Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / edac / ghes_edac.c
blob6d1ddecbf0da3606c73e1139cd01c92af1ede5ba
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * GHES/EDAC Linux driver
5 * Copyright (c) 2013 by Mauro Carvalho Chehab
7 * Red Hat Inc. https://www.redhat.com
8 */
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>
18 struct ghes_pvt {
19 struct mem_ctl_info *mci;
21 /* Buffers for the error handling routine */
22 char other_detail[400];
23 char msg[80];
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
37 * from DMI.
39 struct ghes_hw_desc {
40 int num_dimms;
41 struct dimm_info *dimms;
42 } ghes_hw;
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 {
62 u8 type;
63 u8 length;
64 u16 handle;
65 u16 phys_mem_array_handle;
66 u16 mem_err_info_handle;
67 u16 total_width;
68 u16 data_width;
69 u16 size;
70 u8 form_factor;
71 u8 device_set;
72 u8 device_locator;
73 u8 bank_locator;
74 u8 memory_type;
75 u16 type_detail;
76 u16 speed;
77 u8 manufacturer;
78 u8 serial_number;
79 u8 asset_tag;
80 u8 part_number;
81 u8 attributes;
82 u32 extended_size;
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)
92 return dimm;
95 return NULL;
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);
118 } else {
119 if (entry->size & BIT(15))
120 dimm->nr_pages = MiB_TO_PAGES((entry->size & 0x7fff) << 10);
121 else
122 dimm->nr_pages = MiB_TO_PAGES(entry->size);
125 switch (entry->memory_type) {
126 case 0x12:
127 if (entry->type_detail & BIT(13))
128 dimm->mtype = MEM_RDDR;
129 else
130 dimm->mtype = MEM_DDR;
131 break;
132 case 0x13:
133 if (entry->type_detail & BIT(13))
134 dimm->mtype = MEM_RDDR2;
135 else
136 dimm->mtype = MEM_DDR2;
137 break;
138 case 0x14:
139 dimm->mtype = MEM_FB_DDR2;
140 break;
141 case 0x18:
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;
146 else
147 dimm->mtype = MEM_DDR3;
148 break;
149 case 0x1a:
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;
154 else
155 dimm->mtype = MEM_DDR4;
156 break;
157 default:
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;
166 else
167 dimm->mtype = MEM_UNKNOWN;
171 * Actually, we can only detect if the memory has bits for
172 * checksum or not
174 if (entry->total_width == entry->data_width)
175 dimm->edac_mode = EDAC_NONE;
176 else
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;
201 struct dimm_info *d;
203 if (dh->type != DMI_ENTRY_MEM_DEVICE)
204 return;
206 /* Enlarge the array with additional 16 */
207 if (!hw->num_dimms || !(hw->num_dimms % 16)) {
208 struct dimm_info *new;
210 new = krealloc_array(hw->dimms, hw->num_dimms + 16,
211 sizeof(struct dimm_info), GFP_KERNEL);
212 if (!new) {
213 WARN_ON_ONCE(1);
214 return;
217 hw->dimms = new;
220 d = &hw->dimms[hw->num_dimms];
221 d->idx = hw->num_dimms;
223 assign_dmi_dimm_info(d, entry);
225 hw->num_dimms++;
228 static void ghes_scan_system(void)
230 if (system_scanned)
231 return;
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;
243 unsigned long flags;
244 char *p;
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
249 * know.
251 if (WARN_ON_ONCE(in_nmi()))
252 return;
254 spin_lock_irqsave(&ghes_lock, flags);
256 pvt = ghes_pvt;
257 if (!pvt)
258 goto unlock;
260 mci = pvt->mci;
261 e = &mci->error_desc;
263 /* Cleans the error report buffer */
264 memset(e, 0, sizeof (*e));
265 e->error_count = 1;
266 e->grain = 1;
267 e->msg = pvt->msg;
268 e->other_detail = pvt->other_detail;
269 e->top_layer = -1;
270 e->mid_layer = -1;
271 e->low_layer = -1;
272 *pvt->other_detail = '\0';
273 *pvt->msg = '\0';
275 switch (sev) {
276 case GHES_SEV_CORRECTED:
277 e->type = HW_EVENT_ERR_CORRECTED;
278 break;
279 case GHES_SEV_RECOVERABLE:
280 e->type = HW_EVENT_ERR_UNCORRECTED;
281 break;
282 case GHES_SEV_PANIC:
283 e->type = HW_EVENT_ERR_FATAL;
284 break;
285 default:
286 case GHES_SEV_NO:
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) {
295 p = pvt->msg;
296 switch (mem_err->error_type) {
297 case 0:
298 p += sprintf(p, "Unknown");
299 break;
300 case 1:
301 p += sprintf(p, "No error");
302 break;
303 case 2:
304 p += sprintf(p, "Single-bit ECC");
305 break;
306 case 3:
307 p += sprintf(p, "Multi-bit ECC");
308 break;
309 case 4:
310 p += sprintf(p, "Single-symbol ChipKill ECC");
311 break;
312 case 5:
313 p += sprintf(p, "Multi-symbol ChipKill ECC");
314 break;
315 case 6:
316 p += sprintf(p, "Master abort");
317 break;
318 case 7:
319 p += sprintf(p, "Target abort");
320 break;
321 case 8:
322 p += sprintf(p, "Parity Error");
323 break;
324 case 9:
325 p += sprintf(p, "Watchdog timeout");
326 break;
327 case 10:
328 p += sprintf(p, "Invalid address");
329 break;
330 case 11:
331 p += sprintf(p, "Mirror Broken");
332 break;
333 case 12:
334 p += sprintf(p, "Memory Sparing");
335 break;
336 case 13:
337 p += sprintf(p, "Scrub corrected error");
338 break;
339 case 14:
340 p += sprintf(p, "Scrub uncorrected error");
341 break;
342 case 15:
343 p += sprintf(p, "Physical Memory Map-out event");
344 break;
345 default:
346 p += sprintf(p, "reserved error (%d)",
347 mem_err->error_type);
349 } else {
350 strcpy(pvt->msg, "unknown error");
353 /* Error address */
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);
359 /* Error grain */
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 */
364 p = 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_BANK_GROUP)
376 p += sprintf(p, "bank_group:%d ",
377 mem_err->bank >> CPER_MEM_BANK_GROUP_SHIFT);
378 if (mem_err->validation_bits & CPER_MEM_VALID_BANK_ADDRESS)
379 p += sprintf(p, "bank_address:%d ",
380 mem_err->bank & CPER_MEM_BANK_ADDRESS_MASK);
381 if (mem_err->validation_bits & (CPER_MEM_VALID_ROW | CPER_MEM_VALID_ROW_EXT)) {
382 u32 row = mem_err->row;
384 row |= cper_get_mem_extension(mem_err->validation_bits, mem_err->extended);
385 p += sprintf(p, "row:%d ", row);
387 if (mem_err->validation_bits & CPER_MEM_VALID_COLUMN)
388 p += sprintf(p, "col:%d ", mem_err->column);
389 if (mem_err->validation_bits & CPER_MEM_VALID_BIT_POSITION)
390 p += sprintf(p, "bit_pos:%d ", mem_err->bit_pos);
391 if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) {
392 const char *bank = NULL, *device = NULL;
393 struct dimm_info *dimm;
395 dmi_memdev_name(mem_err->mem_dev_handle, &bank, &device);
396 if (bank != NULL && device != NULL)
397 p += sprintf(p, "DIMM location:%s %s ", bank, device);
398 else
399 p += sprintf(p, "DIMM DMI handle: 0x%.4x ",
400 mem_err->mem_dev_handle);
402 dimm = find_dimm_by_handle(mci, mem_err->mem_dev_handle);
403 if (dimm) {
404 e->top_layer = dimm->idx;
405 strcpy(e->label, dimm->label);
408 if (mem_err->validation_bits & CPER_MEM_VALID_CHIP_ID)
409 p += sprintf(p, "chipID: %d ",
410 mem_err->extended >> CPER_MEM_CHIP_ID_SHIFT);
411 if (p > e->location)
412 *(p - 1) = '\0';
414 if (!*e->label)
415 strcpy(e->label, "unknown memory");
417 /* All other fields are mapped on e->other_detail */
418 p = pvt->other_detail;
419 p += snprintf(p, sizeof(pvt->other_detail),
420 "APEI location: %s ", e->location);
421 if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_STATUS) {
422 u64 status = mem_err->error_status;
424 p += sprintf(p, "status(0x%016llx): ", (long long)status);
425 switch ((status >> 8) & 0xff) {
426 case 1:
427 p += sprintf(p, "Error detected internal to the component ");
428 break;
429 case 16:
430 p += sprintf(p, "Error detected in the bus ");
431 break;
432 case 4:
433 p += sprintf(p, "Storage error in DRAM memory ");
434 break;
435 case 5:
436 p += sprintf(p, "Storage error in TLB ");
437 break;
438 case 6:
439 p += sprintf(p, "Storage error in cache ");
440 break;
441 case 7:
442 p += sprintf(p, "Error in one or more functional units ");
443 break;
444 case 8:
445 p += sprintf(p, "component failed self test ");
446 break;
447 case 9:
448 p += sprintf(p, "Overflow or undervalue of internal queue ");
449 break;
450 case 17:
451 p += sprintf(p, "Virtual address not found on IO-TLB or IO-PDIR ");
452 break;
453 case 18:
454 p += sprintf(p, "Improper access error ");
455 break;
456 case 19:
457 p += sprintf(p, "Access to a memory address which is not mapped to any component ");
458 break;
459 case 20:
460 p += sprintf(p, "Loss of Lockstep ");
461 break;
462 case 21:
463 p += sprintf(p, "Response not associated with a request ");
464 break;
465 case 22:
466 p += sprintf(p, "Bus parity error - must also set the A, C, or D Bits ");
467 break;
468 case 23:
469 p += sprintf(p, "Detection of a PATH_ERROR ");
470 break;
471 case 25:
472 p += sprintf(p, "Bus operation timeout ");
473 break;
474 case 26:
475 p += sprintf(p, "A read was issued to data that has been poisoned ");
476 break;
477 default:
478 p += sprintf(p, "reserved ");
479 break;
482 if (mem_err->validation_bits & CPER_MEM_VALID_REQUESTOR_ID)
483 p += sprintf(p, "requestorID: 0x%016llx ",
484 (long long)mem_err->requestor_id);
485 if (mem_err->validation_bits & CPER_MEM_VALID_RESPONDER_ID)
486 p += sprintf(p, "responderID: 0x%016llx ",
487 (long long)mem_err->responder_id);
488 if (mem_err->validation_bits & CPER_MEM_VALID_TARGET_ID)
489 p += sprintf(p, "targetID: 0x%016llx ",
490 (long long)mem_err->responder_id);
491 if (p > pvt->other_detail)
492 *(p - 1) = '\0';
494 edac_raw_mc_handle_error(e);
496 unlock:
497 spin_unlock_irqrestore(&ghes_lock, flags);
501 * Known systems that are safe to enable this module.
503 static struct acpi_platform_list plat_list[] = {
504 {"HPE ", "Server ", 0, ACPI_SIG_FADT, all_versions},
505 { } /* End */
508 int ghes_edac_register(struct ghes *ghes, struct device *dev)
510 bool fake = false;
511 struct mem_ctl_info *mci;
512 struct ghes_pvt *pvt;
513 struct edac_mc_layer layers[1];
514 unsigned long flags;
515 int idx = -1;
516 int rc = 0;
518 if (IS_ENABLED(CONFIG_X86)) {
519 /* Check if safe to enable on this system */
520 idx = acpi_match_platform_list(plat_list);
521 if (!force_load && idx < 0)
522 return -ENODEV;
523 } else {
524 force_load = true;
525 idx = 0;
528 /* finish another registration/unregistration instance first */
529 mutex_lock(&ghes_reg_mutex);
532 * We have only one logical memory controller to which all DIMMs belong.
534 if (refcount_inc_not_zero(&ghes_refcount))
535 goto unlock;
537 ghes_scan_system();
539 /* Check if we've got a bogus BIOS */
540 if (!ghes_hw.num_dimms) {
541 fake = true;
542 ghes_hw.num_dimms = 1;
545 layers[0].type = EDAC_MC_LAYER_ALL_MEM;
546 layers[0].size = ghes_hw.num_dimms;
547 layers[0].is_virt_csrow = true;
549 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(struct ghes_pvt));
550 if (!mci) {
551 pr_info("Can't allocate memory for EDAC data\n");
552 rc = -ENOMEM;
553 goto unlock;
556 pvt = mci->pvt_info;
557 pvt->mci = mci;
559 mci->pdev = dev;
560 mci->mtype_cap = MEM_FLAG_EMPTY;
561 mci->edac_ctl_cap = EDAC_FLAG_NONE;
562 mci->edac_cap = EDAC_FLAG_NONE;
563 mci->mod_name = "ghes_edac.c";
564 mci->ctl_name = "ghes_edac";
565 mci->dev_name = "ghes";
567 if (fake) {
568 pr_info("This system has a very crappy BIOS: It doesn't even list the DIMMS.\n");
569 pr_info("Its SMBIOS info is wrong. It is doubtful that the error report would\n");
570 pr_info("work on such system. Use this driver with caution\n");
571 } else if (idx < 0) {
572 pr_info("This EDAC driver relies on BIOS to enumerate memory and get error reports.\n");
573 pr_info("Unfortunately, not all BIOSes reflect the memory layout correctly.\n");
574 pr_info("So, the end result of using this driver varies from vendor to vendor.\n");
575 pr_info("If you find incorrect reports, please contact your hardware vendor\n");
576 pr_info("to correct its BIOS.\n");
577 pr_info("This system has %d DIMM sockets.\n", ghes_hw.num_dimms);
580 if (!fake) {
581 struct dimm_info *src, *dst;
582 int i = 0;
584 mci_for_each_dimm(mci, dst) {
585 src = &ghes_hw.dimms[i];
587 dst->idx = src->idx;
588 dst->smbios_handle = src->smbios_handle;
589 dst->nr_pages = src->nr_pages;
590 dst->mtype = src->mtype;
591 dst->edac_mode = src->edac_mode;
592 dst->dtype = src->dtype;
593 dst->grain = src->grain;
596 * If no src->label, preserve default label assigned
597 * from EDAC core.
599 if (strlen(src->label))
600 memcpy(dst->label, src->label, sizeof(src->label));
602 i++;
605 } else {
606 struct dimm_info *dimm = edac_get_dimm(mci, 0, 0, 0);
608 dimm->nr_pages = 1;
609 dimm->grain = 128;
610 dimm->mtype = MEM_UNKNOWN;
611 dimm->dtype = DEV_UNKNOWN;
612 dimm->edac_mode = EDAC_SECDED;
615 rc = edac_mc_add_mc(mci);
616 if (rc < 0) {
617 pr_info("Can't register with the EDAC core\n");
618 edac_mc_free(mci);
619 rc = -ENODEV;
620 goto unlock;
623 spin_lock_irqsave(&ghes_lock, flags);
624 ghes_pvt = pvt;
625 spin_unlock_irqrestore(&ghes_lock, flags);
627 /* only set on success */
628 refcount_set(&ghes_refcount, 1);
630 unlock:
632 /* Not needed anymore */
633 kfree(ghes_hw.dimms);
634 ghes_hw.dimms = NULL;
636 mutex_unlock(&ghes_reg_mutex);
638 return rc;
641 void ghes_edac_unregister(struct ghes *ghes)
643 struct mem_ctl_info *mci;
644 unsigned long flags;
646 if (!force_load)
647 return;
649 mutex_lock(&ghes_reg_mutex);
651 system_scanned = false;
652 memset(&ghes_hw, 0, sizeof(struct ghes_hw_desc));
654 if (!refcount_dec_and_test(&ghes_refcount))
655 goto unlock;
658 * Wait for the irq handler being finished.
660 spin_lock_irqsave(&ghes_lock, flags);
661 mci = ghes_pvt ? ghes_pvt->mci : NULL;
662 ghes_pvt = NULL;
663 spin_unlock_irqrestore(&ghes_lock, flags);
665 if (!mci)
666 goto unlock;
668 mci = edac_mc_del_mc(mci->pdev);
669 if (mci)
670 edac_mc_free(mci);
672 unlock:
673 mutex_unlock(&ghes_reg_mutex);