xref: /linux/arch/powerpc/kernel/rtasd.c (revision 9a87ffc99ec8eb8d35eed7c4f816d75f5cc9662e)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
4  *
5  * Communication to userspace based on kernel/printk.c
6  */
7 
8 #include <linux/types.h>
9 #include <linux/errno.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/of.h>
13 #include <linux/poll.h>
14 #include <linux/proc_fs.h>
15 #include <linux/init.h>
16 #include <linux/vmalloc.h>
17 #include <linux/spinlock.h>
18 #include <linux/cpu.h>
19 #include <linux/workqueue.h>
20 #include <linux/slab.h>
21 #include <linux/topology.h>
22 
23 #include <linux/uaccess.h>
24 #include <asm/io.h>
25 #include <asm/rtas.h>
26 #include <asm/nvram.h>
27 #include <linux/atomic.h>
28 #include <asm/machdep.h>
29 #include <asm/topology.h>
30 
31 
32 static DEFINE_SPINLOCK(rtasd_log_lock);
33 
34 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
35 
36 static char *rtas_log_buf;
37 static unsigned long rtas_log_start;
38 static unsigned long rtas_log_size;
39 
40 static int surveillance_timeout = -1;
41 
42 static unsigned int rtas_error_log_max;
43 static unsigned int rtas_error_log_buffer_max;
44 
45 /* RTAS service tokens */
46 static unsigned int event_scan;
47 static unsigned int rtas_event_scan_rate;
48 
49 static bool full_rtas_msgs;
50 
51 /* Stop logging to nvram after first fatal error */
52 static int logging_enabled; /* Until we initialize everything,
53                              * make sure we don't try logging
54                              * anything */
55 static int error_log_cnt;
56 
57 /*
58  * Since we use 32 bit RTAS, the physical address of this must be below
59  * 4G or else bad things happen. Allocate this in the kernel data and
60  * make it big enough.
61  */
62 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
63 
64 static char *rtas_type[] = {
65 	"Unknown", "Retry", "TCE Error", "Internal Device Failure",
66 	"Timeout", "Data Parity", "Address Parity", "Cache Parity",
67 	"Address Invalid", "ECC Uncorrected", "ECC Corrupted",
68 };
69 
rtas_event_type(int type)70 static char *rtas_event_type(int type)
71 {
72 	if ((type > 0) && (type < 11))
73 		return rtas_type[type];
74 
75 	switch (type) {
76 		case RTAS_TYPE_EPOW:
77 			return "EPOW";
78 		case RTAS_TYPE_PLATFORM:
79 			return "Platform Error";
80 		case RTAS_TYPE_IO:
81 			return "I/O Event";
82 		case RTAS_TYPE_INFO:
83 			return "Platform Information Event";
84 		case RTAS_TYPE_DEALLOC:
85 			return "Resource Deallocation Event";
86 		case RTAS_TYPE_DUMP:
87 			return "Dump Notification Event";
88 		case RTAS_TYPE_PRRN:
89 			return "Platform Resource Reassignment Event";
90 		case RTAS_TYPE_HOTPLUG:
91 			return "Hotplug Event";
92 	}
93 
94 	return rtas_type[0];
95 }
96 
97 /* To see this info, grep RTAS /var/log/messages and each entry
98  * will be collected together with obvious begin/end.
99  * There will be a unique identifier on the begin and end lines.
100  * This will persist across reboots.
101  *
102  * format of error logs returned from RTAS:
103  * bytes	(size)	: contents
104  * --------------------------------------------------------
105  * 0-7		(8)	: rtas_error_log
106  * 8-47		(40)	: extended info
107  * 48-51	(4)	: vendor id
108  * 52-1023 (vendor specific) : location code and debug data
109  */
printk_log_rtas(char * buf,int len)110 static void printk_log_rtas(char *buf, int len)
111 {
112 
113 	int i,j,n = 0;
114 	int perline = 16;
115 	char buffer[64];
116 	char * str = "RTAS event";
117 
118 	if (full_rtas_msgs) {
119 		printk(RTAS_DEBUG "%d -------- %s begin --------\n",
120 		       error_log_cnt, str);
121 
122 		/*
123 		 * Print perline bytes on each line, each line will start
124 		 * with RTAS and a changing number, so syslogd will
125 		 * print lines that are otherwise the same.  Separate every
126 		 * 4 bytes with a space.
127 		 */
128 		for (i = 0; i < len; i++) {
129 			j = i % perline;
130 			if (j == 0) {
131 				memset(buffer, 0, sizeof(buffer));
132 				n = sprintf(buffer, "RTAS %d:", i/perline);
133 			}
134 
135 			if ((i % 4) == 0)
136 				n += sprintf(buffer+n, " ");
137 
138 			n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
139 
140 			if (j == (perline-1))
141 				printk(KERN_DEBUG "%s\n", buffer);
142 		}
143 		if ((i % perline) != 0)
144 			printk(KERN_DEBUG "%s\n", buffer);
145 
146 		printk(RTAS_DEBUG "%d -------- %s end ----------\n",
147 		       error_log_cnt, str);
148 	} else {
149 		struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
150 
151 		printk(RTAS_DEBUG "event: %d, Type: %s (%d), Severity: %d\n",
152 		       error_log_cnt,
153 		       rtas_event_type(rtas_error_type(errlog)),
154 		       rtas_error_type(errlog),
155 		       rtas_error_severity(errlog));
156 	}
157 }
158 
log_rtas_len(char * buf)159 static int log_rtas_len(char * buf)
160 {
161 	int len;
162 	struct rtas_error_log *err;
163 	uint32_t extended_log_length;
164 
165 	/* rtas fixed header */
166 	len = 8;
167 	err = (struct rtas_error_log *)buf;
168 	extended_log_length = rtas_error_extended_log_length(err);
169 	if (rtas_error_extended(err) && extended_log_length) {
170 
171 		/* extended header */
172 		len += extended_log_length;
173 	}
174 
175 	if (rtas_error_log_max == 0)
176 		rtas_error_log_max = rtas_get_error_log_max();
177 
178 	if (len > rtas_error_log_max)
179 		len = rtas_error_log_max;
180 
181 	return len;
182 }
183 
184 /*
185  * First write to nvram, if fatal error, that is the only
186  * place we log the info.  The error will be picked up
187  * on the next reboot by rtasd.  If not fatal, run the
188  * method for the type of error.  Currently, only RTAS
189  * errors have methods implemented, but in the future
190  * there might be a need to store data in nvram before a
191  * call to panic().
192  *
193  * XXX We write to nvram periodically, to indicate error has
194  * been written and sync'd, but there is a possibility
195  * that if we don't shutdown correctly, a duplicate error
196  * record will be created on next reboot.
197  */
pSeries_log_error(char * buf,unsigned int err_type,int fatal)198 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
199 {
200 	unsigned long offset;
201 	unsigned long s;
202 	int len = 0;
203 
204 	pr_debug("rtasd: logging event\n");
205 	if (buf == NULL)
206 		return;
207 
208 	spin_lock_irqsave(&rtasd_log_lock, s);
209 
210 	/* get length and increase count */
211 	switch (err_type & ERR_TYPE_MASK) {
212 	case ERR_TYPE_RTAS_LOG:
213 		len = log_rtas_len(buf);
214 		if (!(err_type & ERR_FLAG_BOOT))
215 			error_log_cnt++;
216 		break;
217 	case ERR_TYPE_KERNEL_PANIC:
218 	default:
219 		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
220 		spin_unlock_irqrestore(&rtasd_log_lock, s);
221 		return;
222 	}
223 
224 #ifdef CONFIG_PPC64
225 	/* Write error to NVRAM */
226 	if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
227 		nvram_write_error_log(buf, len, err_type, error_log_cnt);
228 #endif /* CONFIG_PPC64 */
229 
230 	/*
231 	 * rtas errors can occur during boot, and we do want to capture
232 	 * those somewhere, even if nvram isn't ready (why not?), and even
233 	 * if rtasd isn't ready. Put them into the boot log, at least.
234 	 */
235 	if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
236 		printk_log_rtas(buf, len);
237 
238 	/* Check to see if we need to or have stopped logging */
239 	if (fatal || !logging_enabled) {
240 		logging_enabled = 0;
241 		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
242 		spin_unlock_irqrestore(&rtasd_log_lock, s);
243 		return;
244 	}
245 
246 	/* call type specific method for error */
247 	switch (err_type & ERR_TYPE_MASK) {
248 	case ERR_TYPE_RTAS_LOG:
249 		offset = rtas_error_log_buffer_max *
250 			((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
251 
252 		/* First copy over sequence number */
253 		memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
254 
255 		/* Second copy over error log data */
256 		offset += sizeof(int);
257 		memcpy(&rtas_log_buf[offset], buf, len);
258 
259 		if (rtas_log_size < LOG_NUMBER)
260 			rtas_log_size += 1;
261 		else
262 			rtas_log_start += 1;
263 
264 		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
265 		spin_unlock_irqrestore(&rtasd_log_lock, s);
266 		wake_up_interruptible(&rtas_log_wait);
267 		break;
268 	case ERR_TYPE_KERNEL_PANIC:
269 	default:
270 		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
271 		spin_unlock_irqrestore(&rtasd_log_lock, s);
272 		return;
273 	}
274 }
275 
handle_rtas_event(const struct rtas_error_log * log)276 static void handle_rtas_event(const struct rtas_error_log *log)
277 {
278 	if (!machine_is(pseries))
279 		return;
280 
281 	if (rtas_error_type(log) == RTAS_TYPE_PRRN)
282 		pr_info_ratelimited("Platform resource reassignment ignored.\n");
283 }
284 
rtas_log_open(struct inode * inode,struct file * file)285 static int rtas_log_open(struct inode * inode, struct file * file)
286 {
287 	return 0;
288 }
289 
rtas_log_release(struct inode * inode,struct file * file)290 static int rtas_log_release(struct inode * inode, struct file * file)
291 {
292 	return 0;
293 }
294 
295 /* This will check if all events are logged, if they are then, we
296  * know that we can safely clear the events in NVRAM.
297  * Next we'll sit and wait for something else to log.
298  */
rtas_log_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)299 static ssize_t rtas_log_read(struct file * file, char __user * buf,
300 			 size_t count, loff_t *ppos)
301 {
302 	int error;
303 	char *tmp;
304 	unsigned long s;
305 	unsigned long offset;
306 
307 	if (!buf || count < rtas_error_log_buffer_max)
308 		return -EINVAL;
309 
310 	count = rtas_error_log_buffer_max;
311 
312 	if (!access_ok(buf, count))
313 		return -EFAULT;
314 
315 	tmp = kmalloc(count, GFP_KERNEL);
316 	if (!tmp)
317 		return -ENOMEM;
318 
319 	spin_lock_irqsave(&rtasd_log_lock, s);
320 
321 	/* if it's 0, then we know we got the last one (the one in NVRAM) */
322 	while (rtas_log_size == 0) {
323 		if (file->f_flags & O_NONBLOCK) {
324 			spin_unlock_irqrestore(&rtasd_log_lock, s);
325 			error = -EAGAIN;
326 			goto out;
327 		}
328 
329 		if (!logging_enabled) {
330 			spin_unlock_irqrestore(&rtasd_log_lock, s);
331 			error = -ENODATA;
332 			goto out;
333 		}
334 #ifdef CONFIG_PPC64
335 		nvram_clear_error_log();
336 #endif /* CONFIG_PPC64 */
337 
338 		spin_unlock_irqrestore(&rtasd_log_lock, s);
339 		error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
340 		if (error)
341 			goto out;
342 		spin_lock_irqsave(&rtasd_log_lock, s);
343 	}
344 
345 	offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
346 	memcpy(tmp, &rtas_log_buf[offset], count);
347 
348 	rtas_log_start += 1;
349 	rtas_log_size -= 1;
350 	spin_unlock_irqrestore(&rtasd_log_lock, s);
351 
352 	error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
353 out:
354 	kfree(tmp);
355 	return error;
356 }
357 
rtas_log_poll(struct file * file,poll_table * wait)358 static __poll_t rtas_log_poll(struct file *file, poll_table * wait)
359 {
360 	poll_wait(file, &rtas_log_wait, wait);
361 	if (rtas_log_size)
362 		return EPOLLIN | EPOLLRDNORM;
363 	return 0;
364 }
365 
366 static const struct proc_ops rtas_log_proc_ops = {
367 	.proc_read	= rtas_log_read,
368 	.proc_poll	= rtas_log_poll,
369 	.proc_open	= rtas_log_open,
370 	.proc_release	= rtas_log_release,
371 	.proc_lseek	= noop_llseek,
372 };
373 
enable_surveillance(int timeout)374 static int enable_surveillance(int timeout)
375 {
376 	int error;
377 
378 	error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
379 
380 	if (error == 0)
381 		return 0;
382 
383 	if (error == -EINVAL) {
384 		printk(KERN_DEBUG "rtasd: surveillance not supported\n");
385 		return 0;
386 	}
387 
388 	printk(KERN_ERR "rtasd: could not update surveillance\n");
389 	return -1;
390 }
391 
do_event_scan(void)392 static void do_event_scan(void)
393 {
394 	int error;
395 	do {
396 		memset(logdata, 0, rtas_error_log_max);
397 		error = rtas_call(event_scan, 4, 1, NULL,
398 				  RTAS_EVENT_SCAN_ALL_EVENTS, 0,
399 				  __pa(logdata), rtas_error_log_max);
400 		if (error == -1) {
401 			printk(KERN_ERR "event-scan failed\n");
402 			break;
403 		}
404 
405 		if (error == 0) {
406 			if (rtas_error_type((struct rtas_error_log *)logdata) !=
407 			    RTAS_TYPE_PRRN)
408 				pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
409 						  0);
410 			handle_rtas_event((struct rtas_error_log *)logdata);
411 		}
412 
413 	} while(error == 0);
414 }
415 
416 static void rtas_event_scan(struct work_struct *w);
417 static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
418 
419 /*
420  * Delay should be at least one second since some machines have problems if
421  * we call event-scan too quickly.
422  */
423 static unsigned long event_scan_delay = 1*HZ;
424 static int first_pass = 1;
425 
rtas_event_scan(struct work_struct * w)426 static void rtas_event_scan(struct work_struct *w)
427 {
428 	unsigned int cpu;
429 
430 	do_event_scan();
431 
432 	cpus_read_lock();
433 
434 	/* raw_ OK because just using CPU as starting point. */
435 	cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
436         if (cpu >= nr_cpu_ids) {
437 		cpu = cpumask_first(cpu_online_mask);
438 
439 		if (first_pass) {
440 			first_pass = 0;
441 			event_scan_delay = 30*HZ/rtas_event_scan_rate;
442 
443 			if (surveillance_timeout != -1) {
444 				pr_debug("rtasd: enabling surveillance\n");
445 				enable_surveillance(surveillance_timeout);
446 				pr_debug("rtasd: surveillance enabled\n");
447 			}
448 		}
449 	}
450 
451 	schedule_delayed_work_on(cpu, &event_scan_work,
452 		__round_jiffies_relative(event_scan_delay, cpu));
453 
454 	cpus_read_unlock();
455 }
456 
457 #ifdef CONFIG_PPC64
retrieve_nvram_error_log(void)458 static void __init retrieve_nvram_error_log(void)
459 {
460 	unsigned int err_type ;
461 	int rc ;
462 
463 	/* See if we have any error stored in NVRAM */
464 	memset(logdata, 0, rtas_error_log_max);
465 	rc = nvram_read_error_log(logdata, rtas_error_log_max,
466 	                          &err_type, &error_log_cnt);
467 	/* We can use rtas_log_buf now */
468 	logging_enabled = 1;
469 	if (!rc) {
470 		if (err_type != ERR_FLAG_ALREADY_LOGGED) {
471 			pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
472 		}
473 	}
474 }
475 #else /* CONFIG_PPC64 */
retrieve_nvram_error_log(void)476 static void __init retrieve_nvram_error_log(void)
477 {
478 }
479 #endif /* CONFIG_PPC64 */
480 
start_event_scan(void)481 static void __init start_event_scan(void)
482 {
483 	printk(KERN_DEBUG "RTAS daemon started\n");
484 	pr_debug("rtasd: will sleep for %d milliseconds\n",
485 		 (30000 / rtas_event_scan_rate));
486 
487 	/* Retrieve errors from nvram if any */
488 	retrieve_nvram_error_log();
489 
490 	schedule_delayed_work_on(cpumask_first(cpu_online_mask),
491 				 &event_scan_work, event_scan_delay);
492 }
493 
494 /* Cancel the rtas event scan work */
rtas_cancel_event_scan(void)495 void rtas_cancel_event_scan(void)
496 {
497 	cancel_delayed_work_sync(&event_scan_work);
498 }
499 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
500 
rtas_event_scan_init(void)501 static int __init rtas_event_scan_init(void)
502 {
503 	int err;
504 
505 	if (!machine_is(pseries) && !machine_is(chrp))
506 		return 0;
507 
508 	/* No RTAS */
509 	event_scan = rtas_function_token(RTAS_FN_EVENT_SCAN);
510 	if (event_scan == RTAS_UNKNOWN_SERVICE) {
511 		printk(KERN_INFO "rtasd: No event-scan on system\n");
512 		return -ENODEV;
513 	}
514 
515 	err = of_property_read_u32(rtas.dev, "rtas-event-scan-rate", &rtas_event_scan_rate);
516 	if (err) {
517 		printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
518 		return -ENODEV;
519 	}
520 
521 	if (!rtas_event_scan_rate) {
522 		/* Broken firmware: take a rate of zero to mean don't scan */
523 		printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
524 		return 0;
525 	}
526 
527 	/* Make room for the sequence number */
528 	rtas_error_log_max = rtas_get_error_log_max();
529 	rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
530 
531 	rtas_log_buf = vmalloc(array_size(LOG_NUMBER,
532 					  rtas_error_log_buffer_max));
533 	if (!rtas_log_buf) {
534 		printk(KERN_ERR "rtasd: no memory\n");
535 		return -ENOMEM;
536 	}
537 
538 	start_event_scan();
539 
540 	return 0;
541 }
542 arch_initcall(rtas_event_scan_init);
543 
rtas_init(void)544 static int __init rtas_init(void)
545 {
546 	struct proc_dir_entry *entry;
547 
548 	if (!machine_is(pseries) && !machine_is(chrp))
549 		return 0;
550 
551 	if (!rtas_log_buf)
552 		return -ENODEV;
553 
554 	entry = proc_create("powerpc/rtas/error_log", 0400, NULL,
555 			    &rtas_log_proc_ops);
556 	if (!entry)
557 		printk(KERN_ERR "Failed to create error_log proc entry\n");
558 
559 	return 0;
560 }
561 __initcall(rtas_init);
562 
surveillance_setup(char * str)563 static int __init surveillance_setup(char *str)
564 {
565 	int i;
566 
567 	/* We only do surveillance on pseries */
568 	if (!machine_is(pseries))
569 		return 0;
570 
571 	if (get_option(&str,&i)) {
572 		if (i >= 0 && i <= 255)
573 			surveillance_timeout = i;
574 	}
575 
576 	return 1;
577 }
578 __setup("surveillance=", surveillance_setup);
579 
rtasmsgs_setup(char * str)580 static int __init rtasmsgs_setup(char *str)
581 {
582 	return (kstrtobool(str, &full_rtas_msgs) == 0);
583 }
584 __setup("rtasmsgs=", rtasmsgs_setup);
585