.type = ND_ARS_PERSISTENT;
1677 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1678 ars_start.type = ND_ARS_VOLATILE;
1682 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1683 sizeof(ars_start), &cmd_rc);
1690 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1693 struct nd_cmd_ars_start ars_start;
1694 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1695 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1697 memset(&ars_start, 0, sizeof(ars_start));
1698 ars_start.address = ars_status->restart_address;
1699 ars_start.length = ars_status->restart_length;
1700 ars_start.type = ars_status->type;
1701 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1702 sizeof(ars_start), &cmd_rc);
1708 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1710 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1711 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1714 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1715 acpi_desc->ars_status_size, &cmd_rc);
1721 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1722 struct nd_cmd_ars_status *ars_status)
1727 for (i = 0; i < ars_status->num_records; i++) {
1728 rc = nvdimm_bus_add_poison(nvdimm_bus,
1729 ars_status->records[i].err_address,
1730 ars_status->records[i].length);
1738 static void acpi_nfit_remove_resource(void *data)
1740 struct resource *res = data;
1742 remove_resource(res);
1745 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1746 struct nd_region_desc *ndr_desc)
1748 struct resource *res, *nd_res = ndr_desc->res;
1751 /* No operation if the region is already registered as PMEM */
1752 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1753 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1754 if (is_pmem == REGION_INTERSECTS)
1757 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1761 res->name = "Persistent Memory";
1762 res->start = nd_res->start;
1763 res->end = nd_res->end;
1764 res->flags = IORESOURCE_MEM;
1765 res->desc = IORES_DESC_PERSISTENT_MEMORY;
1767 ret = insert_resource(&iomem_resource, res);
1771 ret = devm_add_action_or_reset(acpi_desc->dev,
1772 acpi_nfit_remove_resource,
1780 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1781 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1782 struct acpi_nfit_memory_map *memdev,
1783 struct nfit_spa *nfit_spa)
1785 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1786 memdev->device_handle);
1787 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1788 struct nd_blk_region_desc *ndbr_desc;
1789 struct nfit_mem *nfit_mem;
1793 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1794 spa->range_index, memdev->device_handle);
1798 nd_mapping->nvdimm = nvdimm;
1799 switch (nfit_spa_type(spa)) {
1801 case NFIT_SPA_VOLATILE:
1802 nd_mapping->start = memdev->address;
1803 nd_mapping->size = memdev->region_size;
1806 nfit_mem = nvdimm_provider_data(nvdimm);
1807 if (!nfit_mem || !nfit_mem->bdw) {
1808 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1809 spa->range_index, nvdimm_name(nvdimm));
1811 nd_mapping->size = nfit_mem->bdw->capacity;
1812 nd_mapping->start = nfit_mem->bdw->start_address;
1813 ndr_desc->num_lanes = nfit_mem->bdw->windows;
1817 ndr_desc->nd_mapping = nd_mapping;
1818 ndr_desc->num_mappings = blk_valid;
1819 ndbr_desc = to_blk_region_desc(ndr_desc);
1820 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1821 ndbr_desc->do_io = acpi_desc->blk_do_io;
1822 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1824 if (!nfit_spa->nd_region)
1832 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
1834 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
1835 nfit_spa_type(spa) == NFIT_SPA_VCD ||
1836 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
1837 nfit_spa_type(spa) == NFIT_SPA_PCD);
1840 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1841 struct nfit_spa *nfit_spa)
1843 static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1844 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1845 struct nd_blk_region_desc ndbr_desc;
1846 struct nd_region_desc *ndr_desc;
1847 struct nfit_memdev *nfit_memdev;
1848 struct nvdimm_bus *nvdimm_bus;
1849 struct resource res;
1852 if (nfit_spa->nd_region)
1855 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
1856 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
1861 memset(&res, 0, sizeof(res));
1862 memset(&nd_mappings, 0, sizeof(nd_mappings));
1863 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
1864 res.start = spa->address;
1865 res.end = res.start + spa->length - 1;
1866 ndr_desc = &ndbr_desc.ndr_desc;
1867 ndr_desc->res = &res;
1868 ndr_desc->provider_data = nfit_spa;
1869 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
1870 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
1871 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
1872 spa->proximity_domain);
1874 ndr_desc->numa_node = NUMA_NO_NODE;
1876 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1877 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1878 struct nd_mapping *nd_mapping;
1880 if (memdev->range_index != spa->range_index)
1882 if (count >= ND_MAX_MAPPINGS) {
1883 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
1884 spa->range_index, ND_MAX_MAPPINGS);
1887 nd_mapping = &nd_mappings[count++];
1888 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
1894 ndr_desc->nd_mapping = nd_mappings;
1895 ndr_desc->num_mappings = count;
1896 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
1900 nvdimm_bus = acpi_desc->nvdimm_bus;
1901 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
1902 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
1904 dev_warn(acpi_desc->dev,
1905 "failed to insert pmem resource to iomem: %d\n",
1910 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
1912 if (!nfit_spa->nd_region)
1914 } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
1915 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
1917 if (!nfit_spa->nd_region)
1919 } else if (nfit_spa_is_virtual(spa)) {
1920 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
1922 if (!nfit_spa->nd_region)
1928 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
1929 nfit_spa->spa->range_index);
1933 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
1936 struct device *dev = acpi_desc->dev;
1937 struct nd_cmd_ars_status *ars_status;
1939 if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
1940 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
1944 if (acpi_desc->ars_status)
1945 devm_kfree(dev, acpi_desc->ars_status);
1946 acpi_desc->ars_status = NULL;
1947 ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
1950 acpi_desc->ars_status = ars_status;
1951 acpi_desc->ars_status_size = max_ars;
1955 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
1956 struct nfit_spa *nfit_spa)
1958 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1961 if (!nfit_spa->max_ars) {
1962 struct nd_cmd_ars_cap ars_cap;
1964 memset(&ars_cap, 0, sizeof(ars_cap));
1965 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
1968 nfit_spa->max_ars = ars_cap.max_ars_out;
1969 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
1970 /* check that the supported scrub types match the spa type */
1971 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
1972 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
1974 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
1975 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
1979 if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
1982 rc = ars_get_status(acpi_desc);
1983 if (rc < 0 && rc != -ENOSPC)
1986 if (ars_status_process_records(acpi_desc->nvdimm_bus,
1987 acpi_desc->ars_status))
1993 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
1994 struct nfit_spa *nfit_spa)
1996 struct acpi_nfit_system_address *spa = nfit_spa->spa;
1997 unsigned int overflow_retry = scrub_overflow_abort;
1998 u64 init_ars_start = 0, init_ars_len = 0;
1999 struct device *dev = acpi_desc->dev;
2000 unsigned int tmo = scrub_timeout;
2003 if (nfit_spa->ars_done || !nfit_spa->nd_region)
2006 rc = ars_start(acpi_desc, nfit_spa);
2008 * If we timed out the initial scan we'll still be busy here,
2009 * and will wait another timeout before giving up permanently.
2011 if (rc < 0 && rc != -EBUSY)
2015 u64 ars_start, ars_len;
2017 if (acpi_desc->cancel)
2019 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2022 if (rc == -EBUSY && !tmo) {
2023 dev_warn(dev, "range %d ars timeout, aborting\n",
2030 * Note, entries may be appended to the list
2031 * while the lock is dropped, but the workqueue
2032 * being active prevents entries being deleted /
2035 mutex_unlock(&acpi_desc->init_mutex);
2038 mutex_lock(&acpi_desc->init_mutex);
2042 /* we got some results, but there are more pending... */
2043 if (rc == -ENOSPC && overflow_retry--) {
2044 if (!init_ars_len) {
2045 init_ars_len = acpi_desc->ars_status->length;
2046 init_ars_start = acpi_desc->ars_status->address;
2048 rc = ars_continue(acpi_desc);
2052 dev_warn(dev, "range %d ars continuation failed\n",
2058 ars_start = init_ars_start;
2059 ars_len = init_ars_len;
2061 ars_start = acpi_desc->ars_status->address;
2062 ars_len = acpi_desc->ars_status->length;
2064 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2065 spa->range_index, ars_start, ars_len);
2066 /* notify the region about new poison entries */
2067 nvdimm_region_notify(nfit_spa->nd_region,
2068 NVDIMM_REVALIDATE_POISON);
2073 static void acpi_nfit_scrub(struct work_struct *work)
2076 u64 init_scrub_length = 0;
2077 struct nfit_spa *nfit_spa;
2078 u64 init_scrub_address = 0;
2079 bool init_ars_done = false;
2080 struct acpi_nfit_desc *acpi_desc;
2081 unsigned int tmo = scrub_timeout;
2082 unsigned int overflow_retry = scrub_overflow_abort;
2084 acpi_desc = container_of(work, typeof(*acpi_desc), work);
2085 dev = acpi_desc->dev;
2088 * We scrub in 2 phases. The first phase waits for any platform
2089 * firmware initiated scrubs to complete and then we go search for the
2090 * affected spa regions to mark them scanned. In the second phase we
2091 * initiate a directed scrub for every range that was not scrubbed in
2095 /* process platform firmware initiated scrubs */
2097 mutex_lock(&acpi_desc->init_mutex);
2098 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2099 struct nd_cmd_ars_status *ars_status;
2100 struct acpi_nfit_system_address *spa;
2101 u64 ars_start, ars_len;
2104 if (acpi_desc->cancel)
2107 if (nfit_spa->nd_region)
2110 if (init_ars_done) {
2112 * No need to re-query, we're now just
2113 * reconciling all the ranges covered by the
2118 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2120 if (rc == -ENOTTY) {
2121 /* no ars capability, just register spa and move on */
2122 acpi_nfit_register_region(acpi_desc, nfit_spa);
2126 if (rc == -EBUSY && !tmo) {
2127 /* fallthrough to directed scrub in phase 2 */
2128 dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2130 } else if (rc == -EBUSY) {
2131 mutex_unlock(&acpi_desc->init_mutex);
2137 /* we got some results, but there are more pending... */
2138 if (rc == -ENOSPC && overflow_retry--) {
2139 ars_status = acpi_desc->ars_status;
2141 * Record the original scrub range, so that we
2142 * can recall all the ranges impacted by the
2145 if (!init_scrub_length) {
2146 init_scrub_length = ars_status->length;
2147 init_scrub_address = ars_status->address;
2149 rc = ars_continue(acpi_desc);
2151 mutex_unlock(&acpi_desc->init_mutex);
2158 * Initial scrub failed, we'll give it one more
2164 /* We got some final results, record completed ranges */
2165 ars_status = acpi_desc->ars_status;
2166 if (init_scrub_length) {
2167 ars_start = init_scrub_address;
2168 ars_len = ars_start + init_scrub_length;
2170 ars_start = ars_status->address;
2171 ars_len = ars_status->length;
2173 spa = nfit_spa->spa;
2175 if (!init_ars_done) {
2176 init_ars_done = true;
2177 dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2178 ars_start, ars_len);
2180 if (ars_start <= spa->address && ars_start + ars_len
2181 >= spa->address + spa->length)
2182 acpi_nfit_register_region(acpi_desc, nfit_spa);
2186 * For all the ranges not covered by an initial scrub we still
2187 * want to see if there are errors, but it's ok to discover them
2190 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2192 * Flag all the ranges that still need scrubbing, but
2193 * register them now to make data available.
2195 if (nfit_spa->nd_region)
2196 nfit_spa->ars_done = 1;
2198 acpi_nfit_register_region(acpi_desc, nfit_spa);
2201 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2202 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2203 mutex_unlock(&acpi_desc->init_mutex);
2206 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2208 struct nfit_spa *nfit_spa;
2211 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2212 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2213 /* BLK regions don't need to wait for ars results */
2214 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2219 queue_work(nfit_wq, &acpi_desc->work);
2223 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2224 struct nfit_table_prev *prev)
2226 struct device *dev = acpi_desc->dev;
2228 if (!list_empty(&prev->spas) ||
2229 !list_empty(&prev->memdevs) ||
2230 !list_empty(&prev->dcrs) ||
2231 !list_empty(&prev->bdws) ||
2232 !list_empty(&prev->idts) ||
2233 !list_empty(&prev->flushes)) {
2234 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2240 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2242 struct device *dev = acpi_desc->dev;
2243 struct nfit_table_prev prev;
2248 mutex_lock(&acpi_desc->init_mutex);
2250 INIT_LIST_HEAD(&prev.spas);
2251 INIT_LIST_HEAD(&prev.memdevs);
2252 INIT_LIST_HEAD(&prev.dcrs);
2253 INIT_LIST_HEAD(&prev.bdws);
2254 INIT_LIST_HEAD(&prev.idts);
2255 INIT_LIST_HEAD(&prev.flushes);
2257 list_cut_position(&prev.spas, &acpi_desc->spas,
2258 acpi_desc->spas.prev);
2259 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2260 acpi_desc->memdevs.prev);
2261 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2262 acpi_desc->dcrs.prev);
2263 list_cut_position(&prev.bdws, &acpi_desc->bdws,
2264 acpi_desc->bdws.prev);
2265 list_cut_position(&prev.idts, &acpi_desc->idts,
2266 acpi_desc->idts.prev);
2267 list_cut_position(&prev.flushes, &acpi_desc->flushes,
2268 acpi_desc->flushes.prev);
2270 data = (u8 *) acpi_desc->nfit;
2272 while (!IS_ERR_OR_NULL(data))
2273 data = add_table(acpi_desc, &prev, data, end);
2276 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2282 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2286 rc = nfit_mem_init(acpi_desc);
2290 acpi_nfit_init_dsms(acpi_desc);
2292 rc = acpi_nfit_register_dimms(acpi_desc);
2296 rc = acpi_nfit_register_regions(acpi_desc);
2299 mutex_unlock(&acpi_desc->init_mutex);
2302 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2304 struct acpi_nfit_flush_work {
2305 struct work_struct work;
2306 struct completion cmp;
2309 static void flush_probe(struct work_struct *work)
2311 struct acpi_nfit_flush_work *flush;
2313 flush = container_of(work, typeof(*flush), work);
2314 complete(&flush->cmp);
2317 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2319 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2320 struct device *dev = acpi_desc->dev;
2321 struct acpi_nfit_flush_work flush;
2323 /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2328 * Scrub work could take 10s of seconds, userspace may give up so we
2329 * need to be interruptible while waiting.
2331 INIT_WORK_ONSTACK(&flush.work, flush_probe);
2332 COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2333 queue_work(nfit_wq, &flush.work);
2334 return wait_for_completion_interruptible(&flush.cmp);
2337 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2338 struct nvdimm *nvdimm, unsigned int cmd)
2340 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2344 if (cmd != ND_CMD_ARS_START)
2348 * The kernel and userspace may race to initiate a scrub, but
2349 * the scrub thread is prepared to lose that initial race. It
2350 * just needs guarantees that any ars it initiates are not
2351 * interrupted by any intervening start reqeusts from userspace.
2353 if (work_busy(&acpi_desc->work))
2359 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2361 struct nvdimm_bus_descriptor *nd_desc;
2363 dev_set_drvdata(dev, acpi_desc);
2364 acpi_desc->dev = dev;
2365 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2366 nd_desc = &acpi_desc->nd_desc;
2367 nd_desc->provider_name = "ACPI.NFIT";
2368 nd_desc->ndctl = acpi_nfit_ctl;
2369 nd_desc->flush_probe = acpi_nfit_flush_probe;
2370 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2371 nd_desc->attr_groups = acpi_nfit_attribute_groups;
2373 INIT_LIST_HEAD(&acpi_desc->spas);
2374 INIT_LIST_HEAD(&acpi_desc->dcrs);
2375 INIT_LIST_HEAD(&acpi_desc->bdws);
2376 INIT_LIST_HEAD(&acpi_desc->idts);
2377 INIT_LIST_HEAD(&acpi_desc->flushes);
2378 INIT_LIST_HEAD(&acpi_desc->memdevs);
2379 INIT_LIST_HEAD(&acpi_desc->dimms);
2380 mutex_init(&acpi_desc->init_mutex);
2381 INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2383 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2385 static int acpi_nfit_add(struct acpi_device *adev)
2387 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2388 struct acpi_nfit_desc *acpi_desc;
2389 struct device *dev = &adev->dev;
2390 struct acpi_table_header *tbl;
2391 acpi_status status = AE_OK;
2395 status = acpi_get_table_with_size(ACPI_SIG_NFIT, 0, &tbl, &sz);
2396 if (ACPI_FAILURE(status)) {
2397 /* This is ok, we could have an nvdimm hotplugged later */
2398 dev_dbg(dev, "failed to find NFIT at startup\n");
2402 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2405 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2406 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2407 if (!acpi_desc->nvdimm_bus)
2411 * Save the acpi header for later and then skip it,
2412 * making nfit point to the first nfit table header.
2414 acpi_desc->acpi_header = *tbl;
2415 acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2416 sz -= sizeof(struct acpi_table_nfit);
2418 /* Evaluate _FIT and override with that if present */
2419 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2420 if (ACPI_SUCCESS(status) && buf.length > 0) {
2421 union acpi_object *obj;
2423 * Adjust for the acpi_object header of the _FIT
2426 if (obj->type == ACPI_TYPE_BUFFER) {
2428 (struct acpi_nfit_header *)obj->buffer.pointer;
2429 sz = obj->buffer.length;
2431 dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2432 __func__, (int) obj->type);
2435 rc = acpi_nfit_init(acpi_desc, sz);
2437 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2443 static int acpi_nfit_remove(struct acpi_device *adev)
2445 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2447 acpi_desc->cancel = 1;
2448 flush_workqueue(nfit_wq);
2449 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2453 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2455 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2456 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2457 struct acpi_nfit_header *nfit_saved;
2458 union acpi_object *obj;
2459 struct device *dev = &adev->dev;
2463 dev_dbg(dev, "%s: event: %d\n", __func__, event);
2467 /* dev->driver may be null if we're being removed */
2468 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2473 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2476 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2477 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2478 if (!acpi_desc->nvdimm_bus)
2482 * Finish previous registration before considering new
2485 flush_workqueue(nfit_wq);
2489 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2490 if (ACPI_FAILURE(status)) {
2491 dev_err(dev, "failed to evaluate _FIT\n");
2495 nfit_saved = acpi_desc->nfit;
2497 if (obj->type == ACPI_TYPE_BUFFER) {
2499 (struct acpi_nfit_header *)obj->buffer.pointer;
2500 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2502 /* Merge failed, restore old nfit, and exit */
2503 acpi_desc->nfit = nfit_saved;
2504 dev_err(dev, "failed to merge updated NFIT\n");
2507 /* Bad _FIT, restore old nfit */
2508 dev_err(dev, "Invalid _FIT\n");
2516 static const struct acpi_device_id acpi_nfit_ids[] = {
2520 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2522 static struct acpi_driver acpi_nfit_driver = {
2523 .name = KBUILD_MODNAME,
2524 .ids = acpi_nfit_ids,
2526 .add = acpi_nfit_add,
2527 .remove = acpi_nfit_remove,
2528 .notify = acpi_nfit_notify,
2532 static __init int nfit_init(void)
2534 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2535 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2536 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2537 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2538 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2539 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2540 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2542 acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2543 acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2544 acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2545 acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2546 acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2547 acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2548 acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2549 acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2550 acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2551 acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2552 acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2553 acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2554 acpi_str_to_uuid(UUID_NFIT_DIMM_N_MSFT, nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
2556 nfit_wq = create_singlethread_workqueue("nfit");
2560 return acpi_bus_register_driver(&acpi_nfit_driver);
2563 static __exit void nfit_exit(void)
2565 acpi_bus_unregister_driver(&acpi_nfit_driver);
2566 destroy_workqueue(nfit_wq);
2569 module_init(nfit_init);
2570 module_exit(nfit_exit);
2571 MODULE_LICENSE("GPL v2");
2572 MODULE_AUTHOR("Intel Corporation");