1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * APEI Boot Error Record Table (BERT) support
  4 *
  5 * Copyright 2011 Intel Corp.
  6 *   Author: Huang Ying <ying.huang@intel.com>
  7 *
  8 * Under normal circumstances, when a hardware error occurs, the error
  9 * handler receives control and processes the error. This gives OSPM a
 10 * chance to process the error condition, report it, and optionally attempt
 11 * recovery. In some cases, the system is unable to process an error.
 12 * For example, system firmware or a management controller may choose to
 13 * reset the system or the system might experience an uncontrolled crash
 14 * or reset.The boot error source is used to report unhandled errors that
 15 * occurred in a previous boot. This mechanism is described in the BERT
 16 * table.
 17 *
 18 * For more information about BERT, please refer to ACPI Specification
 19 * version 4.0, section 17.3.1
 20 */
 21
 22#include <linux/kernel.h>
 23#include <linux/module.h>
 24#include <linux/init.h>
 25#include <linux/acpi.h>
 26#include <linux/io.h>
 27
 28#include "apei-internal.h"
 29
 30#undef pr_fmt
 31#define pr_fmt(fmt) "BERT: " fmt
 32
 33#define ACPI_BERT_PRINT_MAX_RECORDS 5
 34#define ACPI_BERT_PRINT_MAX_LEN 1024
 35
 36static int bert_disable;
 37
 38/*
 39 * Print "all" the error records in the BERT table, but avoid huge spam to
 40 * the console if the BIOS included oversize records, or too many records.
 41 * Skipping some records here does not lose anything because the full
 42 * data is available to user tools in:
 43 *	/sys/firmware/acpi/tables/data/BERT
 44 */
 45static void __init bert_print_all(struct acpi_bert_region *region,
 46				  unsigned int region_len)
 47{
 48	struct acpi_hest_generic_status *estatus =
 49		(struct acpi_hest_generic_status *)region;
 50	int remain = region_len;
 51	int printed = 0, skipped = 0;
 52	u32 estatus_len;
 53
 54	while (remain >= sizeof(struct acpi_bert_region)) {
 55		estatus_len = cper_estatus_len(estatus);
 56		if (remain < estatus_len) {
 57			pr_err(FW_BUG "Truncated status block (length: %u).\n",
 58			       estatus_len);
 59			break;
 60		}
 61
 62		/* No more error records. */
 63		if (!estatus->block_status)
 64			break;
 65
 66		if (cper_estatus_check(estatus)) {
 67			pr_err(FW_BUG "Invalid error record.\n");
 68			break;
 69		}
 70
 71		if (estatus_len < ACPI_BERT_PRINT_MAX_LEN &&
 72		    printed < ACPI_BERT_PRINT_MAX_RECORDS) {
 73			pr_info_once("Error records from previous boot:\n");
 74			cper_estatus_print(KERN_INFO HW_ERR, estatus);
 75			printed++;
 76		} else {
 77			skipped++;
 78		}
 79
 80		/*
 81		 * Because the boot error source is "one-time polled" type,
 82		 * clear Block Status of current Generic Error Status Block,
 83		 * once it's printed.
 84		 */
 85		estatus->block_status = 0;
 86
 87		estatus = (void *)estatus + estatus_len;
 88		remain -= estatus_len;
 89	}
 90
 91	if (skipped)
 92		pr_info(HW_ERR "Skipped %d error records\n", skipped);
 93
 94	if (printed + skipped)
 95		pr_info("Total records found: %d\n", printed + skipped);
 96}
 97
 98static int __init setup_bert_disable(char *str)
 99{
100	bert_disable = 1;
101
102	return 1;
103}
104__setup("bert_disable", setup_bert_disable);
105
106static int __init bert_check_table(struct acpi_table_bert *bert_tab)
107{
108	if (bert_tab->header.length < sizeof(struct acpi_table_bert) ||
109	    bert_tab->region_length < sizeof(struct acpi_bert_region))
110		return -EINVAL;
111
112	return 0;
113}
114
115static int __init bert_init(void)
116{
117	struct apei_resources bert_resources;
118	struct acpi_bert_region *boot_error_region;
119	struct acpi_table_bert *bert_tab;
120	unsigned int region_len;
121	acpi_status status;
122	int rc = 0;
123
124	if (acpi_disabled)
125		return 0;
126
127	if (bert_disable) {
128		pr_info("Boot Error Record Table support is disabled.\n");
129		return 0;
130	}
131
132	status = acpi_get_table(ACPI_SIG_BERT, 0, (struct acpi_table_header **)&bert_tab);
133	if (status == AE_NOT_FOUND)
134		return 0;
135
136	if (ACPI_FAILURE(status)) {
137		pr_err("get table failed, %s.\n", acpi_format_exception(status));
138		return -EINVAL;
139	}
140
141	rc = bert_check_table(bert_tab);
142	if (rc) {
143		pr_err(FW_BUG "table invalid.\n");
144		goto out_put_bert_tab;
145	}
146
147	region_len = bert_tab->region_length;
148	apei_resources_init(&bert_resources);
149	rc = apei_resources_add(&bert_resources, bert_tab->address,
150				region_len, true);
151	if (rc)
152		goto out_put_bert_tab;
153	rc = apei_resources_request(&bert_resources, "APEI BERT");
154	if (rc)
155		goto out_fini;
156	boot_error_region = ioremap_cache(bert_tab->address, region_len);
157	if (boot_error_region) {
158		bert_print_all(boot_error_region, region_len);
159		iounmap(boot_error_region);
160	} else {
161		rc = -ENOMEM;
162	}
163
164	apei_resources_release(&bert_resources);
165out_fini:
166	apei_resources_fini(&bert_resources);
167out_put_bert_tab:
168	acpi_put_table((struct acpi_table_header *)bert_tab);
169
170	return rc;
171}
172
173late_initcall(bert_init);