xref: /openbmc/linux/drivers/acpi/acpi_extlog.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Extended Error Log driver
4  *
5  * Copyright (C) 2013 Intel Corp.
6  * Author: Chen, Gong <gong.chen@intel.com>
7  */
8 
9 #include <linux/module.h>
10 #include <linux/acpi.h>
11 #include <linux/cper.h>
12 #include <linux/ratelimit.h>
13 #include <linux/edac.h>
14 #include <linux/ras.h>
15 #include <asm/cpu.h>
16 #include <asm/mce.h>
17 
18 #include "apei/apei-internal.h"
19 #include <ras/ras_event.h>
20 
21 #define EXT_ELOG_ENTRY_MASK	GENMASK_ULL(51, 0) /* elog entry address mask */
22 
23 #define EXTLOG_DSM_REV		0x0
24 #define	EXTLOG_FN_ADDR		0x1
25 
26 #define FLAG_OS_OPTIN		BIT(0)
27 #define ELOG_ENTRY_VALID	(1ULL<<63)
28 #define ELOG_ENTRY_LEN		0x1000
29 
30 #define EMCA_BUG \
31 	"Can not request iomem region <0x%016llx-0x%016llx> - eMCA disabled\n"
32 
33 struct extlog_l1_head {
34 	u32 ver;	/* Header Version */
35 	u32 hdr_len;	/* Header Length */
36 	u64 total_len;	/* entire L1 Directory length including this header */
37 	u64 elog_base;	/* MCA Error Log Directory base address */
38 	u64 elog_len;	/* MCA Error Log Directory length */
39 	u32 flags;	/* bit 0 - OS/VMM Opt-in */
40 	u8  rev0[12];
41 	u32 entries;	/* Valid L1 Directory entries per logical processor */
42 	u8  rev1[12];
43 };
44 
45 static int old_edac_report_status;
46 
47 static u8 extlog_dsm_uuid[] __initdata = "663E35AF-CC10-41A4-88EA-5470AF055295";
48 
49 /* L1 table related physical address */
50 static u64 elog_base;
51 static size_t elog_size;
52 static u64 l1_dirbase;
53 static size_t l1_size;
54 
55 /* L1 table related virtual address */
56 static void __iomem *extlog_l1_addr;
57 static void __iomem *elog_addr;
58 
59 static void *elog_buf;
60 
61 static u64 *l1_entry_base;
62 static u32 l1_percpu_entry;
63 
64 #define ELOG_IDX(cpu, bank) \
65 	(cpu_physical_id(cpu) * l1_percpu_entry + (bank))
66 
67 #define ELOG_ENTRY_DATA(idx) \
68 	(*(l1_entry_base + (idx)))
69 
70 #define ELOG_ENTRY_ADDR(phyaddr) \
71 	(phyaddr - elog_base + (u8 *)elog_addr)
72 
73 static struct acpi_hest_generic_status *extlog_elog_entry_check(int cpu, int bank)
74 {
75 	int idx;
76 	u64 data;
77 	struct acpi_hest_generic_status *estatus;
78 
79 	WARN_ON(cpu < 0);
80 	idx = ELOG_IDX(cpu, bank);
81 	data = ELOG_ENTRY_DATA(idx);
82 	if ((data & ELOG_ENTRY_VALID) == 0)
83 		return NULL;
84 
85 	data &= EXT_ELOG_ENTRY_MASK;
86 	estatus = (struct acpi_hest_generic_status *)ELOG_ENTRY_ADDR(data);
87 
88 	/* if no valid data in elog entry, just return */
89 	if (estatus->block_status == 0)
90 		return NULL;
91 
92 	return estatus;
93 }
94 
95 static void __print_extlog_rcd(const char *pfx,
96 			       struct acpi_hest_generic_status *estatus, int cpu)
97 {
98 	static atomic_t seqno;
99 	unsigned int curr_seqno;
100 	char pfx_seq[64];
101 
102 	if (!pfx) {
103 		if (estatus->error_severity <= CPER_SEV_CORRECTED)
104 			pfx = KERN_INFO;
105 		else
106 			pfx = KERN_ERR;
107 	}
108 	curr_seqno = atomic_inc_return(&seqno);
109 	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno);
110 	printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu);
111 	cper_estatus_print(pfx_seq, estatus);
112 }
113 
114 static int print_extlog_rcd(const char *pfx,
115 			    struct acpi_hest_generic_status *estatus, int cpu)
116 {
117 	/* Not more than 2 messages every 5 seconds */
118 	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
119 	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
120 	struct ratelimit_state *ratelimit;
121 
122 	if (estatus->error_severity == CPER_SEV_CORRECTED ||
123 	    (estatus->error_severity == CPER_SEV_INFORMATIONAL))
124 		ratelimit = &ratelimit_corrected;
125 	else
126 		ratelimit = &ratelimit_uncorrected;
127 	if (__ratelimit(ratelimit)) {
128 		__print_extlog_rcd(pfx, estatus, cpu);
129 		return 0;
130 	}
131 
132 	return 1;
133 }
134 
135 static int extlog_print(struct notifier_block *nb, unsigned long val,
136 			void *data)
137 {
138 	struct mce *mce = (struct mce *)data;
139 	int	bank = mce->bank;
140 	int	cpu = mce->extcpu;
141 	struct acpi_hest_generic_status *estatus, *tmp;
142 	struct acpi_hest_generic_data *gdata;
143 	const guid_t *fru_id = &guid_null;
144 	char *fru_text = "";
145 	guid_t *sec_type;
146 	static u32 err_seq;
147 
148 	estatus = extlog_elog_entry_check(cpu, bank);
149 	if (estatus == NULL)
150 		return NOTIFY_DONE;
151 
152 	memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
153 	/* clear record status to enable BIOS to update it again */
154 	estatus->block_status = 0;
155 
156 	tmp = (struct acpi_hest_generic_status *)elog_buf;
157 
158 	if (!ras_userspace_consumers()) {
159 		print_extlog_rcd(NULL, tmp, cpu);
160 		goto out;
161 	}
162 
163 	/* log event via trace */
164 	err_seq++;
165 	gdata = (struct acpi_hest_generic_data *)(tmp + 1);
166 	if (gdata->validation_bits & CPER_SEC_VALID_FRU_ID)
167 		fru_id = (guid_t *)gdata->fru_id;
168 	if (gdata->validation_bits & CPER_SEC_VALID_FRU_TEXT)
169 		fru_text = gdata->fru_text;
170 	sec_type = (guid_t *)gdata->section_type;
171 	if (guid_equal(sec_type, &CPER_SEC_PLATFORM_MEM)) {
172 		struct cper_sec_mem_err *mem = (void *)(gdata + 1);
173 		if (gdata->error_data_length >= sizeof(*mem))
174 			trace_extlog_mem_event(mem, err_seq, fru_id, fru_text,
175 					       (u8)gdata->error_severity);
176 	}
177 
178 out:
179 	return NOTIFY_STOP;
180 }
181 
182 static bool __init extlog_get_l1addr(void)
183 {
184 	guid_t guid;
185 	acpi_handle handle;
186 	union acpi_object *obj;
187 
188 	if (guid_parse(extlog_dsm_uuid, &guid))
189 		return false;
190 	if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
191 		return false;
192 	if (!acpi_check_dsm(handle, &guid, EXTLOG_DSM_REV, 1 << EXTLOG_FN_ADDR))
193 		return false;
194 	obj = acpi_evaluate_dsm_typed(handle, &guid, EXTLOG_DSM_REV,
195 				      EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER);
196 	if (!obj) {
197 		return false;
198 	} else {
199 		l1_dirbase = obj->integer.value;
200 		ACPI_FREE(obj);
201 	}
202 
203 	/* Spec says L1 directory must be 4K aligned, bail out if it isn't */
204 	if (l1_dirbase & ((1 << 12) - 1)) {
205 		pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
206 			l1_dirbase);
207 		return false;
208 	}
209 
210 	return true;
211 }
212 static struct notifier_block extlog_mce_dec = {
213 	.notifier_call	= extlog_print,
214 	.priority	= MCE_PRIO_EXTLOG,
215 };
216 
217 static int __init extlog_init(void)
218 {
219 	struct extlog_l1_head *l1_head;
220 	void __iomem *extlog_l1_hdr;
221 	size_t l1_hdr_size;
222 	struct resource *r;
223 	u64 cap;
224 	int rc;
225 
226 	rdmsrl(MSR_IA32_MCG_CAP, cap);
227 
228 	if (!(cap & MCG_ELOG_P) || !extlog_get_l1addr())
229 		return -ENODEV;
230 
231 	if (edac_get_report_status() == EDAC_REPORTING_FORCE) {
232 		pr_warn("Not loading eMCA, error reporting force-enabled through EDAC.\n");
233 		return -EPERM;
234 	}
235 
236 	rc = -EINVAL;
237 	/* get L1 header to fetch necessary information */
238 	l1_hdr_size = sizeof(struct extlog_l1_head);
239 	r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR");
240 	if (!r) {
241 		pr_warn(FW_BUG EMCA_BUG,
242 			(unsigned long long)l1_dirbase,
243 			(unsigned long long)l1_dirbase + l1_hdr_size);
244 		goto err;
245 	}
246 
247 	extlog_l1_hdr = acpi_os_map_iomem(l1_dirbase, l1_hdr_size);
248 	l1_head = (struct extlog_l1_head *)extlog_l1_hdr;
249 	l1_size = l1_head->total_len;
250 	l1_percpu_entry = l1_head->entries;
251 	elog_base = l1_head->elog_base;
252 	elog_size = l1_head->elog_len;
253 	acpi_os_unmap_iomem(extlog_l1_hdr, l1_hdr_size);
254 	release_mem_region(l1_dirbase, l1_hdr_size);
255 
256 	/* remap L1 header again based on completed information */
257 	r = request_mem_region(l1_dirbase, l1_size, "L1 Table");
258 	if (!r) {
259 		pr_warn(FW_BUG EMCA_BUG,
260 			(unsigned long long)l1_dirbase,
261 			(unsigned long long)l1_dirbase + l1_size);
262 		goto err;
263 	}
264 	extlog_l1_addr = acpi_os_map_iomem(l1_dirbase, l1_size);
265 	l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size);
266 
267 	/* remap elog table */
268 	r = request_mem_region(elog_base, elog_size, "Elog Table");
269 	if (!r) {
270 		pr_warn(FW_BUG EMCA_BUG,
271 			(unsigned long long)elog_base,
272 			(unsigned long long)elog_base + elog_size);
273 		goto err_release_l1_dir;
274 	}
275 	elog_addr = acpi_os_map_iomem(elog_base, elog_size);
276 
277 	rc = -ENOMEM;
278 	/* allocate buffer to save elog record */
279 	elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL);
280 	if (elog_buf == NULL)
281 		goto err_release_elog;
282 
283 	/*
284 	 * eMCA event report method has higher priority than EDAC method,
285 	 * unless EDAC event report method is mandatory.
286 	 */
287 	old_edac_report_status = edac_get_report_status();
288 	edac_set_report_status(EDAC_REPORTING_DISABLED);
289 	mce_register_decode_chain(&extlog_mce_dec);
290 	/* enable OS to be involved to take over management from BIOS */
291 	((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN;
292 
293 	return 0;
294 
295 err_release_elog:
296 	if (elog_addr)
297 		acpi_os_unmap_iomem(elog_addr, elog_size);
298 	release_mem_region(elog_base, elog_size);
299 err_release_l1_dir:
300 	if (extlog_l1_addr)
301 		acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
302 	release_mem_region(l1_dirbase, l1_size);
303 err:
304 	pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
305 	return rc;
306 }
307 
308 static void __exit extlog_exit(void)
309 {
310 	edac_set_report_status(old_edac_report_status);
311 	mce_unregister_decode_chain(&extlog_mce_dec);
312 	((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
313 	if (extlog_l1_addr)
314 		acpi_os_unmap_iomem(extlog_l1_addr, l1_size);
315 	if (elog_addr)
316 		acpi_os_unmap_iomem(elog_addr, elog_size);
317 	release_mem_region(elog_base, elog_size);
318 	release_mem_region(l1_dirbase, l1_size);
319 	kfree(elog_buf);
320 }
321 
322 module_init(extlog_init);
323 module_exit(extlog_exit);
324 
325 MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
326 MODULE_DESCRIPTION("Extended MCA Error Log Driver");
327 MODULE_LICENSE("GPL");
328