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