xref: /linux/drivers/edac/ghes_edac.c (revision da5b2ad1c2f18834cb1ce429e2e5a5cf5cbdf21b) 
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 #include <linux/notifier.h>
18 
19 #define OTHER_DETAIL_LEN	400
20 
21 struct ghes_pvt {
22 	struct mem_ctl_info *mci;
23 
24 	/* Buffers for the error handling routine */
25 	char other_detail[OTHER_DETAIL_LEN];
26 	char msg[80];
27 };
28 
29 static refcount_t ghes_refcount = REFCOUNT_INIT(0);
30 
31 /*
32  * Access to ghes_pvt must be protected by ghes_lock. The spinlock
33  * also provides the necessary (implicit) memory barrier for the SMP
34  * case to make the pointer visible on another CPU.
35  */
36 static struct ghes_pvt *ghes_pvt;
37 
38 /*
39  * This driver's representation of the system hardware, as collected
40  * from DMI.
41  */
42 static struct ghes_hw_desc {
43 	int num_dimms;
44 	struct dimm_info *dimms;
45 } ghes_hw;
46 
47 /* GHES registration mutex */
48 static DEFINE_MUTEX(ghes_reg_mutex);
49 
50 /*
51  * Sync with other, potentially concurrent callers of
52  * ghes_edac_report_mem_error(). We don't know what the
53  * "inventive" firmware would do.
54  */
55 static DEFINE_SPINLOCK(ghes_lock);
56 
57 static bool system_scanned;
58 
59 static struct list_head *ghes_devs;
60 
61 /* Memory Device - Type 17 of SMBIOS spec */
62 struct memdev_dmi_entry {
63 	u8 type;
64 	u8 length;
65 	u16 handle;
66 	u16 phys_mem_array_handle;
67 	u16 mem_err_info_handle;
68 	u16 total_width;
69 	u16 data_width;
70 	u16 size;
71 	u8 form_factor;
72 	u8 device_set;
73 	u8 device_locator;
74 	u8 bank_locator;
75 	u8 memory_type;
76 	u16 type_detail;
77 	u16 speed;
78 	u8 manufacturer;
79 	u8 serial_number;
80 	u8 asset_tag;
81 	u8 part_number;
82 	u8 attributes;
83 	u32 extended_size;
84 	u16 conf_mem_clk_speed;
85 } __attribute__((__packed__));
86 
87 static struct dimm_info *find_dimm_by_handle(struct mem_ctl_info *mci, u16 handle)
88 {
89 	struct dimm_info *dimm;
90 
91 	mci_for_each_dimm(mci, dimm) {
92 		if (dimm->smbios_handle == handle)
93 			return dimm;
94 	}
95 
96 	return NULL;
97 }
98 
99 static void dimm_setup_label(struct dimm_info *dimm, u16 handle)
100 {
101 	const char *bank = NULL, *device = NULL;
102 
103 	dmi_memdev_name(handle, &bank, &device);
104 
105 	/*
106 	 * Set to a NULL string when both bank and device are zero. In this case,
107 	 * the label assigned by default will be preserved.
108 	 */
109 	snprintf(dimm->label, sizeof(dimm->label), "%s%s%s",
110 		 (bank && *bank) ? bank : "",
111 		 (bank && *bank && device && *device) ? " " : "",
112 		 (device && *device) ? device : "");
113 }
114 
115 static void assign_dmi_dimm_info(struct dimm_info *dimm, struct memdev_dmi_entry *entry)
116 {
117 	u16 rdr_mask = BIT(7) | BIT(13);
118 
119 	if (entry->size == 0xffff) {
120 		pr_info("Can't get DIMM%i size\n", dimm->idx);
121 		dimm->nr_pages = MiB_TO_PAGES(32);/* Unknown */
122 	} else if (entry->size == 0x7fff) {
123 		dimm->nr_pages = MiB_TO_PAGES(entry->extended_size);
124 	} else {
125 		if (entry->size & BIT(15))
126 			dimm->nr_pages = MiB_TO_PAGES((entry->size & 0x7fff) << 10);
127 		else
128 			dimm->nr_pages = MiB_TO_PAGES(entry->size);
129 	}
130 
131 	switch (entry->memory_type) {
132 	case 0x12:
133 		if (entry->type_detail & BIT(13))
134 			dimm->mtype = MEM_RDDR;
135 		else
136 			dimm->mtype = MEM_DDR;
137 		break;
138 	case 0x13:
139 		if (entry->type_detail & BIT(13))
140 			dimm->mtype = MEM_RDDR2;
141 		else
142 			dimm->mtype = MEM_DDR2;
143 		break;
144 	case 0x14:
145 		dimm->mtype = MEM_FB_DDR2;
146 		break;
147 	case 0x18:
148 		if (entry->type_detail & BIT(12))
149 			dimm->mtype = MEM_NVDIMM;
150 		else if (entry->type_detail & BIT(13))
151 			dimm->mtype = MEM_RDDR3;
152 		else
153 			dimm->mtype = MEM_DDR3;
154 		break;
155 	case 0x1a:
156 		if (entry->type_detail & BIT(12))
157 			dimm->mtype = MEM_NVDIMM;
158 		else if (entry->type_detail & BIT(13))
159 			dimm->mtype = MEM_RDDR4;
160 		else
161 			dimm->mtype = MEM_DDR4;
162 		break;
163 	default:
164 		if (entry->type_detail & BIT(6))
165 			dimm->mtype = MEM_RMBS;
166 		else if ((entry->type_detail & rdr_mask) == rdr_mask)
167 			dimm->mtype = MEM_RDR;
168 		else if (entry->type_detail & BIT(7))
169 			dimm->mtype = MEM_SDR;
170 		else if (entry->type_detail & BIT(9))
171 			dimm->mtype = MEM_EDO;
172 		else
173 			dimm->mtype = MEM_UNKNOWN;
174 	}
175 
176 	/*
177 	 * Actually, we can only detect if the memory has bits for
178 	 * checksum or not
179 	 */
180 	if (entry->total_width == entry->data_width)
181 		dimm->edac_mode = EDAC_NONE;
182 	else
183 		dimm->edac_mode = EDAC_SECDED;
184 
185 	dimm->dtype = DEV_UNKNOWN;
186 	dimm->grain = 128;		/* Likely, worse case */
187 
188 	dimm_setup_label(dimm, entry->handle);
189 
190 	if (dimm->nr_pages) {
191 		edac_dbg(1, "DIMM%i: %s size = %d MB%s\n",
192 			dimm->idx, edac_mem_types[dimm->mtype],
193 			PAGES_TO_MiB(dimm->nr_pages),
194 			(dimm->edac_mode != EDAC_NONE) ? "(ECC)" : "");
195 		edac_dbg(2, "\ttype %d, detail 0x%02x, width %d(total %d)\n",
196 			entry->memory_type, entry->type_detail,
197 			entry->total_width, entry->data_width);
198 	}
199 
200 	dimm->smbios_handle = entry->handle;
201 }
202 
203 static void enumerate_dimms(const struct dmi_header *dh, void *arg)
204 {
205 	struct memdev_dmi_entry *entry = (struct memdev_dmi_entry *)dh;
206 	struct ghes_hw_desc *hw = (struct ghes_hw_desc *)arg;
207 	struct dimm_info *d;
208 
209 	if (dh->type != DMI_ENTRY_MEM_DEVICE)
210 		return;
211 
212 	/* Enlarge the array with additional 16 */
213 	if (!hw->num_dimms || !(hw->num_dimms % 16)) {
214 		struct dimm_info *new;
215 
216 		new = krealloc_array(hw->dimms, hw->num_dimms + 16,
217 				     sizeof(struct dimm_info), GFP_KERNEL);
218 		if (!new) {
219 			WARN_ON_ONCE(1);
220 			return;
221 		}
222 
223 		hw->dimms = new;
224 	}
225 
226 	d = &hw->dimms[hw->num_dimms];
227 	d->idx = hw->num_dimms;
228 
229 	assign_dmi_dimm_info(d, entry);
230 
231 	hw->num_dimms++;
232 }
233 
234 static void ghes_scan_system(void)
235 {
236 	if (system_scanned)
237 		return;
238 
239 	dmi_walk(enumerate_dimms, &ghes_hw);
240 
241 	system_scanned = true;
242 }
243 
244 static int print_mem_error_other_detail(const struct cper_sec_mem_err *mem, char *msg,
245 					const char *location, unsigned int len)
246 {
247 	u32 n;
248 
249 	if (!msg)
250 		return 0;
251 
252 	n = 0;
253 	len -= 1;
254 
255 	n += scnprintf(msg + n, len - n, "APEI location: %s ", location);
256 
257 	if (!(mem->validation_bits & CPER_MEM_VALID_ERROR_STATUS))
258 		goto out;
259 
260 	n += scnprintf(msg + n, len - n, "status(0x%016llx): ", mem->error_status);
261 	n += scnprintf(msg + n, len - n, "%s ", cper_mem_err_status_str(mem->error_status));
262 
263 out:
264 	msg[n] = '\0';
265 
266 	return n;
267 }
268 
269 static int ghes_edac_report_mem_error(struct notifier_block *nb,
270 				      unsigned long val, void *data)
271 {
272 	struct cper_sec_mem_err *mem_err = (struct cper_sec_mem_err *)data;
273 	struct cper_mem_err_compact cmem;
274 	struct edac_raw_error_desc *e;
275 	struct mem_ctl_info *mci;
276 	unsigned long sev = val;
277 	struct ghes_pvt *pvt;
278 	unsigned long flags;
279 	char *p;
280 
281 	/*
282 	 * We can do the locking below because GHES defers error processing
283 	 * from NMI to IRQ context. Whenever that changes, we'd at least
284 	 * know.
285 	 */
286 	if (WARN_ON_ONCE(in_nmi()))
287 		return NOTIFY_OK;
288 
289 	spin_lock_irqsave(&ghes_lock, flags);
290 
291 	pvt = ghes_pvt;
292 	if (!pvt)
293 		goto unlock;
294 
295 	mci = pvt->mci;
296 	e = &mci->error_desc;
297 
298 	/* Cleans the error report buffer */
299 	memset(e, 0, sizeof (*e));
300 	e->error_count = 1;
301 	e->grain = 1;
302 	e->msg = pvt->msg;
303 	e->other_detail = pvt->other_detail;
304 	e->top_layer = -1;
305 	e->mid_layer = -1;
306 	e->low_layer = -1;
307 	*pvt->other_detail = '\0';
308 	*pvt->msg = '\0';
309 
310 	switch (sev) {
311 	case GHES_SEV_CORRECTED:
312 		e->type = HW_EVENT_ERR_CORRECTED;
313 		break;
314 	case GHES_SEV_RECOVERABLE:
315 		e->type = HW_EVENT_ERR_UNCORRECTED;
316 		break;
317 	case GHES_SEV_PANIC:
318 		e->type = HW_EVENT_ERR_FATAL;
319 		break;
320 	default:
321 	case GHES_SEV_NO:
322 		e->type = HW_EVENT_ERR_INFO;
323 	}
324 
325 	edac_dbg(1, "error validation_bits: 0x%08llx\n",
326 		 (long long)mem_err->validation_bits);
327 
328 	/* Error type, mapped on e->msg */
329 	if (mem_err->validation_bits & CPER_MEM_VALID_ERROR_TYPE) {
330 		u8 etype = mem_err->error_type;
331 
332 		p = pvt->msg;
333 		p += snprintf(p, sizeof(pvt->msg), "%s", cper_mem_err_type_str(etype));
334 	} else {
335 		strcpy(pvt->msg, "unknown error");
336 	}
337 
338 	/* Error address */
339 	if (mem_err->validation_bits & CPER_MEM_VALID_PA) {
340 		e->page_frame_number = PHYS_PFN(mem_err->physical_addr);
341 		e->offset_in_page = offset_in_page(mem_err->physical_addr);
342 	}
343 
344 	/* Error grain */
345 	if (mem_err->validation_bits & CPER_MEM_VALID_PA_MASK)
346 		e->grain = ~mem_err->physical_addr_mask + 1;
347 
348 	/* Memory error location, mapped on e->location */
349 	p = e->location;
350 	cper_mem_err_pack(mem_err, &cmem);
351 	p += cper_mem_err_location(&cmem, p);
352 
353 	if (mem_err->validation_bits & CPER_MEM_VALID_MODULE_HANDLE) {
354 		struct dimm_info *dimm;
355 
356 		p += cper_dimm_err_location(&cmem, p);
357 		dimm = find_dimm_by_handle(mci, mem_err->mem_dev_handle);
358 		if (dimm) {
359 			e->top_layer = dimm->idx;
360 			strcpy(e->label, dimm->label);
361 		}
362 	}
363 	if (p > e->location)
364 		*(p - 1) = '\0';
365 
366 	if (!*e->label)
367 		strcpy(e->label, "unknown memory");
368 
369 	/* All other fields are mapped on e->other_detail */
370 	p = pvt->other_detail;
371 	p += print_mem_error_other_detail(mem_err, p, e->location, OTHER_DETAIL_LEN);
372 	if (p > pvt->other_detail)
373 		*(p - 1) = '\0';
374 
375 	edac_raw_mc_handle_error(e);
376 
377 unlock:
378 	spin_unlock_irqrestore(&ghes_lock, flags);
379 
380 	return NOTIFY_OK;
381 }
382 
383 static struct notifier_block ghes_edac_mem_err_nb = {
384 	.notifier_call	= ghes_edac_report_mem_error,
385 	.priority	= 0,
386 };
387 
388 static int ghes_edac_register(struct device *dev)
389 {
390 	bool fake = false;
391 	struct mem_ctl_info *mci;
392 	struct ghes_pvt *pvt;
393 	struct edac_mc_layer layers[1];
394 	unsigned long flags;
395 	int rc = 0;
396 
397 	/* finish another registration/unregistration instance first */
398 	mutex_lock(&ghes_reg_mutex);
399 
400 	/*
401 	 * We have only one logical memory controller to which all DIMMs belong.
402 	 */
403 	if (refcount_inc_not_zero(&ghes_refcount))
404 		goto unlock;
405 
406 	ghes_scan_system();
407 
408 	/* Check if we've got a bogus BIOS */
409 	if (!ghes_hw.num_dimms) {
410 		fake = true;
411 		ghes_hw.num_dimms = 1;
412 	}
413 
414 	layers[0].type = EDAC_MC_LAYER_ALL_MEM;
415 	layers[0].size = ghes_hw.num_dimms;
416 	layers[0].is_virt_csrow = true;
417 
418 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, sizeof(struct ghes_pvt));
419 	if (!mci) {
420 		pr_info("Can't allocate memory for EDAC data\n");
421 		rc = -ENOMEM;
422 		goto unlock;
423 	}
424 
425 	pvt		= mci->pvt_info;
426 	pvt->mci	= mci;
427 
428 	mci->pdev = dev;
429 	mci->mtype_cap = MEM_FLAG_EMPTY;
430 	mci->edac_ctl_cap = EDAC_FLAG_NONE;
431 	mci->edac_cap = EDAC_FLAG_NONE;
432 	mci->mod_name = "ghes_edac.c";
433 	mci->ctl_name = "ghes_edac";
434 	mci->dev_name = "ghes";
435 
436 	if (fake) {
437 		pr_info("This system has a very crappy BIOS: It doesn't even list the DIMMS.\n");
438 		pr_info("Its SMBIOS info is wrong. It is doubtful that the error report would\n");
439 		pr_info("work on such system. Use this driver with caution\n");
440 	}
441 
442 	pr_info("This system has %d DIMM sockets.\n", ghes_hw.num_dimms);
443 
444 	if (!fake) {
445 		struct dimm_info *src, *dst;
446 		int i = 0;
447 
448 		mci_for_each_dimm(mci, dst) {
449 			src = &ghes_hw.dimms[i];
450 
451 			dst->idx	   = src->idx;
452 			dst->smbios_handle = src->smbios_handle;
453 			dst->nr_pages	   = src->nr_pages;
454 			dst->mtype	   = src->mtype;
455 			dst->edac_mode	   = src->edac_mode;
456 			dst->dtype	   = src->dtype;
457 			dst->grain	   = src->grain;
458 
459 			/*
460 			 * If no src->label, preserve default label assigned
461 			 * from EDAC core.
462 			 */
463 			if (strlen(src->label))
464 				memcpy(dst->label, src->label, sizeof(src->label));
465 
466 			i++;
467 		}
468 
469 	} else {
470 		struct dimm_info *dimm = edac_get_dimm(mci, 0, 0, 0);
471 
472 		dimm->nr_pages = 1;
473 		dimm->grain = 128;
474 		dimm->mtype = MEM_UNKNOWN;
475 		dimm->dtype = DEV_UNKNOWN;
476 		dimm->edac_mode = EDAC_SECDED;
477 	}
478 
479 	rc = edac_mc_add_mc(mci);
480 	if (rc < 0) {
481 		pr_info("Can't register with the EDAC core\n");
482 		edac_mc_free(mci);
483 		rc = -ENODEV;
484 		goto unlock;
485 	}
486 
487 	spin_lock_irqsave(&ghes_lock, flags);
488 	ghes_pvt = pvt;
489 	spin_unlock_irqrestore(&ghes_lock, flags);
490 
491 	ghes_register_report_chain(&ghes_edac_mem_err_nb);
492 
493 	/* only set on success */
494 	refcount_set(&ghes_refcount, 1);
495 
496 unlock:
497 
498 	/* Not needed anymore */
499 	kfree(ghes_hw.dimms);
500 	ghes_hw.dimms = NULL;
501 
502 	mutex_unlock(&ghes_reg_mutex);
503 
504 	return rc;
505 }
506 
507 static void ghes_edac_unregister(struct ghes *ghes)
508 {
509 	struct mem_ctl_info *mci;
510 	unsigned long flags;
511 
512 	mutex_lock(&ghes_reg_mutex);
513 
514 	system_scanned = false;
515 	memset(&ghes_hw, 0, sizeof(struct ghes_hw_desc));
516 
517 	if (!refcount_dec_and_test(&ghes_refcount))
518 		goto unlock;
519 
520 	/*
521 	 * Wait for the irq handler being finished.
522 	 */
523 	spin_lock_irqsave(&ghes_lock, flags);
524 	mci = ghes_pvt ? ghes_pvt->mci : NULL;
525 	ghes_pvt = NULL;
526 	spin_unlock_irqrestore(&ghes_lock, flags);
527 
528 	if (!mci)
529 		goto unlock;
530 
531 	mci = edac_mc_del_mc(mci->pdev);
532 	if (mci)
533 		edac_mc_free(mci);
534 
535 	ghes_unregister_report_chain(&ghes_edac_mem_err_nb);
536 
537 unlock:
538 	mutex_unlock(&ghes_reg_mutex);
539 }
540 
541 static int __init ghes_edac_init(void)
542 {
543 	struct ghes *g, *g_tmp;
544 
545 	ghes_devs = ghes_get_devices();
546 	if (!ghes_devs)
547 		return -ENODEV;
548 
549 	if (list_empty(ghes_devs)) {
550 		pr_info("GHES probing device list is empty\n");
551 		return -ENODEV;
552 	}
553 
554 	list_for_each_entry_safe(g, g_tmp, ghes_devs, elist) {
555 		ghes_edac_register(g->dev);
556 	}
557 
558 	return 0;
559 }
560 module_init(ghes_edac_init);
561 
562 static void __exit ghes_edac_exit(void)
563 {
564 	struct ghes *g, *g_tmp;
565 
566 	list_for_each_entry_safe(g, g_tmp, ghes_devs, elist) {
567 		ghes_edac_unregister(g);
568 	}
569 }
570 module_exit(ghes_edac_exit);
571 
572 MODULE_LICENSE("GPL");
573 MODULE_DESCRIPTION("Output ACPI APEI/GHES BIOS detected errors via EDAC");
574