1/*
  2 * Extended Error Log driver
  3 *
  4 * Copyright (C) 2013 Intel Corp.
  5 * Author: Chen, Gong <gong.chen@intel.com>
  6 *
  7 * This file is licensed under GPLv2.
  8 */
  9
 10#include <linux/module.h>
 11#include <linux/acpi.h>
 12#include <linux/cper.h>
 13#include <linux/ratelimit.h>
 14#include <linux/edac.h>
 15#include <asm/cpu.h>
 16#include <asm/mce.h>
 17
 18#include "apei/apei-internal.h"
 19
 20#define EXT_ELOG_ENTRY_MASK	GENMASK_ULL(51, 0) /* elog entry address mask */
 21
 22#define EXTLOG_DSM_REV		0x0
 23#define	EXTLOG_FN_ADDR		0x1
 24
 25#define FLAG_OS_OPTIN		BIT(0)
 26#define ELOG_ENTRY_VALID	(1ULL<<63)
 27#define ELOG_ENTRY_LEN		0x1000
 28
 29#define EMCA_BUG \
 30	"Can not request iomem region <0x%016llx-0x%016llx> - eMCA disabled\n"
 31
 32struct extlog_l1_head {
 33	u32 ver;	/* Header Version */
 34	u32 hdr_len;	/* Header Length */
 35	u64 total_len;	/* entire L1 Directory length including this header */
 36	u64 elog_base;	/* MCA Error Log Directory base address */
 37	u64 elog_len;	/* MCA Error Log Directory length */
 38	u32 flags;	/* bit 0 - OS/VMM Opt-in */
 39	u8  rev0[12];
 40	u32 entries;	/* Valid L1 Directory entries per logical processor */
 41	u8  rev1[12];
 42};
 43
 44static int old_edac_report_status;
 45
 46static u8 extlog_dsm_uuid[] __initdata = "663E35AF-CC10-41A4-88EA-5470AF055295";
 47
 48/* L1 table related physical address */
 49static u64 elog_base;
 50static size_t elog_size;
 51static u64 l1_dirbase;
 52static size_t l1_size;
 53
 54/* L1 table related virtual address */
 55static void __iomem *extlog_l1_addr;
 56static void __iomem *elog_addr;
 57
 58static void *elog_buf;
 59
 60static u64 *l1_entry_base;
 61static u32 l1_percpu_entry;
 62
 63#define ELOG_IDX(cpu, bank) \
 64	(cpu_physical_id(cpu) * l1_percpu_entry + (bank))
 65
 66#define ELOG_ENTRY_DATA(idx) \
 67	(*(l1_entry_base + (idx)))
 68
 69#define ELOG_ENTRY_ADDR(phyaddr) \
 70	(phyaddr - elog_base + (u8 *)elog_addr)
 71
 72static struct acpi_generic_status *extlog_elog_entry_check(int cpu, int bank)
 73{
 74	int idx;
 75	u64 data;
 76	struct acpi_generic_status *estatus;
 77
 78	WARN_ON(cpu < 0);
 79	idx = ELOG_IDX(cpu, bank);
 80	data = ELOG_ENTRY_DATA(idx);
 81	if ((data & ELOG_ENTRY_VALID) == 0)
 82		return NULL;
 83
 84	data &= EXT_ELOG_ENTRY_MASK;
 85	estatus = (struct acpi_generic_status *)ELOG_ENTRY_ADDR(data);
 86
 87	/* if no valid data in elog entry, just return */
 88	if (estatus->block_status == 0)
 89		return NULL;
 90
 91	return estatus;
 92}
 93
 94static void __print_extlog_rcd(const char *pfx,
 95			       struct acpi_generic_status *estatus, int cpu)
 96{
 97	static atomic_t seqno;
 98	unsigned int curr_seqno;
 99	char pfx_seq[64];
100
101	if (!pfx) {
102		if (estatus->error_severity <= CPER_SEV_CORRECTED)
103			pfx = KERN_INFO;
104		else
105			pfx = KERN_ERR;
106	}
107	curr_seqno = atomic_inc_return(&seqno);
108	snprintf(pfx_seq, sizeof(pfx_seq), "%s{%u}", pfx, curr_seqno);
109	printk("%s""Hardware error detected on CPU%d\n", pfx_seq, cpu);
110	cper_estatus_print(pfx_seq, estatus);
111}
112
113static int print_extlog_rcd(const char *pfx,
114			    struct acpi_generic_status *estatus, int cpu)
115{
116	/* Not more than 2 messages every 5 seconds */
117	static DEFINE_RATELIMIT_STATE(ratelimit_corrected, 5*HZ, 2);
118	static DEFINE_RATELIMIT_STATE(ratelimit_uncorrected, 5*HZ, 2);
119	struct ratelimit_state *ratelimit;
120
121	if (estatus->error_severity == CPER_SEV_CORRECTED ||
122	    (estatus->error_severity == CPER_SEV_INFORMATIONAL))
123		ratelimit = &ratelimit_corrected;
124	else
125		ratelimit = &ratelimit_uncorrected;
126	if (__ratelimit(ratelimit)) {
127		__print_extlog_rcd(pfx, estatus, cpu);
128		return 0;
129	}
130
131	return 1;
132}
133
134static int extlog_print(struct notifier_block *nb, unsigned long val,
135			void *data)
136{
137	struct mce *mce = (struct mce *)data;
138	int	bank = mce->bank;
139	int	cpu = mce->extcpu;
140	struct acpi_generic_status *estatus;
141	int rc;
142
143	estatus = extlog_elog_entry_check(cpu, bank);
144	if (estatus == NULL)
145		return NOTIFY_DONE;
146
147	memcpy(elog_buf, (void *)estatus, ELOG_ENTRY_LEN);
148	/* clear record status to enable BIOS to update it again */
149	estatus->block_status = 0;
150
151	rc = print_extlog_rcd(NULL, (struct acpi_generic_status *)elog_buf, cpu);
152
153	return NOTIFY_STOP;
154}
155
156static bool __init extlog_get_l1addr(void)
157{
158	u8 uuid[16];
159	acpi_handle handle;
160	union acpi_object *obj;
161
162	acpi_str_to_uuid(extlog_dsm_uuid, uuid);
163
164	if (ACPI_FAILURE(acpi_get_handle(NULL, "\\_SB", &handle)))
165		return false;
166	if (!acpi_check_dsm(handle, uuid, EXTLOG_DSM_REV, 1 << EXTLOG_FN_ADDR))
167		return false;
168	obj = acpi_evaluate_dsm_typed(handle, uuid, EXTLOG_DSM_REV,
169				      EXTLOG_FN_ADDR, NULL, ACPI_TYPE_INTEGER);
170	if (!obj) {
171		return false;
172	} else {
173		l1_dirbase = obj->integer.value;
174		ACPI_FREE(obj);
175	}
176
177	/* Spec says L1 directory must be 4K aligned, bail out if it isn't */
178	if (l1_dirbase & ((1 << 12) - 1)) {
179		pr_warn(FW_BUG "L1 Directory is invalid at physical %llx\n",
180			l1_dirbase);
181		return false;
182	}
183
184	return true;
185}
186static struct notifier_block extlog_mce_dec = {
187	.notifier_call	= extlog_print,
188};
189
190static int __init extlog_init(void)
191{
192	struct extlog_l1_head *l1_head;
193	void __iomem *extlog_l1_hdr;
194	size_t l1_hdr_size;
195	struct resource *r;
196	u64 cap;
197	int rc;
198
199	if (get_edac_report_status() == EDAC_REPORTING_FORCE) {
200		pr_warn("Not loading eMCA, error reporting force-enabled through EDAC.\n");
201		return -EPERM;
202	}
203
204	rc = -ENODEV;
205	rdmsrl(MSR_IA32_MCG_CAP, cap);
206	if (!(cap & MCG_ELOG_P))
207		return rc;
208
209	if (!extlog_get_l1addr())
210		return rc;
211
212	rc = -EINVAL;
213	/* get L1 header to fetch necessary information */
214	l1_hdr_size = sizeof(struct extlog_l1_head);
215	r = request_mem_region(l1_dirbase, l1_hdr_size, "L1 DIR HDR");
216	if (!r) {
217		pr_warn(FW_BUG EMCA_BUG,
218			(unsigned long long)l1_dirbase,
219			(unsigned long long)l1_dirbase + l1_hdr_size);
220		goto err;
221	}
222
223	extlog_l1_hdr = acpi_os_map_memory(l1_dirbase, l1_hdr_size);
224	l1_head = (struct extlog_l1_head *)extlog_l1_hdr;
225	l1_size = l1_head->total_len;
226	l1_percpu_entry = l1_head->entries;
227	elog_base = l1_head->elog_base;
228	elog_size = l1_head->elog_len;
229	acpi_os_unmap_memory(extlog_l1_hdr, l1_hdr_size);
230	release_mem_region(l1_dirbase, l1_hdr_size);
231
232	/* remap L1 header again based on completed information */
233	r = request_mem_region(l1_dirbase, l1_size, "L1 Table");
234	if (!r) {
235		pr_warn(FW_BUG EMCA_BUG,
236			(unsigned long long)l1_dirbase,
237			(unsigned long long)l1_dirbase + l1_size);
238		goto err;
239	}
240	extlog_l1_addr = acpi_os_map_memory(l1_dirbase, l1_size);
241	l1_entry_base = (u64 *)((u8 *)extlog_l1_addr + l1_hdr_size);
242
243	/* remap elog table */
244	r = request_mem_region(elog_base, elog_size, "Elog Table");
245	if (!r) {
246		pr_warn(FW_BUG EMCA_BUG,
247			(unsigned long long)elog_base,
248			(unsigned long long)elog_base + elog_size);
249		goto err_release_l1_dir;
250	}
251	elog_addr = acpi_os_map_memory(elog_base, elog_size);
252
253	rc = -ENOMEM;
254	/* allocate buffer to save elog record */
255	elog_buf = kmalloc(ELOG_ENTRY_LEN, GFP_KERNEL);
256	if (elog_buf == NULL)
257		goto err_release_elog;
258
259	/*
260	 * eMCA event report method has higher priority than EDAC method,
261	 * unless EDAC event report method is mandatory.
262	 */
263	old_edac_report_status = get_edac_report_status();
264	set_edac_report_status(EDAC_REPORTING_DISABLED);
265	mce_register_decode_chain(&extlog_mce_dec);
266	/* enable OS to be involved to take over management from BIOS */
267	((struct extlog_l1_head *)extlog_l1_addr)->flags |= FLAG_OS_OPTIN;
268
269	return 0;
270
271err_release_elog:
272	if (elog_addr)
273		acpi_os_unmap_memory(elog_addr, elog_size);
274	release_mem_region(elog_base, elog_size);
275err_release_l1_dir:
276	if (extlog_l1_addr)
277		acpi_os_unmap_memory(extlog_l1_addr, l1_size);
278	release_mem_region(l1_dirbase, l1_size);
279err:
280	pr_warn(FW_BUG "Extended error log disabled because of problems parsing f/w tables\n");
281	return rc;
282}
283
284static void __exit extlog_exit(void)
285{
286	set_edac_report_status(old_edac_report_status);
287	mce_unregister_decode_chain(&extlog_mce_dec);
288	((struct extlog_l1_head *)extlog_l1_addr)->flags &= ~FLAG_OS_OPTIN;
289	if (extlog_l1_addr)
290		acpi_os_unmap_memory(extlog_l1_addr, l1_size);
291	if (elog_addr)
292		acpi_os_unmap_memory(elog_addr, elog_size);
293	release_mem_region(elog_base, elog_size);
294	release_mem_region(l1_dirbase, l1_size);
295	kfree(elog_buf);
296}
297
298module_init(extlog_init);
299module_exit(extlog_exit);
300
301MODULE_AUTHOR("Chen, Gong <gong.chen@intel.com>");
302MODULE_DESCRIPTION("Extended MCA Error Log Driver");
303MODULE_LICENSE("GPL");