1 // SPDX-License-Identifier: GPL-2.0-or-later
3  * FRU (Field-Replaceable Unit) Memory Poison Manager
15  * - FRU memory poison section and memory poison descriptor definitions are not yet
19  * - Platforms based on AMD MI300 systems will be the first to use these structures.
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
33  * - Boot time memory retirement may occur later than ideal due to dependencies
37  * - Enough memory should be pre-allocated for each FRU record to be able to hold
40  *   should allow the Platform to update the stored record in-place. Otherwise,
175 	return rec->sec_desc.section_length - sizeof(struct cper_section_descriptor);  in get_fmp_len()
184 		if (rec->fmp.fru_id == fru_id)  in get_fru_record()
224 	checksum = do_fmp_checksum(&rec->fmp, len);  in update_record_on_storage()
227 	rec->fmp.checksum = -checksum;  in update_record_on_storage()
231 	ret = erst_write(&rec->hdr);  in update_record_on_storage()
233 		pr_warn("Storage update failed for FRU 0x%016llx\n", rec->fmp.fru_id);  in update_record_on_storage()
235 		if (ret == -ENOSPC)  in update_record_on_storage()
244 	if (!(rec->fmp.validation_bits & FMP_VALID_LIST_ENTRIES))  in rec_has_valid_entries()
247 	if (!(rec->fmp.validation_bits & FMP_VALID_LIST))  in rec_has_valid_entries()
254  * Row retirement is done on MI300 systems, and some bits are 'don't
256  * includes all column bits and the row[13] bit.
266 	 * Also, order the checks from most->least likely to fail to shortcut the code.  in fpds_equal()
268 	if (MASK_ADDR(old->addr) != MASK_ADDR(new->addr))  in fpds_equal()
271 	if (old->hw_id != new->hw_id)  in fpds_equal()
274 	if (old->addr_type != new->addr_type)  in fpds_equal()
277 	if (old->hw_id_type != new->hw_id_type)  in fpds_equal()
287 	for (i = 0; i < rec->fmp.nr_entries; i++) {  in rec_has_fpd()
288 		struct cper_fru_poison_desc *fpd_i = &rec->entries[i];  in rec_has_fpd()
307 		pr_warn_once("FRU descriptor entry %d out-of-bounds (max: %d)\n",  in save_spa()
326 		pr_warn_once("spa_entries[] index out-of-bounds\n");  in save_spa()
349 	struct cper_sec_fru_mem_poison *fmp = &rec->fmp;  in update_fru_record()
357 	fpd.timestamp	= m->time;  in update_fru_record()
359 	fpd.hw_id	= m->ipid;  in update_fru_record()
361 	fpd.addr	= m->addr;  in update_fru_record()
371 	if (rec->fmp.nr_entries >= max_nr_entries) {  in update_fru_record()
372 		pr_warn("Exceeded number of entries for FRU 0x%016llx\n", rec->fmp.fru_id);  in update_fru_record()
376 	entry  = fmp->nr_entries;  in update_fru_record()
379 	save_spa(rec, entry, m->addr, m->ipid, m->extcpu);  in update_fru_record()
380 	fpd_dest  = &rec->entries[entry];  in update_fru_record()
383 	fmp->nr_entries		 = entry + 1;  in update_fru_record()
384 	fmp->validation_bits	|= FMP_VALID_LIST_ENTRIES;  in update_fru_record()
385 	fmp->validation_bits	|= FMP_VALID_LIST;  in update_fru_record()
387 	pr_debug("Updated FRU 0x%016llx entry #%u\n", fmp->fru_id, entry);  in update_fru_record()
416 	retire_dram_row(m->addr, m->ipid, m->extcpu);  in fru_handle_mem_poison()
422 	rec = get_fru_record(m->ppin);  in fru_handle_mem_poison()
438 	struct cper_sec_fru_mem_poison *fmp = &rec->fmp;  in retire_mem_fmp()
441 	for (i = 0; i < fmp->nr_entries; i++) {  in retire_mem_fmp()
442 		struct cper_fru_poison_desc *fpd = &rec->entries[i];  in retire_mem_fmp()
445 		if (fpd->hw_id_type != FPD_HW_ID_TYPE_MCA_IPID)  in retire_mem_fmp()
448 		if (fpd->addr_type != FPD_ADDR_TYPE_MCA_ADDR)  in retire_mem_fmp()
453 			if (topology_ppin(cpu) == fmp->fru_id) {  in retire_mem_fmp()
463 		retire_dram_row(fpd->addr, fpd->hw_id, err_cpu);  in retire_mem_fmp()
464 		save_spa(rec, i, fpd->addr, fpd->hw_id, err_cpu);  in retire_mem_fmp()
484 	struct cper_section_descriptor	*sec_desc = &rec->sec_desc;  in set_rec_fields()
485 	struct cper_record_header	*hdr	  = &rec->hdr;  in set_rec_fields()
489 	 * Update the record lengths and keep everything else as-is.  in set_rec_fields()
491 	if (hdr->record_length && hdr->record_length < max_rec_len) {  in set_rec_fields()
493 			 hdr->record_id, hdr->record_length, max_rec_len);  in set_rec_fields()
497 	memcpy(hdr->signature, CPER_SIG_RECORD, CPER_SIG_SIZE);  in set_rec_fields()
498 	hdr->revision			= CPER_RECORD_REV;  in set_rec_fields()
499 	hdr->signature_end		= CPER_SIG_END;  in set_rec_fields()
505 	hdr->section_count		= 1;  in set_rec_fields()
508 	hdr->error_severity		= CPER_SEV_RECOVERABLE;  in set_rec_fields()
510 	hdr->validation_bits		= 0;  in set_rec_fields()
511 	hdr->creator_id			= CPER_CREATOR_FMP;  in set_rec_fields()
512 	hdr->notification_type		= CPER_NOTIFY_MCE;  in set_rec_fields()
513 	hdr->record_id			= cper_next_record_id();  in set_rec_fields()
514 	hdr->flags			= CPER_HW_ERROR_FLAGS_PREVERR;  in set_rec_fields()
516 	sec_desc->section_offset	= sizeof(struct cper_record_header);  in set_rec_fields()
517 	sec_desc->revision		= CPER_SEC_REV;  in set_rec_fields()
518 	sec_desc->validation_bits	= 0;  in set_rec_fields()
519 	sec_desc->flags			= CPER_SEC_PRIMARY;  in set_rec_fields()
520 	sec_desc->section_type		= CPER_SECTION_TYPE_FMP;  in set_rec_fields()
521 	sec_desc->section_severity	= CPER_SEV_RECOVERABLE;  in set_rec_fields()
524 	hdr->record_length		= max_rec_len;  in set_rec_fields()
525 	sec_desc->section_length	= max_rec_len - sizeof(struct cper_record_header);  in set_rec_fields()
537 		if (rec->hdr.record_length == max_rec_len)  in save_new_records()
540 		if (!rec->hdr.record_length)  in save_new_records()
557 		erst_clear(rec->hdr.record_id);  in save_new_records()
566 	struct cper_sec_fru_mem_poison *fmp = &rec->fmp;  in fmp_is_usable()
569 	pr_debug("Validation bits: 0x%016llx\n", fmp->validation_bits);  in fmp_is_usable()
571 	if (!(fmp->validation_bits & FMP_VALID_ARCH_TYPE)) {  in fmp_is_usable()
576 	if (fmp->fru_arch_type != FMP_ARCH_TYPE_X86_CPUID_1_EAX) {  in fmp_is_usable()
581 	if (!(fmp->validation_bits & FMP_VALID_ARCH)) {  in fmp_is_usable()
587 	if (fmp->fru_arch != cpuid) {  in fmp_is_usable()
589 			 fmp->fru_arch, cpuid);  in fmp_is_usable()
593 	if (!(fmp->validation_bits & FMP_VALID_ID_TYPE)) {  in fmp_is_usable()
598 	if (fmp->fru_id_type != FMP_ID_TYPE_X86_PPIN) {  in fmp_is_usable()
603 	if (!(fmp->validation_bits & FMP_VALID_ID)) {  in fmp_is_usable()
613 	struct cper_sec_fru_mem_poison *fmp = &rec->fmp;  in fmp_is_valid()
623 	checksum = do_fmp_checksum(fmp, len) + fmp->checksum;  in fmp_is_valid()
645 	new = get_fru_record(old->fmp.fru_id);  in get_valid_record()
656  * - If it was not created by this module, then ignore it.
657  * - If it is valid, then copy its data to the local cache.
658  * - If it is not valid, then erase it.
669 		ret = -ENOMEM;  in get_saved_records()
686 		len = erst_read_record(record_id, &old->hdr, FMPM_MAX_REC_LEN,  in get_saved_records()
700 			saved_nr_entries  = len - sizeof(struct fru_rec);  in get_saved_records()
706 			ret = -EINVAL;  in get_saved_records()
723 	struct cper_sec_fru_mem_poison *fmp = &rec->fmp;  in set_fmp_fields()
725 	fmp->fru_arch_type    = FMP_ARCH_TYPE_X86_CPUID_1_EAX;  in set_fmp_fields()
726 	fmp->validation_bits |= FMP_VALID_ARCH_TYPE;  in set_fmp_fields()
729 	fmp->fru_arch	      = cpuid_eax(1);  in set_fmp_fields()
730 	fmp->validation_bits |= FMP_VALID_ARCH;  in set_fmp_fields()
732 	fmp->fru_id_type      = FMP_ID_TYPE_X86_PPIN;  in set_fmp_fields()
733 	fmp->validation_bits |= FMP_VALID_ID_TYPE;  in set_fmp_fields()
735 	fmp->fru_id	      = topology_ppin(cpu);  in set_fmp_fields()
736 	fmp->validation_bits |= FMP_VALID_ID;  in set_fmp_fields()
759 			ret = -ENODEV;  in init_fmps()
774 		return -ENODEV;  in get_system_info()
778 		return -ENODEV;  in get_system_info()
784 		return -ENODEV;  in get_system_info()
789 		return -ENODEV;  in get_system_info()
824 		ret = -ENOMEM;  in allocate_records()
831 			ret = -ENOMEM;  in allocate_records()
838 		ret = -ENOMEM;  in allocate_records()
848 	while (--i >= 0)  in allocate_records()
878 #define U64_PAD		(LONG_WIDTH - U64_WIDTH)
879 #define TS_PAD		(LONG_WIDTH - TIMESTAMP_WIDTH)
888 		seq_printf(f, "%-*s", SHORT_WIDTH, "fru_idx");  in fmpm_show()
889 		seq_printf(f, "%-*s", LONG_WIDTH,  "fru_id");  in fmpm_show()
890 		seq_printf(f, "%-*s", SHORT_WIDTH, "entry");  in fmpm_show()
891 		seq_printf(f, "%-*s", LONG_WIDTH,  "timestamp");  in fmpm_show()
892 		seq_printf(f, "%-*s", LONG_WIDTH,  "hw_id");  in fmpm_show()
893 		seq_printf(f, "%-*s", LONG_WIDTH,  "addr");  in fmpm_show()
894 		seq_printf(f, "%-*s", LONG_WIDTH,  "spa");  in fmpm_show()
898 	spa_entry = line - 1;  in fmpm_show()
906 	seq_printf(f, "%-*u",		SHORT_WIDTH, fru_idx);  in fmpm_show()
907 	seq_printf(f, "0x%016llx%-*s",	rec->fmp.fru_id, U64_PAD, "");  in fmpm_show()
908 	seq_printf(f, "%-*u",		SHORT_WIDTH, entry);  in fmpm_show()
912 	if (entry >= rec->fmp.nr_entries) {  in fmpm_show()
913 		seq_printf(f, "%-*s", LONG_WIDTH, "*");  in fmpm_show()
914 		seq_printf(f, "%-*s", LONG_WIDTH, "*");  in fmpm_show()
915 		seq_printf(f, "%-*s", LONG_WIDTH, "*");  in fmpm_show()
916 		seq_printf(f, "%-*s", LONG_WIDTH, "*");  in fmpm_show()
920 	fpd = &rec->entries[entry];  in fmpm_show()
922 	seq_printf(f, "%ptT%-*s",	&fpd->timestamp, TS_PAD,  "");  in fmpm_show()
923 	seq_printf(f, "0x%016llx%-*s",	fpd->hw_id,	 U64_PAD, "");  in fmpm_show()
924 	seq_printf(f, "0x%016llx%-*s",	fpd->addr,	 U64_PAD, "");  in fmpm_show()
927 		seq_printf(f, "%-*s", LONG_WIDTH, "*");  in fmpm_show()
929 		seq_printf(f, "0x%016llx%-*s", spa_entries[spa_entry], U64_PAD, "");  in fmpm_show()
985 		ret = -ENODEV;  in fru_mem_poison_init()
991 		ret = -ENODEV;  in fru_mem_poison_init()