Commit | Line | Data |
---|---|---|
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> | |
0287ebed | 22 | #include <linux/delay.h> |
1da177e4 LT |
23 | |
24 | #include <asm/uaccess.h> | |
25 | #include <asm/io.h> | |
26 | #include <asm/rtas.h> | |
27 | #include <asm/prom.h> | |
28 | #include <asm/nvram.h> | |
29 | #include <asm/atomic.h> | |
e8222502 | 30 | #include <asm/machdep.h> |
1da177e4 LT |
31 | |
32 | #if 0 | |
33 | #define DEBUG(A...) printk(KERN_ERR A) | |
34 | #else | |
35 | #define DEBUG(A...) | |
36 | #endif | |
37 | ||
38 | static DEFINE_SPINLOCK(rtasd_log_lock); | |
39 | ||
40 | DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait); | |
41 | ||
42 | static char *rtas_log_buf; | |
43 | static unsigned long rtas_log_start; | |
44 | static unsigned long rtas_log_size; | |
45 | ||
46 | static int surveillance_timeout = -1; | |
47 | static unsigned int rtas_event_scan_rate; | |
48 | static unsigned int rtas_error_log_max; | |
49 | static unsigned int rtas_error_log_buffer_max; | |
50 | ||
51 | static int full_rtas_msgs = 0; | |
52 | ||
53 | extern int no_logging; | |
54 | ||
55 | volatile int error_log_cnt = 0; | |
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 int get_eventscan_parms(void); | |
65 | ||
66 | static 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 | ||
72 | static 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 | */ | |
108 | static 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 | ||
155 | static 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_log_length) { | |
164 | ||
165 | /* extended header */ | |
166 | len += err->extended_log_length; | |
167 | } | |
168 | ||
169 | if (rtas_error_log_max == 0) { | |
170 | get_eventscan_parms(); | |
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 | */ | |
192 | void 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 | DEBUG("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 | spin_unlock_irqrestore(&rtasd_log_lock, s); | |
214 | return; | |
215 | } | |
216 | ||
217 | /* Write error to NVRAM */ | |
218 | if (!no_logging && !(err_type & ERR_FLAG_BOOT)) | |
219 | nvram_write_error_log(buf, len, err_type); | |
220 | ||
221 | /* | |
222 | * rtas errors can occur during boot, and we do want to capture | |
223 | * those somewhere, even if nvram isn't ready (why not?), and even | |
224 | * if rtasd isn't ready. Put them into the boot log, at least. | |
225 | */ | |
226 | if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG) | |
227 | printk_log_rtas(buf, len); | |
228 | ||
229 | /* Check to see if we need to or have stopped logging */ | |
230 | if (fatal || no_logging) { | |
231 | no_logging = 1; | |
232 | spin_unlock_irqrestore(&rtasd_log_lock, s); | |
233 | return; | |
234 | } | |
235 | ||
236 | /* call type specific method for error */ | |
237 | switch (err_type & ERR_TYPE_MASK) { | |
238 | case ERR_TYPE_RTAS_LOG: | |
239 | offset = rtas_error_log_buffer_max * | |
240 | ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK); | |
241 | ||
242 | /* First copy over sequence number */ | |
243 | memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int)); | |
244 | ||
245 | /* Second copy over error log data */ | |
246 | offset += sizeof(int); | |
247 | memcpy(&rtas_log_buf[offset], buf, len); | |
248 | ||
249 | if (rtas_log_size < LOG_NUMBER) | |
250 | rtas_log_size += 1; | |
251 | else | |
252 | rtas_log_start += 1; | |
253 | ||
254 | spin_unlock_irqrestore(&rtasd_log_lock, s); | |
255 | wake_up_interruptible(&rtas_log_wait); | |
256 | break; | |
257 | case ERR_TYPE_KERNEL_PANIC: | |
258 | default: | |
259 | spin_unlock_irqrestore(&rtasd_log_lock, s); | |
260 | return; | |
261 | } | |
262 | ||
263 | } | |
264 | ||
265 | ||
266 | static int rtas_log_open(struct inode * inode, struct file * file) | |
267 | { | |
268 | return 0; | |
269 | } | |
270 | ||
271 | static int rtas_log_release(struct inode * inode, struct file * file) | |
272 | { | |
273 | return 0; | |
274 | } | |
275 | ||
276 | /* This will check if all events are logged, if they are then, we | |
277 | * know that we can safely clear the events in NVRAM. | |
278 | * Next we'll sit and wait for something else to log. | |
279 | */ | |
280 | static ssize_t rtas_log_read(struct file * file, char __user * buf, | |
281 | size_t count, loff_t *ppos) | |
282 | { | |
283 | int error; | |
284 | char *tmp; | |
285 | unsigned long s; | |
286 | unsigned long offset; | |
287 | ||
288 | if (!buf || count < rtas_error_log_buffer_max) | |
289 | return -EINVAL; | |
290 | ||
291 | count = rtas_error_log_buffer_max; | |
292 | ||
293 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
294 | return -EFAULT; | |
295 | ||
296 | tmp = kmalloc(count, GFP_KERNEL); | |
297 | if (!tmp) | |
298 | return -ENOMEM; | |
299 | ||
300 | ||
301 | spin_lock_irqsave(&rtasd_log_lock, s); | |
302 | /* if it's 0, then we know we got the last one (the one in NVRAM) */ | |
303 | if (rtas_log_size == 0 && !no_logging) | |
304 | nvram_clear_error_log(); | |
305 | spin_unlock_irqrestore(&rtasd_log_lock, s); | |
306 | ||
307 | ||
308 | error = wait_event_interruptible(rtas_log_wait, rtas_log_size); | |
309 | if (error) | |
310 | goto out; | |
311 | ||
312 | spin_lock_irqsave(&rtasd_log_lock, s); | |
313 | offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK); | |
314 | memcpy(tmp, &rtas_log_buf[offset], count); | |
315 | ||
316 | rtas_log_start += 1; | |
317 | rtas_log_size -= 1; | |
318 | spin_unlock_irqrestore(&rtasd_log_lock, s); | |
319 | ||
320 | error = copy_to_user(buf, tmp, count) ? -EFAULT : count; | |
321 | out: | |
322 | kfree(tmp); | |
323 | return error; | |
324 | } | |
325 | ||
326 | static unsigned int rtas_log_poll(struct file *file, poll_table * wait) | |
327 | { | |
328 | poll_wait(file, &rtas_log_wait, wait); | |
329 | if (rtas_log_size) | |
330 | return POLLIN | POLLRDNORM; | |
331 | return 0; | |
332 | } | |
333 | ||
5dfe4c96 | 334 | const struct file_operations proc_rtas_log_operations = { |
1da177e4 LT |
335 | .read = rtas_log_read, |
336 | .poll = rtas_log_poll, | |
337 | .open = rtas_log_open, | |
338 | .release = rtas_log_release, | |
339 | }; | |
340 | ||
341 | static int enable_surveillance(int timeout) | |
342 | { | |
343 | int error; | |
344 | ||
345 | error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout); | |
346 | ||
347 | if (error == 0) | |
348 | return 0; | |
349 | ||
350 | if (error == -EINVAL) { | |
90ddfebe | 351 | printk(KERN_DEBUG "rtasd: surveillance not supported\n"); |
1da177e4 LT |
352 | return 0; |
353 | } | |
354 | ||
355 | printk(KERN_ERR "rtasd: could not update surveillance\n"); | |
356 | return -1; | |
357 | } | |
358 | ||
359 | static int get_eventscan_parms(void) | |
360 | { | |
361 | struct device_node *node; | |
954a46e2 | 362 | const int *ip; |
1da177e4 LT |
363 | |
364 | node = of_find_node_by_path("/rtas"); | |
365 | ||
e2eb6392 | 366 | ip = of_get_property(node, "rtas-event-scan-rate", NULL); |
1da177e4 LT |
367 | if (ip == NULL) { |
368 | printk(KERN_ERR "rtasd: no rtas-event-scan-rate\n"); | |
369 | of_node_put(node); | |
370 | return -1; | |
371 | } | |
372 | rtas_event_scan_rate = *ip; | |
373 | DEBUG("rtas-event-scan-rate %d\n", rtas_event_scan_rate); | |
374 | ||
375 | /* Make room for the sequence number */ | |
376 | rtas_error_log_max = rtas_get_error_log_max(); | |
377 | rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int); | |
378 | ||
379 | of_node_put(node); | |
380 | ||
381 | return 0; | |
382 | } | |
383 | ||
384 | static void do_event_scan(int event_scan) | |
385 | { | |
386 | int error; | |
387 | do { | |
388 | memset(logdata, 0, rtas_error_log_max); | |
389 | error = rtas_call(event_scan, 4, 1, NULL, | |
390 | RTAS_EVENT_SCAN_ALL_EVENTS, 0, | |
391 | __pa(logdata), rtas_error_log_max); | |
392 | if (error == -1) { | |
393 | printk(KERN_ERR "event-scan failed\n"); | |
394 | break; | |
395 | } | |
396 | ||
397 | if (error == 0) | |
398 | pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0); | |
399 | ||
400 | } while(error == 0); | |
401 | } | |
402 | ||
403 | static void do_event_scan_all_cpus(long delay) | |
404 | { | |
405 | int cpu; | |
406 | ||
407 | lock_cpu_hotplug(); | |
408 | cpu = first_cpu(cpu_online_map); | |
409 | for (;;) { | |
410 | set_cpus_allowed(current, cpumask_of_cpu(cpu)); | |
411 | do_event_scan(rtas_token("event-scan")); | |
412 | set_cpus_allowed(current, CPU_MASK_ALL); | |
413 | ||
414 | /* Drop hotplug lock, and sleep for the specified delay */ | |
415 | unlock_cpu_hotplug(); | |
0287ebed | 416 | msleep_interruptible(delay); |
1da177e4 LT |
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 | ||
426 | static 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 | ||
90ddfebe | 443 | printk(KERN_DEBUG "RTAS daemon started\n"); |
1da177e4 | 444 | |
0287ebed | 445 | DEBUG("will sleep for %d milliseconds\n", (30000/rtas_event_scan_rate)); |
1da177e4 LT |
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. */ | |
0287ebed | 462 | do_event_scan_all_cpus(1000); |
1da177e4 LT |
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 (;;) | |
0287ebed | 474 | do_event_scan_all_cpus(30000/rtas_event_scan_rate); |
1da177e4 LT |
475 | |
476 | error: | |
477 | /* Should delete proc entries */ | |
478 | return -EINVAL; | |
479 | } | |
480 | ||
481 | static int __init rtas_init(void) | |
482 | { | |
483 | struct proc_dir_entry *entry; | |
484 | ||
e8222502 | 485 | if (!machine_is(pseries)) |
799d6046 PM |
486 | return 0; |
487 | ||
488 | /* No RTAS */ | |
1da177e4 | 489 | if (rtas_token("event-scan") == RTAS_UNKNOWN_SERVICE) { |
90ddfebe | 490 | printk(KERN_DEBUG "rtasd: no event-scan on system\n"); |
49c28e4e | 491 | return -ENODEV; |
1da177e4 LT |
492 | } |
493 | ||
494 | entry = create_proc_entry("ppc64/rtas/error_log", S_IRUSR, NULL); | |
495 | if (entry) | |
496 | entry->proc_fops = &proc_rtas_log_operations; | |
497 | else | |
498 | printk(KERN_ERR "Failed to create error_log proc entry\n"); | |
499 | ||
500 | if (kernel_thread(rtasd, NULL, CLONE_FS) < 0) | |
501 | printk(KERN_ERR "Failed to start RTAS daemon\n"); | |
502 | ||
503 | return 0; | |
504 | } | |
505 | ||
506 | static int __init surveillance_setup(char *str) | |
507 | { | |
508 | int i; | |
509 | ||
510 | if (get_option(&str,&i)) { | |
511 | if (i >= 0 && i <= 255) | |
512 | surveillance_timeout = i; | |
513 | } | |
514 | ||
515 | return 1; | |
516 | } | |
517 | ||
518 | static int __init rtasmsgs_setup(char *str) | |
519 | { | |
520 | if (strcmp(str, "on") == 0) | |
521 | full_rtas_msgs = 1; | |
522 | else if (strcmp(str, "off") == 0) | |
523 | full_rtas_msgs = 0; | |
524 | ||
525 | return 1; | |
526 | } | |
527 | __initcall(rtas_init); | |
528 | __setup("surveillance=", surveillance_setup); | |
529 | __setup("rtasmsgs=", rtasmsgs_setup); |