]> git.ipfire.org Git - people/arne_f/kernel.git/blame - arch/ppc64/kernel/rtasd.c
Linux-2.6.12-rc2
[people/arne_f/kernel.git] / arch / ppc64 / kernel / rtasd.c
CommitLineData
1da177e4
LT
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
23#include <asm/uaccess.h>
24#include <asm/io.h>
25#include <asm/rtas.h>
26#include <asm/prom.h>
27#include <asm/nvram.h>
28#include <asm/atomic.h>
29#include <asm/systemcfg.h>
30
31#if 0
32#define DEBUG(A...) printk(KERN_ERR A)
33#else
34#define DEBUG(A...)
35#endif
36
37static DEFINE_SPINLOCK(rtasd_log_lock);
38
39DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
40
41static char *rtas_log_buf;
42static unsigned long rtas_log_start;
43static unsigned long rtas_log_size;
44
45static int surveillance_timeout = -1;
46static unsigned int rtas_event_scan_rate;
47static unsigned int rtas_error_log_max;
48static unsigned int rtas_error_log_buffer_max;
49
50static int full_rtas_msgs = 0;
51
52extern int no_logging;
53
54volatile int error_log_cnt = 0;
55
56/*
57 * Since we use 32 bit RTAS, the physical address of this must be below
58 * 4G or else bad things happen. Allocate this in the kernel data and
59 * make it big enough.
60 */
61static unsigned char logdata[RTAS_ERROR_LOG_MAX];
62
63static int get_eventscan_parms(void);
64
65static char *rtas_type[] = {
66 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
67 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
68 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
69};
70
71static char *rtas_event_type(int type)
72{
73 if ((type > 0) && (type < 11))
74 return rtas_type[type];
75
76 switch (type) {
77 case RTAS_TYPE_EPOW:
78 return "EPOW";
79 case RTAS_TYPE_PLATFORM:
80 return "Platform Error";
81 case RTAS_TYPE_IO:
82 return "I/O Event";
83 case RTAS_TYPE_INFO:
84 return "Platform Information Event";
85 case RTAS_TYPE_DEALLOC:
86 return "Resource Deallocation Event";
87 case RTAS_TYPE_DUMP:
88 return "Dump Notification Event";
89 }
90
91 return rtas_type[0];
92}
93
94/* To see this info, grep RTAS /var/log/messages and each entry
95 * will be collected together with obvious begin/end.
96 * There will be a unique identifier on the begin and end lines.
97 * This will persist across reboots.
98 *
99 * format of error logs returned from RTAS:
100 * bytes (size) : contents
101 * --------------------------------------------------------
102 * 0-7 (8) : rtas_error_log
103 * 8-47 (40) : extended info
104 * 48-51 (4) : vendor id
105 * 52-1023 (vendor specific) : location code and debug data
106 */
107static void printk_log_rtas(char *buf, int len)
108{
109
110 int i,j,n = 0;
111 int perline = 16;
112 char buffer[64];
113 char * str = "RTAS event";
114
115 if (full_rtas_msgs) {
116 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
117 error_log_cnt, str);
118
119 /*
120 * Print perline bytes on each line, each line will start
121 * with RTAS and a changing number, so syslogd will
122 * print lines that are otherwise the same. Separate every
123 * 4 bytes with a space.
124 */
125 for (i = 0; i < len; i++) {
126 j = i % perline;
127 if (j == 0) {
128 memset(buffer, 0, sizeof(buffer));
129 n = sprintf(buffer, "RTAS %d:", i/perline);
130 }
131
132 if ((i % 4) == 0)
133 n += sprintf(buffer+n, " ");
134
135 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
136
137 if (j == (perline-1))
138 printk(KERN_DEBUG "%s\n", buffer);
139 }
140 if ((i % perline) != 0)
141 printk(KERN_DEBUG "%s\n", buffer);
142
143 printk(RTAS_DEBUG "%d -------- %s end ----------\n",
144 error_log_cnt, str);
145 } else {
146 struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
147
148 printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
149 error_log_cnt, rtas_event_type(errlog->type),
150 errlog->severity);
151 }
152}
153
154static int log_rtas_len(char * buf)
155{
156 int len;
157 struct rtas_error_log *err;
158
159 /* rtas fixed header */
160 len = 8;
161 err = (struct rtas_error_log *)buf;
162 if (err->extended_log_length) {
163
164 /* extended header */
165 len += err->extended_log_length;
166 }
167
168 if (rtas_error_log_max == 0) {
169 get_eventscan_parms();
170 }
171 if (len > rtas_error_log_max)
172 len = rtas_error_log_max;
173
174 return len;
175}
176
177/*
178 * First write to nvram, if fatal error, that is the only
179 * place we log the info. The error will be picked up
180 * on the next reboot by rtasd. If not fatal, run the
181 * method for the type of error. Currently, only RTAS
182 * errors have methods implemented, but in the future
183 * there might be a need to store data in nvram before a
184 * call to panic().
185 *
186 * XXX We write to nvram periodically, to indicate error has
187 * been written and sync'd, but there is a possibility
188 * that if we don't shutdown correctly, a duplicate error
189 * record will be created on next reboot.
190 */
191void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
192{
193 unsigned long offset;
194 unsigned long s;
195 int len = 0;
196
197 DEBUG("logging event\n");
198 if (buf == NULL)
199 return;
200
201 spin_lock_irqsave(&rtasd_log_lock, s);
202
203 /* get length and increase count */
204 switch (err_type & ERR_TYPE_MASK) {
205 case ERR_TYPE_RTAS_LOG:
206 len = log_rtas_len(buf);
207 if (!(err_type & ERR_FLAG_BOOT))
208 error_log_cnt++;
209 break;
210 case ERR_TYPE_KERNEL_PANIC:
211 default:
212 spin_unlock_irqrestore(&rtasd_log_lock, s);
213 return;
214 }
215
216 /* Write error to NVRAM */
217 if (!no_logging && !(err_type & ERR_FLAG_BOOT))
218 nvram_write_error_log(buf, len, err_type);
219
220 /*
221 * rtas errors can occur during boot, and we do want to capture
222 * those somewhere, even if nvram isn't ready (why not?), and even
223 * if rtasd isn't ready. Put them into the boot log, at least.
224 */
225 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
226 printk_log_rtas(buf, len);
227
228 /* Check to see if we need to or have stopped logging */
229 if (fatal || no_logging) {
230 no_logging = 1;
231 spin_unlock_irqrestore(&rtasd_log_lock, s);
232 return;
233 }
234
235 /* call type specific method for error */
236 switch (err_type & ERR_TYPE_MASK) {
237 case ERR_TYPE_RTAS_LOG:
238 offset = rtas_error_log_buffer_max *
239 ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
240
241 /* First copy over sequence number */
242 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
243
244 /* Second copy over error log data */
245 offset += sizeof(int);
246 memcpy(&rtas_log_buf[offset], buf, len);
247
248 if (rtas_log_size < LOG_NUMBER)
249 rtas_log_size += 1;
250 else
251 rtas_log_start += 1;
252
253 spin_unlock_irqrestore(&rtasd_log_lock, s);
254 wake_up_interruptible(&rtas_log_wait);
255 break;
256 case ERR_TYPE_KERNEL_PANIC:
257 default:
258 spin_unlock_irqrestore(&rtasd_log_lock, s);
259 return;
260 }
261
262}
263
264
265static int rtas_log_open(struct inode * inode, struct file * file)
266{
267 return 0;
268}
269
270static int rtas_log_release(struct inode * inode, struct file * file)
271{
272 return 0;
273}
274
275/* This will check if all events are logged, if they are then, we
276 * know that we can safely clear the events in NVRAM.
277 * Next we'll sit and wait for something else to log.
278 */
279static ssize_t rtas_log_read(struct file * file, char __user * buf,
280 size_t count, loff_t *ppos)
281{
282 int error;
283 char *tmp;
284 unsigned long s;
285 unsigned long offset;
286
287 if (!buf || count < rtas_error_log_buffer_max)
288 return -EINVAL;
289
290 count = rtas_error_log_buffer_max;
291
292 if (!access_ok(VERIFY_WRITE, buf, count))
293 return -EFAULT;
294
295 tmp = kmalloc(count, GFP_KERNEL);
296 if (!tmp)
297 return -ENOMEM;
298
299
300 spin_lock_irqsave(&rtasd_log_lock, s);
301 /* if it's 0, then we know we got the last one (the one in NVRAM) */
302 if (rtas_log_size == 0 && !no_logging)
303 nvram_clear_error_log();
304 spin_unlock_irqrestore(&rtasd_log_lock, s);
305
306
307 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
308 if (error)
309 goto out;
310
311 spin_lock_irqsave(&rtasd_log_lock, s);
312 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
313 memcpy(tmp, &rtas_log_buf[offset], count);
314
315 rtas_log_start += 1;
316 rtas_log_size -= 1;
317 spin_unlock_irqrestore(&rtasd_log_lock, s);
318
319 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
320out:
321 kfree(tmp);
322 return error;
323}
324
325static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
326{
327 poll_wait(file, &rtas_log_wait, wait);
328 if (rtas_log_size)
329 return POLLIN | POLLRDNORM;
330 return 0;
331}
332
333struct file_operations proc_rtas_log_operations = {
334 .read = rtas_log_read,
335 .poll = rtas_log_poll,
336 .open = rtas_log_open,
337 .release = rtas_log_release,
338};
339
340static int enable_surveillance(int timeout)
341{
342 int error;
343
344 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
345
346 if (error == 0)
347 return 0;
348
349 if (error == -EINVAL) {
350 printk(KERN_INFO "rtasd: surveillance not supported\n");
351 return 0;
352 }
353
354 printk(KERN_ERR "rtasd: could not update surveillance\n");
355 return -1;
356}
357
358static int get_eventscan_parms(void)
359{
360 struct device_node *node;
361 int *ip;
362
363 node = of_find_node_by_path("/rtas");
364
365 ip = (int *)get_property(node, "rtas-event-scan-rate", NULL);
366 if (ip == NULL) {
367 printk(KERN_ERR "rtasd: no rtas-event-scan-rate\n");
368 of_node_put(node);
369 return -1;
370 }
371 rtas_event_scan_rate = *ip;
372 DEBUG("rtas-event-scan-rate %d\n", rtas_event_scan_rate);
373
374 /* Make room for the sequence number */
375 rtas_error_log_max = rtas_get_error_log_max();
376 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
377
378 of_node_put(node);
379
380 return 0;
381}
382
383static void do_event_scan(int event_scan)
384{
385 int error;
386 do {
387 memset(logdata, 0, rtas_error_log_max);
388 error = rtas_call(event_scan, 4, 1, NULL,
389 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
390 __pa(logdata), rtas_error_log_max);
391 if (error == -1) {
392 printk(KERN_ERR "event-scan failed\n");
393 break;
394 }
395
396 if (error == 0)
397 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
398
399 } while(error == 0);
400}
401
402static void do_event_scan_all_cpus(long delay)
403{
404 int cpu;
405
406 lock_cpu_hotplug();
407 cpu = first_cpu(cpu_online_map);
408 for (;;) {
409 set_cpus_allowed(current, cpumask_of_cpu(cpu));
410 do_event_scan(rtas_token("event-scan"));
411 set_cpus_allowed(current, CPU_MASK_ALL);
412
413 /* Drop hotplug lock, and sleep for the specified delay */
414 unlock_cpu_hotplug();
415 set_current_state(TASK_INTERRUPTIBLE);
416 schedule_timeout(delay);
417 lock_cpu_hotplug();
418
419 cpu = next_cpu(cpu, cpu_online_map);
420 if (cpu == NR_CPUS)
421 break;
422 }
423 unlock_cpu_hotplug();
424}
425
426static int rtasd(void *unused)
427{
428 unsigned int err_type;
429 int event_scan = rtas_token("event-scan");
430 int rc;
431
432 daemonize("rtasd");
433
434 if (event_scan == RTAS_UNKNOWN_SERVICE || get_eventscan_parms() == -1)
435 goto error;
436
437 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
438 if (!rtas_log_buf) {
439 printk(KERN_ERR "rtasd: no memory\n");
440 goto error;
441 }
442
443 printk(KERN_ERR "RTAS daemon started\n");
444
445 DEBUG("will sleep for %d jiffies\n", (HZ*60/rtas_event_scan_rate) / 2);
446
447 /* See if we have any error stored in NVRAM */
448 memset(logdata, 0, rtas_error_log_max);
449
450 rc = nvram_read_error_log(logdata, rtas_error_log_max, &err_type);
451
452 /* We can use rtas_log_buf now */
453 no_logging = 0;
454
455 if (!rc) {
456 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
457 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
458 }
459 }
460
461 /* First pass. */
462 do_event_scan_all_cpus(HZ);
463
464 if (surveillance_timeout != -1) {
465 DEBUG("enabling surveillance\n");
466 enable_surveillance(surveillance_timeout);
467 DEBUG("surveillance enabled\n");
468 }
469
470 /* Delay should be at least one second since some
471 * machines have problems if we call event-scan too
472 * quickly. */
473 for (;;)
474 do_event_scan_all_cpus((HZ*60/rtas_event_scan_rate) / 2);
475
476error:
477 /* Should delete proc entries */
478 return -EINVAL;
479}
480
481static int __init rtas_init(void)
482{
483 struct proc_dir_entry *entry;
484
485 /* No RTAS, only warn if we are on a pSeries box */
486 if (rtas_token("event-scan") == RTAS_UNKNOWN_SERVICE) {
487 if (systemcfg->platform & PLATFORM_PSERIES)
488 printk(KERN_ERR "rtasd: no event-scan on system\n");
489 return 1;
490 }
491
492 entry = create_proc_entry("ppc64/rtas/error_log", S_IRUSR, NULL);
493 if (entry)
494 entry->proc_fops = &proc_rtas_log_operations;
495 else
496 printk(KERN_ERR "Failed to create error_log proc entry\n");
497
498 if (kernel_thread(rtasd, NULL, CLONE_FS) < 0)
499 printk(KERN_ERR "Failed to start RTAS daemon\n");
500
501 return 0;
502}
503
504static int __init surveillance_setup(char *str)
505{
506 int i;
507
508 if (get_option(&str,&i)) {
509 if (i >= 0 && i <= 255)
510 surveillance_timeout = i;
511 }
512
513 return 1;
514}
515
516static int __init rtasmsgs_setup(char *str)
517{
518 if (strcmp(str, "on") == 0)
519 full_rtas_msgs = 1;
520 else if (strcmp(str, "off") == 0)
521 full_rtas_msgs = 0;
522
523 return 1;
524}
525__initcall(rtas_init);
526__setup("surveillance=", surveillance_setup);
527__setup("rtasmsgs=", rtasmsgs_setup);