1 // SPDX-License-Identifier: GPL-2.0-only
3 * GHES/EDAC Linux driver
5 * Copyright (c) 2013 by Mauro Carvalho Chehab
7 * Red Hat Inc. http://www.redhat.com
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <acpi/ghes.h>
13 #include <linux/edac.h>
14 #include <linux/dmi.h>
15 #include "edac_module.h"
16 #include <ras/ras_event.h>
19 struct mem_ctl_info *mci;
21 /* Buffers for the error handling routine */
22 char other_detail[400];
26 static refcount_t ghes_refcount = REFCOUNT_INIT(0);
29 * Access to ghes_pvt must be protected by ghes_lock. The spinlock
30 * also provides the necessary (implicit) memory barrier for the SMP
31 * case to make the pointer visible on another CPU.
33 static struct ghes_pvt *ghes_pvt;
36 * This driver's representation of the system hardware, as collected
41 struct dimm_info *dimms;
44 /* GHES registration mutex */
45 static DEFINE_MUTEX(ghes_reg_mutex);
48 * Sync with other, potentially concurrent callers of
49 * ghes_edac_report_mem_error(). We don't know what the
50 * "inventive" firmware would do.
52 static DEFINE_SPINLOCK(ghes_lock);
54 /* "ghes_edac.force_load=1" skips the platform check */
55 static bool __read_mostly force_load;
56 module_param(force_load, bool, 0);
58 static bool system_scanned;
60 /* Memory Device - Type 17 of SMBIOS spec */
61 struct memdev_dmi_entry {
65 u16 phys_mem_array_handle;
66 u16 mem_err_info_handle;
83 u16 conf_mem_clk_speed;
84 } __attribute__((__packed__));
86 static struct dimm_info *find_dimm_by_handle(struct mem_ctl_info *mci, u16 handle)
88 struct dimm_info *dimm;
90 mci_for_each_dimm(mci, dimm) {
91 if (dimm->smbios_handle == handle)
98 static void dimm_setup_label(struct dimm_info *dimm, u16 handle)
100 const char *bank = NULL, *device = NULL;
102 dmi_memdev_name(handle, &bank, &device);
104 /* both strings must be non-zero */
105 if (bank && *bank && device && *device)
106 snprintf(dimm->label, sizeof(dimm->label), "%s %s", bank, device);
109 static void assign_dmi_dimm_info(struct dimm_info *dimm, struct memdev_dmi_entry *entry)
111 u16 rdr_mask = BIT(7) | BIT(13);
113 if (entry->size == 0xffff) {
114 pr_info("Can't get DIMM%i size\n", dimm->idx);
115 dimm->nr_pages = MiB_TO_PAGES(32);/* Unknown */
116 } else if (entry->size == 0x7fff) {
117 dimm->nr_pages = MiB_TO_PAGES(entry->extended_size);
119 if (entry->size & BIT(15))
120 dimm->nr_pages = MiB_TO_PAGES((entry->size & 0x7fff) << 10);
122 dimm->nr_pages = MiB_TO_PAGES(entry->size);
125 switch (entry->memory_type) {
127 if (entry->type_detail & BIT(13))
128 dimm->mtype = MEM_RDDR;
130 dimm->mtype = MEM_DDR;
133 if (entry->type_detail & BIT(13))
134 dimm->mtype = MEM_RDDR2;
136 dimm->mtype = MEM_DDR2;
139 dimm->mtype = MEM_FB_DDR2;
142 if (entry->type_detail & BIT(12))
143 dimm->mtype = MEM_NVDIMM;
144 else if (entry->type_detail & BIT(13))
145 dimm->mtype = MEM_RDDR3;
147 dimm->mtype = MEM_DDR3;
150 if (entry->type_detail & BIT(12))
151 dimm->mtype = MEM_NVDIMM;
152 else if (entry->type_detail & BIT(13))
153 dimm->mtype = MEM_RDDR4;
155 dimm->mtype = MEM_DDR4;
158 if (entry->type_detail & BIT(6))
159 dimm->mtype = MEM_RMBS;
160 else if ((entry->type_detail & rdr_mask) == rdr_mask)
161 dimm->mtype = MEM_RDR;
162 else if (entry->type_detail & BIT(7))
163 dimm->mtype = MEM_SDR;
164 else if (entry->type_detail & BIT(9))
165 dimm->mtype = MEM_EDO;
167 dimm->mtype = MEM_UNKNOWN;
171 * Actually, we can only detect if the memory has bits for
174 if (entry->total_width == entry->data_width)
175 dimm->edac_mode = EDAC_NONE;
177 dimm->edac_mode = EDAC_SECDED;
179 dimm->dtype = DEV_UNKNOWN;
180 dimm->grain = 128; /* Likely, worse case */
182 dimm_setup_label(dimm, entry->handle);
184 if (dimm->nr_pages) {
185 edac_dbg(1, "DIMM%i: %s size = %d MB%s\n",
186 dimm->idx, edac_mem_types[dimm->mtype],
187 PAGES_TO_MiB(dimm->nr_pages),
188 (dimm->edac_mode != EDAC_NONE) ? "(ECC)" : "");
189 edac_dbg(2, "\ttype %d, detail 0x%02x, width %d(total %d)\n",
190 entry->memory_type, entry->type_detail,
191 entry->total_width, entry->data_width);
194 dimm->smbios_handle = entry->handle;
197 static void enumerate_dimms(const struct dmi_header *dh, void *arg)
199 struct memdev_dmi_entry *entry = (struct memdev_dmi_entry *)dh;
200 struct ghes_hw_desc *hw = (struct ghes_hw_desc *)arg;
203 if (dh->type != DMI_ENTRY_MEM_DEVICE)
206 /* Enlarge the array with additional 16 */
207 if (!hw->num_dimms || !(hw->num_dimms % 16)) {
208 struct dimm_info *new;
210 new = krealloc(hw->dimms, (hw->num_dimms + 16) * sizeof(struct dimm_info),
220 d = &hw->dimms[hw->num_dimms];
221 d->idx = hw->num_dimms;
223 assign_dmi_dimm_info(d, entry);
228 static void ghes_scan_system(void)
233 dmi_walk(enumerate_dimms, &ghes_hw);
235 system_scanned = true;
238 void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err)
240 struct edac_raw_error_desc *e;
241 struct mem_ctl_info *mci;
242 struct ghes_pvt *pvt;
247 * We can do the locking below because GHES defers error processing
248 * from NMI to IRQ context. Whenever that changes, we'd at least
251 if (WARN_ON_ONCE(in_nmi()))
254 spin_lock_irqsave(&ghes_lock, flags);
261 e = &mci->error_desc;
263 /* Cleans the error report buffer */
264 memset(e, 0, sizeof (*e));
268 e->other_detail = pvt->other_detail;
272 *pvt->other_detail = '\0';
276 case GHES_SEV_CORRECTED:
277 e->type = HW_EVENT_ERR_CORRECTED;
279 case GHES_SEV_RECOVERABLE:
280 e->type = HW_EVENT_ERR_UNCORRECTED;
283 e->type = HW_EVENT_ERR_FATAL;
287 e->type = HW_EVENT_ERR_INFO;
290 edac_dbg(1, "error validation_bits: 0x%08llx\n",
291 (long long)mem_err->validation_bits);
293 /* Error type, mapped on e->msg */
294 if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) {
296 switch (mem_err->error_type) {
298 p += sprintf(p, "Unknown");
301 p += sprintf(p, "No error");
304 p += sprintf(p, "Single-bit ECC");
307 p += sprintf(p, "Multi-bit ECC");
310 p += sprintf(p, "Single-symbol ChipKill ECC");
313 p += sprintf(p, "Multi-symbol ChipKill ECC");
316 p += sprintf(p, "Master abort");
319 p += sprintf(p, "Target abort");
322 p += sprintf(p, "Parity Error");
325 p += sprintf(p, "Watchdog timeout");
328 p += sprintf(p, "Invalid address");
331 p += sprintf(p, "Mirror Broken");
334 p += sprintf(p, "Memory Sparing");
337 p += sprintf(p, "Scrub corrected error");
340 p += sprintf(p, "Scrub uncorrected error");
343 p += sprintf(p, "Physical Memory Map-out event");
346 p += sprintf(p, "reserved error (%d)",
347 mem_err->error_type);
350 strcpy(pvt->msg, "unknown error");
354 if (mem_err->validation_bits & CPER_MEM_VALID_PA) {
355 e->page_frame_number = PHYS_PFN(mem_err->physical_addr);
356 e->offset_in_page = offset_in_page(mem_err->physical_addr);
360 if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK)
361 e->grain = ~mem_err->physical_addr_mask + 1;
363 /* Memory error location, mapped on e->location */
365 if (mem_err->validation_bits & CPER_MEM_VALID_NODE)
366 p += sprintf(p, "node:%d ", mem_err->node);
367 if (mem_err->validation_bits & CPER_MEM_VALID_CARD)
368 p += sprintf(p, "card:%d ", mem_err->card);
369 if (mem_err->validation_bits & CPER_MEM_VALID_MODULE)
370 p += sprintf(p, "module:%d ", mem_err->module);
371 if (mem_err->validation_bits & CPER_MEM_VALID_RANK_NUMBER)
372 p += sprintf(p, "rank:%d ", mem_err->rank);
373 if (mem_err->validation_bits & CPER_MEM_VALID_BANK)
374 p += sprintf(p, "bank:%d ", mem_err->bank);
375 if (mem_err->validation_bits & CPER_MEM_VALID_ROW)
376 p += sprintf(p, "row:%d ", mem_err->row);
377 if (mem_err->validation_bits & CPER_MEM_VALID_COLUMN)
378 p += sprintf(p, "col:%d ", mem_err->column);
379 if (mem_err->validation_bits & CPER_MEM_VALID_BIT_POSITION)
380 p += sprintf(p, "bit_pos:%d ", mem_err->bit_pos);
381 if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) {
382 const char *bank = NULL, *device = NULL;
383 struct dimm_info *dimm;
385 dmi_memdev_name(mem_err->mem_dev_handle, &bank, &device);
386 if (bank != NULL && device != NULL)
387 p += sprintf(p, "DIMM location:%s %s ", bank, device);
389 p += sprintf(p, "DIMM DMI handle: 0x%.4x ",
390 mem_err->mem_dev_handle);
392 dimm = find_dimm_by_handle(mci, mem_err->mem_dev_handle);
394 e->top_layer = dimm->idx;
395 strcpy(e->label, dimm->label);
402 strcpy(e->label, "unknown memory");
404 /* All other fields are mapped on e->other_detail */
405 p = pvt->other_detail;
406 p += snprintf(p, sizeof(pvt->other_detail),
407 "APEI location: %s ", e->location);
408 if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_STATUS) {
409 u64 status = mem_err->error_status;
411 p += sprintf(p, "status(0x%016llx): ", (long long)status);
412 switch ((status >> 8) & 0xff) {
414 p += sprintf(p, "Error detected internal to the component ");
417 p += sprintf(p, "Error detected in the bus ");
420 p += sprintf(p, "Storage error in DRAM memory ");
423 p += sprintf(p, "Storage error in TLB ");
426 p += sprintf(p, "Storage error in cache ");
429 p += sprintf(p, "Error in one or more functional units ");
432 p += sprintf(p, "component failed self test ");
435 p += sprintf(p, "Overflow or undervalue of internal queue ");
438 p += sprintf(p, "Virtual address not found on IO-TLB or IO-PDIR ");
441 p += sprintf(p, "Improper access error ");
444 p += sprintf(p, "Access to a memory address which is not mapped to any component ");
447 p += sprintf(p, "Loss of Lockstep ");
450 p += sprintf(p, "Response not associated with a request ");
453 p += sprintf(p, "Bus parity error - must also set the A, C, or D Bits ");
456 p += sprintf(p, "Detection of a PATH_ERROR ");
459 p += sprintf(p, "Bus operation timeout ");
462 p += sprintf(p, "A read was issued to data that has been poisoned ");
465 p += sprintf(p, "reserved ");
469 if (mem_err->validation_bits & CPER_MEM_VALID_REQUESTOR_ID)
470 p += sprintf(p, "requestorID: 0x%016llx ",
471 (long long)mem_err->requestor_id);
472 if (mem_err->validation_bits & CPER_MEM_VALID_RESPONDER_ID)
473 p += sprintf(p, "responderID: 0x%016llx ",
474 (long long)mem_err->responder_id);
475 if (mem_err->validation_bits & CPER_MEM_VALID_TARGET_ID)
476 p += sprintf(p, "targetID: 0x%016llx ",
477 (long long)mem_err->responder_id);
478 if (p > pvt->other_detail)
481 edac_raw_mc_handle_error(e);
484 spin_unlock_irqrestore(&ghes_lock, flags);
488 * Known systems that are safe to enable this module.
490 static struct acpi_platform_list plat_list[] = {
491 {"HPE ", "Server ", 0, ACPI_SIG_FADT, all_versions},
495 int ghes_edac_register(struct ghes *ghes, struct device *dev)
498 struct mem_ctl_info *mci;
499 struct ghes_pvt *pvt;
500 struct edac_mc_layer layers[1];
505 if (IS_ENABLED(CONFIG_X86)) {
506 /* Check if safe to enable on this system */
507 idx = acpi_match_platform_list(plat_list);
508 if (!force_load && idx < 0)
514 /* finish another registration/unregistration instance first */
515 mutex_lock(&ghes_reg_mutex);
518 * We have only one logical memory controller to which all DIMMs belong.
520 if (refcount_inc_not_zero(&ghes_refcount))
525 /* Check if we've got a bogus BIOS */
526 if (!ghes_hw.num_dimms) {
528 ghes_hw.num_dimms = 1;
531 layers[0].type = EDAC_MC_LAYER_ALL_MEM;
532 layers[0].size = ghes_hw.num_dimms;
533 layers[0].is_virt_csrow = true;
535 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(struct ghes_pvt));
537 pr_info("Can't allocate memory for EDAC data\n");
546 mci->mtype_cap = MEM_FLAG_EMPTY;
547 mci->edac_ctl_cap = EDAC_FLAG_NONE;
548 mci->edac_cap = EDAC_FLAG_NONE;
549 mci->mod_name = "ghes_edac.c";
550 mci->ctl_name = "ghes_edac";
551 mci->dev_name = "ghes";
554 pr_info("This system has a very crappy BIOS: It doesn't even list the DIMMS.\n");
555 pr_info("Its SMBIOS info is wrong. It is doubtful that the error report would\n");
556 pr_info("work on such system. Use this driver with caution\n");
557 } else if (idx < 0) {
558 pr_info("This EDAC driver relies on BIOS to enumerate memory and get error reports.\n");
559 pr_info("Unfortunately, not all BIOSes reflect the memory layout correctly.\n");
560 pr_info("So, the end result of using this driver varies from vendor to vendor.\n");
561 pr_info("If you find incorrect reports, please contact your hardware vendor\n");
562 pr_info("to correct its BIOS.\n");
563 pr_info("This system has %d DIMM sockets.\n", ghes_hw.num_dimms);
567 struct dimm_info *src, *dst;
570 mci_for_each_dimm(mci, dst) {
571 src = &ghes_hw.dimms[i];
574 dst->smbios_handle = src->smbios_handle;
575 dst->nr_pages = src->nr_pages;
576 dst->mtype = src->mtype;
577 dst->edac_mode = src->edac_mode;
578 dst->dtype = src->dtype;
579 dst->grain = src->grain;
582 * If no src->label, preserve default label assigned
585 if (strlen(src->label))
586 memcpy(dst->label, src->label, sizeof(src->label));
592 struct dimm_info *dimm = edac_get_dimm(mci, 0, 0, 0);
596 dimm->mtype = MEM_UNKNOWN;
597 dimm->dtype = DEV_UNKNOWN;
598 dimm->edac_mode = EDAC_SECDED;
601 rc = edac_mc_add_mc(mci);
603 pr_info("Can't register with the EDAC core\n");
609 spin_lock_irqsave(&ghes_lock, flags);
611 spin_unlock_irqrestore(&ghes_lock, flags);
613 /* only set on success */
614 refcount_set(&ghes_refcount, 1);
618 /* Not needed anymore */
619 kfree(ghes_hw.dimms);
620 ghes_hw.dimms = NULL;
622 mutex_unlock(&ghes_reg_mutex);
627 void ghes_edac_unregister(struct ghes *ghes)
629 struct mem_ctl_info *mci;
632 mutex_lock(&ghes_reg_mutex);
634 system_scanned = false;
636 if (!refcount_dec_and_test(&ghes_refcount))
640 * Wait for the irq handler being finished.
642 spin_lock_irqsave(&ghes_lock, flags);
643 mci = ghes_pvt ? ghes_pvt->mci : NULL;
645 spin_unlock_irqrestore(&ghes_lock, flags);
650 mci = edac_mc_del_mc(mci->pdev);
655 mutex_unlock(&ghes_reg_mutex);