Loading...
1/*
2 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Communication to userspace based on kernel/printk.c
10 */
11
12#include <linux/types.h>
13#include <linux/errno.h>
14#include <linux/sched.h>
15#include <linux/kernel.h>
16#include <linux/poll.h>
17#include <linux/proc_fs.h>
18#include <linux/init.h>
19#include <linux/vmalloc.h>
20#include <linux/spinlock.h>
21#include <linux/cpu.h>
22#include <linux/workqueue.h>
23#include <linux/slab.h>
24#include <linux/topology.h>
25
26#include <linux/uaccess.h>
27#include <asm/io.h>
28#include <asm/rtas.h>
29#include <asm/prom.h>
30#include <asm/nvram.h>
31#include <linux/atomic.h>
32#include <asm/machdep.h>
33#include <asm/topology.h>
34
35
36static DEFINE_SPINLOCK(rtasd_log_lock);
37
38static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
39
40static char *rtas_log_buf;
41static unsigned long rtas_log_start;
42static unsigned long rtas_log_size;
43
44static int surveillance_timeout = -1;
45
46static unsigned int rtas_error_log_max;
47static unsigned int rtas_error_log_buffer_max;
48
49/* RTAS service tokens */
50static unsigned int event_scan;
51static unsigned int rtas_event_scan_rate;
52
53static bool full_rtas_msgs;
54
55/* Stop logging to nvram after first fatal error */
56static int logging_enabled; /* Until we initialize everything,
57 * make sure we don't try logging
58 * anything */
59static int error_log_cnt;
60
61/*
62 * Since we use 32 bit RTAS, the physical address of this must be below
63 * 4G or else bad things happen. Allocate this in the kernel data and
64 * make it big enough.
65 */
66static unsigned char logdata[RTAS_ERROR_LOG_MAX];
67
68static char *rtas_type[] = {
69 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
70 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
71 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
72};
73
74static char *rtas_event_type(int type)
75{
76 if ((type > 0) && (type < 11))
77 return rtas_type[type];
78
79 switch (type) {
80 case RTAS_TYPE_EPOW:
81 return "EPOW";
82 case RTAS_TYPE_PLATFORM:
83 return "Platform Error";
84 case RTAS_TYPE_IO:
85 return "I/O Event";
86 case RTAS_TYPE_INFO:
87 return "Platform Information Event";
88 case RTAS_TYPE_DEALLOC:
89 return "Resource Deallocation Event";
90 case RTAS_TYPE_DUMP:
91 return "Dump Notification Event";
92 case RTAS_TYPE_PRRN:
93 return "Platform Resource Reassignment 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, Severity: %d\n",
154 error_log_cnt, rtas_event_type(rtas_error_type(errlog)),
155 rtas_error_severity(errlog));
156 }
157}
158
159static 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 */
198void 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
276#ifdef CONFIG_PPC_PSERIES
277static s32 prrn_update_scope;
278
279static void prrn_work_fn(struct work_struct *work)
280{
281 /*
282 * For PRRN, we must pass the negative of the scope value in
283 * the RTAS event.
284 */
285 pseries_devicetree_update(-prrn_update_scope);
286 numa_update_cpu_topology(false);
287}
288
289static DECLARE_WORK(prrn_work, prrn_work_fn);
290
291static void prrn_schedule_update(u32 scope)
292{
293 flush_work(&prrn_work);
294 prrn_update_scope = scope;
295 schedule_work(&prrn_work);
296}
297
298static void handle_rtas_event(const struct rtas_error_log *log)
299{
300 if (rtas_error_type(log) != RTAS_TYPE_PRRN || !prrn_is_enabled())
301 return;
302
303 /* For PRRN Events the extended log length is used to denote
304 * the scope for calling rtas update-nodes.
305 */
306 prrn_schedule_update(rtas_error_extended_log_length(log));
307}
308
309#else
310
311static void handle_rtas_event(const struct rtas_error_log *log)
312{
313 return;
314}
315
316#endif
317
318static int rtas_log_open(struct inode * inode, struct file * file)
319{
320 return 0;
321}
322
323static int rtas_log_release(struct inode * inode, struct file * file)
324{
325 return 0;
326}
327
328/* This will check if all events are logged, if they are then, we
329 * know that we can safely clear the events in NVRAM.
330 * Next we'll sit and wait for something else to log.
331 */
332static ssize_t rtas_log_read(struct file * file, char __user * buf,
333 size_t count, loff_t *ppos)
334{
335 int error;
336 char *tmp;
337 unsigned long s;
338 unsigned long offset;
339
340 if (!buf || count < rtas_error_log_buffer_max)
341 return -EINVAL;
342
343 count = rtas_error_log_buffer_max;
344
345 if (!access_ok(VERIFY_WRITE, buf, count))
346 return -EFAULT;
347
348 tmp = kmalloc(count, GFP_KERNEL);
349 if (!tmp)
350 return -ENOMEM;
351
352 spin_lock_irqsave(&rtasd_log_lock, s);
353
354 /* if it's 0, then we know we got the last one (the one in NVRAM) */
355 while (rtas_log_size == 0) {
356 if (file->f_flags & O_NONBLOCK) {
357 spin_unlock_irqrestore(&rtasd_log_lock, s);
358 error = -EAGAIN;
359 goto out;
360 }
361
362 if (!logging_enabled) {
363 spin_unlock_irqrestore(&rtasd_log_lock, s);
364 error = -ENODATA;
365 goto out;
366 }
367#ifdef CONFIG_PPC64
368 nvram_clear_error_log();
369#endif /* CONFIG_PPC64 */
370
371 spin_unlock_irqrestore(&rtasd_log_lock, s);
372 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
373 if (error)
374 goto out;
375 spin_lock_irqsave(&rtasd_log_lock, s);
376 }
377
378 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
379 memcpy(tmp, &rtas_log_buf[offset], count);
380
381 rtas_log_start += 1;
382 rtas_log_size -= 1;
383 spin_unlock_irqrestore(&rtasd_log_lock, s);
384
385 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
386out:
387 kfree(tmp);
388 return error;
389}
390
391static __poll_t rtas_log_poll(struct file *file, poll_table * wait)
392{
393 poll_wait(file, &rtas_log_wait, wait);
394 if (rtas_log_size)
395 return EPOLLIN | EPOLLRDNORM;
396 return 0;
397}
398
399static const struct file_operations proc_rtas_log_operations = {
400 .read = rtas_log_read,
401 .poll = rtas_log_poll,
402 .open = rtas_log_open,
403 .release = rtas_log_release,
404 .llseek = noop_llseek,
405};
406
407static int enable_surveillance(int timeout)
408{
409 int error;
410
411 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
412
413 if (error == 0)
414 return 0;
415
416 if (error == -EINVAL) {
417 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
418 return 0;
419 }
420
421 printk(KERN_ERR "rtasd: could not update surveillance\n");
422 return -1;
423}
424
425static void do_event_scan(void)
426{
427 int error;
428 do {
429 memset(logdata, 0, rtas_error_log_max);
430 error = rtas_call(event_scan, 4, 1, NULL,
431 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
432 __pa(logdata), rtas_error_log_max);
433 if (error == -1) {
434 printk(KERN_ERR "event-scan failed\n");
435 break;
436 }
437
438 if (error == 0) {
439 if (rtas_error_type((struct rtas_error_log *)logdata) !=
440 RTAS_TYPE_PRRN)
441 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
442 0);
443 handle_rtas_event((struct rtas_error_log *)logdata);
444 }
445
446 } while(error == 0);
447}
448
449static void rtas_event_scan(struct work_struct *w);
450static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
451
452/*
453 * Delay should be at least one second since some machines have problems if
454 * we call event-scan too quickly.
455 */
456static unsigned long event_scan_delay = 1*HZ;
457static int first_pass = 1;
458
459static void rtas_event_scan(struct work_struct *w)
460{
461 unsigned int cpu;
462
463 do_event_scan();
464
465 get_online_cpus();
466
467 /* raw_ OK because just using CPU as starting point. */
468 cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
469 if (cpu >= nr_cpu_ids) {
470 cpu = cpumask_first(cpu_online_mask);
471
472 if (first_pass) {
473 first_pass = 0;
474 event_scan_delay = 30*HZ/rtas_event_scan_rate;
475
476 if (surveillance_timeout != -1) {
477 pr_debug("rtasd: enabling surveillance\n");
478 enable_surveillance(surveillance_timeout);
479 pr_debug("rtasd: surveillance enabled\n");
480 }
481 }
482 }
483
484 schedule_delayed_work_on(cpu, &event_scan_work,
485 __round_jiffies_relative(event_scan_delay, cpu));
486
487 put_online_cpus();
488}
489
490#ifdef CONFIG_PPC64
491static void retrieve_nvram_error_log(void)
492{
493 unsigned int err_type ;
494 int rc ;
495
496 /* See if we have any error stored in NVRAM */
497 memset(logdata, 0, rtas_error_log_max);
498 rc = nvram_read_error_log(logdata, rtas_error_log_max,
499 &err_type, &error_log_cnt);
500 /* We can use rtas_log_buf now */
501 logging_enabled = 1;
502 if (!rc) {
503 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
504 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
505 }
506 }
507}
508#else /* CONFIG_PPC64 */
509static void retrieve_nvram_error_log(void)
510{
511}
512#endif /* CONFIG_PPC64 */
513
514static void start_event_scan(void)
515{
516 printk(KERN_DEBUG "RTAS daemon started\n");
517 pr_debug("rtasd: will sleep for %d milliseconds\n",
518 (30000 / rtas_event_scan_rate));
519
520 /* Retrieve errors from nvram if any */
521 retrieve_nvram_error_log();
522
523 schedule_delayed_work_on(cpumask_first(cpu_online_mask),
524 &event_scan_work, event_scan_delay);
525}
526
527/* Cancel the rtas event scan work */
528void rtas_cancel_event_scan(void)
529{
530 cancel_delayed_work_sync(&event_scan_work);
531}
532EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
533
534static int __init rtas_event_scan_init(void)
535{
536 if (!machine_is(pseries) && !machine_is(chrp))
537 return 0;
538
539 /* No RTAS */
540 event_scan = rtas_token("event-scan");
541 if (event_scan == RTAS_UNKNOWN_SERVICE) {
542 printk(KERN_INFO "rtasd: No event-scan on system\n");
543 return -ENODEV;
544 }
545
546 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
547 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
548 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
549 return -ENODEV;
550 }
551
552 if (!rtas_event_scan_rate) {
553 /* Broken firmware: take a rate of zero to mean don't scan */
554 printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
555 return 0;
556 }
557
558 /* Make room for the sequence number */
559 rtas_error_log_max = rtas_get_error_log_max();
560 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
561
562 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
563 if (!rtas_log_buf) {
564 printk(KERN_ERR "rtasd: no memory\n");
565 return -ENOMEM;
566 }
567
568 start_event_scan();
569
570 return 0;
571}
572arch_initcall(rtas_event_scan_init);
573
574static int __init rtas_init(void)
575{
576 struct proc_dir_entry *entry;
577
578 if (!machine_is(pseries) && !machine_is(chrp))
579 return 0;
580
581 if (!rtas_log_buf)
582 return -ENODEV;
583
584 entry = proc_create("powerpc/rtas/error_log", 0400, NULL,
585 &proc_rtas_log_operations);
586 if (!entry)
587 printk(KERN_ERR "Failed to create error_log proc entry\n");
588
589 return 0;
590}
591__initcall(rtas_init);
592
593static int __init surveillance_setup(char *str)
594{
595 int i;
596
597 /* We only do surveillance on pseries */
598 if (!machine_is(pseries))
599 return 0;
600
601 if (get_option(&str,&i)) {
602 if (i >= 0 && i <= 255)
603 surveillance_timeout = i;
604 }
605
606 return 1;
607}
608__setup("surveillance=", surveillance_setup);
609
610static int __init rtasmsgs_setup(char *str)
611{
612 return (kstrtobool(str, &full_rtas_msgs) == 0);
613}
614__setup("rtasmsgs=", rtasmsgs_setup);
1/*
2 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Communication to userspace based on kernel/printk.c
10 */
11
12#include <linux/types.h>
13#include <linux/errno.h>
14#include <linux/sched.h>
15#include <linux/kernel.h>
16#include <linux/poll.h>
17#include <linux/proc_fs.h>
18#include <linux/init.h>
19#include <linux/vmalloc.h>
20#include <linux/spinlock.h>
21#include <linux/cpu.h>
22#include <linux/workqueue.h>
23#include <linux/slab.h>
24
25#include <asm/uaccess.h>
26#include <asm/io.h>
27#include <asm/rtas.h>
28#include <asm/prom.h>
29#include <asm/nvram.h>
30#include <linux/atomic.h>
31#include <asm/machdep.h>
32#include <asm/topology.h>
33
34
35static DEFINE_SPINLOCK(rtasd_log_lock);
36
37static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
38
39static char *rtas_log_buf;
40static unsigned long rtas_log_start;
41static unsigned long rtas_log_size;
42
43static int surveillance_timeout = -1;
44
45static unsigned int rtas_error_log_max;
46static unsigned int rtas_error_log_buffer_max;
47
48/* RTAS service tokens */
49static unsigned int event_scan;
50static unsigned int rtas_event_scan_rate;
51
52static bool full_rtas_msgs;
53
54/* Stop logging to nvram after first fatal error */
55static int logging_enabled; /* Until we initialize everything,
56 * make sure we don't try logging
57 * anything */
58static int error_log_cnt;
59
60/*
61 * Since we use 32 bit RTAS, the physical address of this must be below
62 * 4G or else bad things happen. Allocate this in the kernel data and
63 * make it big enough.
64 */
65static unsigned char logdata[RTAS_ERROR_LOG_MAX];
66
67static char *rtas_type[] = {
68 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
69 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
70 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
71};
72
73static char *rtas_event_type(int type)
74{
75 if ((type > 0) && (type < 11))
76 return rtas_type[type];
77
78 switch (type) {
79 case RTAS_TYPE_EPOW:
80 return "EPOW";
81 case RTAS_TYPE_PLATFORM:
82 return "Platform Error";
83 case RTAS_TYPE_IO:
84 return "I/O Event";
85 case RTAS_TYPE_INFO:
86 return "Platform Information Event";
87 case RTAS_TYPE_DEALLOC:
88 return "Resource Deallocation Event";
89 case RTAS_TYPE_DUMP:
90 return "Dump Notification Event";
91 case RTAS_TYPE_PRRN:
92 return "Platform Resource Reassignment Event";
93 }
94
95 return rtas_type[0];
96}
97
98/* To see this info, grep RTAS /var/log/messages and each entry
99 * will be collected together with obvious begin/end.
100 * There will be a unique identifier on the begin and end lines.
101 * This will persist across reboots.
102 *
103 * format of error logs returned from RTAS:
104 * bytes (size) : contents
105 * --------------------------------------------------------
106 * 0-7 (8) : rtas_error_log
107 * 8-47 (40) : extended info
108 * 48-51 (4) : vendor id
109 * 52-1023 (vendor specific) : location code and debug data
110 */
111static void printk_log_rtas(char *buf, int len)
112{
113
114 int i,j,n = 0;
115 int perline = 16;
116 char buffer[64];
117 char * str = "RTAS event";
118
119 if (full_rtas_msgs) {
120 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
121 error_log_cnt, str);
122
123 /*
124 * Print perline bytes on each line, each line will start
125 * with RTAS and a changing number, so syslogd will
126 * print lines that are otherwise the same. Separate every
127 * 4 bytes with a space.
128 */
129 for (i = 0; i < len; i++) {
130 j = i % perline;
131 if (j == 0) {
132 memset(buffer, 0, sizeof(buffer));
133 n = sprintf(buffer, "RTAS %d:", i/perline);
134 }
135
136 if ((i % 4) == 0)
137 n += sprintf(buffer+n, " ");
138
139 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
140
141 if (j == (perline-1))
142 printk(KERN_DEBUG "%s\n", buffer);
143 }
144 if ((i % perline) != 0)
145 printk(KERN_DEBUG "%s\n", buffer);
146
147 printk(RTAS_DEBUG "%d -------- %s end ----------\n",
148 error_log_cnt, str);
149 } else {
150 struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
151
152 printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
153 error_log_cnt, rtas_event_type(rtas_error_type(errlog)),
154 rtas_error_severity(errlog));
155 }
156}
157
158static int log_rtas_len(char * buf)
159{
160 int len;
161 struct rtas_error_log *err;
162 uint32_t extended_log_length;
163
164 /* rtas fixed header */
165 len = 8;
166 err = (struct rtas_error_log *)buf;
167 extended_log_length = rtas_error_extended_log_length(err);
168 if (rtas_error_extended(err) && extended_log_length) {
169
170 /* extended header */
171 len += extended_log_length;
172 }
173
174 if (rtas_error_log_max == 0)
175 rtas_error_log_max = rtas_get_error_log_max();
176
177 if (len > rtas_error_log_max)
178 len = rtas_error_log_max;
179
180 return len;
181}
182
183/*
184 * First write to nvram, if fatal error, that is the only
185 * place we log the info. The error will be picked up
186 * on the next reboot by rtasd. If not fatal, run the
187 * method for the type of error. Currently, only RTAS
188 * errors have methods implemented, but in the future
189 * there might be a need to store data in nvram before a
190 * call to panic().
191 *
192 * XXX We write to nvram periodically, to indicate error has
193 * been written and sync'd, but there is a possibility
194 * that if we don't shutdown correctly, a duplicate error
195 * record will be created on next reboot.
196 */
197void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
198{
199 unsigned long offset;
200 unsigned long s;
201 int len = 0;
202
203 pr_debug("rtasd: logging event\n");
204 if (buf == NULL)
205 return;
206
207 spin_lock_irqsave(&rtasd_log_lock, s);
208
209 /* get length and increase count */
210 switch (err_type & ERR_TYPE_MASK) {
211 case ERR_TYPE_RTAS_LOG:
212 len = log_rtas_len(buf);
213 if (!(err_type & ERR_FLAG_BOOT))
214 error_log_cnt++;
215 break;
216 case ERR_TYPE_KERNEL_PANIC:
217 default:
218 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
219 spin_unlock_irqrestore(&rtasd_log_lock, s);
220 return;
221 }
222
223#ifdef CONFIG_PPC64
224 /* Write error to NVRAM */
225 if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
226 nvram_write_error_log(buf, len, err_type, error_log_cnt);
227#endif /* CONFIG_PPC64 */
228
229 /*
230 * rtas errors can occur during boot, and we do want to capture
231 * those somewhere, even if nvram isn't ready (why not?), and even
232 * if rtasd isn't ready. Put them into the boot log, at least.
233 */
234 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
235 printk_log_rtas(buf, len);
236
237 /* Check to see if we need to or have stopped logging */
238 if (fatal || !logging_enabled) {
239 logging_enabled = 0;
240 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
241 spin_unlock_irqrestore(&rtasd_log_lock, s);
242 return;
243 }
244
245 /* call type specific method for error */
246 switch (err_type & ERR_TYPE_MASK) {
247 case ERR_TYPE_RTAS_LOG:
248 offset = rtas_error_log_buffer_max *
249 ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
250
251 /* First copy over sequence number */
252 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
253
254 /* Second copy over error log data */
255 offset += sizeof(int);
256 memcpy(&rtas_log_buf[offset], buf, len);
257
258 if (rtas_log_size < LOG_NUMBER)
259 rtas_log_size += 1;
260 else
261 rtas_log_start += 1;
262
263 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
264 spin_unlock_irqrestore(&rtasd_log_lock, s);
265 wake_up_interruptible(&rtas_log_wait);
266 break;
267 case ERR_TYPE_KERNEL_PANIC:
268 default:
269 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
270 spin_unlock_irqrestore(&rtasd_log_lock, s);
271 return;
272 }
273}
274
275#ifdef CONFIG_PPC_PSERIES
276static s32 prrn_update_scope;
277
278static void prrn_work_fn(struct work_struct *work)
279{
280 /*
281 * For PRRN, we must pass the negative of the scope value in
282 * the RTAS event.
283 */
284 pseries_devicetree_update(-prrn_update_scope);
285}
286
287static DECLARE_WORK(prrn_work, prrn_work_fn);
288
289static void prrn_schedule_update(u32 scope)
290{
291 flush_work(&prrn_work);
292 prrn_update_scope = scope;
293 schedule_work(&prrn_work);
294}
295
296static void handle_rtas_event(const struct rtas_error_log *log)
297{
298 if (rtas_error_type(log) != RTAS_TYPE_PRRN || !prrn_is_enabled())
299 return;
300
301 /* For PRRN Events the extended log length is used to denote
302 * the scope for calling rtas update-nodes.
303 */
304 prrn_schedule_update(rtas_error_extended_log_length(log));
305}
306
307#else
308
309static void handle_rtas_event(const struct rtas_error_log *log)
310{
311 return;
312}
313
314#endif
315
316static int rtas_log_open(struct inode * inode, struct file * file)
317{
318 return 0;
319}
320
321static int rtas_log_release(struct inode * inode, struct file * file)
322{
323 return 0;
324}
325
326/* This will check if all events are logged, if they are then, we
327 * know that we can safely clear the events in NVRAM.
328 * Next we'll sit and wait for something else to log.
329 */
330static ssize_t rtas_log_read(struct file * file, char __user * buf,
331 size_t count, loff_t *ppos)
332{
333 int error;
334 char *tmp;
335 unsigned long s;
336 unsigned long offset;
337
338 if (!buf || count < rtas_error_log_buffer_max)
339 return -EINVAL;
340
341 count = rtas_error_log_buffer_max;
342
343 if (!access_ok(VERIFY_WRITE, buf, count))
344 return -EFAULT;
345
346 tmp = kmalloc(count, GFP_KERNEL);
347 if (!tmp)
348 return -ENOMEM;
349
350 spin_lock_irqsave(&rtasd_log_lock, s);
351
352 /* if it's 0, then we know we got the last one (the one in NVRAM) */
353 while (rtas_log_size == 0) {
354 if (file->f_flags & O_NONBLOCK) {
355 spin_unlock_irqrestore(&rtasd_log_lock, s);
356 error = -EAGAIN;
357 goto out;
358 }
359
360 if (!logging_enabled) {
361 spin_unlock_irqrestore(&rtasd_log_lock, s);
362 error = -ENODATA;
363 goto out;
364 }
365#ifdef CONFIG_PPC64
366 nvram_clear_error_log();
367#endif /* CONFIG_PPC64 */
368
369 spin_unlock_irqrestore(&rtasd_log_lock, s);
370 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
371 if (error)
372 goto out;
373 spin_lock_irqsave(&rtasd_log_lock, s);
374 }
375
376 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
377 memcpy(tmp, &rtas_log_buf[offset], count);
378
379 rtas_log_start += 1;
380 rtas_log_size -= 1;
381 spin_unlock_irqrestore(&rtasd_log_lock, s);
382
383 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
384out:
385 kfree(tmp);
386 return error;
387}
388
389static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
390{
391 poll_wait(file, &rtas_log_wait, wait);
392 if (rtas_log_size)
393 return POLLIN | POLLRDNORM;
394 return 0;
395}
396
397static const struct file_operations proc_rtas_log_operations = {
398 .read = rtas_log_read,
399 .poll = rtas_log_poll,
400 .open = rtas_log_open,
401 .release = rtas_log_release,
402 .llseek = noop_llseek,
403};
404
405static int enable_surveillance(int timeout)
406{
407 int error;
408
409 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
410
411 if (error == 0)
412 return 0;
413
414 if (error == -EINVAL) {
415 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
416 return 0;
417 }
418
419 printk(KERN_ERR "rtasd: could not update surveillance\n");
420 return -1;
421}
422
423static void do_event_scan(void)
424{
425 int error;
426 do {
427 memset(logdata, 0, rtas_error_log_max);
428 error = rtas_call(event_scan, 4, 1, NULL,
429 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
430 __pa(logdata), rtas_error_log_max);
431 if (error == -1) {
432 printk(KERN_ERR "event-scan failed\n");
433 break;
434 }
435
436 if (error == 0) {
437 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
438 handle_rtas_event((struct rtas_error_log *)logdata);
439 }
440
441 } while(error == 0);
442}
443
444static void rtas_event_scan(struct work_struct *w);
445DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
446
447/*
448 * Delay should be at least one second since some machines have problems if
449 * we call event-scan too quickly.
450 */
451static unsigned long event_scan_delay = 1*HZ;
452static int first_pass = 1;
453
454static void rtas_event_scan(struct work_struct *w)
455{
456 unsigned int cpu;
457
458 do_event_scan();
459
460 get_online_cpus();
461
462 /* raw_ OK because just using CPU as starting point. */
463 cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
464 if (cpu >= nr_cpu_ids) {
465 cpu = cpumask_first(cpu_online_mask);
466
467 if (first_pass) {
468 first_pass = 0;
469 event_scan_delay = 30*HZ/rtas_event_scan_rate;
470
471 if (surveillance_timeout != -1) {
472 pr_debug("rtasd: enabling surveillance\n");
473 enable_surveillance(surveillance_timeout);
474 pr_debug("rtasd: surveillance enabled\n");
475 }
476 }
477 }
478
479 schedule_delayed_work_on(cpu, &event_scan_work,
480 __round_jiffies_relative(event_scan_delay, cpu));
481
482 put_online_cpus();
483}
484
485#ifdef CONFIG_PPC64
486static void retreive_nvram_error_log(void)
487{
488 unsigned int err_type ;
489 int rc ;
490
491 /* See if we have any error stored in NVRAM */
492 memset(logdata, 0, rtas_error_log_max);
493 rc = nvram_read_error_log(logdata, rtas_error_log_max,
494 &err_type, &error_log_cnt);
495 /* We can use rtas_log_buf now */
496 logging_enabled = 1;
497 if (!rc) {
498 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
499 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
500 }
501 }
502}
503#else /* CONFIG_PPC64 */
504static void retreive_nvram_error_log(void)
505{
506}
507#endif /* CONFIG_PPC64 */
508
509static void start_event_scan(void)
510{
511 printk(KERN_DEBUG "RTAS daemon started\n");
512 pr_debug("rtasd: will sleep for %d milliseconds\n",
513 (30000 / rtas_event_scan_rate));
514
515 /* Retrieve errors from nvram if any */
516 retreive_nvram_error_log();
517
518 schedule_delayed_work_on(cpumask_first(cpu_online_mask),
519 &event_scan_work, event_scan_delay);
520}
521
522/* Cancel the rtas event scan work */
523void rtas_cancel_event_scan(void)
524{
525 cancel_delayed_work_sync(&event_scan_work);
526}
527EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
528
529static int __init rtas_init(void)
530{
531 struct proc_dir_entry *entry;
532
533 if (!machine_is(pseries) && !machine_is(chrp))
534 return 0;
535
536 /* No RTAS */
537 event_scan = rtas_token("event-scan");
538 if (event_scan == RTAS_UNKNOWN_SERVICE) {
539 printk(KERN_INFO "rtasd: No event-scan on system\n");
540 return -ENODEV;
541 }
542
543 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
544 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
545 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
546 return -ENODEV;
547 }
548
549 if (!rtas_event_scan_rate) {
550 /* Broken firmware: take a rate of zero to mean don't scan */
551 printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
552 return 0;
553 }
554
555 /* Make room for the sequence number */
556 rtas_error_log_max = rtas_get_error_log_max();
557 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
558
559 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
560 if (!rtas_log_buf) {
561 printk(KERN_ERR "rtasd: no memory\n");
562 return -ENOMEM;
563 }
564
565 entry = proc_create("powerpc/rtas/error_log", S_IRUSR, NULL,
566 &proc_rtas_log_operations);
567 if (!entry)
568 printk(KERN_ERR "Failed to create error_log proc entry\n");
569
570 start_event_scan();
571
572 return 0;
573}
574__initcall(rtas_init);
575
576static int __init surveillance_setup(char *str)
577{
578 int i;
579
580 /* We only do surveillance on pseries */
581 if (!machine_is(pseries))
582 return 0;
583
584 if (get_option(&str,&i)) {
585 if (i >= 0 && i <= 255)
586 surveillance_timeout = i;
587 }
588
589 return 1;
590}
591__setup("surveillance=", surveillance_setup);
592
593static int __init rtasmsgs_setup(char *str)
594{
595 return (kstrtobool(str, &full_rtas_msgs) == 0);
596}
597__setup("rtasmsgs=", rtasmsgs_setup);