GNU Linux-libre 5.10.153-gnu1
[releases.git] / drivers / edac / ghes_edac.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * GHES/EDAC Linux driver
4  *
5  * Copyright (c) 2013 by Mauro Carvalho Chehab
6  *
7  * Red Hat Inc. https://www.redhat.com
8  */
9
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
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>
17
18 struct ghes_pvt {
19         struct mem_ctl_info *mci;
20
21         /* Buffers for the error handling routine */
22         char other_detail[400];
23         char msg[80];
24 };
25
26 static refcount_t ghes_refcount = REFCOUNT_INIT(0);
27
28 /*
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.
32  */
33 static struct ghes_pvt *ghes_pvt;
34
35 /*
36  * This driver's representation of the system hardware, as collected
37  * from DMI.
38  */
39 struct ghes_hw_desc {
40         int num_dimms;
41         struct dimm_info *dimms;
42 } ghes_hw;
43
44 /* GHES registration mutex */
45 static DEFINE_MUTEX(ghes_reg_mutex);
46
47 /*
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.
51  */
52 static DEFINE_SPINLOCK(ghes_lock);
53
54 /* "ghes_edac.force_load=1" skips the platform check */
55 static bool __read_mostly force_load;
56 module_param(force_load, bool, 0);
57
58 static bool system_scanned;
59
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__));
85
86 static struct dimm_info *find_dimm_by_handle(struct mem_ctl_info *mci, u16 handle)
87 {
88         struct dimm_info *dimm;
89
90         mci_for_each_dimm(mci, dimm) {
91                 if (dimm->smbios_handle == handle)
92                         return dimm;
93         }
94
95         return NULL;
96 }
97
98 static void dimm_setup_label(struct dimm_info *dimm, u16 handle)
99 {
100         const char *bank = NULL, *device = NULL;
101
102         dmi_memdev_name(handle, &bank, &device);
103
104         /*
105          * Set to a NULL string when both bank and device are zero. In this case,
106          * the label assigned by default will be preserved.
107          */
108         snprintf(dimm->label, sizeof(dimm->label), "%s%s%s",
109                  (bank && *bank) ? bank : "",
110                  (bank && *bank && device && *device) ? " " : "",
111                  (device && *device) ? device : "");
112 }
113
114 static void assign_dmi_dimm_info(struct dimm_info *dimm, struct memdev_dmi_entry *entry)
115 {
116         u16 rdr_mask = BIT(7) | BIT(13);
117
118         if (entry->size == 0xffff) {
119                 pr_info("Can't get DIMM%i size\n", dimm->idx);
120                 dimm->nr_pages = MiB_TO_PAGES(32);/* Unknown */
121         } else if (entry->size == 0x7fff) {
122                 dimm->nr_pages = MiB_TO_PAGES(entry->extended_size);
123         } else {
124                 if (entry->size & BIT(15))
125                         dimm->nr_pages = MiB_TO_PAGES((entry->size & 0x7fff) << 10);
126                 else
127                         dimm->nr_pages = MiB_TO_PAGES(entry->size);
128         }
129
130         switch (entry->memory_type) {
131         case 0x12:
132                 if (entry->type_detail & BIT(13))
133                         dimm->mtype = MEM_RDDR;
134                 else
135                         dimm->mtype = MEM_DDR;
136                 break;
137         case 0x13:
138                 if (entry->type_detail & BIT(13))
139                         dimm->mtype = MEM_RDDR2;
140                 else
141                         dimm->mtype = MEM_DDR2;
142                 break;
143         case 0x14:
144                 dimm->mtype = MEM_FB_DDR2;
145                 break;
146         case 0x18:
147                 if (entry->type_detail & BIT(12))
148                         dimm->mtype = MEM_NVDIMM;
149                 else if (entry->type_detail & BIT(13))
150                         dimm->mtype = MEM_RDDR3;
151                 else
152                         dimm->mtype = MEM_DDR3;
153                 break;
154         case 0x1a:
155                 if (entry->type_detail & BIT(12))
156                         dimm->mtype = MEM_NVDIMM;
157                 else if (entry->type_detail & BIT(13))
158                         dimm->mtype = MEM_RDDR4;
159                 else
160                         dimm->mtype = MEM_DDR4;
161                 break;
162         default:
163                 if (entry->type_detail & BIT(6))
164                         dimm->mtype = MEM_RMBS;
165                 else if ((entry->type_detail & rdr_mask) == rdr_mask)
166                         dimm->mtype = MEM_RDR;
167                 else if (entry->type_detail & BIT(7))
168                         dimm->mtype = MEM_SDR;
169                 else if (entry->type_detail & BIT(9))
170                         dimm->mtype = MEM_EDO;
171                 else
172                         dimm->mtype = MEM_UNKNOWN;
173         }
174
175         /*
176          * Actually, we can only detect if the memory has bits for
177          * checksum or not
178          */
179         if (entry->total_width == entry->data_width)
180                 dimm->edac_mode = EDAC_NONE;
181         else
182                 dimm->edac_mode = EDAC_SECDED;
183
184         dimm->dtype = DEV_UNKNOWN;
185         dimm->grain = 128;              /* Likely, worse case */
186
187         dimm_setup_label(dimm, entry->handle);
188
189         if (dimm->nr_pages) {
190                 edac_dbg(1, "DIMM%i: %s size = %d MB%s\n",
191                         dimm->idx, edac_mem_types[dimm->mtype],
192                         PAGES_TO_MiB(dimm->nr_pages),
193                         (dimm->edac_mode != EDAC_NONE) ? "(ECC)" : "");
194                 edac_dbg(2, "\ttype %d, detail 0x%02x, width %d(total %d)\n",
195                         entry->memory_type, entry->type_detail,
196                         entry->total_width, entry->data_width);
197         }
198
199         dimm->smbios_handle = entry->handle;
200 }
201
202 static void enumerate_dimms(const struct dmi_header *dh, void *arg)
203 {
204         struct memdev_dmi_entry *entry = (struct memdev_dmi_entry *)dh;
205         struct ghes_hw_desc *hw = (struct ghes_hw_desc *)arg;
206         struct dimm_info *d;
207
208         if (dh->type != DMI_ENTRY_MEM_DEVICE)
209                 return;
210
211         /* Enlarge the array with additional 16 */
212         if (!hw->num_dimms || !(hw->num_dimms % 16)) {
213                 struct dimm_info *new;
214
215                 new = krealloc(hw->dimms, (hw->num_dimms + 16) * sizeof(struct dimm_info),
216                                 GFP_KERNEL);
217                 if (!new) {
218                         WARN_ON_ONCE(1);
219                         return;
220                 }
221
222                 hw->dimms = new;
223         }
224
225         d = &hw->dimms[hw->num_dimms];
226         d->idx = hw->num_dimms;
227
228         assign_dmi_dimm_info(d, entry);
229
230         hw->num_dimms++;
231 }
232
233 static void ghes_scan_system(void)
234 {
235         if (system_scanned)
236                 return;
237
238         dmi_walk(enumerate_dimms, &ghes_hw);
239
240         system_scanned = true;
241 }
242
243 void ghes_edac_report_mem_error(int sev, struct cper_sec_mem_err *mem_err)
244 {
245         struct edac_raw_error_desc *e;
246         struct mem_ctl_info *mci;
247         struct ghes_pvt *pvt;
248         unsigned long flags;
249         char *p;
250
251         /*
252          * We can do the locking below because GHES defers error processing
253          * from NMI to IRQ context. Whenever that changes, we'd at least
254          * know.
255          */
256         if (WARN_ON_ONCE(in_nmi()))
257                 return;
258
259         spin_lock_irqsave(&ghes_lock, flags);
260
261         pvt = ghes_pvt;
262         if (!pvt)
263                 goto unlock;
264
265         mci = pvt->mci;
266         e = &mci->error_desc;
267
268         /* Cleans the error report buffer */
269         memset(e, 0, sizeof (*e));
270         e->error_count = 1;
271         e->grain = 1;
272         e->msg = pvt->msg;
273         e->other_detail = pvt->other_detail;
274         e->top_layer = -1;
275         e->mid_layer = -1;
276         e->low_layer = -1;
277         *pvt->other_detail = '\0';
278         *pvt->msg = '\0';
279
280         switch (sev) {
281         case GHES_SEV_CORRECTED:
282                 e->type = HW_EVENT_ERR_CORRECTED;
283                 break;
284         case GHES_SEV_RECOVERABLE:
285                 e->type = HW_EVENT_ERR_UNCORRECTED;
286                 break;
287         case GHES_SEV_PANIC:
288                 e->type = HW_EVENT_ERR_FATAL;
289                 break;
290         default:
291         case GHES_SEV_NO:
292                 e->type = HW_EVENT_ERR_INFO;
293         }
294
295         edac_dbg(1, "error validation_bits: 0x%08llx\n",
296                  (long long)mem_err->validation_bits);
297
298         /* Error type, mapped on e->msg */
299         if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) {
300                 p = pvt->msg;
301                 switch (mem_err->error_type) {
302                 case 0:
303                         p += sprintf(p, "Unknown");
304                         break;
305                 case 1:
306                         p += sprintf(p, "No error");
307                         break;
308                 case 2:
309                         p += sprintf(p, "Single-bit ECC");
310                         break;
311                 case 3:
312                         p += sprintf(p, "Multi-bit ECC");
313                         break;
314                 case 4:
315                         p += sprintf(p, "Single-symbol ChipKill ECC");
316                         break;
317                 case 5:
318                         p += sprintf(p, "Multi-symbol ChipKill ECC");
319                         break;
320                 case 6:
321                         p += sprintf(p, "Master abort");
322                         break;
323                 case 7:
324                         p += sprintf(p, "Target abort");
325                         break;
326                 case 8:
327                         p += sprintf(p, "Parity Error");
328                         break;
329                 case 9:
330                         p += sprintf(p, "Watchdog timeout");
331                         break;
332                 case 10:
333                         p += sprintf(p, "Invalid address");
334                         break;
335                 case 11:
336                         p += sprintf(p, "Mirror Broken");
337                         break;
338                 case 12:
339                         p += sprintf(p, "Memory Sparing");
340                         break;
341                 case 13:
342                         p += sprintf(p, "Scrub corrected error");
343                         break;
344                 case 14:
345                         p += sprintf(p, "Scrub uncorrected error");
346                         break;
347                 case 15:
348                         p += sprintf(p, "Physical Memory Map-out event");
349                         break;
350                 default:
351                         p += sprintf(p, "reserved error (%d)",
352                                      mem_err->error_type);
353                 }
354         } else {
355                 strcpy(pvt->msg, "unknown error");
356         }
357
358         /* Error address */
359         if (mem_err->validation_bits & CPER_MEM_VALID_PA) {
360                 e->page_frame_number = PHYS_PFN(mem_err->physical_addr);
361                 e->offset_in_page = offset_in_page(mem_err->physical_addr);
362         }
363
364         /* Error grain */
365         if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK)
366                 e->grain = ~mem_err->physical_addr_mask + 1;
367
368         /* Memory error location, mapped on e->location */
369         p = e->location;
370         if (mem_err->validation_bits & CPER_MEM_VALID_NODE)
371                 p += sprintf(p, "node:%d ", mem_err->node);
372         if (mem_err->validation_bits & CPER_MEM_VALID_CARD)
373                 p += sprintf(p, "card:%d ", mem_err->card);
374         if (mem_err->validation_bits & CPER_MEM_VALID_MODULE)
375                 p += sprintf(p, "module:%d ", mem_err->module);
376         if (mem_err->validation_bits & CPER_MEM_VALID_RANK_NUMBER)
377                 p += sprintf(p, "rank:%d ", mem_err->rank);
378         if (mem_err->validation_bits & CPER_MEM_VALID_BANK)
379                 p += sprintf(p, "bank:%d ", mem_err->bank);
380         if (mem_err->validation_bits & CPER_MEM_VALID_BANK_GROUP)
381                 p += sprintf(p, "bank_group:%d ",
382                              mem_err->bank >> CPER_MEM_BANK_GROUP_SHIFT);
383         if (mem_err->validation_bits & CPER_MEM_VALID_BANK_ADDRESS)
384                 p += sprintf(p, "bank_address:%d ",
385                              mem_err->bank & CPER_MEM_BANK_ADDRESS_MASK);
386         if (mem_err->validation_bits & (CPER_MEM_VALID_ROW | CPER_MEM_VALID_ROW_EXT)) {
387                 u32 row = mem_err->row;
388
389                 row |= cper_get_mem_extension(mem_err->validation_bits, mem_err->extended);
390                 p += sprintf(p, "row:%d ", row);
391         }
392         if (mem_err->validation_bits & CPER_MEM_VALID_COLUMN)
393                 p += sprintf(p, "col:%d ", mem_err->column);
394         if (mem_err->validation_bits & CPER_MEM_VALID_BIT_POSITION)
395                 p += sprintf(p, "bit_pos:%d ", mem_err->bit_pos);
396         if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) {
397                 const char *bank = NULL, *device = NULL;
398                 struct dimm_info *dimm;
399
400                 dmi_memdev_name(mem_err->mem_dev_handle, &bank, &device);
401                 if (bank != NULL && device != NULL)
402                         p += sprintf(p, "DIMM location:%s %s ", bank, device);
403                 else
404                         p += sprintf(p, "DIMM DMI handle: 0x%.4x ",
405                                      mem_err->mem_dev_handle);
406
407                 dimm = find_dimm_by_handle(mci, mem_err->mem_dev_handle);
408                 if (dimm) {
409                         e->top_layer = dimm->idx;
410                         strcpy(e->label, dimm->label);
411                 }
412         }
413         if (mem_err->validation_bits & CPER_MEM_VALID_CHIP_ID)
414                 p += sprintf(p, "chipID: %d ",
415                              mem_err->extended >> CPER_MEM_CHIP_ID_SHIFT);
416         if (p > e->location)
417                 *(p - 1) = '\0';
418
419         if (!*e->label)
420                 strcpy(e->label, "unknown memory");
421
422         /* All other fields are mapped on e->other_detail */
423         p = pvt->other_detail;
424         p += snprintf(p, sizeof(pvt->other_detail),
425                 "APEI location: %s ", e->location);
426         if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_STATUS) {
427                 u64 status = mem_err->error_status;
428
429                 p += sprintf(p, "status(0x%016llx): ", (long long)status);
430                 switch ((status >> 8) & 0xff) {
431                 case 1:
432                         p += sprintf(p, "Error detected internal to the component ");
433                         break;
434                 case 16:
435                         p += sprintf(p, "Error detected in the bus ");
436                         break;
437                 case 4:
438                         p += sprintf(p, "Storage error in DRAM memory ");
439                         break;
440                 case 5:
441                         p += sprintf(p, "Storage error in TLB ");
442                         break;
443                 case 6:
444                         p += sprintf(p, "Storage error in cache ");
445                         break;
446                 case 7:
447                         p += sprintf(p, "Error in one or more functional units ");
448                         break;
449                 case 8:
450                         p += sprintf(p, "component failed self test ");
451                         break;
452                 case 9:
453                         p += sprintf(p, "Overflow or undervalue of internal queue ");
454                         break;
455                 case 17:
456                         p += sprintf(p, "Virtual address not found on IO-TLB or IO-PDIR ");
457                         break;
458                 case 18:
459                         p += sprintf(p, "Improper access error ");
460                         break;
461                 case 19:
462                         p += sprintf(p, "Access to a memory address which is not mapped to any component ");
463                         break;
464                 case 20:
465                         p += sprintf(p, "Loss of Lockstep ");
466                         break;
467                 case 21:
468                         p += sprintf(p, "Response not associated with a request ");
469                         break;
470                 case 22:
471                         p += sprintf(p, "Bus parity error - must also set the A, C, or D Bits ");
472                         break;
473                 case 23:
474                         p += sprintf(p, "Detection of a PATH_ERROR ");
475                         break;
476                 case 25:
477                         p += sprintf(p, "Bus operation timeout ");
478                         break;
479                 case 26:
480                         p += sprintf(p, "A read was issued to data that has been poisoned ");
481                         break;
482                 default:
483                         p += sprintf(p, "reserved ");
484                         break;
485                 }
486         }
487         if (mem_err->validation_bits & CPER_MEM_VALID_REQUESTOR_ID)
488                 p += sprintf(p, "requestorID: 0x%016llx ",
489                              (long long)mem_err->requestor_id);
490         if (mem_err->validation_bits & CPER_MEM_VALID_RESPONDER_ID)
491                 p += sprintf(p, "responderID: 0x%016llx ",
492                              (long long)mem_err->responder_id);
493         if (mem_err->validation_bits & CPER_MEM_VALID_TARGET_ID)
494                 p += sprintf(p, "targetID: 0x%016llx ",
495                              (long long)mem_err->responder_id);
496         if (p > pvt->other_detail)
497                 *(p - 1) = '\0';
498
499         edac_raw_mc_handle_error(e);
500
501 unlock:
502         spin_unlock_irqrestore(&ghes_lock, flags);
503 }
504
505 /*
506  * Known systems that are safe to enable this module.
507  */
508 static struct acpi_platform_list plat_list[] = {
509         {"HPE   ", "Server  ", 0, ACPI_SIG_FADT, all_versions},
510         { } /* End */
511 };
512
513 int ghes_edac_register(struct ghes *ghes, struct device *dev)
514 {
515         bool fake = false;
516         struct mem_ctl_info *mci;
517         struct ghes_pvt *pvt;
518         struct edac_mc_layer layers[1];
519         unsigned long flags;
520         int idx = -1;
521         int rc = 0;
522
523         if (IS_ENABLED(CONFIG_X86)) {
524                 /* Check if safe to enable on this system */
525                 idx = acpi_match_platform_list(plat_list);
526                 if (!force_load && idx < 0)
527                         return -ENODEV;
528         } else {
529                 force_load = true;
530                 idx = 0;
531         }
532
533         /* finish another registration/unregistration instance first */
534         mutex_lock(&ghes_reg_mutex);
535
536         /*
537          * We have only one logical memory controller to which all DIMMs belong.
538          */
539         if (refcount_inc_not_zero(&ghes_refcount))
540                 goto unlock;
541
542         ghes_scan_system();
543
544         /* Check if we've got a bogus BIOS */
545         if (!ghes_hw.num_dimms) {
546                 fake = true;
547                 ghes_hw.num_dimms = 1;
548         }
549
550         layers[0].type = EDAC_MC_LAYER_ALL_MEM;
551         layers[0].size = ghes_hw.num_dimms;
552         layers[0].is_virt_csrow = true;
553
554         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(struct ghes_pvt));
555         if (!mci) {
556                 pr_info("Can't allocate memory for EDAC data\n");
557                 rc = -ENOMEM;
558                 goto unlock;
559         }
560
561         pvt             = mci->pvt_info;
562         pvt->mci        = mci;
563
564         mci->pdev = dev;
565         mci->mtype_cap = MEM_FLAG_EMPTY;
566         mci->edac_ctl_cap = EDAC_FLAG_NONE;
567         mci->edac_cap = EDAC_FLAG_NONE;
568         mci->mod_name = "ghes_edac.c";
569         mci->ctl_name = "ghes_edac";
570         mci->dev_name = "ghes";
571
572         if (fake) {
573                 pr_info("This system has a very crappy BIOS: It doesn't even list the DIMMS.\n");
574                 pr_info("Its SMBIOS info is wrong. It is doubtful that the error report would\n");
575                 pr_info("work on such system. Use this driver with caution\n");
576         } else if (idx < 0) {
577                 pr_info("This EDAC driver relies on BIOS to enumerate memory and get error reports.\n");
578                 pr_info("Unfortunately, not all BIOSes reflect the memory layout correctly.\n");
579                 pr_info("So, the end result of using this driver varies from vendor to vendor.\n");
580                 pr_info("If you find incorrect reports, please contact your hardware vendor\n");
581                 pr_info("to correct its BIOS.\n");
582                 pr_info("This system has %d DIMM sockets.\n", ghes_hw.num_dimms);
583         }
584
585         if (!fake) {
586                 struct dimm_info *src, *dst;
587                 int i = 0;
588
589                 mci_for_each_dimm(mci, dst) {
590                         src = &ghes_hw.dimms[i];
591
592                         dst->idx           = src->idx;
593                         dst->smbios_handle = src->smbios_handle;
594                         dst->nr_pages      = src->nr_pages;
595                         dst->mtype         = src->mtype;
596                         dst->edac_mode     = src->edac_mode;
597                         dst->dtype         = src->dtype;
598                         dst->grain         = src->grain;
599
600                         /*
601                          * If no src->label, preserve default label assigned
602                          * from EDAC core.
603                          */
604                         if (strlen(src->label))
605                                 memcpy(dst->label, src->label, sizeof(src->label));
606
607                         i++;
608                 }
609
610         } else {
611                 struct dimm_info *dimm = edac_get_dimm(mci, 0, 0, 0);
612
613                 dimm->nr_pages = 1;
614                 dimm->grain = 128;
615                 dimm->mtype = MEM_UNKNOWN;
616                 dimm->dtype = DEV_UNKNOWN;
617                 dimm->edac_mode = EDAC_SECDED;
618         }
619
620         rc = edac_mc_add_mc(mci);
621         if (rc < 0) {
622                 pr_info("Can't register with the EDAC core\n");
623                 edac_mc_free(mci);
624                 rc = -ENODEV;
625                 goto unlock;
626         }
627
628         spin_lock_irqsave(&ghes_lock, flags);
629         ghes_pvt = pvt;
630         spin_unlock_irqrestore(&ghes_lock, flags);
631
632         /* only set on success */
633         refcount_set(&ghes_refcount, 1);
634
635 unlock:
636
637         /* Not needed anymore */
638         kfree(ghes_hw.dimms);
639         ghes_hw.dimms = NULL;
640
641         mutex_unlock(&ghes_reg_mutex);
642
643         return rc;
644 }
645
646 void ghes_edac_unregister(struct ghes *ghes)
647 {
648         struct mem_ctl_info *mci;
649         unsigned long flags;
650
651         if (!force_load)
652                 return;
653
654         mutex_lock(&ghes_reg_mutex);
655
656         system_scanned = false;
657         memset(&ghes_hw, 0, sizeof(struct ghes_hw_desc));
658
659         if (!refcount_dec_and_test(&ghes_refcount))
660                 goto unlock;
661
662         /*
663          * Wait for the irq handler being finished.
664          */
665         spin_lock_irqsave(&ghes_lock, flags);
666         mci = ghes_pvt ? ghes_pvt->mci : NULL;
667         ghes_pvt = NULL;
668         spin_unlock_irqrestore(&ghes_lock, flags);
669
670         if (!mci)
671                 goto unlock;
672
673         mci = edac_mc_del_mc(mci->pdev);
674         if (mci)
675                 edac_mc_free(mci);
676
677 unlock:
678         mutex_unlock(&ghes_reg_mutex);
679 }