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
36 static DEFINE_SPINLOCK(rtasd_log_lock);
37
38 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
39
40 static char *rtas_log_buf;
41 static unsigned long rtas_log_start;
42 static unsigned long rtas_log_size;
43
44 static int surveillance_timeout = -1;
45
46 static unsigned int rtas_error_log_max;
47 static unsigned int rtas_error_log_buffer_max;
48
49 /* RTAS service tokens */
50 static unsigned int event_scan;
51 static unsigned int rtas_event_scan_rate;
52
53 static bool full_rtas_msgs;
54
55 /* Stop logging to nvram after first fatal error */
56 static int logging_enabled; /* Until we initialize everything,
57 * make sure we don't try logging
58 * anything */
59 static 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 */
66 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
67
68 static 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
rtas_event_type(int type)74 static 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 */
printk_log_rtas(char * buf,int len)112 static 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
log_rtas_len(char * buf)159 static 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 */
pSeries_log_error(char * buf,unsigned int err_type,int fatal)198 void 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
handle_prrn_event(s32 scope)277 static void handle_prrn_event(s32 scope)
278 {
279 /*
280 * For PRRN, we must pass the negative of the scope value in
281 * the RTAS event.
282 */
283 pseries_devicetree_update(-scope);
284 numa_update_cpu_topology(false);
285 }
286
handle_rtas_event(const struct rtas_error_log * log)287 static void handle_rtas_event(const struct rtas_error_log *log)
288 {
289 if (rtas_error_type(log) != RTAS_TYPE_PRRN || !prrn_is_enabled())
290 return;
291
292 /* For PRRN Events the extended log length is used to denote
293 * the scope for calling rtas update-nodes.
294 */
295 handle_prrn_event(rtas_error_extended_log_length(log));
296 }
297
298 #else
299
handle_rtas_event(const struct rtas_error_log * log)300 static void handle_rtas_event(const struct rtas_error_log *log)
301 {
302 return;
303 }
304
305 #endif
306
rtas_log_open(struct inode * inode,struct file * file)307 static int rtas_log_open(struct inode * inode, struct file * file)
308 {
309 return 0;
310 }
311
rtas_log_release(struct inode * inode,struct file * file)312 static int rtas_log_release(struct inode * inode, struct file * file)
313 {
314 return 0;
315 }
316
317 /* This will check if all events are logged, if they are then, we
318 * know that we can safely clear the events in NVRAM.
319 * Next we'll sit and wait for something else to log.
320 */
rtas_log_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)321 static ssize_t rtas_log_read(struct file * file, char __user * buf,
322 size_t count, loff_t *ppos)
323 {
324 int error;
325 char *tmp;
326 unsigned long s;
327 unsigned long offset;
328
329 if (!buf || count < rtas_error_log_buffer_max)
330 return -EINVAL;
331
332 count = rtas_error_log_buffer_max;
333
334 if (!access_ok(VERIFY_WRITE, buf, count))
335 return -EFAULT;
336
337 tmp = kmalloc(count, GFP_KERNEL);
338 if (!tmp)
339 return -ENOMEM;
340
341 spin_lock_irqsave(&rtasd_log_lock, s);
342
343 /* if it's 0, then we know we got the last one (the one in NVRAM) */
344 while (rtas_log_size == 0) {
345 if (file->f_flags & O_NONBLOCK) {
346 spin_unlock_irqrestore(&rtasd_log_lock, s);
347 error = -EAGAIN;
348 goto out;
349 }
350
351 if (!logging_enabled) {
352 spin_unlock_irqrestore(&rtasd_log_lock, s);
353 error = -ENODATA;
354 goto out;
355 }
356 #ifdef CONFIG_PPC64
357 nvram_clear_error_log();
358 #endif /* CONFIG_PPC64 */
359
360 spin_unlock_irqrestore(&rtasd_log_lock, s);
361 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
362 if (error)
363 goto out;
364 spin_lock_irqsave(&rtasd_log_lock, s);
365 }
366
367 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
368 memcpy(tmp, &rtas_log_buf[offset], count);
369
370 rtas_log_start += 1;
371 rtas_log_size -= 1;
372 spin_unlock_irqrestore(&rtasd_log_lock, s);
373
374 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
375 out:
376 kfree(tmp);
377 return error;
378 }
379
rtas_log_poll(struct file * file,poll_table * wait)380 static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
381 {
382 poll_wait(file, &rtas_log_wait, wait);
383 if (rtas_log_size)
384 return POLLIN | POLLRDNORM;
385 return 0;
386 }
387
388 static const struct file_operations proc_rtas_log_operations = {
389 .read = rtas_log_read,
390 .poll = rtas_log_poll,
391 .open = rtas_log_open,
392 .release = rtas_log_release,
393 .llseek = noop_llseek,
394 };
395
enable_surveillance(int timeout)396 static int enable_surveillance(int timeout)
397 {
398 int error;
399
400 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
401
402 if (error == 0)
403 return 0;
404
405 if (error == -EINVAL) {
406 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
407 return 0;
408 }
409
410 printk(KERN_ERR "rtasd: could not update surveillance\n");
411 return -1;
412 }
413
do_event_scan(void)414 static void do_event_scan(void)
415 {
416 int error;
417 do {
418 memset(logdata, 0, rtas_error_log_max);
419 error = rtas_call(event_scan, 4, 1, NULL,
420 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
421 __pa(logdata), rtas_error_log_max);
422 if (error == -1) {
423 printk(KERN_ERR "event-scan failed\n");
424 break;
425 }
426
427 if (error == 0) {
428 if (rtas_error_type((struct rtas_error_log *)logdata) !=
429 RTAS_TYPE_PRRN)
430 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
431 0);
432 handle_rtas_event((struct rtas_error_log *)logdata);
433 }
434
435 } while(error == 0);
436 }
437
438 static void rtas_event_scan(struct work_struct *w);
439 static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
440
441 /*
442 * Delay should be at least one second since some machines have problems if
443 * we call event-scan too quickly.
444 */
445 static unsigned long event_scan_delay = 1*HZ;
446 static int first_pass = 1;
447
rtas_event_scan(struct work_struct * w)448 static void rtas_event_scan(struct work_struct *w)
449 {
450 unsigned int cpu;
451
452 do_event_scan();
453
454 get_online_cpus();
455
456 /* raw_ OK because just using CPU as starting point. */
457 cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
458 if (cpu >= nr_cpu_ids) {
459 cpu = cpumask_first(cpu_online_mask);
460
461 if (first_pass) {
462 first_pass = 0;
463 event_scan_delay = 30*HZ/rtas_event_scan_rate;
464
465 if (surveillance_timeout != -1) {
466 pr_debug("rtasd: enabling surveillance\n");
467 enable_surveillance(surveillance_timeout);
468 pr_debug("rtasd: surveillance enabled\n");
469 }
470 }
471 }
472
473 schedule_delayed_work_on(cpu, &event_scan_work,
474 __round_jiffies_relative(event_scan_delay, cpu));
475
476 put_online_cpus();
477 }
478
479 #ifdef CONFIG_PPC64
retrieve_nvram_error_log(void)480 static void retrieve_nvram_error_log(void)
481 {
482 unsigned int err_type ;
483 int rc ;
484
485 /* See if we have any error stored in NVRAM */
486 memset(logdata, 0, rtas_error_log_max);
487 rc = nvram_read_error_log(logdata, rtas_error_log_max,
488 &err_type, &error_log_cnt);
489 /* We can use rtas_log_buf now */
490 logging_enabled = 1;
491 if (!rc) {
492 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
493 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
494 }
495 }
496 }
497 #else /* CONFIG_PPC64 */
retrieve_nvram_error_log(void)498 static void retrieve_nvram_error_log(void)
499 {
500 }
501 #endif /* CONFIG_PPC64 */
502
start_event_scan(void)503 static void start_event_scan(void)
504 {
505 printk(KERN_DEBUG "RTAS daemon started\n");
506 pr_debug("rtasd: will sleep for %d milliseconds\n",
507 (30000 / rtas_event_scan_rate));
508
509 /* Retrieve errors from nvram if any */
510 retrieve_nvram_error_log();
511
512 schedule_delayed_work_on(cpumask_first(cpu_online_mask),
513 &event_scan_work, event_scan_delay);
514 }
515
516 /* Cancel the rtas event scan work */
rtas_cancel_event_scan(void)517 void rtas_cancel_event_scan(void)
518 {
519 cancel_delayed_work_sync(&event_scan_work);
520 }
521 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
522
rtas_event_scan_init(void)523 static int __init rtas_event_scan_init(void)
524 {
525 if (!machine_is(pseries) && !machine_is(chrp))
526 return 0;
527
528 /* No RTAS */
529 event_scan = rtas_token("event-scan");
530 if (event_scan == RTAS_UNKNOWN_SERVICE) {
531 printk(KERN_INFO "rtasd: No event-scan on system\n");
532 return -ENODEV;
533 }
534
535 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
536 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
537 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
538 return -ENODEV;
539 }
540
541 if (!rtas_event_scan_rate) {
542 /* Broken firmware: take a rate of zero to mean don't scan */
543 printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
544 return 0;
545 }
546
547 /* Make room for the sequence number */
548 rtas_error_log_max = rtas_get_error_log_max();
549 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
550
551 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
552 if (!rtas_log_buf) {
553 printk(KERN_ERR "rtasd: no memory\n");
554 return -ENOMEM;
555 }
556
557 start_event_scan();
558
559 return 0;
560 }
561 arch_initcall(rtas_event_scan_init);
562
rtas_init(void)563 static int __init rtas_init(void)
564 {
565 struct proc_dir_entry *entry;
566
567 if (!machine_is(pseries) && !machine_is(chrp))
568 return 0;
569
570 if (!rtas_log_buf)
571 return -ENODEV;
572
573 entry = proc_create("powerpc/rtas/error_log", S_IRUSR, NULL,
574 &proc_rtas_log_operations);
575 if (!entry)
576 printk(KERN_ERR "Failed to create error_log proc entry\n");
577
578 return 0;
579 }
580 __initcall(rtas_init);
581
surveillance_setup(char * str)582 static int __init surveillance_setup(char *str)
583 {
584 int i;
585
586 /* We only do surveillance on pseries */
587 if (!machine_is(pseries))
588 return 0;
589
590 if (get_option(&str,&i)) {
591 if (i >= 0 && i <= 255)
592 surveillance_timeout = i;
593 }
594
595 return 1;
596 }
597 __setup("surveillance=", surveillance_setup);
598
rtasmsgs_setup(char * str)599 static int __init rtasmsgs_setup(char *str)
600 {
601 return (kstrtobool(str, &full_rtas_msgs) == 0);
602 }
603 __setup("rtasmsgs=", rtasmsgs_setup);
604