1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * FRU (Field-Replaceable Unit) Memory Poison Manager
5 * Copyright (c) 2024, Advanced Micro Devices, Inc.
13 * Implementation notes, assumptions, and limitations:
15 * - FRU memory poison section and memory poison descriptor definitions are not yet
16 * included in the UEFI specification. So they are defined here. Afterwards, they
17 * may be moved to linux/cper.h, if appropriate.
19 * - Platforms based on AMD MI300 systems will be the first to use these structures.
20 * There are a number of assumptions made here that will need to be generalized
21 * to support other platforms.
23 * AMD MI300-based platform(s) assumptions:
24 * - Memory errors are reported through x86 MCA.
25 * - The entire DRAM row containing a memory error should be retired.
26 * - There will be (1) FRU memory poison section per CPER.
27 * - The FRU will be the CPU package (processor socket).
28 * - The default number of memory poison descriptor entries should be (8).
29 * - The platform will use ACPI ERST for persistent storage.
30 * - All FRU records should be saved to persistent storage. Module init will
31 * fail if any FRU record is not successfully written.
33 * - Boot time memory retirement may occur later than ideal due to dependencies
34 * on other libraries and drivers. This leaves a gap where bad memory may be
35 * accessed during early boot stages.
37 * - Enough memory should be pre-allocated for each FRU record to be able to hold
38 * the expected number of descriptor entries. This, mostly empty, record is
39 * written to storage during init time. Subsequent writes to the same record
40 * should allow the Platform to update the stored record in-place. Otherwise,
41 * if the record is extended, then the Platform may need to perform costly memory
42 * management operations on the storage. For example, the Platform may spend time
43 * in Firmware copying and invalidating memory on a relatively slow SPI ROM.
46 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
48 #include <linux/cper.h>
49 #include <linux/ras.h>
50 #include <linux/cpu.h>
52 #include <acpi/apei.h>
54 #include <asm/cpu_device_id.h>
57 #include "../debugfs.h"
59 #include "atl/internal.h"
61 #define INVALID_CPU UINT_MAX
64 #define FMP_VALID_ARCH_TYPE BIT_ULL(0)
65 #define FMP_VALID_ARCH BIT_ULL(1)
66 #define FMP_VALID_ID_TYPE BIT_ULL(2)
67 #define FMP_VALID_ID BIT_ULL(3)
68 #define FMP_VALID_LIST_ENTRIES BIT_ULL(4)
69 #define FMP_VALID_LIST BIT_ULL(5)
71 /* FRU Architecture Types */
72 #define FMP_ARCH_TYPE_X86_CPUID_1_EAX 0
75 #define FMP_ID_TYPE_X86_PPIN 0
77 /* FRU Memory Poison Section */
78 struct cper_sec_fru_mem_poison {
88 /* FRU Descriptor ID Types */
89 #define FPD_HW_ID_TYPE_MCA_IPID 0
91 /* FRU Descriptor Address Types */
92 #define FPD_ADDR_TYPE_MCA_ADDR 0
94 /* Memory Poison Descriptor */
95 struct cper_fru_poison_desc {
103 /* Collection of headers and sections for easy pointer use. */
105 struct cper_record_header hdr;
106 struct cper_section_descriptor sec_desc;
107 struct cper_sec_fru_mem_poison fmp;
108 struct cper_fru_poison_desc entries[];
112 * Pointers to the complete CPER record of each FRU.
114 * Memory allocation will include padded space for descriptor entries.
116 static struct fru_rec **fru_records;
118 /* system physical addresses array */
119 static u64 *spa_entries;
121 static struct dentry *fmpm_dfs_dir;
122 static struct dentry *fmpm_dfs_entries;
124 #define CPER_CREATOR_FMP \
125 GUID_INIT(0xcd5c2993, 0xf4b2, 0x41b2, 0xb5, 0xd4, 0xf9, 0xc3, \
126 0xa0, 0x33, 0x08, 0x75)
128 #define CPER_SECTION_TYPE_FMP \
129 GUID_INIT(0x5e4706c1, 0x5356, 0x48c6, 0x93, 0x0b, 0x52, 0xf2, \
130 0x12, 0x0a, 0x44, 0x58)
133 * DOC: max_nr_entries (byte)
134 * Maximum number of descriptor entries possible for each FRU.
136 * Values between '1' and '255' are valid.
137 * No input or '0' will default to FMPM_DEFAULT_MAX_NR_ENTRIES.
139 static u8 max_nr_entries;
140 module_param(max_nr_entries, byte, 0644);
141 MODULE_PARM_DESC(max_nr_entries,
142 "Maximum number of memory poison descriptor entries per FRU");
144 #define FMPM_DEFAULT_MAX_NR_ENTRIES 8
146 /* Maximum number of FRUs in the system. */
147 #define FMPM_MAX_NR_FRU 256
148 static unsigned int max_nr_fru;
150 /* Total length of record including headers and list of descriptor entries. */
151 static size_t max_rec_len;
153 #define FMPM_MAX_REC_LEN (sizeof(struct fru_rec) + (sizeof(struct cper_fru_poison_desc) * 255))
155 /* Total number of SPA entries across all FRUs. */
156 static unsigned int spa_nr_entries;
159 * Protect the local records cache in fru_records and prevent concurrent
160 * writes to storage. This is only needed after init once notifier block
161 * registration is done.
163 * The majority of a record is fixed at module init and will not change
164 * during run time. The entries within a record will be updated as new
165 * errors are reported. The mutex should be held whenever the entries are
166 * accessed during run time.
168 static DEFINE_MUTEX(fmpm_update_mutex);
170 #define for_each_fru(i, rec) \
171 for (i = 0; rec = fru_records[i], i < max_nr_fru; i++)
173 static inline u32 get_fmp_len(struct fru_rec *rec)
175 return rec->sec_desc.section_length - sizeof(struct cper_section_descriptor);
178 static struct fru_rec *get_fru_record(u64 fru_id)
183 for_each_fru(i, rec) {
184 if (rec->fmp.fru_id == fru_id)
188 pr_debug("Record not found for FRU 0x%016llx\n", fru_id);
194 * Sum up all bytes within the FRU Memory Poison Section including the Memory
195 * Poison Descriptor entries.
197 * Don't include the old checksum here. It's a u32 value, so summing each of its
198 * bytes will give the wrong total.
200 static u32 do_fmp_checksum(struct cper_sec_fru_mem_poison *fmp, u32 len)
205 /* Skip old checksum. */
206 buf = (u8 *)fmp + sizeof(u32);
210 checksum += (u8)(*(buf++));
215 static int update_record_on_storage(struct fru_rec *rec)
220 /* Calculate a new checksum. */
221 len = get_fmp_len(rec);
223 /* Get the current total. */
224 checksum = do_fmp_checksum(&rec->fmp, len);
226 /* Use the complement value. */
227 rec->fmp.checksum = -checksum;
229 pr_debug("Writing to storage\n");
231 ret = erst_write(&rec->hdr);
233 pr_warn("Storage update failed for FRU 0x%016llx\n", rec->fmp.fru_id);
236 pr_warn("Not enough space on storage\n");
242 static bool rec_has_valid_entries(struct fru_rec *rec)
244 if (!(rec->fmp.validation_bits & FMP_VALID_LIST_ENTRIES))
247 if (!(rec->fmp.validation_bits & FMP_VALID_LIST))
253 static bool fpds_equal(struct cper_fru_poison_desc *old, struct cper_fru_poison_desc *new)
256 * Ignore timestamp field.
257 * The same physical error may be reported multiple times due to stuck bits, etc.
259 * Also, order the checks from most->least likely to fail to shortcut the code.
261 if (old->addr != new->addr)
264 if (old->hw_id != new->hw_id)
267 if (old->addr_type != new->addr_type)
270 if (old->hw_id_type != new->hw_id_type)
276 static bool rec_has_fpd(struct fru_rec *rec, struct cper_fru_poison_desc *fpd)
280 for (i = 0; i < rec->fmp.nr_entries; i++) {
281 struct cper_fru_poison_desc *fpd_i = &rec->entries[i];
283 if (fpds_equal(fpd_i, fpd)) {
284 pr_debug("Found duplicate record\n");
292 static void save_spa(struct fru_rec *rec, unsigned int entry,
293 u64 addr, u64 id, unsigned int cpu)
295 unsigned int i, fru_idx, spa_entry;
296 struct atl_err a_err;
299 if (entry >= max_nr_entries) {
300 pr_warn_once("FRU descriptor entry %d out-of-bounds (max: %d)\n",
301 entry, max_nr_entries);
305 /* spa_nr_entries is always multiple of max_nr_entries */
306 for (i = 0; i < spa_nr_entries; i += max_nr_entries) {
307 fru_idx = i / max_nr_entries;
308 if (fru_records[fru_idx] == rec)
312 if (i >= spa_nr_entries) {
313 pr_warn_once("FRU record %d not found\n", i);
317 spa_entry = i + entry;
318 if (spa_entry >= spa_nr_entries) {
319 pr_warn_once("spa_entries[] index out-of-bounds\n");
323 memset(&a_err, 0, sizeof(struct atl_err));
329 spa = amd_convert_umc_mca_addr_to_sys_addr(&a_err);
330 if (IS_ERR_VALUE(spa)) {
331 pr_debug("Failed to get system address\n");
335 spa_entries[spa_entry] = spa;
336 pr_debug("fru_idx: %u, entry: %u, spa_entry: %u, spa: 0x%016llx\n",
337 fru_idx, entry, spa_entry, spa_entries[spa_entry]);
340 static void update_fru_record(struct fru_rec *rec, struct mce *m)
342 struct cper_sec_fru_mem_poison *fmp = &rec->fmp;
343 struct cper_fru_poison_desc fpd, *fpd_dest;
346 mutex_lock(&fmpm_update_mutex);
348 memset(&fpd, 0, sizeof(struct cper_fru_poison_desc));
350 fpd.timestamp = m->time;
351 fpd.hw_id_type = FPD_HW_ID_TYPE_MCA_IPID;
353 fpd.addr_type = FPD_ADDR_TYPE_MCA_ADDR;
356 /* This is the first entry, so just save it. */
357 if (!rec_has_valid_entries(rec))
360 /* Ignore already recorded errors. */
361 if (rec_has_fpd(rec, &fpd))
364 if (rec->fmp.nr_entries >= max_nr_entries) {
365 pr_warn("Exceeded number of entries for FRU 0x%016llx\n", rec->fmp.fru_id);
369 entry = fmp->nr_entries;
372 save_spa(rec, entry, m->addr, m->ipid, m->extcpu);
373 fpd_dest = &rec->entries[entry];
374 memcpy(fpd_dest, &fpd, sizeof(struct cper_fru_poison_desc));
376 fmp->nr_entries = entry + 1;
377 fmp->validation_bits |= FMP_VALID_LIST_ENTRIES;
378 fmp->validation_bits |= FMP_VALID_LIST;
380 pr_debug("Updated FRU 0x%016llx entry #%u\n", fmp->fru_id, entry);
382 update_record_on_storage(rec);
385 mutex_unlock(&fmpm_update_mutex);
388 static void retire_dram_row(u64 addr, u64 id, u32 cpu)
390 struct atl_err a_err;
392 memset(&a_err, 0, sizeof(struct atl_err));
398 amd_retire_dram_row(&a_err);
401 static int fru_handle_mem_poison(struct notifier_block *nb, unsigned long val, void *data)
403 struct mce *m = (struct mce *)data;
406 if (!mce_is_memory_error(m))
409 retire_dram_row(m->addr, m->ipid, m->extcpu);
412 * An invalid FRU ID should not happen on real errors. But it
413 * could happen from software error injection, etc.
415 rec = get_fru_record(m->ppin);
419 update_fru_record(rec, m);
424 static struct notifier_block fru_mem_poison_nb = {
425 .notifier_call = fru_handle_mem_poison,
426 .priority = MCE_PRIO_LOWEST,
429 static void retire_mem_fmp(struct fru_rec *rec)
431 struct cper_sec_fru_mem_poison *fmp = &rec->fmp;
434 for (i = 0; i < fmp->nr_entries; i++) {
435 struct cper_fru_poison_desc *fpd = &rec->entries[i];
436 unsigned int err_cpu = INVALID_CPU;
438 if (fpd->hw_id_type != FPD_HW_ID_TYPE_MCA_IPID)
441 if (fpd->addr_type != FPD_ADDR_TYPE_MCA_ADDR)
445 for_each_online_cpu(cpu) {
446 if (topology_ppin(cpu) == fmp->fru_id) {
453 if (err_cpu == INVALID_CPU)
456 retire_dram_row(fpd->addr, fpd->hw_id, err_cpu);
457 save_spa(rec, i, fpd->addr, fpd->hw_id, err_cpu);
461 static void retire_mem_records(void)
466 for_each_fru(i, rec) {
467 if (!rec_has_valid_entries(rec))
474 /* Set the CPER Record Header and CPER Section Descriptor fields. */
475 static void set_rec_fields(struct fru_rec *rec)
477 struct cper_section_descriptor *sec_desc = &rec->sec_desc;
478 struct cper_record_header *hdr = &rec->hdr;
481 * This is a saved record created with fewer max_nr_entries.
482 * Update the record lengths and keep everything else as-is.
484 if (hdr->record_length && hdr->record_length < max_rec_len) {
485 pr_debug("Growing record 0x%016llx from %u to %zu bytes\n",
486 hdr->record_id, hdr->record_length, max_rec_len);
490 memcpy(hdr->signature, CPER_SIG_RECORD, CPER_SIG_SIZE);
491 hdr->revision = CPER_RECORD_REV;
492 hdr->signature_end = CPER_SIG_END;
495 * Currently, it is assumed that there is one FRU Memory Poison
496 * section per CPER. But this may change for other implementations.
498 hdr->section_count = 1;
500 /* The logged errors are recoverable. Otherwise, they'd never make it here. */
501 hdr->error_severity = CPER_SEV_RECOVERABLE;
503 hdr->validation_bits = 0;
504 hdr->creator_id = CPER_CREATOR_FMP;
505 hdr->notification_type = CPER_NOTIFY_MCE;
506 hdr->record_id = cper_next_record_id();
507 hdr->flags = CPER_HW_ERROR_FLAGS_PREVERR;
509 sec_desc->section_offset = sizeof(struct cper_record_header);
510 sec_desc->revision = CPER_SEC_REV;
511 sec_desc->validation_bits = 0;
512 sec_desc->flags = CPER_SEC_PRIMARY;
513 sec_desc->section_type = CPER_SECTION_TYPE_FMP;
514 sec_desc->section_severity = CPER_SEV_RECOVERABLE;
517 hdr->record_length = max_rec_len;
518 sec_desc->section_length = max_rec_len - sizeof(struct cper_record_header);
521 static int save_new_records(void)
523 DECLARE_BITMAP(new_records, FMPM_MAX_NR_FRU);
528 for_each_fru(i, rec) {
529 /* No need to update saved records that match the current record size. */
530 if (rec->hdr.record_length == max_rec_len)
533 if (!rec->hdr.record_length)
534 set_bit(i, new_records);
538 ret = update_record_on_storage(rec);
546 for_each_fru(i, rec) {
547 if (!test_bit(i, new_records))
550 erst_clear(rec->hdr.record_id);
556 /* Check that the record matches expected types for the current system.*/
557 static bool fmp_is_usable(struct fru_rec *rec)
559 struct cper_sec_fru_mem_poison *fmp = &rec->fmp;
562 pr_debug("Validation bits: 0x%016llx\n", fmp->validation_bits);
564 if (!(fmp->validation_bits & FMP_VALID_ARCH_TYPE)) {
565 pr_debug("Arch type unknown\n");
569 if (fmp->fru_arch_type != FMP_ARCH_TYPE_X86_CPUID_1_EAX) {
570 pr_debug("Arch type not 'x86 Family/Model/Stepping'\n");
574 if (!(fmp->validation_bits & FMP_VALID_ARCH)) {
575 pr_debug("Arch value unknown\n");
579 cpuid = cpuid_eax(1);
580 if (fmp->fru_arch != cpuid) {
581 pr_debug("Arch value mismatch: record = 0x%016llx, system = 0x%016llx\n",
582 fmp->fru_arch, cpuid);
586 if (!(fmp->validation_bits & FMP_VALID_ID_TYPE)) {
587 pr_debug("FRU ID type unknown\n");
591 if (fmp->fru_id_type != FMP_ID_TYPE_X86_PPIN) {
592 pr_debug("FRU ID type is not 'x86 PPIN'\n");
596 if (!(fmp->validation_bits & FMP_VALID_ID)) {
597 pr_debug("FRU ID value unknown\n");
604 static bool fmp_is_valid(struct fru_rec *rec)
606 struct cper_sec_fru_mem_poison *fmp = &rec->fmp;
609 len = get_fmp_len(rec);
610 if (len < sizeof(struct cper_sec_fru_mem_poison)) {
611 pr_debug("fmp length is too small\n");
615 /* Checksum must sum to zero for the entire section. */
616 checksum = do_fmp_checksum(fmp, len) + fmp->checksum;
618 pr_debug("fmp checksum failed: sum = 0x%x\n", checksum);
619 print_hex_dump_debug("fmp record: ", DUMP_PREFIX_NONE, 16, 1, fmp, len, false);
623 if (!fmp_is_usable(rec))
629 static struct fru_rec *get_valid_record(struct fru_rec *old)
633 if (!fmp_is_valid(old)) {
634 pr_debug("Ignoring invalid record\n");
638 new = get_fru_record(old->fmp.fru_id);
640 pr_debug("Ignoring record for absent FRU\n");
646 * Fetch saved records from persistent storage.
648 * For each found record:
649 * - If it was not created by this module, then ignore it.
650 * - If it is valid, then copy its data to the local cache.
651 * - If it is not valid, then erase it.
653 static int get_saved_records(void)
655 struct fru_rec *old, *new;
660 old = kmalloc(FMPM_MAX_REC_LEN, GFP_KERNEL);
666 ret = erst_get_record_id_begin(&pos);
670 while (!erst_get_record_id_next(&pos, &record_id)) {
671 if (record_id == APEI_ERST_INVALID_RECORD_ID)
674 * Make sure to clear temporary buffer between reads to avoid
675 * leftover data from records of various sizes.
677 memset(old, 0, FMPM_MAX_REC_LEN);
679 len = erst_read_record(record_id, &old->hdr, FMPM_MAX_REC_LEN,
680 sizeof(struct fru_rec), &CPER_CREATOR_FMP);
684 new = get_valid_record(old);
686 erst_clear(record_id);
690 if (len > max_rec_len) {
691 unsigned int saved_nr_entries;
693 saved_nr_entries = len - sizeof(struct fru_rec);
694 saved_nr_entries /= sizeof(struct cper_fru_poison_desc);
696 pr_warn("Saved record found with %u entries.\n", saved_nr_entries);
697 pr_warn("Please increase max_nr_entries to %u.\n", saved_nr_entries);
703 /* Restore the record */
704 memcpy(new, old, len);
708 erst_get_record_id_end();
714 static void set_fmp_fields(struct fru_rec *rec, unsigned int cpu)
716 struct cper_sec_fru_mem_poison *fmp = &rec->fmp;
718 fmp->fru_arch_type = FMP_ARCH_TYPE_X86_CPUID_1_EAX;
719 fmp->validation_bits |= FMP_VALID_ARCH_TYPE;
721 /* Assume all CPUs in the system have the same value for now. */
722 fmp->fru_arch = cpuid_eax(1);
723 fmp->validation_bits |= FMP_VALID_ARCH;
725 fmp->fru_id_type = FMP_ID_TYPE_X86_PPIN;
726 fmp->validation_bits |= FMP_VALID_ID_TYPE;
728 fmp->fru_id = topology_ppin(cpu);
729 fmp->validation_bits |= FMP_VALID_ID;
732 static int init_fmps(void)
738 for_each_fru(i, rec) {
739 unsigned int fru_cpu = INVALID_CPU;
742 for_each_online_cpu(cpu) {
743 if (topology_physical_package_id(cpu) == i) {
750 if (fru_cpu == INVALID_CPU) {
751 pr_debug("Failed to find matching CPU for FRU #%u\n", i);
756 set_fmp_fields(rec, fru_cpu);
762 static int get_system_info(void)
764 /* Only load on MI300A systems for now. */
765 if (!(boot_cpu_data.x86_model >= 0x90 &&
766 boot_cpu_data.x86_model <= 0x9f))
769 if (!cpu_feature_enabled(X86_FEATURE_AMD_PPIN)) {
770 pr_debug("PPIN feature not available\n");
774 /* Use CPU socket as FRU for MI300 systems. */
775 max_nr_fru = topology_max_packages();
779 if (max_nr_fru > FMPM_MAX_NR_FRU) {
780 pr_warn("Too many FRUs to manage: found: %u, max: %u\n",
781 max_nr_fru, FMPM_MAX_NR_FRU);
786 max_nr_entries = FMPM_DEFAULT_MAX_NR_ENTRIES;
788 spa_nr_entries = max_nr_fru * max_nr_entries;
790 max_rec_len = sizeof(struct fru_rec);
791 max_rec_len += sizeof(struct cper_fru_poison_desc) * max_nr_entries;
793 pr_info("max FRUs: %u, max entries: %u, max record length: %lu\n",
794 max_nr_fru, max_nr_entries, max_rec_len);
799 static void free_records(void)
811 static int allocate_records(void)
815 fru_records = kcalloc(max_nr_fru, sizeof(struct fru_rec *), GFP_KERNEL);
821 for (i = 0; i < max_nr_fru; i++) {
822 fru_records[i] = kzalloc(max_rec_len, GFP_KERNEL);
823 if (!fru_records[i]) {
829 spa_entries = kcalloc(spa_nr_entries, sizeof(u64), GFP_KERNEL);
835 for (i = 0; i < spa_nr_entries; i++)
836 spa_entries[i] = INVALID_SPA;
842 kfree(fru_records[i]);
849 static void *fmpm_start(struct seq_file *f, loff_t *pos)
851 if (*pos >= (spa_nr_entries + 1))
856 static void *fmpm_next(struct seq_file *f, void *data, loff_t *pos)
858 if (++(*pos) >= (spa_nr_entries + 1))
863 static void fmpm_stop(struct seq_file *f, void *data)
867 #define SHORT_WIDTH 8
869 #define TIMESTAMP_WIDTH 19
870 #define LONG_WIDTH 24
871 #define U64_PAD (LONG_WIDTH - U64_WIDTH)
872 #define TS_PAD (LONG_WIDTH - TIMESTAMP_WIDTH)
873 static int fmpm_show(struct seq_file *f, void *data)
875 unsigned int fru_idx, entry, spa_entry, line;
876 struct cper_fru_poison_desc *fpd;
879 line = *(loff_t *)data;
881 seq_printf(f, "%-*s", SHORT_WIDTH, "fru_idx");
882 seq_printf(f, "%-*s", LONG_WIDTH, "fru_id");
883 seq_printf(f, "%-*s", SHORT_WIDTH, "entry");
884 seq_printf(f, "%-*s", LONG_WIDTH, "timestamp");
885 seq_printf(f, "%-*s", LONG_WIDTH, "hw_id");
886 seq_printf(f, "%-*s", LONG_WIDTH, "addr");
887 seq_printf(f, "%-*s", LONG_WIDTH, "spa");
891 spa_entry = line - 1;
892 fru_idx = spa_entry / max_nr_entries;
893 entry = spa_entry % max_nr_entries;
895 rec = fru_records[fru_idx];
899 seq_printf(f, "%-*u", SHORT_WIDTH, fru_idx);
900 seq_printf(f, "0x%016llx%-*s", rec->fmp.fru_id, U64_PAD, "");
901 seq_printf(f, "%-*u", SHORT_WIDTH, entry);
903 mutex_lock(&fmpm_update_mutex);
905 if (entry >= rec->fmp.nr_entries) {
906 seq_printf(f, "%-*s", LONG_WIDTH, "*");
907 seq_printf(f, "%-*s", LONG_WIDTH, "*");
908 seq_printf(f, "%-*s", LONG_WIDTH, "*");
909 seq_printf(f, "%-*s", LONG_WIDTH, "*");
913 fpd = &rec->entries[entry];
915 seq_printf(f, "%ptT%-*s", &fpd->timestamp, TS_PAD, "");
916 seq_printf(f, "0x%016llx%-*s", fpd->hw_id, U64_PAD, "");
917 seq_printf(f, "0x%016llx%-*s", fpd->addr, U64_PAD, "");
919 if (spa_entries[spa_entry] == INVALID_SPA)
920 seq_printf(f, "%-*s", LONG_WIDTH, "*");
922 seq_printf(f, "0x%016llx%-*s", spa_entries[spa_entry], U64_PAD, "");
925 mutex_unlock(&fmpm_update_mutex);
932 static const struct seq_operations fmpm_seq_ops = {
939 static int fmpm_open(struct inode *inode, struct file *file)
941 return seq_open(file, &fmpm_seq_ops);
944 static const struct file_operations fmpm_fops = {
946 .release = seq_release,
951 static void setup_debugfs(void)
953 struct dentry *dfs = ras_get_debugfs_root();
958 fmpm_dfs_dir = debugfs_create_dir("fmpm", dfs);
962 fmpm_dfs_entries = debugfs_create_file("entries", 0400, fmpm_dfs_dir, NULL, &fmpm_fops);
963 if (!fmpm_dfs_entries)
964 debugfs_remove(fmpm_dfs_dir);
967 static const struct x86_cpu_id fmpm_cpuids[] = {
968 X86_MATCH_VENDOR_FAM(AMD, 0x19, NULL),
971 MODULE_DEVICE_TABLE(x86cpu, fmpm_cpuids);
973 static int __init fru_mem_poison_init(void)
977 if (!x86_match_cpu(fmpm_cpuids)) {
983 pr_debug("ERST not available\n");
988 ret = get_system_info();
992 ret = allocate_records();
1000 ret = get_saved_records();
1004 ret = save_new_records();
1010 retire_mem_records();
1012 mce_register_decode_chain(&fru_mem_poison_nb);
1014 pr_info("FRU Memory Poison Manager initialized\n");
1023 static void __exit fru_mem_poison_exit(void)
1025 mce_unregister_decode_chain(&fru_mem_poison_nb);
1026 debugfs_remove(fmpm_dfs_dir);
1030 module_init(fru_mem_poison_init);
1031 module_exit(fru_mem_poison_exit);
1033 MODULE_LICENSE("GPL");
1034 MODULE_DESCRIPTION("FRU Memory Poison Manager");