2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/list_sort.h>
14 #include <linux/libnvdimm.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/ndctl.h>
18 #include <linux/sysfs.h>
19 #include <linux/delay.h>
20 #include <linux/list.h>
21 #include <linux/acpi.h>
22 #include <linux/sort.h>
25 #include <asm/cacheflush.h>
29 * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
32 #include <linux/io-64-nonatomic-hi-lo.h>
34 static bool force_enable_dimms;
35 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
36 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
38 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
39 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
40 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
42 /* after three payloads of overflow, it's dead jim */
43 static unsigned int scrub_overflow_abort = 3;
44 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
45 MODULE_PARM_DESC(scrub_overflow_abort,
46 "Number of times we overflow ARS results before abort");
48 static bool disable_vendor_specific;
49 module_param(disable_vendor_specific, bool, S_IRUGO);
50 MODULE_PARM_DESC(disable_vendor_specific,
51 "Limit commands to the publicly specified set");
53 static unsigned long override_dsm_mask;
54 module_param(override_dsm_mask, ulong, S_IRUGO);
55 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
57 static int default_dsm_family = -1;
58 module_param(default_dsm_family, int, S_IRUGO);
59 MODULE_PARM_DESC(default_dsm_family,
60 "Try this DSM type first when identifying NVDIMM family");
62 LIST_HEAD(acpi_descs);
63 DEFINE_MUTEX(acpi_desc_lock);
65 static struct workqueue_struct *nfit_wq;
67 struct nfit_table_prev {
68 struct list_head spas;
69 struct list_head memdevs;
70 struct list_head dcrs;
71 struct list_head bdws;
72 struct list_head idts;
73 struct list_head flushes;
76 static guid_t nfit_uuid[NFIT_UUID_MAX];
78 const guid_t *to_nfit_uuid(enum nfit_uuids id)
80 return &nfit_uuid[id];
82 EXPORT_SYMBOL(to_nfit_uuid);
84 static struct acpi_nfit_desc *to_acpi_nfit_desc(
85 struct nvdimm_bus_descriptor *nd_desc)
87 return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
90 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
92 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
95 * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
98 if (!nd_desc->provider_name
99 || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
102 return to_acpi_device(acpi_desc->dev);
105 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
107 struct nd_cmd_clear_error *clear_err;
108 struct nd_cmd_ars_status *ars_status;
113 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
120 /* No supported scan types for this range */
121 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
122 if ((status >> 16 & flags) == 0)
125 case ND_CMD_ARS_START:
126 /* ARS is in progress */
127 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
134 case ND_CMD_ARS_STATUS:
139 /* Check extended status (Upper two bytes) */
140 if (status == NFIT_ARS_STATUS_DONE)
143 /* ARS is in progress */
144 if (status == NFIT_ARS_STATUS_BUSY)
147 /* No ARS performed for the current boot */
148 if (status == NFIT_ARS_STATUS_NONE)
152 * ARS interrupted, either we overflowed or some other
153 * agent wants the scan to stop. If we didn't overflow
154 * then just continue with the returned results.
156 if (status == NFIT_ARS_STATUS_INTR) {
157 if (ars_status->out_length >= 40 && (ars_status->flags
158 & NFIT_ARS_F_OVERFLOW))
167 case ND_CMD_CLEAR_ERROR:
171 if (!clear_err->cleared)
173 if (clear_err->length > clear_err->cleared)
174 return clear_err->cleared;
180 /* all other non-zero status results in an error */
186 static int xlat_nvdimm_status(void *buf, unsigned int cmd, u32 status)
189 case ND_CMD_GET_CONFIG_SIZE:
190 if (status >> 16 & ND_CONFIG_LOCKED)
197 /* all other non-zero status results in an error */
203 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
207 return xlat_bus_status(buf, cmd, status);
208 return xlat_nvdimm_status(buf, cmd, status);
211 static int cmd_to_func(struct nfit_mem *nfit_mem, unsigned int cmd,
212 struct nd_cmd_pkg *call_pkg)
217 if (nfit_mem && nfit_mem->family != call_pkg->nd_family)
220 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
221 if (call_pkg->nd_reserved2[i])
223 return call_pkg->nd_command;
226 /* In the !call_pkg case, bus commands == bus functions */
230 /* Linux ND commands == NVDIMM_FAMILY_INTEL function numbers */
231 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
235 * Force function number validation to fail since 0 is never
236 * published as a valid function in dsm_mask.
241 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
242 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
244 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
245 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
246 union acpi_object in_obj, in_buf, *out_obj;
247 const struct nd_cmd_desc *desc = NULL;
248 struct device *dev = acpi_desc->dev;
249 struct nd_cmd_pkg *call_pkg = NULL;
250 const char *cmd_name, *dimm_name;
251 unsigned long cmd_mask, dsm_mask;
252 u32 offset, fw_status = 0;
260 if (cmd == ND_CMD_CALL)
262 func = cmd_to_func(nfit_mem, cmd, call_pkg);
267 struct acpi_device *adev = nfit_mem->adev;
272 dimm_name = nvdimm_name(nvdimm);
273 cmd_name = nvdimm_cmd_name(cmd);
274 cmd_mask = nvdimm_cmd_mask(nvdimm);
275 dsm_mask = nfit_mem->dsm_mask;
276 desc = nd_cmd_dimm_desc(cmd);
277 guid = to_nfit_uuid(nfit_mem->family);
278 handle = adev->handle;
280 struct acpi_device *adev = to_acpi_dev(acpi_desc);
282 cmd_name = nvdimm_bus_cmd_name(cmd);
283 cmd_mask = nd_desc->cmd_mask;
284 dsm_mask = nd_desc->bus_dsm_mask;
285 desc = nd_cmd_bus_desc(cmd);
286 guid = to_nfit_uuid(NFIT_DEV_BUS);
287 handle = adev->handle;
291 if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
295 * Check for a valid command. For ND_CMD_CALL, we also have to
296 * make sure that the DSM function is supported.
298 if (cmd == ND_CMD_CALL && !test_bit(func, &dsm_mask))
300 else if (!test_bit(cmd, &cmd_mask))
303 in_obj.type = ACPI_TYPE_PACKAGE;
304 in_obj.package.count = 1;
305 in_obj.package.elements = &in_buf;
306 in_buf.type = ACPI_TYPE_BUFFER;
307 in_buf.buffer.pointer = buf;
308 in_buf.buffer.length = 0;
310 /* libnvdimm has already validated the input envelope */
311 for (i = 0; i < desc->in_num; i++)
312 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
316 /* skip over package wrapper */
317 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
318 in_buf.buffer.length = call_pkg->nd_size_in;
321 dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
322 __func__, dimm_name, cmd, func, in_buf.buffer.length);
323 print_hex_dump_debug("nvdimm in ", DUMP_PREFIX_OFFSET, 4, 4,
324 in_buf.buffer.pointer,
325 min_t(u32, 256, in_buf.buffer.length), true);
327 out_obj = acpi_evaluate_dsm(handle, guid, 1, func, &in_obj);
329 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
334 if (out_obj->type != ACPI_TYPE_BUFFER) {
335 dev_dbg(dev, "%s unexpected output object type cmd: %s type: %d\n",
336 dimm_name, cmd_name, out_obj->type);
342 call_pkg->nd_fw_size = out_obj->buffer.length;
343 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
344 out_obj->buffer.pointer,
345 min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
349 * Need to support FW function w/o known size in advance.
350 * Caller can determine required size based upon nd_fw_size.
351 * If we return an error (like elsewhere) then caller wouldn't
352 * be able to rely upon data returned to make calculation.
359 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__, dimm_name,
360 cmd_name, out_obj->buffer.length);
361 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
362 out_obj->buffer.pointer,
363 min_t(u32, 128, out_obj->buffer.length), true);
365 for (i = 0, offset = 0; i < desc->out_num; i++) {
366 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
367 (u32 *) out_obj->buffer.pointer,
368 out_obj->buffer.length - offset);
370 if (offset + out_size > out_obj->buffer.length) {
371 dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
372 __func__, dimm_name, cmd_name, i);
376 if (in_buf.buffer.length + offset + out_size > buf_len) {
377 dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
378 __func__, dimm_name, cmd_name, i);
382 memcpy(buf + in_buf.buffer.length + offset,
383 out_obj->buffer.pointer + offset, out_size);
388 * Set fw_status for all the commands with a known format to be
389 * later interpreted by xlat_status().
391 if (i >= 1 && ((cmd >= ND_CMD_ARS_CAP && cmd <= ND_CMD_CLEAR_ERROR)
392 || (cmd >= ND_CMD_SMART && cmd <= ND_CMD_VENDOR)))
393 fw_status = *(u32 *) out_obj->buffer.pointer;
395 if (offset + in_buf.buffer.length < buf_len) {
398 * status valid, return the number of bytes left
399 * unfilled in the output buffer
401 rc = buf_len - offset - in_buf.buffer.length;
403 *cmd_rc = xlat_status(nvdimm, buf, cmd,
406 dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
407 __func__, dimm_name, cmd_name, buf_len,
414 *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
422 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
424 static const char *spa_type_name(u16 type)
426 static const char *to_name[] = {
427 [NFIT_SPA_VOLATILE] = "volatile",
428 [NFIT_SPA_PM] = "pmem",
429 [NFIT_SPA_DCR] = "dimm-control-region",
430 [NFIT_SPA_BDW] = "block-data-window",
431 [NFIT_SPA_VDISK] = "volatile-disk",
432 [NFIT_SPA_VCD] = "volatile-cd",
433 [NFIT_SPA_PDISK] = "persistent-disk",
434 [NFIT_SPA_PCD] = "persistent-cd",
438 if (type > NFIT_SPA_PCD)
441 return to_name[type];
444 int nfit_spa_type(struct acpi_nfit_system_address *spa)
448 for (i = 0; i < NFIT_UUID_MAX; i++)
449 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
454 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
455 struct nfit_table_prev *prev,
456 struct acpi_nfit_system_address *spa)
458 struct device *dev = acpi_desc->dev;
459 struct nfit_spa *nfit_spa;
461 if (spa->header.length != sizeof(*spa))
464 list_for_each_entry(nfit_spa, &prev->spas, list) {
465 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
466 list_move_tail(&nfit_spa->list, &acpi_desc->spas);
471 nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
475 INIT_LIST_HEAD(&nfit_spa->list);
476 memcpy(nfit_spa->spa, spa, sizeof(*spa));
477 list_add_tail(&nfit_spa->list, &acpi_desc->spas);
478 dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
480 spa_type_name(nfit_spa_type(spa)));
484 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
485 struct nfit_table_prev *prev,
486 struct acpi_nfit_memory_map *memdev)
488 struct device *dev = acpi_desc->dev;
489 struct nfit_memdev *nfit_memdev;
491 if (memdev->header.length != sizeof(*memdev))
494 list_for_each_entry(nfit_memdev, &prev->memdevs, list)
495 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
496 list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
500 nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
504 INIT_LIST_HEAD(&nfit_memdev->list);
505 memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
506 list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
507 dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
508 __func__, memdev->device_handle, memdev->range_index,
509 memdev->region_index, memdev->flags);
514 * An implementation may provide a truncated control region if no block windows
517 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
519 if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
524 return offsetof(struct acpi_nfit_control_region, window_size);
527 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
528 struct nfit_table_prev *prev,
529 struct acpi_nfit_control_region *dcr)
531 struct device *dev = acpi_desc->dev;
532 struct nfit_dcr *nfit_dcr;
534 if (!sizeof_dcr(dcr))
537 list_for_each_entry(nfit_dcr, &prev->dcrs, list)
538 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
539 list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
543 nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
547 INIT_LIST_HEAD(&nfit_dcr->list);
548 memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
549 list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
550 dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
551 dcr->region_index, dcr->windows);
555 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
556 struct nfit_table_prev *prev,
557 struct acpi_nfit_data_region *bdw)
559 struct device *dev = acpi_desc->dev;
560 struct nfit_bdw *nfit_bdw;
562 if (bdw->header.length != sizeof(*bdw))
564 list_for_each_entry(nfit_bdw, &prev->bdws, list)
565 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
566 list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
570 nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
574 INIT_LIST_HEAD(&nfit_bdw->list);
575 memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
576 list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
577 dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
578 bdw->region_index, bdw->windows);
582 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
584 if (idt->header.length < sizeof(*idt))
586 return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
589 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
590 struct nfit_table_prev *prev,
591 struct acpi_nfit_interleave *idt)
593 struct device *dev = acpi_desc->dev;
594 struct nfit_idt *nfit_idt;
596 if (!sizeof_idt(idt))
599 list_for_each_entry(nfit_idt, &prev->idts, list) {
600 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
603 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
604 list_move_tail(&nfit_idt->list, &acpi_desc->idts);
609 nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
613 INIT_LIST_HEAD(&nfit_idt->list);
614 memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
615 list_add_tail(&nfit_idt->list, &acpi_desc->idts);
616 dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
617 idt->interleave_index, idt->line_count);
621 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
623 if (flush->header.length < sizeof(*flush))
625 return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
628 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
629 struct nfit_table_prev *prev,
630 struct acpi_nfit_flush_address *flush)
632 struct device *dev = acpi_desc->dev;
633 struct nfit_flush *nfit_flush;
635 if (!sizeof_flush(flush))
638 list_for_each_entry(nfit_flush, &prev->flushes, list) {
639 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
642 if (memcmp(nfit_flush->flush, flush,
643 sizeof_flush(flush)) == 0) {
644 list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
649 nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
650 + sizeof_flush(flush), GFP_KERNEL);
653 INIT_LIST_HEAD(&nfit_flush->list);
654 memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
655 list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
656 dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
657 flush->device_handle, flush->hint_count);
661 static void *add_table(struct acpi_nfit_desc *acpi_desc,
662 struct nfit_table_prev *prev, void *table, const void *end)
664 struct device *dev = acpi_desc->dev;
665 struct acpi_nfit_header *hdr;
666 void *err = ERR_PTR(-ENOMEM);
673 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
679 case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
680 if (!add_spa(acpi_desc, prev, table))
683 case ACPI_NFIT_TYPE_MEMORY_MAP:
684 if (!add_memdev(acpi_desc, prev, table))
687 case ACPI_NFIT_TYPE_CONTROL_REGION:
688 if (!add_dcr(acpi_desc, prev, table))
691 case ACPI_NFIT_TYPE_DATA_REGION:
692 if (!add_bdw(acpi_desc, prev, table))
695 case ACPI_NFIT_TYPE_INTERLEAVE:
696 if (!add_idt(acpi_desc, prev, table))
699 case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
700 if (!add_flush(acpi_desc, prev, table))
703 case ACPI_NFIT_TYPE_SMBIOS:
704 dev_dbg(dev, "%s: smbios\n", __func__);
707 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
711 return table + hdr->length;
714 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
715 struct nfit_mem *nfit_mem)
717 u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
718 u16 dcr = nfit_mem->dcr->region_index;
719 struct nfit_spa *nfit_spa;
721 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
722 u16 range_index = nfit_spa->spa->range_index;
723 int type = nfit_spa_type(nfit_spa->spa);
724 struct nfit_memdev *nfit_memdev;
726 if (type != NFIT_SPA_BDW)
729 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
730 if (nfit_memdev->memdev->range_index != range_index)
732 if (nfit_memdev->memdev->device_handle != device_handle)
734 if (nfit_memdev->memdev->region_index != dcr)
737 nfit_mem->spa_bdw = nfit_spa->spa;
742 dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
743 nfit_mem->spa_dcr->range_index);
744 nfit_mem->bdw = NULL;
747 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
748 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
750 u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
751 struct nfit_memdev *nfit_memdev;
752 struct nfit_bdw *nfit_bdw;
753 struct nfit_idt *nfit_idt;
754 u16 idt_idx, range_index;
756 list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
757 if (nfit_bdw->bdw->region_index != dcr)
759 nfit_mem->bdw = nfit_bdw->bdw;
766 nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
768 if (!nfit_mem->spa_bdw)
771 range_index = nfit_mem->spa_bdw->range_index;
772 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
773 if (nfit_memdev->memdev->range_index != range_index ||
774 nfit_memdev->memdev->region_index != dcr)
776 nfit_mem->memdev_bdw = nfit_memdev->memdev;
777 idt_idx = nfit_memdev->memdev->interleave_index;
778 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
779 if (nfit_idt->idt->interleave_index != idt_idx)
781 nfit_mem->idt_bdw = nfit_idt->idt;
788 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
789 struct acpi_nfit_system_address *spa)
791 struct nfit_mem *nfit_mem, *found;
792 struct nfit_memdev *nfit_memdev;
793 int type = spa ? nfit_spa_type(spa) : 0;
805 * This loop runs in two modes, when a dimm is mapped the loop
806 * adds memdev associations to an existing dimm, or creates a
807 * dimm. In the unmapped dimm case this loop sweeps for memdev
808 * instances with an invalid / zero range_index and adds those
809 * dimms without spa associations.
811 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
812 struct nfit_flush *nfit_flush;
813 struct nfit_dcr *nfit_dcr;
817 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
819 if (!spa && nfit_memdev->memdev->range_index)
822 dcr = nfit_memdev->memdev->region_index;
823 device_handle = nfit_memdev->memdev->device_handle;
824 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
825 if (__to_nfit_memdev(nfit_mem)->device_handle
834 nfit_mem = devm_kzalloc(acpi_desc->dev,
835 sizeof(*nfit_mem), GFP_KERNEL);
838 INIT_LIST_HEAD(&nfit_mem->list);
839 nfit_mem->acpi_desc = acpi_desc;
840 list_add(&nfit_mem->list, &acpi_desc->dimms);
843 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
844 if (nfit_dcr->dcr->region_index != dcr)
847 * Record the control region for the dimm. For
848 * the ACPI 6.1 case, where there are separate
849 * control regions for the pmem vs blk
850 * interfaces, be sure to record the extended
854 nfit_mem->dcr = nfit_dcr->dcr;
855 else if (nfit_mem->dcr->windows == 0
856 && nfit_dcr->dcr->windows)
857 nfit_mem->dcr = nfit_dcr->dcr;
861 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
862 struct acpi_nfit_flush_address *flush;
865 if (nfit_flush->flush->device_handle != device_handle)
867 nfit_mem->nfit_flush = nfit_flush;
868 flush = nfit_flush->flush;
869 nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
871 * sizeof(struct resource), GFP_KERNEL);
872 if (!nfit_mem->flush_wpq)
874 for (i = 0; i < flush->hint_count; i++) {
875 struct resource *res = &nfit_mem->flush_wpq[i];
877 res->start = flush->hint_address[i];
878 res->end = res->start + 8 - 1;
883 if (dcr && !nfit_mem->dcr) {
884 dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
885 spa->range_index, dcr);
889 if (type == NFIT_SPA_DCR) {
890 struct nfit_idt *nfit_idt;
893 /* multiple dimms may share a SPA when interleaved */
894 nfit_mem->spa_dcr = spa;
895 nfit_mem->memdev_dcr = nfit_memdev->memdev;
896 idt_idx = nfit_memdev->memdev->interleave_index;
897 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
898 if (nfit_idt->idt->interleave_index != idt_idx)
900 nfit_mem->idt_dcr = nfit_idt->idt;
903 nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
904 } else if (type == NFIT_SPA_PM) {
906 * A single dimm may belong to multiple SPA-PM
907 * ranges, record at least one in addition to
910 nfit_mem->memdev_pmem = nfit_memdev->memdev;
912 nfit_mem->memdev_dcr = nfit_memdev->memdev;
918 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
920 struct nfit_mem *a = container_of(_a, typeof(*a), list);
921 struct nfit_mem *b = container_of(_b, typeof(*b), list);
922 u32 handleA, handleB;
924 handleA = __to_nfit_memdev(a)->device_handle;
925 handleB = __to_nfit_memdev(b)->device_handle;
926 if (handleA < handleB)
928 else if (handleA > handleB)
933 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
935 struct nfit_spa *nfit_spa;
940 * For each SPA-DCR or SPA-PMEM address range find its
941 * corresponding MEMDEV(s). From each MEMDEV find the
942 * corresponding DCR. Then, if we're operating on a SPA-DCR,
943 * try to find a SPA-BDW and a corresponding BDW that references
944 * the DCR. Throw it all into an nfit_mem object. Note, that
947 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
948 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
954 * If a DIMM has failed to be mapped into SPA there will be no
955 * SPA entries above. Find and register all the unmapped DIMMs
956 * for reporting and recovery purposes.
958 rc = __nfit_mem_init(acpi_desc, NULL);
962 list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
967 static ssize_t bus_dsm_mask_show(struct device *dev,
968 struct device_attribute *attr, char *buf)
970 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
971 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
973 return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
975 static struct device_attribute dev_attr_bus_dsm_mask =
976 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
978 static ssize_t revision_show(struct device *dev,
979 struct device_attribute *attr, char *buf)
981 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
982 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
983 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
985 return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
987 static DEVICE_ATTR_RO(revision);
989 static ssize_t hw_error_scrub_show(struct device *dev,
990 struct device_attribute *attr, char *buf)
992 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
993 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
994 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
996 return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1000 * The 'hw_error_scrub' attribute can have the following values written to it:
1001 * '0': Switch to the default mode where an exception will only insert
1002 * the address of the memory error into the poison and badblocks lists.
1003 * '1': Enable a full scrub to happen if an exception for a memory error is
1006 static ssize_t hw_error_scrub_store(struct device *dev,
1007 struct device_attribute *attr, const char *buf, size_t size)
1009 struct nvdimm_bus_descriptor *nd_desc;
1013 rc = kstrtol(buf, 0, &val);
1018 nd_desc = dev_get_drvdata(dev);
1020 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1023 case HW_ERROR_SCRUB_ON:
1024 acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1026 case HW_ERROR_SCRUB_OFF:
1027 acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1039 static DEVICE_ATTR_RW(hw_error_scrub);
1042 * This shows the number of full Address Range Scrubs that have been
1043 * completed since driver load time. Userspace can wait on this using
1044 * select/poll etc. A '+' at the end indicates an ARS is in progress
1046 static ssize_t scrub_show(struct device *dev,
1047 struct device_attribute *attr, char *buf)
1049 struct nvdimm_bus_descriptor *nd_desc;
1050 ssize_t rc = -ENXIO;
1053 nd_desc = dev_get_drvdata(dev);
1055 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1057 mutex_lock(&acpi_desc->init_mutex);
1058 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1059 work_busy(&acpi_desc->work)
1060 && !acpi_desc->cancel ? "+\n" : "\n");
1061 mutex_unlock(&acpi_desc->init_mutex);
1067 static ssize_t scrub_store(struct device *dev,
1068 struct device_attribute *attr, const char *buf, size_t size)
1070 struct nvdimm_bus_descriptor *nd_desc;
1074 rc = kstrtol(buf, 0, &val);
1081 nd_desc = dev_get_drvdata(dev);
1083 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1085 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1092 static DEVICE_ATTR_RW(scrub);
1094 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1096 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1097 const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1098 | 1 << ND_CMD_ARS_STATUS;
1100 return (nd_desc->cmd_mask & mask) == mask;
1103 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1105 struct device *dev = container_of(kobj, struct device, kobj);
1106 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1108 if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1113 static struct attribute *acpi_nfit_attributes[] = {
1114 &dev_attr_revision.attr,
1115 &dev_attr_scrub.attr,
1116 &dev_attr_hw_error_scrub.attr,
1117 &dev_attr_bus_dsm_mask.attr,
1121 static const struct attribute_group acpi_nfit_attribute_group = {
1123 .attrs = acpi_nfit_attributes,
1124 .is_visible = nfit_visible,
1127 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1128 &nvdimm_bus_attribute_group,
1129 &acpi_nfit_attribute_group,
1133 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1135 struct nvdimm *nvdimm = to_nvdimm(dev);
1136 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1138 return __to_nfit_memdev(nfit_mem);
1141 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1143 struct nvdimm *nvdimm = to_nvdimm(dev);
1144 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1146 return nfit_mem->dcr;
1149 static ssize_t handle_show(struct device *dev,
1150 struct device_attribute *attr, char *buf)
1152 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1154 return sprintf(buf, "%#x\n", memdev->device_handle);
1156 static DEVICE_ATTR_RO(handle);
1158 static ssize_t phys_id_show(struct device *dev,
1159 struct device_attribute *attr, char *buf)
1161 struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1163 return sprintf(buf, "%#x\n", memdev->physical_id);
1165 static DEVICE_ATTR_RO(phys_id);
1167 static ssize_t vendor_show(struct device *dev,
1168 struct device_attribute *attr, char *buf)
1170 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1172 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1174 static DEVICE_ATTR_RO(vendor);
1176 static ssize_t rev_id_show(struct device *dev,
1177 struct device_attribute *attr, char *buf)
1179 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1181 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1183 static DEVICE_ATTR_RO(rev_id);
1185 static ssize_t device_show(struct device *dev,
1186 struct device_attribute *attr, char *buf)
1188 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1190 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1192 static DEVICE_ATTR_RO(device);
1194 static ssize_t subsystem_vendor_show(struct device *dev,
1195 struct device_attribute *attr, char *buf)
1197 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1199 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1201 static DEVICE_ATTR_RO(subsystem_vendor);
1203 static ssize_t subsystem_rev_id_show(struct device *dev,
1204 struct device_attribute *attr, char *buf)
1206 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1208 return sprintf(buf, "0x%04x\n",
1209 be16_to_cpu(dcr->subsystem_revision_id));
1211 static DEVICE_ATTR_RO(subsystem_rev_id);
1213 static ssize_t subsystem_device_show(struct device *dev,
1214 struct device_attribute *attr, char *buf)
1216 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1218 return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1220 static DEVICE_ATTR_RO(subsystem_device);
1222 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1224 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1227 if (nfit_mem->memdev_pmem)
1229 if (nfit_mem->memdev_bdw)
1234 static ssize_t format_show(struct device *dev,
1235 struct device_attribute *attr, char *buf)
1237 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1239 return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1241 static DEVICE_ATTR_RO(format);
1243 static ssize_t format1_show(struct device *dev,
1244 struct device_attribute *attr, char *buf)
1247 ssize_t rc = -ENXIO;
1248 struct nfit_mem *nfit_mem;
1249 struct nfit_memdev *nfit_memdev;
1250 struct acpi_nfit_desc *acpi_desc;
1251 struct nvdimm *nvdimm = to_nvdimm(dev);
1252 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1254 nfit_mem = nvdimm_provider_data(nvdimm);
1255 acpi_desc = nfit_mem->acpi_desc;
1256 handle = to_nfit_memdev(dev)->device_handle;
1258 /* assumes DIMMs have at most 2 published interface codes */
1259 mutex_lock(&acpi_desc->init_mutex);
1260 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1261 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1262 struct nfit_dcr *nfit_dcr;
1264 if (memdev->device_handle != handle)
1267 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1268 if (nfit_dcr->dcr->region_index != memdev->region_index)
1270 if (nfit_dcr->dcr->code == dcr->code)
1272 rc = sprintf(buf, "0x%04x\n",
1273 le16_to_cpu(nfit_dcr->dcr->code));
1279 mutex_unlock(&acpi_desc->init_mutex);
1282 static DEVICE_ATTR_RO(format1);
1284 static ssize_t formats_show(struct device *dev,
1285 struct device_attribute *attr, char *buf)
1287 struct nvdimm *nvdimm = to_nvdimm(dev);
1289 return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1291 static DEVICE_ATTR_RO(formats);
1293 static ssize_t serial_show(struct device *dev,
1294 struct device_attribute *attr, char *buf)
1296 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1298 return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1300 static DEVICE_ATTR_RO(serial);
1302 static ssize_t family_show(struct device *dev,
1303 struct device_attribute *attr, char *buf)
1305 struct nvdimm *nvdimm = to_nvdimm(dev);
1306 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1308 if (nfit_mem->family < 0)
1310 return sprintf(buf, "%d\n", nfit_mem->family);
1312 static DEVICE_ATTR_RO(family);
1314 static ssize_t dsm_mask_show(struct device *dev,
1315 struct device_attribute *attr, char *buf)
1317 struct nvdimm *nvdimm = to_nvdimm(dev);
1318 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1320 if (nfit_mem->family < 0)
1322 return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1324 static DEVICE_ATTR_RO(dsm_mask);
1326 static ssize_t flags_show(struct device *dev,
1327 struct device_attribute *attr, char *buf)
1329 u16 flags = to_nfit_memdev(dev)->flags;
1331 return sprintf(buf, "%s%s%s%s%s%s%s\n",
1332 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1333 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1334 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1335 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1336 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1337 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1338 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1340 static DEVICE_ATTR_RO(flags);
1342 static ssize_t id_show(struct device *dev,
1343 struct device_attribute *attr, char *buf)
1345 struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1347 if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1348 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1349 be16_to_cpu(dcr->vendor_id),
1350 dcr->manufacturing_location,
1351 be16_to_cpu(dcr->manufacturing_date),
1352 be32_to_cpu(dcr->serial_number));
1354 return sprintf(buf, "%04x-%08x\n",
1355 be16_to_cpu(dcr->vendor_id),
1356 be32_to_cpu(dcr->serial_number));
1358 static DEVICE_ATTR_RO(id);
1360 static struct attribute *acpi_nfit_dimm_attributes[] = {
1361 &dev_attr_handle.attr,
1362 &dev_attr_phys_id.attr,
1363 &dev_attr_vendor.attr,
1364 &dev_attr_device.attr,
1365 &dev_attr_rev_id.attr,
1366 &dev_attr_subsystem_vendor.attr,
1367 &dev_attr_subsystem_device.attr,
1368 &dev_attr_subsystem_rev_id.attr,
1369 &dev_attr_format.attr,
1370 &dev_attr_formats.attr,
1371 &dev_attr_format1.attr,
1372 &dev_attr_serial.attr,
1373 &dev_attr_flags.attr,
1375 &dev_attr_family.attr,
1376 &dev_attr_dsm_mask.attr,
1380 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1381 struct attribute *a, int n)
1383 struct device *dev = container_of(kobj, struct device, kobj);
1384 struct nvdimm *nvdimm = to_nvdimm(dev);
1386 if (!to_nfit_dcr(dev)) {
1387 /* Without a dcr only the memdev attributes can be surfaced */
1388 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1389 || a == &dev_attr_flags.attr
1390 || a == &dev_attr_family.attr
1391 || a == &dev_attr_dsm_mask.attr)
1396 if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1401 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1403 .attrs = acpi_nfit_dimm_attributes,
1404 .is_visible = acpi_nfit_dimm_attr_visible,
1407 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1408 &nvdimm_attribute_group,
1409 &nd_device_attribute_group,
1410 &acpi_nfit_dimm_attribute_group,
1414 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1417 struct nfit_mem *nfit_mem;
1419 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1420 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1421 return nfit_mem->nvdimm;
1426 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1428 struct nfit_mem *nfit_mem;
1429 struct acpi_nfit_desc *acpi_desc;
1431 dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1434 if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1435 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1440 acpi_desc = dev_get_drvdata(dev->parent);
1445 * If we successfully retrieved acpi_desc, then we know nfit_mem data
1448 nfit_mem = dev_get_drvdata(dev);
1449 if (nfit_mem && nfit_mem->flags_attr)
1450 sysfs_notify_dirent(nfit_mem->flags_attr);
1452 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1454 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1456 struct acpi_device *adev = data;
1457 struct device *dev = &adev->dev;
1459 device_lock(dev->parent);
1460 __acpi_nvdimm_notify(dev, event);
1461 device_unlock(dev->parent);
1464 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1465 struct nfit_mem *nfit_mem, u32 device_handle)
1467 struct acpi_device *adev, *adev_dimm;
1468 struct device *dev = acpi_desc->dev;
1469 unsigned long dsm_mask;
1474 /* nfit test assumes 1:1 relationship between commands and dsms */
1475 nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1476 nfit_mem->family = NVDIMM_FAMILY_INTEL;
1477 adev = to_acpi_dev(acpi_desc);
1481 adev_dimm = acpi_find_child_device(adev, device_handle, false);
1482 nfit_mem->adev = adev_dimm;
1484 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1486 return force_enable_dimms ? 0 : -ENODEV;
1489 if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1490 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1491 dev_err(dev, "%s: notification registration failed\n",
1492 dev_name(&adev_dimm->dev));
1496 * Record nfit_mem for the notification path to track back to
1497 * the nfit sysfs attributes for this dimm device object.
1499 dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1502 * Until standardization materializes we need to consider 4
1503 * different command sets. Note, that checking for function0 (bit0)
1504 * tells us if any commands are reachable through this GUID.
1506 for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1507 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1508 if (family < 0 || i == default_dsm_family)
1511 /* limit the supported commands to those that are publicly documented */
1512 nfit_mem->family = family;
1513 if (override_dsm_mask && !disable_vendor_specific)
1514 dsm_mask = override_dsm_mask;
1515 else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1517 if (disable_vendor_specific)
1518 dsm_mask &= ~(1 << ND_CMD_VENDOR);
1519 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1520 dsm_mask = 0x1c3c76;
1521 } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1523 if (disable_vendor_specific)
1524 dsm_mask &= ~(1 << 8);
1525 } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1526 dsm_mask = 0xffffffff;
1528 dev_dbg(dev, "unknown dimm command family\n");
1529 nfit_mem->family = -1;
1530 /* DSMs are optional, continue loading the driver... */
1535 * Function 0 is the command interrogation function, don't
1536 * export it to potential userspace use, and enable it to be
1537 * used as an error value in acpi_nfit_ctl().
1541 guid = to_nfit_uuid(nfit_mem->family);
1542 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1543 if (acpi_check_dsm(adev_dimm->handle, guid, 1, 1ULL << i))
1544 set_bit(i, &nfit_mem->dsm_mask);
1549 static void shutdown_dimm_notify(void *data)
1551 struct acpi_nfit_desc *acpi_desc = data;
1552 struct nfit_mem *nfit_mem;
1554 mutex_lock(&acpi_desc->init_mutex);
1556 * Clear out the nfit_mem->flags_attr and shut down dimm event
1559 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1560 struct acpi_device *adev_dimm = nfit_mem->adev;
1562 if (nfit_mem->flags_attr) {
1563 sysfs_put(nfit_mem->flags_attr);
1564 nfit_mem->flags_attr = NULL;
1567 acpi_remove_notify_handler(adev_dimm->handle,
1568 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1569 dev_set_drvdata(&adev_dimm->dev, NULL);
1572 mutex_unlock(&acpi_desc->init_mutex);
1575 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1577 struct nfit_mem *nfit_mem;
1578 int dimm_count = 0, rc;
1579 struct nvdimm *nvdimm;
1581 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1582 struct acpi_nfit_flush_address *flush;
1583 unsigned long flags = 0, cmd_mask;
1584 struct nfit_memdev *nfit_memdev;
1588 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1589 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1595 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1596 set_bit(NDD_ALIASING, &flags);
1598 /* collate flags across all memdevs for this dimm */
1599 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1600 struct acpi_nfit_memory_map *dimm_memdev;
1602 dimm_memdev = __to_nfit_memdev(nfit_mem);
1603 if (dimm_memdev->device_handle
1604 != nfit_memdev->memdev->device_handle)
1606 dimm_memdev->flags |= nfit_memdev->memdev->flags;
1609 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1610 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1611 set_bit(NDD_UNARMED, &flags);
1613 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1618 * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1619 * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1620 * userspace interface.
1622 cmd_mask = 1UL << ND_CMD_CALL;
1623 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1624 cmd_mask |= nfit_mem->dsm_mask;
1626 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1628 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1629 acpi_nfit_dimm_attribute_groups,
1630 flags, cmd_mask, flush ? flush->hint_count : 0,
1631 nfit_mem->flush_wpq);
1635 nfit_mem->nvdimm = nvdimm;
1638 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1641 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1642 nvdimm_name(nvdimm),
1643 mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1644 mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1645 mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1646 mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1647 mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1651 rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1656 * Now that dimms are successfully registered, and async registration
1657 * is flushed, attempt to enable event notification.
1659 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1660 struct kernfs_node *nfit_kernfs;
1662 nvdimm = nfit_mem->nvdimm;
1666 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1668 nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1670 sysfs_put(nfit_kernfs);
1671 if (!nfit_mem->flags_attr)
1672 dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1673 nvdimm_name(nvdimm));
1676 return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1681 * These constants are private because there are no kernel consumers of
1684 enum nfit_aux_cmds {
1685 NFIT_CMD_TRANSLATE_SPA = 5,
1686 NFIT_CMD_ARS_INJECT_SET = 7,
1687 NFIT_CMD_ARS_INJECT_CLEAR = 8,
1688 NFIT_CMD_ARS_INJECT_GET = 9,
1691 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1693 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1694 const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1695 struct acpi_device *adev;
1696 unsigned long dsm_mask;
1699 nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1700 adev = to_acpi_dev(acpi_desc);
1704 for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1705 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1706 set_bit(i, &nd_desc->cmd_mask);
1707 set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1710 (1 << ND_CMD_ARS_CAP) |
1711 (1 << ND_CMD_ARS_START) |
1712 (1 << ND_CMD_ARS_STATUS) |
1713 (1 << ND_CMD_CLEAR_ERROR) |
1714 (1 << NFIT_CMD_TRANSLATE_SPA) |
1715 (1 << NFIT_CMD_ARS_INJECT_SET) |
1716 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1717 (1 << NFIT_CMD_ARS_INJECT_GET);
1718 for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1719 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1720 set_bit(i, &nd_desc->bus_dsm_mask);
1723 static ssize_t range_index_show(struct device *dev,
1724 struct device_attribute *attr, char *buf)
1726 struct nd_region *nd_region = to_nd_region(dev);
1727 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1729 return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1731 static DEVICE_ATTR_RO(range_index);
1733 static ssize_t ecc_unit_size_show(struct device *dev,
1734 struct device_attribute *attr, char *buf)
1736 struct nd_region *nd_region = to_nd_region(dev);
1737 struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1739 return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1741 static DEVICE_ATTR_RO(ecc_unit_size);
1743 static struct attribute *acpi_nfit_region_attributes[] = {
1744 &dev_attr_range_index.attr,
1745 &dev_attr_ecc_unit_size.attr,
1749 static const struct attribute_group acpi_nfit_region_attribute_group = {
1751 .attrs = acpi_nfit_region_attributes,
1754 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1755 &nd_region_attribute_group,
1756 &nd_mapping_attribute_group,
1757 &nd_device_attribute_group,
1758 &nd_numa_attribute_group,
1759 &acpi_nfit_region_attribute_group,
1763 /* enough info to uniquely specify an interleave set */
1764 struct nfit_set_info {
1765 struct nfit_set_info_map {
1772 struct nfit_set_info2 {
1773 struct nfit_set_info_map2 {
1777 u16 manufacturing_date;
1778 u8 manufacturing_location;
1783 static size_t sizeof_nfit_set_info(int num_mappings)
1785 return sizeof(struct nfit_set_info)
1786 + num_mappings * sizeof(struct nfit_set_info_map);
1789 static size_t sizeof_nfit_set_info2(int num_mappings)
1791 return sizeof(struct nfit_set_info2)
1792 + num_mappings * sizeof(struct nfit_set_info_map2);
1795 static int cmp_map_compat(const void *m0, const void *m1)
1797 const struct nfit_set_info_map *map0 = m0;
1798 const struct nfit_set_info_map *map1 = m1;
1800 return memcmp(&map0->region_offset, &map1->region_offset,
1804 static int cmp_map(const void *m0, const void *m1)
1806 const struct nfit_set_info_map *map0 = m0;
1807 const struct nfit_set_info_map *map1 = m1;
1809 if (map0->region_offset < map1->region_offset)
1811 else if (map0->region_offset > map1->region_offset)
1816 static int cmp_map2(const void *m0, const void *m1)
1818 const struct nfit_set_info_map2 *map0 = m0;
1819 const struct nfit_set_info_map2 *map1 = m1;
1821 if (map0->region_offset < map1->region_offset)
1823 else if (map0->region_offset > map1->region_offset)
1828 /* Retrieve the nth entry referencing this spa */
1829 static struct acpi_nfit_memory_map *memdev_from_spa(
1830 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1832 struct nfit_memdev *nfit_memdev;
1834 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1835 if (nfit_memdev->memdev->range_index == range_index)
1837 return nfit_memdev->memdev;
1841 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1842 struct nd_region_desc *ndr_desc,
1843 struct acpi_nfit_system_address *spa)
1845 struct device *dev = acpi_desc->dev;
1846 struct nd_interleave_set *nd_set;
1847 u16 nr = ndr_desc->num_mappings;
1848 struct nfit_set_info2 *info2;
1849 struct nfit_set_info *info;
1852 nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1855 ndr_desc->nd_set = nd_set;
1856 guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
1858 info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1862 info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
1866 for (i = 0; i < nr; i++) {
1867 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1868 struct nfit_set_info_map *map = &info->mapping[i];
1869 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
1870 struct nvdimm *nvdimm = mapping->nvdimm;
1871 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1872 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1873 spa->range_index, i);
1874 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1876 if (!memdev || !nfit_mem->dcr) {
1877 dev_err(dev, "%s: failed to find DCR\n", __func__);
1881 map->region_offset = memdev->region_offset;
1882 map->serial_number = dcr->serial_number;
1884 map2->region_offset = memdev->region_offset;
1885 map2->serial_number = dcr->serial_number;
1886 map2->vendor_id = dcr->vendor_id;
1887 map2->manufacturing_date = dcr->manufacturing_date;
1888 map2->manufacturing_location = dcr->manufacturing_location;
1891 /* v1.1 namespaces */
1892 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1894 nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1896 /* v1.2 namespaces */
1897 sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
1899 nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
1901 /* support v1.1 namespaces created with the wrong sort order */
1902 sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1903 cmp_map_compat, NULL);
1904 nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1906 /* record the result of the sort for the mapping position */
1907 for (i = 0; i < nr; i++) {
1908 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
1911 for (j = 0; j < nr; j++) {
1912 struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
1913 struct nvdimm *nvdimm = mapping->nvdimm;
1914 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1915 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
1917 if (map2->serial_number == dcr->serial_number &&
1918 map2->vendor_id == dcr->vendor_id &&
1919 map2->manufacturing_date == dcr->manufacturing_date &&
1920 map2->manufacturing_location
1921 == dcr->manufacturing_location) {
1922 mapping->position = i;
1928 ndr_desc->nd_set = nd_set;
1929 devm_kfree(dev, info);
1930 devm_kfree(dev, info2);
1935 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1937 struct acpi_nfit_interleave *idt = mmio->idt;
1938 u32 sub_line_offset, line_index, line_offset;
1939 u64 line_no, table_skip_count, table_offset;
1941 line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1942 table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1943 line_offset = idt->line_offset[line_index]
1945 table_offset = table_skip_count * mmio->table_size;
1947 return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1950 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1952 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1953 u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1954 const u32 STATUS_MASK = 0x80000037;
1956 if (mmio->num_lines)
1957 offset = to_interleave_offset(offset, mmio);
1959 return readl(mmio->addr.base + offset) & STATUS_MASK;
1962 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1963 resource_size_t dpa, unsigned int len, unsigned int write)
1966 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1969 BCW_OFFSET_MASK = (1ULL << 48)-1,
1971 BCW_LEN_MASK = (1ULL << 8) - 1,
1975 cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1976 len = len >> L1_CACHE_SHIFT;
1977 cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1978 cmd |= ((u64) write) << BCW_CMD_SHIFT;
1980 offset = nfit_blk->cmd_offset + mmio->size * bw;
1981 if (mmio->num_lines)
1982 offset = to_interleave_offset(offset, mmio);
1984 writeq(cmd, mmio->addr.base + offset);
1985 nvdimm_flush(nfit_blk->nd_region);
1987 if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1988 readq(mmio->addr.base + offset);
1991 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1992 resource_size_t dpa, void *iobuf, size_t len, int rw,
1995 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1996 unsigned int copied = 0;
2000 base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2001 + lane * mmio->size;
2002 write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2007 if (mmio->num_lines) {
2010 offset = to_interleave_offset(base_offset + copied,
2012 div_u64_rem(offset, mmio->line_size, &line_offset);
2013 c = min_t(size_t, len, mmio->line_size - line_offset);
2015 offset = base_offset + nfit_blk->bdw_offset;
2020 memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2022 if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2023 arch_invalidate_pmem((void __force *)
2024 mmio->addr.aperture + offset, c);
2026 memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2034 nvdimm_flush(nfit_blk->nd_region);
2036 rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2040 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2041 resource_size_t dpa, void *iobuf, u64 len, int rw)
2043 struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2044 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2045 struct nd_region *nd_region = nfit_blk->nd_region;
2046 unsigned int lane, copied = 0;
2049 lane = nd_region_acquire_lane(nd_region);
2051 u64 c = min(len, mmio->size);
2053 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2054 iobuf + copied, c, rw, lane);
2061 nd_region_release_lane(nd_region, lane);
2066 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2067 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2070 mmio->num_lines = idt->line_count;
2071 mmio->line_size = idt->line_size;
2072 if (interleave_ways == 0)
2074 mmio->table_size = mmio->num_lines * interleave_ways
2081 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2082 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2084 struct nd_cmd_dimm_flags flags;
2087 memset(&flags, 0, sizeof(flags));
2088 rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2089 sizeof(flags), NULL);
2091 if (rc >= 0 && flags.status == 0)
2092 nfit_blk->dimm_flags = flags.flags;
2093 else if (rc == -ENOTTY) {
2094 /* fall back to a conservative default */
2095 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2103 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2106 struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2107 struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2108 struct nfit_blk_mmio *mmio;
2109 struct nfit_blk *nfit_blk;
2110 struct nfit_mem *nfit_mem;
2111 struct nvdimm *nvdimm;
2114 nvdimm = nd_blk_region_to_dimm(ndbr);
2115 nfit_mem = nvdimm_provider_data(nvdimm);
2116 if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2117 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2118 nfit_mem ? "" : " nfit_mem",
2119 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2120 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2124 nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2127 nd_blk_region_set_provider_data(ndbr, nfit_blk);
2128 nfit_blk->nd_region = to_nd_region(dev);
2130 /* map block aperture memory */
2131 nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2132 mmio = &nfit_blk->mmio[BDW];
2133 mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2134 nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2135 if (!mmio->addr.base) {
2136 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2137 nvdimm_name(nvdimm));
2140 mmio->size = nfit_mem->bdw->size;
2141 mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2142 mmio->idt = nfit_mem->idt_bdw;
2143 mmio->spa = nfit_mem->spa_bdw;
2144 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2145 nfit_mem->memdev_bdw->interleave_ways);
2147 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2148 __func__, nvdimm_name(nvdimm));
2152 /* map block control memory */
2153 nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2154 nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2155 mmio = &nfit_blk->mmio[DCR];
2156 mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2157 nfit_mem->spa_dcr->length);
2158 if (!mmio->addr.base) {
2159 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2160 nvdimm_name(nvdimm));
2163 mmio->size = nfit_mem->dcr->window_size;
2164 mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2165 mmio->idt = nfit_mem->idt_dcr;
2166 mmio->spa = nfit_mem->spa_dcr;
2167 rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2168 nfit_mem->memdev_dcr->interleave_ways);
2170 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2171 __func__, nvdimm_name(nvdimm));
2175 rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2177 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2178 __func__, nvdimm_name(nvdimm));
2182 if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2183 dev_warn(dev, "unable to guarantee persistence of writes\n");
2185 if (mmio->line_size == 0)
2188 if ((u32) nfit_blk->cmd_offset % mmio->line_size
2189 + 8 > mmio->line_size) {
2190 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2192 } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2193 + 8 > mmio->line_size) {
2194 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2201 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2202 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2204 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2205 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2208 cmd->address = spa->address;
2209 cmd->length = spa->length;
2210 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2211 sizeof(*cmd), &cmd_rc);
2217 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2221 struct nd_cmd_ars_start ars_start;
2222 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2223 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2225 memset(&ars_start, 0, sizeof(ars_start));
2226 ars_start.address = spa->address;
2227 ars_start.length = spa->length;
2228 ars_start.flags = acpi_desc->ars_start_flags;
2229 if (nfit_spa_type(spa) == NFIT_SPA_PM)
2230 ars_start.type = ND_ARS_PERSISTENT;
2231 else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2232 ars_start.type = ND_ARS_VOLATILE;
2236 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2237 sizeof(ars_start), &cmd_rc);
2244 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2247 struct nd_cmd_ars_start ars_start;
2248 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2249 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2251 memset(&ars_start, 0, sizeof(ars_start));
2252 ars_start.address = ars_status->restart_address;
2253 ars_start.length = ars_status->restart_length;
2254 ars_start.type = ars_status->type;
2255 ars_start.flags = acpi_desc->ars_start_flags;
2256 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2257 sizeof(ars_start), &cmd_rc);
2263 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2265 struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2266 struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2269 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2270 acpi_desc->ars_status_size, &cmd_rc);
2276 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2277 struct nd_cmd_ars_status *ars_status)
2279 struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2284 * First record starts at 44 byte offset from the start of the
2287 if (ars_status->out_length < 44)
2289 for (i = 0; i < ars_status->num_records; i++) {
2290 /* only process full records */
2291 if (ars_status->out_length
2292 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2294 rc = nvdimm_bus_add_poison(nvdimm_bus,
2295 ars_status->records[i].err_address,
2296 ars_status->records[i].length);
2300 if (i < ars_status->num_records)
2301 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2306 static void acpi_nfit_remove_resource(void *data)
2308 struct resource *res = data;
2310 remove_resource(res);
2313 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2314 struct nd_region_desc *ndr_desc)
2316 struct resource *res, *nd_res = ndr_desc->res;
2319 /* No operation if the region is already registered as PMEM */
2320 is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2321 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2322 if (is_pmem == REGION_INTERSECTS)
2325 res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2329 res->name = "Persistent Memory";
2330 res->start = nd_res->start;
2331 res->end = nd_res->end;
2332 res->flags = IORESOURCE_MEM;
2333 res->desc = IORES_DESC_PERSISTENT_MEMORY;
2335 ret = insert_resource(&iomem_resource, res);
2339 ret = devm_add_action_or_reset(acpi_desc->dev,
2340 acpi_nfit_remove_resource,
2348 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2349 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2350 struct acpi_nfit_memory_map *memdev,
2351 struct nfit_spa *nfit_spa)
2353 struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2354 memdev->device_handle);
2355 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2356 struct nd_blk_region_desc *ndbr_desc;
2357 struct nfit_mem *nfit_mem;
2361 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2362 spa->range_index, memdev->device_handle);
2366 mapping->nvdimm = nvdimm;
2367 switch (nfit_spa_type(spa)) {
2369 case NFIT_SPA_VOLATILE:
2370 mapping->start = memdev->address;
2371 mapping->size = memdev->region_size;
2374 nfit_mem = nvdimm_provider_data(nvdimm);
2375 if (!nfit_mem || !nfit_mem->bdw) {
2376 dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2377 spa->range_index, nvdimm_name(nvdimm));
2381 mapping->size = nfit_mem->bdw->capacity;
2382 mapping->start = nfit_mem->bdw->start_address;
2383 ndr_desc->num_lanes = nfit_mem->bdw->windows;
2384 ndr_desc->mapping = mapping;
2385 ndr_desc->num_mappings = 1;
2386 ndbr_desc = to_blk_region_desc(ndr_desc);
2387 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2388 ndbr_desc->do_io = acpi_desc->blk_do_io;
2389 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2392 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2394 if (!nfit_spa->nd_region)
2402 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2404 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2405 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2406 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2407 nfit_spa_type(spa) == NFIT_SPA_PCD);
2410 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2412 return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2413 nfit_spa_type(spa) == NFIT_SPA_VCD ||
2414 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2417 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2418 struct nfit_spa *nfit_spa)
2420 static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2421 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2422 struct nd_blk_region_desc ndbr_desc;
2423 struct nd_region_desc *ndr_desc;
2424 struct nfit_memdev *nfit_memdev;
2425 struct nvdimm_bus *nvdimm_bus;
2426 struct resource res;
2429 if (nfit_spa->nd_region)
2432 if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2433 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2438 memset(&res, 0, sizeof(res));
2439 memset(&mappings, 0, sizeof(mappings));
2440 memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2441 res.start = spa->address;
2442 res.end = res.start + spa->length - 1;
2443 ndr_desc = &ndbr_desc.ndr_desc;
2444 ndr_desc->res = &res;
2445 ndr_desc->provider_data = nfit_spa;
2446 ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2447 if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2448 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2449 spa->proximity_domain);
2451 ndr_desc->numa_node = NUMA_NO_NODE;
2453 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2454 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2455 struct nd_mapping_desc *mapping;
2457 /* range index 0 == unmapped in SPA or invalid-SPA */
2458 if (memdev->range_index == 0 || spa->range_index == 0)
2460 if (memdev->range_index != spa->range_index)
2462 if (count >= ND_MAX_MAPPINGS) {
2463 dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2464 spa->range_index, ND_MAX_MAPPINGS);
2467 mapping = &mappings[count++];
2468 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2474 ndr_desc->mapping = mappings;
2475 ndr_desc->num_mappings = count;
2476 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2480 nvdimm_bus = acpi_desc->nvdimm_bus;
2481 if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2482 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2484 dev_warn(acpi_desc->dev,
2485 "failed to insert pmem resource to iomem: %d\n",
2490 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2492 if (!nfit_spa->nd_region)
2494 } else if (nfit_spa_is_volatile(spa)) {
2495 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2497 if (!nfit_spa->nd_region)
2499 } else if (nfit_spa_is_virtual(spa)) {
2500 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2502 if (!nfit_spa->nd_region)
2508 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2509 nfit_spa->spa->range_index);
2513 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2516 struct device *dev = acpi_desc->dev;
2517 struct nd_cmd_ars_status *ars_status;
2519 if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2520 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2524 if (acpi_desc->ars_status)
2525 devm_kfree(dev, acpi_desc->ars_status);
2526 acpi_desc->ars_status = NULL;
2527 ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2530 acpi_desc->ars_status = ars_status;
2531 acpi_desc->ars_status_size = max_ars;
2535 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2536 struct nfit_spa *nfit_spa)
2538 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2541 if (!nfit_spa->max_ars) {
2542 struct nd_cmd_ars_cap ars_cap;
2544 memset(&ars_cap, 0, sizeof(ars_cap));
2545 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2548 nfit_spa->max_ars = ars_cap.max_ars_out;
2549 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2550 /* check that the supported scrub types match the spa type */
2551 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2552 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2554 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2555 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2559 if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2562 rc = ars_get_status(acpi_desc);
2563 if (rc < 0 && rc != -ENOSPC)
2566 if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2572 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2573 struct nfit_spa *nfit_spa)
2575 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2576 unsigned int overflow_retry = scrub_overflow_abort;
2577 u64 init_ars_start = 0, init_ars_len = 0;
2578 struct device *dev = acpi_desc->dev;
2579 unsigned int tmo = scrub_timeout;
2582 if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2585 rc = ars_start(acpi_desc, nfit_spa);
2587 * If we timed out the initial scan we'll still be busy here,
2588 * and will wait another timeout before giving up permanently.
2590 if (rc < 0 && rc != -EBUSY)
2594 u64 ars_start, ars_len;
2596 if (acpi_desc->cancel)
2598 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2601 if (rc == -EBUSY && !tmo) {
2602 dev_warn(dev, "range %d ars timeout, aborting\n",
2609 * Note, entries may be appended to the list
2610 * while the lock is dropped, but the workqueue
2611 * being active prevents entries being deleted /
2614 mutex_unlock(&acpi_desc->init_mutex);
2617 mutex_lock(&acpi_desc->init_mutex);
2621 /* we got some results, but there are more pending... */
2622 if (rc == -ENOSPC && overflow_retry--) {
2623 if (!init_ars_len) {
2624 init_ars_len = acpi_desc->ars_status->length;
2625 init_ars_start = acpi_desc->ars_status->address;
2627 rc = ars_continue(acpi_desc);
2631 dev_warn(dev, "range %d ars continuation failed\n",
2637 ars_start = init_ars_start;
2638 ars_len = init_ars_len;
2640 ars_start = acpi_desc->ars_status->address;
2641 ars_len = acpi_desc->ars_status->length;
2643 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2644 spa->range_index, ars_start, ars_len);
2645 /* notify the region about new poison entries */
2646 nvdimm_region_notify(nfit_spa->nd_region,
2647 NVDIMM_REVALIDATE_POISON);
2652 static void acpi_nfit_scrub(struct work_struct *work)
2655 u64 init_scrub_length = 0;
2656 struct nfit_spa *nfit_spa;
2657 u64 init_scrub_address = 0;
2658 bool init_ars_done = false;
2659 struct acpi_nfit_desc *acpi_desc;
2660 unsigned int tmo = scrub_timeout;
2661 unsigned int overflow_retry = scrub_overflow_abort;
2663 acpi_desc = container_of(work, typeof(*acpi_desc), work);
2664 dev = acpi_desc->dev;
2667 * We scrub in 2 phases. The first phase waits for any platform
2668 * firmware initiated scrubs to complete and then we go search for the
2669 * affected spa regions to mark them scanned. In the second phase we
2670 * initiate a directed scrub for every range that was not scrubbed in
2671 * phase 1. If we're called for a 'rescan', we harmlessly pass through
2672 * the first phase, but really only care about running phase 2, where
2673 * regions can be notified of new poison.
2676 /* process platform firmware initiated scrubs */
2678 mutex_lock(&acpi_desc->init_mutex);
2679 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2680 struct nd_cmd_ars_status *ars_status;
2681 struct acpi_nfit_system_address *spa;
2682 u64 ars_start, ars_len;
2685 if (acpi_desc->cancel)
2688 if (nfit_spa->nd_region)
2691 if (init_ars_done) {
2693 * No need to re-query, we're now just
2694 * reconciling all the ranges covered by the
2699 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2701 if (rc == -ENOTTY) {
2702 /* no ars capability, just register spa and move on */
2703 acpi_nfit_register_region(acpi_desc, nfit_spa);
2707 if (rc == -EBUSY && !tmo) {
2708 /* fallthrough to directed scrub in phase 2 */
2709 dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2711 } else if (rc == -EBUSY) {
2712 mutex_unlock(&acpi_desc->init_mutex);
2718 /* we got some results, but there are more pending... */
2719 if (rc == -ENOSPC && overflow_retry--) {
2720 ars_status = acpi_desc->ars_status;
2722 * Record the original scrub range, so that we
2723 * can recall all the ranges impacted by the
2726 if (!init_scrub_length) {
2727 init_scrub_length = ars_status->length;
2728 init_scrub_address = ars_status->address;
2730 rc = ars_continue(acpi_desc);
2732 mutex_unlock(&acpi_desc->init_mutex);
2739 * Initial scrub failed, we'll give it one more
2745 /* We got some final results, record completed ranges */
2746 ars_status = acpi_desc->ars_status;
2747 if (init_scrub_length) {
2748 ars_start = init_scrub_address;
2749 ars_len = ars_start + init_scrub_length;
2751 ars_start = ars_status->address;
2752 ars_len = ars_status->length;
2754 spa = nfit_spa->spa;
2756 if (!init_ars_done) {
2757 init_ars_done = true;
2758 dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2759 ars_start, ars_len);
2761 if (ars_start <= spa->address && ars_start + ars_len
2762 >= spa->address + spa->length)
2763 acpi_nfit_register_region(acpi_desc, nfit_spa);
2767 * For all the ranges not covered by an initial scrub we still
2768 * want to see if there are errors, but it's ok to discover them
2771 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2773 * Flag all the ranges that still need scrubbing, but
2774 * register them now to make data available.
2776 if (!nfit_spa->nd_region) {
2777 nfit_spa->ars_required = 1;
2778 acpi_nfit_register_region(acpi_desc, nfit_spa);
2781 acpi_desc->init_complete = 1;
2783 list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2784 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2785 acpi_desc->scrub_count++;
2786 acpi_desc->ars_start_flags = 0;
2787 if (acpi_desc->scrub_count_state)
2788 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2789 mutex_unlock(&acpi_desc->init_mutex);
2792 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2794 struct nfit_spa *nfit_spa;
2796 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2797 int rc, type = nfit_spa_type(nfit_spa->spa);
2799 /* PMEM and VMEM will be registered by the ARS workqueue */
2800 if (type == NFIT_SPA_PM || type == NFIT_SPA_VOLATILE)
2802 /* BLK apertures belong to BLK region registration below */
2803 if (type == NFIT_SPA_BDW)
2805 /* BLK regions don't need to wait for ARS results */
2806 rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2811 acpi_desc->ars_start_flags = 0;
2812 if (!acpi_desc->cancel)
2813 queue_work(nfit_wq, &acpi_desc->work);
2817 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2818 struct nfit_table_prev *prev)
2820 struct device *dev = acpi_desc->dev;
2822 if (!list_empty(&prev->spas) ||
2823 !list_empty(&prev->memdevs) ||
2824 !list_empty(&prev->dcrs) ||
2825 !list_empty(&prev->bdws) ||
2826 !list_empty(&prev->idts) ||
2827 !list_empty(&prev->flushes)) {
2828 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2834 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
2836 struct device *dev = acpi_desc->dev;
2837 struct kernfs_node *nfit;
2838 struct device *bus_dev;
2840 if (!ars_supported(acpi_desc->nvdimm_bus))
2843 bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2844 nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
2846 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
2849 acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
2851 if (!acpi_desc->scrub_count_state) {
2852 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
2859 static void acpi_nfit_unregister(void *data)
2861 struct acpi_nfit_desc *acpi_desc = data;
2863 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2866 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
2868 struct device *dev = acpi_desc->dev;
2869 struct nfit_table_prev prev;
2873 if (!acpi_desc->nvdimm_bus) {
2874 acpi_nfit_init_dsms(acpi_desc);
2876 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
2877 &acpi_desc->nd_desc);
2878 if (!acpi_desc->nvdimm_bus)
2881 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
2886 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
2890 /* register this acpi_desc for mce notifications */
2891 mutex_lock(&acpi_desc_lock);
2892 list_add_tail(&acpi_desc->list, &acpi_descs);
2893 mutex_unlock(&acpi_desc_lock);
2896 mutex_lock(&acpi_desc->init_mutex);
2898 INIT_LIST_HEAD(&prev.spas);
2899 INIT_LIST_HEAD(&prev.memdevs);
2900 INIT_LIST_HEAD(&prev.dcrs);
2901 INIT_LIST_HEAD(&prev.bdws);
2902 INIT_LIST_HEAD(&prev.idts);
2903 INIT_LIST_HEAD(&prev.flushes);
2905 list_cut_position(&prev.spas, &acpi_desc->spas,
2906 acpi_desc->spas.prev);
2907 list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2908 acpi_desc->memdevs.prev);
2909 list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2910 acpi_desc->dcrs.prev);
2911 list_cut_position(&prev.bdws, &acpi_desc->bdws,
2912 acpi_desc->bdws.prev);
2913 list_cut_position(&prev.idts, &acpi_desc->idts,
2914 acpi_desc->idts.prev);
2915 list_cut_position(&prev.flushes, &acpi_desc->flushes,
2916 acpi_desc->flushes.prev);
2919 while (!IS_ERR_OR_NULL(data))
2920 data = add_table(acpi_desc, &prev, data, end);
2923 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2929 rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2933 rc = nfit_mem_init(acpi_desc);
2937 rc = acpi_nfit_register_dimms(acpi_desc);
2941 rc = acpi_nfit_register_regions(acpi_desc);
2944 mutex_unlock(&acpi_desc->init_mutex);
2947 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2949 struct acpi_nfit_flush_work {
2950 struct work_struct work;
2951 struct completion cmp;
2954 static void flush_probe(struct work_struct *work)
2956 struct acpi_nfit_flush_work *flush;
2958 flush = container_of(work, typeof(*flush), work);
2959 complete(&flush->cmp);
2962 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2964 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2965 struct device *dev = acpi_desc->dev;
2966 struct acpi_nfit_flush_work flush;
2969 /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2973 /* bounce the init_mutex to make init_complete valid */
2974 mutex_lock(&acpi_desc->init_mutex);
2975 if (acpi_desc->cancel || acpi_desc->init_complete) {
2976 mutex_unlock(&acpi_desc->init_mutex);
2981 * Scrub work could take 10s of seconds, userspace may give up so we
2982 * need to be interruptible while waiting.
2984 INIT_WORK_ONSTACK(&flush.work, flush_probe);
2985 init_completion(&flush.cmp);
2986 queue_work(nfit_wq, &flush.work);
2987 mutex_unlock(&acpi_desc->init_mutex);
2989 rc = wait_for_completion_interruptible(&flush.cmp);
2990 cancel_work_sync(&flush.work);
2994 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2995 struct nvdimm *nvdimm, unsigned int cmd)
2997 struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3001 if (cmd != ND_CMD_ARS_START)
3005 * The kernel and userspace may race to initiate a scrub, but
3006 * the scrub thread is prepared to lose that initial race. It
3007 * just needs guarantees that any ars it initiates are not
3008 * interrupted by any intervening start reqeusts from userspace.
3010 if (work_busy(&acpi_desc->work))
3016 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
3018 struct device *dev = acpi_desc->dev;
3019 struct nfit_spa *nfit_spa;
3021 if (work_busy(&acpi_desc->work))
3024 mutex_lock(&acpi_desc->init_mutex);
3025 if (acpi_desc->cancel) {
3026 mutex_unlock(&acpi_desc->init_mutex);
3030 list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3031 struct acpi_nfit_system_address *spa = nfit_spa->spa;
3033 if (nfit_spa_type(spa) != NFIT_SPA_PM)
3036 nfit_spa->ars_required = 1;
3038 acpi_desc->ars_start_flags = flags;
3039 queue_work(nfit_wq, &acpi_desc->work);
3040 dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
3041 mutex_unlock(&acpi_desc->init_mutex);
3046 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3048 struct nvdimm_bus_descriptor *nd_desc;
3050 dev_set_drvdata(dev, acpi_desc);
3051 acpi_desc->dev = dev;
3052 acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3053 nd_desc = &acpi_desc->nd_desc;
3054 nd_desc->provider_name = "ACPI.NFIT";
3055 nd_desc->module = THIS_MODULE;
3056 nd_desc->ndctl = acpi_nfit_ctl;
3057 nd_desc->flush_probe = acpi_nfit_flush_probe;
3058 nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3059 nd_desc->attr_groups = acpi_nfit_attribute_groups;
3061 INIT_LIST_HEAD(&acpi_desc->spas);
3062 INIT_LIST_HEAD(&acpi_desc->dcrs);
3063 INIT_LIST_HEAD(&acpi_desc->bdws);
3064 INIT_LIST_HEAD(&acpi_desc->idts);
3065 INIT_LIST_HEAD(&acpi_desc->flushes);
3066 INIT_LIST_HEAD(&acpi_desc->memdevs);
3067 INIT_LIST_HEAD(&acpi_desc->dimms);
3068 INIT_LIST_HEAD(&acpi_desc->list);
3069 mutex_init(&acpi_desc->init_mutex);
3070 INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3072 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
3074 static void acpi_nfit_put_table(void *table)
3076 acpi_put_table(table);
3079 void acpi_nfit_shutdown(void *data)
3081 struct acpi_nfit_desc *acpi_desc = data;
3082 struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3085 * Destruct under acpi_desc_lock so that nfit_handle_mce does not
3088 mutex_lock(&acpi_desc_lock);
3089 list_del(&acpi_desc->list);
3090 mutex_unlock(&acpi_desc_lock);
3092 mutex_lock(&acpi_desc->init_mutex);
3093 acpi_desc->cancel = 1;
3094 mutex_unlock(&acpi_desc->init_mutex);
3097 * Bounce the nvdimm bus lock to make sure any in-flight
3098 * acpi_nfit_ars_rescan() submissions have had a chance to
3099 * either submit or see ->cancel set.
3101 device_lock(bus_dev);
3102 device_unlock(bus_dev);
3104 flush_workqueue(nfit_wq);
3106 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3108 static int acpi_nfit_add(struct acpi_device *adev)
3110 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3111 struct acpi_nfit_desc *acpi_desc;
3112 struct device *dev = &adev->dev;
3113 struct acpi_table_header *tbl;
3114 acpi_status status = AE_OK;
3118 status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3119 if (ACPI_FAILURE(status)) {
3120 /* This is ok, we could have an nvdimm hotplugged later */
3121 dev_dbg(dev, "failed to find NFIT at startup\n");
3125 rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3130 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3133 acpi_nfit_desc_init(acpi_desc, &adev->dev);
3135 /* Save the acpi header for exporting the revision via sysfs */
3136 acpi_desc->acpi_header = *tbl;
3138 /* Evaluate _FIT and override with that if present */
3139 status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3140 if (ACPI_SUCCESS(status) && buf.length > 0) {
3141 union acpi_object *obj = buf.pointer;
3143 if (obj->type == ACPI_TYPE_BUFFER)
3144 rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3145 obj->buffer.length);
3147 dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
3148 __func__, (int) obj->type);
3151 /* skip over the lead-in header table */
3152 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3153 + sizeof(struct acpi_table_nfit),
3154 sz - sizeof(struct acpi_table_nfit));
3158 return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3161 static int acpi_nfit_remove(struct acpi_device *adev)
3163 /* see acpi_nfit_unregister */
3167 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3169 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3170 struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3171 union acpi_object *obj;
3176 /* dev->driver may be null if we're being removed */
3177 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
3182 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3185 acpi_nfit_desc_init(acpi_desc, dev);
3188 * Finish previous registration before considering new
3191 flush_workqueue(nfit_wq);
3195 status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3196 if (ACPI_FAILURE(status)) {
3197 dev_err(dev, "failed to evaluate _FIT\n");
3202 if (obj->type == ACPI_TYPE_BUFFER) {
3203 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3204 obj->buffer.length);
3206 dev_err(dev, "failed to merge updated NFIT\n");
3208 dev_err(dev, "Invalid _FIT\n");
3212 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3214 struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3215 u8 flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3216 0 : ND_ARS_RETURN_PREV_DATA;
3218 acpi_nfit_ars_rescan(acpi_desc, flags);
3221 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3223 dev_dbg(dev, "%s: event: 0x%x\n", __func__, event);
3226 case NFIT_NOTIFY_UPDATE:
3227 return acpi_nfit_update_notify(dev, handle);
3228 case NFIT_NOTIFY_UC_MEMORY_ERROR:
3229 return acpi_nfit_uc_error_notify(dev, handle);
3234 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3236 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3238 device_lock(&adev->dev);
3239 __acpi_nfit_notify(&adev->dev, adev->handle, event);
3240 device_unlock(&adev->dev);
3243 static const struct acpi_device_id acpi_nfit_ids[] = {
3247 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3249 static struct acpi_driver acpi_nfit_driver = {
3250 .name = KBUILD_MODNAME,
3251 .ids = acpi_nfit_ids,
3253 .add = acpi_nfit_add,
3254 .remove = acpi_nfit_remove,
3255 .notify = acpi_nfit_notify,
3259 static __init int nfit_init(void)
3263 BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3264 BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3265 BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3266 BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3267 BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3268 BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3269 BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3271 guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3272 guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3273 guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3274 guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3275 guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3276 guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3277 guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3278 guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3279 guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3280 guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3281 guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3282 guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3283 guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3285 nfit_wq = create_singlethread_workqueue("nfit");
3289 nfit_mce_register();
3290 ret = acpi_bus_register_driver(&acpi_nfit_driver);
3292 nfit_mce_unregister();
3293 destroy_workqueue(nfit_wq);
3300 static __exit void nfit_exit(void)
3302 nfit_mce_unregister();
3303 acpi_bus_unregister_driver(&acpi_nfit_driver);
3304 destroy_workqueue(nfit_wq);
3305 WARN_ON(!list_empty(&acpi_descs));
3308 module_init(nfit_init);
3309 module_exit(nfit_exit);
3310 MODULE_LICENSE("GPL v2");
3311 MODULE_AUTHOR("Intel Corporation");