added 2.6.29.6 aldebaran kernel
[nao-ulib.git] / kernel / 2.6.29.6-aldebaran-rt / arch / powerpc / platforms / pseries / rtasd.c
blob1d49676598632360f6254374741a9a04709e0018
1 /*
2 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
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.
9 * Communication to userspace based on kernel/printk.c
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/delay.h>
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>
30 #include <asm/machdep.h>
33 static DEFINE_SPINLOCK(rtasd_log_lock);
35 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
37 static char *rtas_log_buf;
38 static unsigned long rtas_log_start;
39 static unsigned long rtas_log_size;
41 static int surveillance_timeout = -1;
42 static unsigned int rtas_error_log_max;
43 static unsigned int rtas_error_log_buffer_max;
45 /* RTAS service tokens */
46 static unsigned int event_scan;
47 static unsigned int rtas_event_scan_rate;
49 static int full_rtas_msgs = 0;
51 /* Stop logging to nvram after first fatal error */
52 static int logging_enabled; /* Until we initialize everything,
53 * make sure we don't try logging
54 * anything */
55 static int error_log_cnt;
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.
62 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
64 static char *rtas_type[] = {
65 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
66 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
67 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
70 static char *rtas_event_type(int type)
72 if ((type > 0) && (type < 11))
73 return rtas_type[type];
75 switch (type) {
76 case RTAS_TYPE_EPOW:
77 return "EPOW";
78 case RTAS_TYPE_PLATFORM:
79 return "Platform Error";
80 case RTAS_TYPE_IO:
81 return "I/O Event";
82 case RTAS_TYPE_INFO:
83 return "Platform Information Event";
84 case RTAS_TYPE_DEALLOC:
85 return "Resource Deallocation Event";
86 case RTAS_TYPE_DUMP:
87 return "Dump Notification Event";
90 return rtas_type[0];
93 /* To see this info, grep RTAS /var/log/messages and each entry
94 * will be collected together with obvious begin/end.
95 * There will be a unique identifier on the begin and end lines.
96 * This will persist across reboots.
98 * format of error logs returned from RTAS:
99 * bytes (size) : contents
100 * --------------------------------------------------------
101 * 0-7 (8) : rtas_error_log
102 * 8-47 (40) : extended info
103 * 48-51 (4) : vendor id
104 * 52-1023 (vendor specific) : location code and debug data
106 static void printk_log_rtas(char *buf, int len)
109 int i,j,n = 0;
110 int perline = 16;
111 char buffer[64];
112 char * str = "RTAS event";
114 if (full_rtas_msgs) {
115 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
116 error_log_cnt, str);
119 * Print perline bytes on each line, each line will start
120 * with RTAS and a changing number, so syslogd will
121 * print lines that are otherwise the same. Separate every
122 * 4 bytes with a space.
124 for (i = 0; i < len; i++) {
125 j = i % perline;
126 if (j == 0) {
127 memset(buffer, 0, sizeof(buffer));
128 n = sprintf(buffer, "RTAS %d:", i/perline);
131 if ((i % 4) == 0)
132 n += sprintf(buffer+n, " ");
134 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
136 if (j == (perline-1))
137 printk(KERN_DEBUG "%s\n", buffer);
139 if ((i % perline) != 0)
140 printk(KERN_DEBUG "%s\n", buffer);
142 printk(RTAS_DEBUG "%d -------- %s end ----------\n",
143 error_log_cnt, str);
144 } else {
145 struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
147 printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
148 error_log_cnt, rtas_event_type(errlog->type),
149 errlog->severity);
153 static int log_rtas_len(char * buf)
155 int len;
156 struct rtas_error_log *err;
158 /* rtas fixed header */
159 len = 8;
160 err = (struct rtas_error_log *)buf;
161 if (err->extended_log_length) {
163 /* extended header */
164 len += err->extended_log_length;
167 if (rtas_error_log_max == 0)
168 rtas_error_log_max = rtas_get_error_log_max();
170 if (len > rtas_error_log_max)
171 len = rtas_error_log_max;
173 return len;
177 * First write to nvram, if fatal error, that is the only
178 * place we log the info. The error will be picked up
179 * on the next reboot by rtasd. If not fatal, run the
180 * method for the type of error. Currently, only RTAS
181 * errors have methods implemented, but in the future
182 * there might be a need to store data in nvram before a
183 * call to panic().
185 * XXX We write to nvram periodically, to indicate error has
186 * been written and sync'd, but there is a possibility
187 * that if we don't shutdown correctly, a duplicate error
188 * record will be created on next reboot.
190 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
192 unsigned long offset;
193 unsigned long s;
194 int len = 0;
196 pr_debug("rtasd: logging event\n");
197 if (buf == NULL)
198 return;
200 spin_lock_irqsave(&rtasd_log_lock, s);
202 /* get length and increase count */
203 switch (err_type & ERR_TYPE_MASK) {
204 case ERR_TYPE_RTAS_LOG:
205 len = log_rtas_len(buf);
206 if (!(err_type & ERR_FLAG_BOOT))
207 error_log_cnt++;
208 break;
209 case ERR_TYPE_KERNEL_PANIC:
210 default:
211 WARN_ON_ONCE_NONRT(!irqs_disabled()); /* @@@ DEBUG @@@ */
212 spin_unlock_irqrestore(&rtasd_log_lock, s);
213 return;
216 /* Write error to NVRAM */
217 if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
218 nvram_write_error_log(buf, len, err_type, error_log_cnt);
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.
225 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
226 printk_log_rtas(buf, len);
228 /* Check to see if we need to or have stopped logging */
229 if (fatal || !logging_enabled) {
230 logging_enabled = 0;
231 WARN_ON_ONCE_NONRT(!irqs_disabled()); /* @@@ DEBUG @@@ */
232 spin_unlock_irqrestore(&rtasd_log_lock, s);
233 return;
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);
242 /* First copy over sequence number */
243 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
245 /* Second copy over error log data */
246 offset += sizeof(int);
247 memcpy(&rtas_log_buf[offset], buf, len);
249 if (rtas_log_size < LOG_NUMBER)
250 rtas_log_size += 1;
251 else
252 rtas_log_start += 1;
254 WARN_ON_ONCE_NONRT(!irqs_disabled()); /* @@@ DEBUG @@@ */
255 spin_unlock_irqrestore(&rtasd_log_lock, s);
256 wake_up_interruptible(&rtas_log_wait);
257 break;
258 case ERR_TYPE_KERNEL_PANIC:
259 default:
260 WARN_ON_ONCE_NONRT(!irqs_disabled()); /* @@@ DEBUG @@@ */
261 spin_unlock_irqrestore(&rtasd_log_lock, s);
262 return;
268 static int rtas_log_open(struct inode * inode, struct file * file)
270 return 0;
273 static int rtas_log_release(struct inode * inode, struct file * file)
275 return 0;
278 /* This will check if all events are logged, if they are then, we
279 * know that we can safely clear the events in NVRAM.
280 * Next we'll sit and wait for something else to log.
282 static ssize_t rtas_log_read(struct file * file, char __user * buf,
283 size_t count, loff_t *ppos)
285 int error;
286 char *tmp;
287 unsigned long s;
288 unsigned long offset;
290 if (!buf || count < rtas_error_log_buffer_max)
291 return -EINVAL;
293 count = rtas_error_log_buffer_max;
295 if (!access_ok(VERIFY_WRITE, buf, count))
296 return -EFAULT;
298 tmp = kmalloc(count, GFP_KERNEL);
299 if (!tmp)
300 return -ENOMEM;
302 spin_lock_irqsave(&rtasd_log_lock, s);
303 /* if it's 0, then we know we got the last one (the one in NVRAM) */
304 while (rtas_log_size == 0) {
305 if (file->f_flags & O_NONBLOCK) {
306 spin_unlock_irqrestore(&rtasd_log_lock, s);
307 error = -EAGAIN;
308 goto out;
311 if (!logging_enabled) {
312 spin_unlock_irqrestore(&rtasd_log_lock, s);
313 error = -ENODATA;
314 goto out;
316 nvram_clear_error_log();
318 spin_unlock_irqrestore(&rtasd_log_lock, s);
319 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
320 if (error)
321 goto out;
322 spin_lock_irqsave(&rtasd_log_lock, s);
325 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
326 memcpy(tmp, &rtas_log_buf[offset], count);
328 rtas_log_start += 1;
329 rtas_log_size -= 1;
330 spin_unlock_irqrestore(&rtasd_log_lock, s);
332 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
333 out:
334 kfree(tmp);
335 return error;
338 static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
340 poll_wait(file, &rtas_log_wait, wait);
341 if (rtas_log_size)
342 return POLLIN | POLLRDNORM;
343 return 0;
346 static const struct file_operations proc_rtas_log_operations = {
347 .read = rtas_log_read,
348 .poll = rtas_log_poll,
349 .open = rtas_log_open,
350 .release = rtas_log_release,
353 static int enable_surveillance(int timeout)
355 int error;
357 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
359 if (error == 0)
360 return 0;
362 if (error == -EINVAL) {
363 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
364 return 0;
367 printk(KERN_ERR "rtasd: could not update surveillance\n");
368 return -1;
371 static void do_event_scan(void)
373 int error;
374 do {
375 memset(logdata, 0, rtas_error_log_max);
376 error = rtas_call(event_scan, 4, 1, NULL,
377 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
378 __pa(logdata), rtas_error_log_max);
379 if (error == -1) {
380 printk(KERN_ERR "event-scan failed\n");
381 break;
384 if (error == 0)
385 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
387 } while(error == 0);
390 static void do_event_scan_all_cpus(long delay)
392 int cpu;
394 get_online_cpus();
395 cpu = first_cpu(cpu_online_map);
396 for (;;) {
397 set_cpus_allowed(current, cpumask_of_cpu(cpu));
398 do_event_scan();
399 set_cpus_allowed(current, CPU_MASK_ALL);
401 /* Drop hotplug lock, and sleep for the specified delay */
402 put_online_cpus();
403 msleep_interruptible(delay);
404 get_online_cpus();
406 cpu = next_cpu(cpu, cpu_online_map);
407 if (cpu == NR_CPUS)
408 break;
410 put_online_cpus();
413 static int rtasd(void *unused)
415 unsigned int err_type;
416 int rc;
418 daemonize("rtasd");
420 printk(KERN_DEBUG "RTAS daemon started\n");
421 pr_debug("rtasd: will sleep for %d milliseconds\n",
422 (30000 / rtas_event_scan_rate));
424 /* See if we have any error stored in NVRAM */
425 memset(logdata, 0, rtas_error_log_max);
426 rc = nvram_read_error_log(logdata, rtas_error_log_max,
427 &err_type, &error_log_cnt);
428 /* We can use rtas_log_buf now */
429 logging_enabled = 1;
431 if (!rc) {
432 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
433 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
437 /* First pass. */
438 do_event_scan_all_cpus(1000);
440 if (surveillance_timeout != -1) {
441 pr_debug("rtasd: enabling surveillance\n");
442 enable_surveillance(surveillance_timeout);
443 pr_debug("rtasd: surveillance enabled\n");
446 /* Delay should be at least one second since some
447 * machines have problems if we call event-scan too
448 * quickly. */
449 for (;;)
450 do_event_scan_all_cpus(30000/rtas_event_scan_rate);
452 return -EINVAL;
455 static int __init rtas_init(void)
457 struct proc_dir_entry *entry;
459 if (!machine_is(pseries))
460 return 0;
462 /* No RTAS */
463 event_scan = rtas_token("event-scan");
464 if (event_scan == RTAS_UNKNOWN_SERVICE) {
465 printk(KERN_DEBUG "rtasd: no event-scan on system\n");
466 return -ENODEV;
469 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
470 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
471 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
472 return -ENODEV;
475 /* Make room for the sequence number */
476 rtas_error_log_max = rtas_get_error_log_max();
477 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
479 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
480 if (!rtas_log_buf) {
481 printk(KERN_ERR "rtasd: no memory\n");
482 return -ENOMEM;
485 entry = proc_create("ppc64/rtas/error_log", S_IRUSR, NULL,
486 &proc_rtas_log_operations);
487 if (!entry)
488 printk(KERN_ERR "Failed to create error_log proc entry\n");
490 if (kernel_thread(rtasd, NULL, CLONE_FS) < 0)
491 printk(KERN_ERR "Failed to start RTAS daemon\n");
493 return 0;
496 static int __init surveillance_setup(char *str)
498 int i;
500 if (get_option(&str,&i)) {
501 if (i >= 0 && i <= 255)
502 surveillance_timeout = i;
505 return 1;
508 static int __init rtasmsgs_setup(char *str)
510 if (strcmp(str, "on") == 0)
511 full_rtas_msgs = 1;
512 else if (strcmp(str, "off") == 0)
513 full_rtas_msgs = 0;
515 return 1;
517 __initcall(rtas_init);
518 __setup("surveillance=", surveillance_setup);
519 __setup("rtasmsgs=", rtasmsgs_setup);