rtasd.c

  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
 32static DEFINE_SPINLOCK(rtasd_log_lock);
 33
 34static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
 35
 36static char *rtas_log_buf;
 37static unsigned long rtas_log_start;
 38static unsigned long rtas_log_size;
 39
 40static int surveillance_timeout = -1;
 41
 42static unsigned int rtas_error_log_max;
 43static unsigned int rtas_error_log_buffer_max;
 44
 45/* RTAS service tokens */
 46static unsigned int event_scan;
 47static unsigned int rtas_event_scan_rate;
 48
 49static bool full_rtas_msgs;
 50
 51/* Stop logging to nvram after first fatal error */
 52static int logging_enabled; /* Until we initialize everything,
 53                             * make sure we don't try logging
 54                             * anything */
 55static 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 */
 62static unsigned char logdata[RTAS_ERROR_LOG_MAX];
 63
 64static 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
 70static 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		case RTAS_TYPE_HVPIPE:
 93			return "Hypervisor Pipe Notification event";
 94	}
 95
 96	return rtas_type[0];
 97}
 98
 99/* To see this info, grep RTAS /var/log/messages and each entry
100 * will be collected together with obvious begin/end.
101 * There will be a unique identifier on the begin and end lines.
102 * This will persist across reboots.
103 *
104 * format of error logs returned from RTAS:
105 * bytes	(size)	: contents
106 * --------------------------------------------------------
107 * 0-7		(8)	: rtas_error_log
108 * 8-47		(40)	: extended info
109 * 48-51	(4)	: vendor id
110 * 52-1023 (vendor specific) : location code and debug data
111 */
112static void printk_log_rtas(char *buf, int len)
113{
114
115	int i,j,n = 0;
116	int perline = 16;
117	char buffer[64];
118	char * str = "RTAS event";
119
120	if (full_rtas_msgs) {
121		printk(RTAS_DEBUG "%d -------- %s begin --------\n",
122		       error_log_cnt, str);
123
124		/*
125		 * Print perline bytes on each line, each line will start
126		 * with RTAS and a changing number, so syslogd will
127		 * print lines that are otherwise the same.  Separate every
128		 * 4 bytes with a space.
129		 */
130		for (i = 0; i < len; i++) {
131			j = i % perline;
132			if (j == 0) {
133				memset(buffer, 0, sizeof(buffer));
134				n = sprintf(buffer, "RTAS %d:", i/perline);
135			}
136
137			if ((i % 4) == 0)
138				n += sprintf(buffer+n, " ");
139
140			n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
141
142			if (j == (perline-1))
143				printk(KERN_DEBUG "%s\n", buffer);
144		}
145		if ((i % perline) != 0)
146			printk(KERN_DEBUG "%s\n", buffer);
147
148		printk(RTAS_DEBUG "%d -------- %s end ----------\n",
149		       error_log_cnt, str);
150	} else {
151		struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
152
153		printk(RTAS_DEBUG "event: %d, Type: %s (%d), Severity: %d\n",
154		       error_log_cnt,
155		       rtas_event_type(rtas_error_type(errlog)),
156		       rtas_error_type(errlog),
157		       rtas_error_severity(errlog));
158	}
159}
160
161static int log_rtas_len(char * buf)
162{
163	int len;
164	struct rtas_error_log *err;
165	uint32_t extended_log_length;
166
167	/* rtas fixed header */
168	len = 8;
169	err = (struct rtas_error_log *)buf;
170	extended_log_length = rtas_error_extended_log_length(err);
171	if (rtas_error_extended(err) && extended_log_length) {
172
173		/* extended header */
174		len += extended_log_length;
175	}
176
177	if (rtas_error_log_max == 0)
178		rtas_error_log_max = rtas_get_error_log_max();
179
180	if (len > rtas_error_log_max)
181		len = rtas_error_log_max;
182
183	return len;
184}
185
186/*
187 * First write to nvram, if fatal error, that is the only
188 * place we log the info.  The error will be picked up
189 * on the next reboot by rtasd.  If not fatal, run the
190 * method for the type of error.  Currently, only RTAS
191 * errors have methods implemented, but in the future
192 * there might be a need to store data in nvram before a
193 * call to panic().
194 *
195 * XXX We write to nvram periodically, to indicate error has
196 * been written and sync'd, but there is a possibility
197 * that if we don't shutdown correctly, a duplicate error
198 * record will be created on next reboot.
199 */
200void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
201{
202	unsigned long offset;
203	unsigned long s;
204	int len = 0;
205
206	pr_debug("rtasd: logging event\n");
207	if (buf == NULL)
208		return;
209
210	spin_lock_irqsave(&rtasd_log_lock, s);
211
212	/* get length and increase count */
213	switch (err_type & ERR_TYPE_MASK) {
214	case ERR_TYPE_RTAS_LOG:
215		len = log_rtas_len(buf);
216		if (!(err_type & ERR_FLAG_BOOT))
217			error_log_cnt++;
218		break;
219	case ERR_TYPE_KERNEL_PANIC:
220	default:
221		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
222		spin_unlock_irqrestore(&rtasd_log_lock, s);
223		return;
224	}
225
226#ifdef CONFIG_PPC64
227	/* Write error to NVRAM */
228	if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
229		nvram_write_error_log(buf, len, err_type, error_log_cnt);
230#endif /* CONFIG_PPC64 */
231
232	/*
233	 * rtas errors can occur during boot, and we do want to capture
234	 * those somewhere, even if nvram isn't ready (why not?), and even
235	 * if rtasd isn't ready. Put them into the boot log, at least.
236	 */
237	if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
238		printk_log_rtas(buf, len);
239
240	/* Check to see if we need to or have stopped logging */
241	if (fatal || !logging_enabled) {
242		logging_enabled = 0;
243		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
244		spin_unlock_irqrestore(&rtasd_log_lock, s);
245		return;
246	}
247
248	/* call type specific method for error */
249	switch (err_type & ERR_TYPE_MASK) {
250	case ERR_TYPE_RTAS_LOG:
251		offset = rtas_error_log_buffer_max *
252			((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
253
254		/* First copy over sequence number */
255		memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
256
257		/* Second copy over error log data */
258		offset += sizeof(int);
259		memcpy(&rtas_log_buf[offset], buf, len);
260
261		if (rtas_log_size < LOG_NUMBER)
262			rtas_log_size += 1;
263		else
264			rtas_log_start += 1;
265
266		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
267		spin_unlock_irqrestore(&rtasd_log_lock, s);
268		wake_up_interruptible(&rtas_log_wait);
269		break;
270	case ERR_TYPE_KERNEL_PANIC:
271	default:
272		WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
273		spin_unlock_irqrestore(&rtasd_log_lock, s);
274		return;
275	}
276}
277
278static void handle_rtas_event(const struct rtas_error_log *log)
279{
280	if (!machine_is(pseries))
281		return;
282
283	if (rtas_error_type(log) == RTAS_TYPE_PRRN)
284		pr_info_ratelimited("Platform resource reassignment ignored.\n");
285}
286
287static int rtas_log_open(struct inode * inode, struct file * file)
288{
289	return 0;
290}
291
292static int rtas_log_release(struct inode * inode, struct file * file)
293{
294	return 0;
295}
296
297/* This will check if all events are logged, if they are then, we
298 * know that we can safely clear the events in NVRAM.
299 * Next we'll sit and wait for something else to log.
300 */
301static ssize_t rtas_log_read(struct file * file, char __user * buf,
302			 size_t count, loff_t *ppos)
303{
304	int error;
305	char *tmp;
306	unsigned long s;
307	unsigned long offset;
308
309	if (!buf || count < rtas_error_log_buffer_max)
310		return -EINVAL;
311
312	count = rtas_error_log_buffer_max;
313
314	if (!access_ok(buf, count))
315		return -EFAULT;
316
317	tmp = kmalloc(count, GFP_KERNEL);
318	if (!tmp)
319		return -ENOMEM;
320
321	spin_lock_irqsave(&rtasd_log_lock, s);
322
323	/* if it's 0, then we know we got the last one (the one in NVRAM) */
324	while (rtas_log_size == 0) {
325		if (file->f_flags & O_NONBLOCK) {
326			spin_unlock_irqrestore(&rtasd_log_lock, s);
327			error = -EAGAIN;
328			goto out;
329		}
330
331		if (!logging_enabled) {
332			spin_unlock_irqrestore(&rtasd_log_lock, s);
333			error = -ENODATA;
334			goto out;
335		}
336#ifdef CONFIG_PPC64
337		nvram_clear_error_log();
338#endif /* CONFIG_PPC64 */
339
340		spin_unlock_irqrestore(&rtasd_log_lock, s);
341		error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
342		if (error)
343			goto out;
344		spin_lock_irqsave(&rtasd_log_lock, s);
345	}
346
347	offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
348	memcpy(tmp, &rtas_log_buf[offset], count);
349
350	rtas_log_start += 1;
351	rtas_log_size -= 1;
352	spin_unlock_irqrestore(&rtasd_log_lock, s);
353
354	error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
355out:
356	kfree(tmp);
357	return error;
358}
359
360static __poll_t rtas_log_poll(struct file *file, poll_table * wait)
361{
362	poll_wait(file, &rtas_log_wait, wait);
363	if (rtas_log_size)
364		return EPOLLIN | EPOLLRDNORM;
365	return 0;
366}
367
368static const struct proc_ops rtas_log_proc_ops = {
369	.proc_read	= rtas_log_read,
370	.proc_poll	= rtas_log_poll,
371	.proc_open	= rtas_log_open,
372	.proc_release	= rtas_log_release,
373	.proc_lseek	= noop_llseek,
374};
375
376static int enable_surveillance(int timeout)
377{
378	int error;
379
380	error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
381
382	if (error == 0)
383		return 0;
384
385	if (error == -EINVAL) {
386		printk(KERN_DEBUG "rtasd: surveillance not supported\n");
387		return 0;
388	}
389
390	printk(KERN_ERR "rtasd: could not update surveillance\n");
391	return -1;
392}
393
394static void do_event_scan(void)
395{
396	int error;
397	do {
398		memset(logdata, 0, rtas_error_log_max);
399		error = rtas_call(event_scan, 4, 1, NULL,
400				  RTAS_EVENT_SCAN_ALL_EVENTS, 0,
401				  __pa(logdata), rtas_error_log_max);
402		if (error == -1) {
403			printk(KERN_ERR "event-scan failed\n");
404			break;
405		}
406
407		if (error == 0) {
408			if (rtas_error_type((struct rtas_error_log *)logdata) !=
409			    RTAS_TYPE_PRRN)
410				pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
411						  0);
412			handle_rtas_event((struct rtas_error_log *)logdata);
413		}
414
415	} while(error == 0);
416}
417
418static void rtas_event_scan(struct work_struct *w);
419static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
420
421/*
422 * Delay should be at least one second since some machines have problems if
423 * we call event-scan too quickly.
424 */
425static unsigned long event_scan_delay = 1*HZ;
426static int first_pass = 1;
427
428static void rtas_event_scan(struct work_struct *w)
429{
430	unsigned int cpu;
431
432	do_event_scan();
433
434	cpus_read_lock();
435
436	/* raw_ OK because just using CPU as starting point. */
437	cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
438        if (cpu >= nr_cpu_ids) {
439		cpu = cpumask_first(cpu_online_mask);
440
441		if (first_pass) {
442			first_pass = 0;
443			event_scan_delay = 30*HZ/rtas_event_scan_rate;
444
445			if (surveillance_timeout != -1) {
446				pr_debug("rtasd: enabling surveillance\n");
447				enable_surveillance(surveillance_timeout);
448				pr_debug("rtasd: surveillance enabled\n");
449			}
450		}
451	}
452
453	schedule_delayed_work_on(cpu, &event_scan_work,
454		__round_jiffies_relative(event_scan_delay, cpu));
455
456	cpus_read_unlock();
457}
458
459#ifdef CONFIG_PPC64
460static void __init retrieve_nvram_error_log(void)
461{
462	unsigned int err_type ;
463	int rc ;
464
465	/* See if we have any error stored in NVRAM */
466	memset(logdata, 0, rtas_error_log_max);
467	rc = nvram_read_error_log(logdata, rtas_error_log_max,
468	                          &err_type, &error_log_cnt);
469	/* We can use rtas_log_buf now */
470	logging_enabled = 1;
471	if (!rc) {
472		if (err_type != ERR_FLAG_ALREADY_LOGGED) {
473			pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
474		}
475	}
476}
477#else /* CONFIG_PPC64 */
478static void __init retrieve_nvram_error_log(void)
479{
480}
481#endif /* CONFIG_PPC64 */
482
483static void __init start_event_scan(void)
484{
485	printk(KERN_DEBUG "RTAS daemon started\n");
486	pr_debug("rtasd: will sleep for %d milliseconds\n",
487		 (30000 / rtas_event_scan_rate));
488
489	/* Retrieve errors from nvram if any */
490	retrieve_nvram_error_log();
491
492	schedule_delayed_work_on(cpumask_first(cpu_online_mask),
493				 &event_scan_work, event_scan_delay);
494}
495
496/* Cancel the rtas event scan work */
497void rtas_cancel_event_scan(void)
498{
499	cancel_delayed_work_sync(&event_scan_work);
500}
501EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
502
503static int __init rtas_event_scan_init(void)
504{
505	int err;
506
507	if (!machine_is(pseries) && !machine_is(chrp))
508		return 0;
509
510	/* No RTAS */
511	event_scan = rtas_function_token(RTAS_FN_EVENT_SCAN);
512	if (event_scan == RTAS_UNKNOWN_SERVICE) {
513		printk(KERN_INFO "rtasd: No event-scan on system\n");
514		return -ENODEV;
515	}
516
517	err = of_property_read_u32(rtas.dev, "rtas-event-scan-rate", &rtas_event_scan_rate);
518	if (err) {
519		printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
520		return -ENODEV;
521	}
522
523	if (!rtas_event_scan_rate) {
524		/* Broken firmware: take a rate of zero to mean don't scan */
525		printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
526		return 0;
527	}
528
529	/* Make room for the sequence number */
530	rtas_error_log_max = rtas_get_error_log_max();
531	rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
532
533	rtas_log_buf = vmalloc(array_size(LOG_NUMBER,
534					  rtas_error_log_buffer_max));
535	if (!rtas_log_buf) {
536		printk(KERN_ERR "rtasd: no memory\n");
537		return -ENOMEM;
538	}
539
540	start_event_scan();
541
542	return 0;
543}
544arch_initcall(rtas_event_scan_init);
545
546static int __init rtas_init(void)
547{
548	struct proc_dir_entry *entry;
549
550	if (!machine_is(pseries) && !machine_is(chrp))
551		return 0;
552
553	if (!rtas_log_buf)
554		return -ENODEV;
555
556	entry = proc_create("powerpc/rtas/error_log", 0400, NULL,
557			    &rtas_log_proc_ops);
558	if (!entry)
559		printk(KERN_ERR "Failed to create error_log proc entry\n");
560
561	return 0;
562}
563__initcall(rtas_init);
564
565static int __init surveillance_setup(char *str)
566{
567	int i;
568
569	/* We only do surveillance on pseries */
570	if (!machine_is(pseries))
571		return 0;
572
573	if (get_option(&str,&i)) {
574		if (i >= 0 && i <= 255)
575			surveillance_timeout = i;
576	}
577
578	return 1;
579}
580__setup("surveillance=", surveillance_setup);
581
582static int __init rtasmsgs_setup(char *str)
583{
584	return (kstrtobool(str, &full_rtas_msgs) == 0);
585}
586__setup("rtasmsgs=", rtasmsgs_setup);