2 * GHES/EDAC Linux driver
4 * This file may be distributed under the terms of the GNU General Public
7 * Copyright (c) 2013 by Mauro Carvalho Chehab
9 * Red Hat Inc. http://www.redhat.com
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <acpi/ghes.h>
15 #include <linux/edac.h>
16 #include <linux/dmi.h>
17 #include "edac_module.h"
18 #include <ras/ras_event.h>
20 struct ghes_edac_pvt {
21 struct list_head list;
23 struct mem_ctl_info *mci;
25 /* Buffers for the error handling routine */
26 char detail_location[240];
27 char other_detail[160];
31 static LIST_HEAD(ghes_reglist);
32 static DEFINE_MUTEX(ghes_edac_lock);
33 static int ghes_edac_mc_num;
36 /* Memory Device - Type 17 of SMBIOS spec */
37 struct memdev_dmi_entry {
41 u16 phys_mem_array_handle;
42 u16 mem_err_info_handle;
59 u16 conf_mem_clk_speed;
60 } __attribute__((__packed__));
62 struct ghes_edac_dimm_fill {
63 struct mem_ctl_info *mci;
67 static void ghes_edac_count_dimms(const struct dmi_header *dh, void *arg)
71 if (dh->type == DMI_ENTRY_MEM_DEVICE)
75 static void ghes_edac_dmidecode(const struct dmi_header *dh, void *arg)
77 struct ghes_edac_dimm_fill *dimm_fill = arg;
78 struct mem_ctl_info *mci = dimm_fill->mci;
80 if (dh->type == DMI_ENTRY_MEM_DEVICE) {
81 struct memdev_dmi_entry *entry = (struct memdev_dmi_entry *)dh;
82 struct dimm_info *dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
84 dimm_fill->count, 0, 0);
86 if (entry->size == 0xffff) {
87 pr_info("Can't get DIMM%i size\n",
89 dimm->nr_pages = MiB_TO_PAGES(32);/* Unknown */
90 } else if (entry->size == 0x7fff) {
91 dimm->nr_pages = MiB_TO_PAGES(entry->extended_size);
93 if (entry->size & 1 << 15)
94 dimm->nr_pages = MiB_TO_PAGES((entry->size &
97 dimm->nr_pages = MiB_TO_PAGES(entry->size);
100 switch (entry->memory_type) {
102 if (entry->type_detail & 1 << 13)
103 dimm->mtype = MEM_RDDR;
105 dimm->mtype = MEM_DDR;
108 if (entry->type_detail & 1 << 13)
109 dimm->mtype = MEM_RDDR2;
111 dimm->mtype = MEM_DDR2;
114 dimm->mtype = MEM_FB_DDR2;
117 if (entry->type_detail & 1 << 13)
118 dimm->mtype = MEM_RDDR3;
120 dimm->mtype = MEM_DDR3;
123 if (entry->type_detail & 1 << 6)
124 dimm->mtype = MEM_RMBS;
125 else if ((entry->type_detail & ((1 << 7) | (1 << 13)))
126 == ((1 << 7) | (1 << 13)))
127 dimm->mtype = MEM_RDR;
128 else if (entry->type_detail & 1 << 7)
129 dimm->mtype = MEM_SDR;
130 else if (entry->type_detail & 1 << 9)
131 dimm->mtype = MEM_EDO;
133 dimm->mtype = MEM_UNKNOWN;
137 * Actually, we can only detect if the memory has bits for
140 if (entry->total_width == entry->data_width)
141 dimm->edac_mode = EDAC_NONE;
143 dimm->edac_mode = EDAC_SECDED;
145 dimm->dtype = DEV_UNKNOWN;
146 dimm->grain = 128; /* Likely, worse case */
149 * FIXME: It shouldn't be hard to also fill the DIMM labels
152 if (dimm->nr_pages) {
153 edac_dbg(1, "DIMM%i: %s size = %d MB%s\n",
154 dimm_fill->count, edac_mem_types[dimm->mtype],
155 PAGES_TO_MiB(dimm->nr_pages),
156 (dimm->edac_mode != EDAC_NONE) ? "(ECC)" : "");
157 edac_dbg(2, "\ttype %d, detail 0x%02x, width %d(total %d)\n",
158 entry->memory_type, entry->type_detail,
159 entry->total_width, entry->data_width);
166 void ghes_edac_report_mem_error(struct ghes *ghes, int sev,
167 struct cper_sec_mem_err *mem_err)
169 enum hw_event_mc_err_type type;
170 struct edac_raw_error_desc *e;
171 struct mem_ctl_info *mci;
172 struct ghes_edac_pvt *pvt = NULL;
176 list_for_each_entry(pvt, &ghes_reglist, list) {
177 if (ghes == pvt->ghes)
181 pr_err("Internal error: Can't find EDAC structure\n");
185 e = &mci->error_desc;
187 /* Cleans the error report buffer */
188 memset(e, 0, sizeof (*e));
191 strcpy(e->label, "unknown label");
193 e->other_detail = pvt->other_detail;
197 *pvt->other_detail = '\0';
201 case GHES_SEV_CORRECTED:
202 type = HW_EVENT_ERR_CORRECTED;
204 case GHES_SEV_RECOVERABLE:
205 type = HW_EVENT_ERR_UNCORRECTED;
208 type = HW_EVENT_ERR_FATAL;
212 type = HW_EVENT_ERR_INFO;
215 edac_dbg(1, "error validation_bits: 0x%08llx\n",
216 (long long)mem_err->validation_bits);
218 /* Error type, mapped on e->msg */
219 if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) {
221 switch (mem_err->error_type) {
223 p += sprintf(p, "Unknown");
226 p += sprintf(p, "No error");
229 p += sprintf(p, "Single-bit ECC");
232 p += sprintf(p, "Multi-bit ECC");
235 p += sprintf(p, "Single-symbol ChipKill ECC");
238 p += sprintf(p, "Multi-symbol ChipKill ECC");
241 p += sprintf(p, "Master abort");
244 p += sprintf(p, "Target abort");
247 p += sprintf(p, "Parity Error");
250 p += sprintf(p, "Watchdog timeout");
253 p += sprintf(p, "Invalid address");
256 p += sprintf(p, "Mirror Broken");
259 p += sprintf(p, "Memory Sparing");
262 p += sprintf(p, "Scrub corrected error");
265 p += sprintf(p, "Scrub uncorrected error");
268 p += sprintf(p, "Physical Memory Map-out event");
271 p += sprintf(p, "reserved error (%d)",
272 mem_err->error_type);
275 strcpy(pvt->msg, "unknown error");
279 if (mem_err->validation_bits & CPER_MEM_VALID_PA) {
280 e->page_frame_number = mem_err->physical_addr >> PAGE_SHIFT;
281 e->offset_in_page = mem_err->physical_addr & ~PAGE_MASK;
285 if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK)
286 e->grain = ~mem_err->physical_addr_mask + 1;
288 /* Memory error location, mapped on e->location */
290 if (mem_err->validation_bits & CPER_MEM_VALID_NODE)
291 p += sprintf(p, "node:%d ", mem_err->node);
292 if (mem_err->validation_bits & CPER_MEM_VALID_CARD)
293 p += sprintf(p, "card:%d ", mem_err->card);
294 if (mem_err->validation_bits & CPER_MEM_VALID_MODULE)
295 p += sprintf(p, "module:%d ", mem_err->module);
296 if (mem_err->validation_bits & CPER_MEM_VALID_RANK_NUMBER)
297 p += sprintf(p, "rank:%d ", mem_err->rank);
298 if (mem_err->validation_bits & CPER_MEM_VALID_BANK)
299 p += sprintf(p, "bank:%d ", mem_err->bank);
300 if (mem_err->validation_bits & CPER_MEM_VALID_ROW)
301 p += sprintf(p, "row:%d ", mem_err->row);
302 if (mem_err->validation_bits & CPER_MEM_VALID_COLUMN)
303 p += sprintf(p, "col:%d ", mem_err->column);
304 if (mem_err->validation_bits & CPER_MEM_VALID_BIT_POSITION)
305 p += sprintf(p, "bit_pos:%d ", mem_err->bit_pos);
306 if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) {
307 const char *bank = NULL, *device = NULL;
308 dmi_memdev_name(mem_err->mem_dev_handle, &bank, &device);
309 if (bank != NULL && device != NULL)
310 p += sprintf(p, "DIMM location:%s %s ", bank, device);
312 p += sprintf(p, "DIMM DMI handle: 0x%.4x ",
313 mem_err->mem_dev_handle);
318 /* All other fields are mapped on e->other_detail */
319 p = pvt->other_detail;
320 if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_STATUS) {
321 u64 status = mem_err->error_status;
323 p += sprintf(p, "status(0x%016llx): ", (long long)status);
324 switch ((status >> 8) & 0xff) {
326 p += sprintf(p, "Error detected internal to the component ");
329 p += sprintf(p, "Error detected in the bus ");
332 p += sprintf(p, "Storage error in DRAM memory ");
335 p += sprintf(p, "Storage error in TLB ");
338 p += sprintf(p, "Storage error in cache ");
341 p += sprintf(p, "Error in one or more functional units ");
344 p += sprintf(p, "component failed self test ");
347 p += sprintf(p, "Overflow or undervalue of internal queue ");
350 p += sprintf(p, "Virtual address not found on IO-TLB or IO-PDIR ");
353 p += sprintf(p, "Improper access error ");
356 p += sprintf(p, "Access to a memory address which is not mapped to any component ");
359 p += sprintf(p, "Loss of Lockstep ");
362 p += sprintf(p, "Response not associated with a request ");
365 p += sprintf(p, "Bus parity error - must also set the A, C, or D Bits ");
368 p += sprintf(p, "Detection of a PATH_ERROR ");
371 p += sprintf(p, "Bus operation timeout ");
374 p += sprintf(p, "A read was issued to data that has been poisoned ");
377 p += sprintf(p, "reserved ");
381 if (mem_err->validation_bits & CPER_MEM_VALID_REQUESTOR_ID)
382 p += sprintf(p, "requestorID: 0x%016llx ",
383 (long long)mem_err->requestor_id);
384 if (mem_err->validation_bits & CPER_MEM_VALID_RESPONDER_ID)
385 p += sprintf(p, "responderID: 0x%016llx ",
386 (long long)mem_err->responder_id);
387 if (mem_err->validation_bits & CPER_MEM_VALID_TARGET_ID)
388 p += sprintf(p, "targetID: 0x%016llx ",
389 (long long)mem_err->responder_id);
390 if (p > pvt->other_detail)
393 /* Sanity-check driver-supplied grain value. */
394 if (WARN_ON_ONCE(!e->grain))
397 grain_bits = fls_long(e->grain - 1);
399 /* Generate the trace event */
400 snprintf(pvt->detail_location, sizeof(pvt->detail_location),
401 "APEI location: %s %s", e->location, e->other_detail);
402 trace_mc_event(type, e->msg, e->label, e->error_count,
403 mci->mc_idx, e->top_layer, e->mid_layer, e->low_layer,
404 (e->page_frame_number << PAGE_SHIFT) | e->offset_in_page,
405 grain_bits, e->syndrome, pvt->detail_location);
407 /* Report the error via EDAC API */
408 edac_raw_mc_handle_error(type, mci, e);
410 EXPORT_SYMBOL_GPL(ghes_edac_report_mem_error);
412 int ghes_edac_register(struct ghes *ghes, struct device *dev)
415 int rc, num_dimm = 0;
416 struct mem_ctl_info *mci;
417 struct edac_mc_layer layers[1];
418 struct ghes_edac_pvt *pvt;
419 struct ghes_edac_dimm_fill dimm_fill;
421 /* Get the number of DIMMs */
422 dmi_walk(ghes_edac_count_dimms, &num_dimm);
424 /* Check if we've got a bogus BIOS */
430 layers[0].type = EDAC_MC_LAYER_ALL_MEM;
431 layers[0].size = num_dimm;
432 layers[0].is_virt_csrow = true;
435 * We need to serialize edac_mc_alloc() and edac_mc_add_mc(),
436 * to avoid duplicated memory controller numbers
438 mutex_lock(&ghes_edac_lock);
439 mci = edac_mc_alloc(ghes_edac_mc_num, ARRAY_SIZE(layers), layers,
442 pr_info("Can't allocate memory for EDAC data\n");
443 mutex_unlock(&ghes_edac_lock);
448 memset(pvt, 0, sizeof(*pvt));
449 list_add_tail(&pvt->list, &ghes_reglist);
454 mci->mtype_cap = MEM_FLAG_EMPTY;
455 mci->edac_ctl_cap = EDAC_FLAG_NONE;
456 mci->edac_cap = EDAC_FLAG_NONE;
457 mci->mod_name = "ghes_edac.c";
458 mci->ctl_name = "ghes_edac";
459 mci->dev_name = "ghes";
461 if (!ghes_edac_mc_num) {
463 pr_info("This EDAC driver relies on BIOS to enumerate memory and get error reports.\n");
464 pr_info("Unfortunately, not all BIOSes reflect the memory layout correctly.\n");
465 pr_info("So, the end result of using this driver varies from vendor to vendor.\n");
466 pr_info("If you find incorrect reports, please contact your hardware vendor\n");
467 pr_info("to correct its BIOS.\n");
468 pr_info("This system has %d DIMM sockets.\n",
471 pr_info("This system has a very crappy BIOS: It doesn't even list the DIMMS.\n");
472 pr_info("Its SMBIOS info is wrong. It is doubtful that the error report would\n");
473 pr_info("work on such system. Use this driver with caution\n");
479 * Fill DIMM info from DMI for the memory controller #0
481 * Keep it in blank for the other memory controllers, as
482 * there's no reliable way to properly credit each DIMM to
483 * the memory controller, as different BIOSes fill the
484 * DMI bank location fields on different ways
486 if (!ghes_edac_mc_num) {
489 dmi_walk(ghes_edac_dmidecode, &dimm_fill);
492 struct dimm_info *dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
493 mci->n_layers, 0, 0, 0);
497 dimm->mtype = MEM_UNKNOWN;
498 dimm->dtype = DEV_UNKNOWN;
499 dimm->edac_mode = EDAC_SECDED;
502 rc = edac_mc_add_mc(mci);
504 pr_info("Can't register at EDAC core\n");
506 mutex_unlock(&ghes_edac_lock);
511 mutex_unlock(&ghes_edac_lock);
514 EXPORT_SYMBOL_GPL(ghes_edac_register);
516 void ghes_edac_unregister(struct ghes *ghes)
518 struct mem_ctl_info *mci;
519 struct ghes_edac_pvt *pvt, *tmp;
521 list_for_each_entry_safe(pvt, tmp, &ghes_reglist, list) {
522 if (ghes == pvt->ghes) {
524 edac_mc_del_mc(mci->pdev);
526 list_del(&pvt->list);
530 EXPORT_SYMBOL_GPL(ghes_edac_unregister);