forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2023-12-11 072de836f53be56a70cecf70b43ae43b7ce17376 
 kernel/drivers/ras/cec.c
....@@ -1,6 +1,10 @@
11 // SPDX-License-Identifier: GPL-2.0
2
+/*
3
+ * Copyright (c) 2017-2019 Borislav Petkov, SUSE Labs.
4
+ */
25 #include <linux/mm.h>
36 #include <linux/gfp.h>
7
+#include <linux/ras.h>
48 #include <linux/kernel.h>
59 #include <linux/workqueue.h>
610 
....@@ -37,9 +41,9 @@
3741  * thus emulate an an LRU-like behavior when deleting elements to free up space
3842  * in the page.
3943  *
40
- * When an element reaches it's max count of count_threshold, we try to poison
41
- * it by assuming that errors triggered count_threshold times in a single page
42
- * are excessive and that page shouldn't be used anymore. count_threshold is
44
+ * When an element reaches it's max count of action_threshold, we try to poison
45
+ * it by assuming that errors triggered action_threshold times in a single page
46
+ * are excessive and that page shouldn't be used anymore. action_threshold is
4347  * initialized to COUNT_MASK which is the maximum.
4448  *
4549  * That error event entry causes cec_add_elem() to return !0 value and thus
....@@ -122,7 +126,7 @@
122126 static u64 dfs_pfn;
123127 
124128 /* Amount of errors after which we offline */
125
-static unsigned int count_threshold = COUNT_MASK;
129
+static u64 action_threshold = COUNT_MASK;
126130 
127131 /* Each element "decays" each decay_interval which is 24hrs by default. */
128132 #define CEC_DECAY_DEFAULT_INTERVAL	24 * 60 * 60	/* 24 hrs */
....@@ -276,12 +280,49 @@
276280 	return pfn;
277281 }
278282 
283
+static bool sanity_check(struct ce_array *ca)
284
+{
285
+	bool ret = false;
286
+	u64 prev = 0;
287
+	int i;
279288 
280
-int cec_add_elem(u64 pfn)
289
+	for (i = 0; i < ca->n; i++) {
290
+		u64 this = PFN(ca->array[i]);
291
+
292
+		if (WARN(prev > this, "prev: 0x%016llx <-> this: 0x%016llx\n", prev, this))
293
+			ret = true;
294
+
295
+		prev = this;
296
+	}
297
+
298
+	if (!ret)
299
+		return ret;
300
+
301
+	pr_info("Sanity check dump:\n{ n: %d\n", ca->n);
302
+	for (i = 0; i < ca->n; i++) {
303
+		u64 this = PFN(ca->array[i]);
304
+
305
+		pr_info(" %03d: [%016llx|%03llx]\n", i, this, FULL_COUNT(ca->array[i]));
306
+	}
307
+	pr_info("}\n");
308
+
309
+	return ret;
310
+}
311
+
312
+/**
313
+ * cec_add_elem - Add an element to the CEC array.
314
+ * @pfn:	page frame number to insert
315
+ *
316
+ * Return values:
317
+ * - <0:	on error
318
+ * -  0:	on success
319
+ * - >0:	when the inserted pfn was offlined
320
+ */
321
+static int cec_add_elem(u64 pfn)
281322 {
282323 	struct ce_array *ca = &ce_arr;
283
-	unsigned int to;
284
-	int count, ret = 0;
324
+	int count, err, ret = 0;
325
+	unsigned int to = 0;
285326 
286327 	/*
287328 	 * We can be called very early on the identify_cpu() path where we are
....@@ -290,15 +331,16 @@
290331 	if (!ce_arr.array || ce_arr.disabled)
291332 		return -ENODEV;
292333 
293
-	ca->ces_entered++;
294
-
295334 	mutex_lock(&ce_mutex);
296335 
336
+	ca->ces_entered++;
337
+
338
+	/* Array full, free the LRU slot. */
297339 	if (ca->n == MAX_ELEMS)
298340 		WARN_ON(!del_lru_elem_unlocked(ca));
299341 
300
-	ret = find_elem(ca, pfn, &to);
301
-	if (ret < 0) {
342
+	err = find_elem(ca, pfn, &to);
343
+	if (err < 0) {
302344 		/*
303345 		 * Shift range [to-end] to make room for one more element.
304346 		 */
....@@ -306,24 +348,17 @@
306348 			(void *)&ca->array[to],
307349 			(ca->n - to) * sizeof(u64));
308350 
309
-		ca->array[to] = (pfn << PAGE_SHIFT) |
310
-				(DECAY_MASK << COUNT_BITS) | 1;
311
-
351
+		ca->array[to] = pfn << PAGE_SHIFT;
312352 		ca->n++;
313
-
314
-		ret = 0;
315
-
316
-		goto decay;
317353 	}
318354 
355
+	/* Add/refresh element generation and increment count */
356
+	ca->array[to] |= DECAY_MASK << COUNT_BITS;
357
+	ca->array[to]++;
358
+
359
+	/* Check action threshold and soft-offline, if reached. */
319360 	count = COUNT(ca->array[to]);
320
-
321
-	if (count < count_threshold) {
322
-		ca->array[to] |= (DECAY_MASK << COUNT_BITS);
323
-		ca->array[to]++;
324
-
325
-		ret = 0;
326
-	} else {
361
+	if (count >= action_threshold) {
327362 		u64 pfn = ca->array[to] >> PAGE_SHIFT;
328363 
329364 		if (!pfn_valid(pfn)) {
....@@ -338,19 +373,20 @@
338373 		del_elem(ca, to);
339374 
340375 		/*
341
-		 * Return a >0 value to denote that we've reached the offlining
342
-		 * threshold.
376
+		 * Return a >0 value to callers, to denote that we've reached
377
+		 * the offlining threshold.
343378 		 */
344379 		ret = 1;
345380 
346381 		goto unlock;
347382 	}
348383 
349
-decay:
350384 	ca->decay_count++;
351385 
352386 	if (ca->decay_count >= CLEAN_ELEMS)
353387 		do_spring_cleaning(ca);
388
+
389
+	WARN_ON_ONCE(sanity_check(ca));
354390 
355391 unlock:
356392 	mutex_unlock(&ce_mutex);
....@@ -378,38 +414,39 @@
378414 
379415 static int decay_interval_set(void *data, u64 val)
380416 {
381
-	*(u64 *)data = val;
382
-
383417 	if (val < CEC_DECAY_MIN_INTERVAL)
384418 		return -EINVAL;
385419 
386420 	if (val > CEC_DECAY_MAX_INTERVAL)
387421 		return -EINVAL;
388422 
423
+	*(u64 *)data   = val;
389424 	decay_interval = val;
390425 
391426 	cec_mod_work(decay_interval);
427
+
392428 	return 0;
393429 }
394430 DEFINE_DEBUGFS_ATTRIBUTE(decay_interval_ops, u64_get, decay_interval_set, "%lld\n");
395431 
396
-static int count_threshold_set(void *data, u64 val)
432
+static int action_threshold_set(void *data, u64 val)
397433 {
398434 	*(u64 *)data = val;
399435 
400436 	if (val > COUNT_MASK)
401437 		val = COUNT_MASK;
402438 
403
-	count_threshold = val;
439
+	action_threshold = val;
404440 
405441 	return 0;
406442 }
407
-DEFINE_DEBUGFS_ATTRIBUTE(count_threshold_ops, u64_get, count_threshold_set, "%lld\n");
443
+DEFINE_DEBUGFS_ATTRIBUTE(action_threshold_ops, u64_get, action_threshold_set, "%lld\n");
408444 
409
-static int array_dump(struct seq_file *m, void *v)
445
+static const char * const bins[] = { "00", "01", "10", "11" };
446
+
447
+static int array_show(struct seq_file *m, void *v)
410448 {
411449 	struct ce_array *ca = &ce_arr;
412
-	u64 prev = 0;
413450 	int i;
414451 
415452 	mutex_lock(&ce_mutex);
....@@ -418,11 +455,8 @@
418455 	for (i = 0; i < ca->n; i++) {
419456 		u64 this = PFN(ca->array[i]);
420457 
421
-		seq_printf(m, " %03d: [%016llx|%03llx]\n", i, this, FULL_COUNT(ca->array[i]));
422
-
423
-		WARN_ON(prev > this);
424
-
425
-		prev = this;
458
+		seq_printf(m, " %3d: [%016llx|%s|%03llx]\n",
459
+			   i, this, bins[DECAY(ca->array[i])], COUNT(ca->array[i]));
426460 	}
427461 
428462 	seq_printf(m, "}\n");
....@@ -435,25 +469,14 @@
435469 	seq_printf(m, "Decay interval: %lld seconds\n", decay_interval);
436470 	seq_printf(m, "Decays: %lld\n", ca->decays_done);
437471 
438
-	seq_printf(m, "Action threshold: %d\n", count_threshold);
472
+	seq_printf(m, "Action threshold: %lld\n", action_threshold);
439473 
440474 	mutex_unlock(&ce_mutex);
441475 
442476 	return 0;
443477 }
444478 
445
-static int array_open(struct inode *inode, struct file *filp)
446
-{
447
-	return single_open(filp, array_dump, NULL);
448
-}
449
-
450
-static const struct file_operations array_ops = {
451
-	.owner	 = THIS_MODULE,
452
-	.open	 = array_open,
453
-	.read	 = seq_read,
454
-	.llseek	 = seq_lseek,
455
-	.release = single_release,
456
-};
479
+DEFINE_SHOW_ATTRIBUTE(array);
457480 
458481 static int __init create_debugfs_nodes(void)
459482 {
....@@ -465,18 +488,6 @@
465488 		return -1;
466489 	}
467490 
468
-	pfn = debugfs_create_file("pfn", S_IRUSR | S_IWUSR, d, &dfs_pfn, &pfn_ops);
469
-	if (!pfn) {
470
-		pr_warn("Error creating pfn debugfs node!\n");
471
-		goto err;
472
-	}
473
-
474
-	array = debugfs_create_file("array", S_IRUSR, d, NULL, &array_ops);
475
-	if (!array) {
476
-		pr_warn("Error creating array debugfs node!\n");
477
-		goto err;
478
-	}
479
-
480491 	decay = debugfs_create_file("decay_interval", S_IRUSR | S_IWUSR, d,
481492 				    &decay_interval, &decay_interval_ops);
482493 	if (!decay) {
....@@ -484,13 +495,27 @@
484495 		goto err;
485496 	}
486497 
487
-	count = debugfs_create_file("count_threshold", S_IRUSR | S_IWUSR, d,
488
-				    &count_threshold, &count_threshold_ops);
498
+	count = debugfs_create_file("action_threshold", S_IRUSR | S_IWUSR, d,
499
+				    &action_threshold, &action_threshold_ops);
489500 	if (!count) {
490
-		pr_warn("Error creating count_threshold debugfs node!\n");
501
+		pr_warn("Error creating action_threshold debugfs node!\n");
491502 		goto err;
492503 	}
493504 
505
+	if (!IS_ENABLED(CONFIG_RAS_CEC_DEBUG))
506
+		return 0;
507
+
508
+	pfn = debugfs_create_file("pfn", S_IRUSR | S_IWUSR, d, &dfs_pfn, &pfn_ops);
509
+	if (!pfn) {
510
+		pr_warn("Error creating pfn debugfs node!\n");
511
+		goto err;
512
+	}
513
+
514
+	array = debugfs_create_file("array", S_IRUSR, d, NULL, &array_fops);
515
+	if (!array) {
516
+		pr_warn("Error creating array debugfs node!\n");
517
+		goto err;
518
+	}
494519 
495520 	return 0;
496521 
....@@ -500,25 +525,57 @@
500525 	return 1;
501526 }
502527 
503
-void __init cec_init(void)
528
+static int cec_notifier(struct notifier_block *nb, unsigned long val,
529
+			void *data)
530
+{
531
+	struct mce *m = (struct mce *)data;
532
+
533
+	if (!m)
534
+		return NOTIFY_DONE;
535
+
536
+	/* We eat only correctable DRAM errors with usable addresses. */
537
+	if (mce_is_memory_error(m) &&
538
+	    mce_is_correctable(m)  &&
539
+	    mce_usable_address(m)) {
540
+		if (!cec_add_elem(m->addr >> PAGE_SHIFT)) {
541
+			m->kflags |= MCE_HANDLED_CEC;
542
+			return NOTIFY_OK;
543
+		}
544
+	}
545
+
546
+	return NOTIFY_DONE;
547
+}
548
+
549
+static struct notifier_block cec_nb = {
550
+	.notifier_call	= cec_notifier,
551
+	.priority	= MCE_PRIO_CEC,
552
+};
553
+
554
+static int __init cec_init(void)
504555 {
505556 	if (ce_arr.disabled)
506
-		return;
557
+		return -ENODEV;
507558 
508559 	ce_arr.array = (void *)get_zeroed_page(GFP_KERNEL);
509560 	if (!ce_arr.array) {
510561 		pr_err("Error allocating CE array page!\n");
511
-		return;
562
+		return -ENOMEM;
512563 	}
513564 
514
-	if (create_debugfs_nodes())
515
-		return;
565
+	if (create_debugfs_nodes()) {
566
+		free_page((unsigned long)ce_arr.array);
567
+		return -ENOMEM;
568
+	}
516569 
517570 	INIT_DELAYED_WORK(&cec_work, cec_work_fn);
518571 	schedule_delayed_work(&cec_work, CEC_DECAY_DEFAULT_INTERVAL);
519572 
573
+	mce_register_decode_chain(&cec_nb);
574
+
520575 	pr_info("Correctable Errors collector initialized.\n");
576
+	return 0;
521577 }
578
+late_initcall(cec_init);
522579 
523580 int __init parse_cec_param(char *str)
524581 {