rtasd.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528
  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. #include <linux/types.h>
  12. #include <linux/errno.h>
  13. #include <linux/sched.h>
  14. #include <linux/kernel.h>
  15. #include <linux/poll.h>
  16. #include <linux/proc_fs.h>
  17. #include <linux/init.h>
  18. #include <linux/vmalloc.h>
  19. #include <linux/spinlock.h>
  20. #include <linux/cpu.h>
  21. #include <linux/delay.h>
  22. #include <asm/uaccess.h>
  23. #include <asm/io.h>
  24. #include <asm/rtas.h>
  25. #include <asm/prom.h>
  26. #include <asm/nvram.h>
  27. #include <asm/atomic.h>
  28. #if 0
  29. #define DEBUG(A...) printk(KERN_ERR A)
  30. #else
  31. #define DEBUG(A...)
  32. #endif
  33. static DEFINE_SPINLOCK(rtasd_log_lock);
  34. DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
  35. static char *rtas_log_buf;
  36. static unsigned long rtas_log_start;
  37. static unsigned long rtas_log_size;
  38. static int surveillance_timeout = -1;
  39. static unsigned int rtas_event_scan_rate;
  40. static unsigned int rtas_error_log_max;
  41. static unsigned int rtas_error_log_buffer_max;
  42. static int full_rtas_msgs = 0;
  43. extern int no_logging;
  44. volatile int error_log_cnt = 0;
  45. /*
  46. * Since we use 32 bit RTAS, the physical address of this must be below
  47. * 4G or else bad things happen. Allocate this in the kernel data and
  48. * make it big enough.
  49. */
  50. static unsigned char logdata[RTAS_ERROR_LOG_MAX];
  51. static int get_eventscan_parms(void);
  52. static char *rtas_type[] = {
  53. "Unknown", "Retry", "TCE Error", "Internal Device Failure",
  54. "Timeout", "Data Parity", "Address Parity", "Cache Parity",
  55. "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
  56. };
  57. static char *rtas_event_type(int type)
  58. {
  59. if ((type > 0) && (type < 11))
  60. return rtas_type[type];
  61. switch (type) {
  62. case RTAS_TYPE_EPOW:
  63. return "EPOW";
  64. case RTAS_TYPE_PLATFORM:
  65. return "Platform Error";
  66. case RTAS_TYPE_IO:
  67. return "I/O Event";
  68. case RTAS_TYPE_INFO:
  69. return "Platform Information Event";
  70. case RTAS_TYPE_DEALLOC:
  71. return "Resource Deallocation Event";
  72. case RTAS_TYPE_DUMP:
  73. return "Dump Notification Event";
  74. }
  75. return rtas_type[0];
  76. }
  77. /* To see this info, grep RTAS /var/log/messages and each entry
  78. * will be collected together with obvious begin/end.
  79. * There will be a unique identifier on the begin and end lines.
  80. * This will persist across reboots.
  81. *
  82. * format of error logs returned from RTAS:
  83. * bytes (size) : contents
  84. * --------------------------------------------------------
  85. * 0-7 (8) : rtas_error_log
  86. * 8-47 (40) : extended info
  87. * 48-51 (4) : vendor id
  88. * 52-1023 (vendor specific) : location code and debug data
  89. */
  90. static void printk_log_rtas(char *buf, int len)
  91. {
  92. int i,j,n = 0;
  93. int perline = 16;
  94. char buffer[64];
  95. char * str = "RTAS event";
  96. if (full_rtas_msgs) {
  97. printk(RTAS_DEBUG "%d -------- %s begin --------\n",
  98. error_log_cnt, str);
  99. /*
  100. * Print perline bytes on each line, each line will start
  101. * with RTAS and a changing number, so syslogd will
  102. * print lines that are otherwise the same. Separate every
  103. * 4 bytes with a space.
  104. */
  105. for (i = 0; i < len; i++) {
  106. j = i % perline;
  107. if (j == 0) {
  108. memset(buffer, 0, sizeof(buffer));
  109. n = sprintf(buffer, "RTAS %d:", i/perline);
  110. }
  111. if ((i % 4) == 0)
  112. n += sprintf(buffer+n, " ");
  113. n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
  114. if (j == (perline-1))
  115. printk(KERN_DEBUG "%s\n", buffer);
  116. }
  117. if ((i % perline) != 0)
  118. printk(KERN_DEBUG "%s\n", buffer);
  119. printk(RTAS_DEBUG "%d -------- %s end ----------\n",
  120. error_log_cnt, str);
  121. } else {
  122. struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
  123. printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
  124. error_log_cnt, rtas_event_type(errlog->type),
  125. errlog->severity);
  126. }
  127. }
  128. static int log_rtas_len(char * buf)
  129. {
  130. int len;
  131. struct rtas_error_log *err;
  132. /* rtas fixed header */
  133. len = 8;
  134. err = (struct rtas_error_log *)buf;
  135. if (err->extended_log_length) {
  136. /* extended header */
  137. len += err->extended_log_length;
  138. }
  139. if (rtas_error_log_max == 0) {
  140. get_eventscan_parms();
  141. }
  142. if (len > rtas_error_log_max)
  143. len = rtas_error_log_max;
  144. return len;
  145. }
  146. /*
  147. * First write to nvram, if fatal error, that is the only
  148. * place we log the info. The error will be picked up
  149. * on the next reboot by rtasd. If not fatal, run the
  150. * method for the type of error. Currently, only RTAS
  151. * errors have methods implemented, but in the future
  152. * there might be a need to store data in nvram before a
  153. * call to panic().
  154. *
  155. * XXX We write to nvram periodically, to indicate error has
  156. * been written and sync'd, but there is a possibility
  157. * that if we don't shutdown correctly, a duplicate error
  158. * record will be created on next reboot.
  159. */
  160. void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
  161. {
  162. unsigned long offset;
  163. unsigned long s;
  164. int len = 0;
  165. DEBUG("logging event\n");
  166. if (buf == NULL)
  167. return;
  168. spin_lock_irqsave(&rtasd_log_lock, s);
  169. /* get length and increase count */
  170. switch (err_type & ERR_TYPE_MASK) {
  171. case ERR_TYPE_RTAS_LOG:
  172. len = log_rtas_len(buf);
  173. if (!(err_type & ERR_FLAG_BOOT))
  174. error_log_cnt++;
  175. break;
  176. case ERR_TYPE_KERNEL_PANIC:
  177. default:
  178. spin_unlock_irqrestore(&rtasd_log_lock, s);
  179. return;
  180. }
  181. /* Write error to NVRAM */
  182. if (!no_logging && !(err_type & ERR_FLAG_BOOT))
  183. nvram_write_error_log(buf, len, err_type);
  184. /*
  185. * rtas errors can occur during boot, and we do want to capture
  186. * those somewhere, even if nvram isn't ready (why not?), and even
  187. * if rtasd isn't ready. Put them into the boot log, at least.
  188. */
  189. if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
  190. printk_log_rtas(buf, len);
  191. /* Check to see if we need to or have stopped logging */
  192. if (fatal || no_logging) {
  193. no_logging = 1;
  194. spin_unlock_irqrestore(&rtasd_log_lock, s);
  195. return;
  196. }
  197. /* call type specific method for error */
  198. switch (err_type & ERR_TYPE_MASK) {
  199. case ERR_TYPE_RTAS_LOG:
  200. offset = rtas_error_log_buffer_max *
  201. ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
  202. /* First copy over sequence number */
  203. memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
  204. /* Second copy over error log data */
  205. offset += sizeof(int);
  206. memcpy(&rtas_log_buf[offset], buf, len);
  207. if (rtas_log_size < LOG_NUMBER)
  208. rtas_log_size += 1;
  209. else
  210. rtas_log_start += 1;
  211. spin_unlock_irqrestore(&rtasd_log_lock, s);
  212. wake_up_interruptible(&rtas_log_wait);
  213. break;
  214. case ERR_TYPE_KERNEL_PANIC:
  215. default:
  216. spin_unlock_irqrestore(&rtasd_log_lock, s);
  217. return;
  218. }
  219. }
  220. static int rtas_log_open(struct inode * inode, struct file * file)
  221. {
  222. return 0;
  223. }
  224. static int rtas_log_release(struct inode * inode, struct file * file)
  225. {
  226. return 0;
  227. }
  228. /* This will check if all events are logged, if they are then, we
  229. * know that we can safely clear the events in NVRAM.
  230. * Next we'll sit and wait for something else to log.
  231. */
  232. static ssize_t rtas_log_read(struct file * file, char __user * buf,
  233. size_t count, loff_t *ppos)
  234. {
  235. int error;
  236. char *tmp;
  237. unsigned long s;
  238. unsigned long offset;
  239. if (!buf || count < rtas_error_log_buffer_max)
  240. return -EINVAL;
  241. count = rtas_error_log_buffer_max;
  242. if (!access_ok(VERIFY_WRITE, buf, count))
  243. return -EFAULT;
  244. tmp = kmalloc(count, GFP_KERNEL);
  245. if (!tmp)
  246. return -ENOMEM;
  247. spin_lock_irqsave(&rtasd_log_lock, s);
  248. /* if it's 0, then we know we got the last one (the one in NVRAM) */
  249. if (rtas_log_size == 0 && !no_logging)
  250. nvram_clear_error_log();
  251. spin_unlock_irqrestore(&rtasd_log_lock, s);
  252. error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
  253. if (error)
  254. goto out;
  255. spin_lock_irqsave(&rtasd_log_lock, s);
  256. offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
  257. memcpy(tmp, &rtas_log_buf[offset], count);
  258. rtas_log_start += 1;
  259. rtas_log_size -= 1;
  260. spin_unlock_irqrestore(&rtasd_log_lock, s);
  261. error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
  262. out:
  263. kfree(tmp);
  264. return error;
  265. }
  266. static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
  267. {
  268. poll_wait(file, &rtas_log_wait, wait);
  269. if (rtas_log_size)
  270. return POLLIN | POLLRDNORM;
  271. return 0;
  272. }
  273. struct file_operations proc_rtas_log_operations = {
  274. .read = rtas_log_read,
  275. .poll = rtas_log_poll,
  276. .open = rtas_log_open,
  277. .release = rtas_log_release,
  278. };
  279. static int enable_surveillance(int timeout)
  280. {
  281. int error;
  282. error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
  283. if (error == 0)
  284. return 0;
  285. if (error == -EINVAL) {
  286. printk(KERN_INFO "rtasd: surveillance not supported\n");
  287. return 0;
  288. }
  289. printk(KERN_ERR "rtasd: could not update surveillance\n");
  290. return -1;
  291. }
  292. static int get_eventscan_parms(void)
  293. {
  294. struct device_node *node;
  295. int *ip;
  296. node = of_find_node_by_path("/rtas");
  297. ip = (int *)get_property(node, "rtas-event-scan-rate", NULL);
  298. if (ip == NULL) {
  299. printk(KERN_ERR "rtasd: no rtas-event-scan-rate\n");
  300. of_node_put(node);
  301. return -1;
  302. }
  303. rtas_event_scan_rate = *ip;
  304. DEBUG("rtas-event-scan-rate %d\n", rtas_event_scan_rate);
  305. /* Make room for the sequence number */
  306. rtas_error_log_max = rtas_get_error_log_max();
  307. rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
  308. of_node_put(node);
  309. return 0;
  310. }
  311. static void do_event_scan(int event_scan)
  312. {
  313. int error;
  314. do {
  315. memset(logdata, 0, rtas_error_log_max);
  316. error = rtas_call(event_scan, 4, 1, NULL,
  317. RTAS_EVENT_SCAN_ALL_EVENTS, 0,
  318. __pa(logdata), rtas_error_log_max);
  319. if (error == -1) {
  320. printk(KERN_ERR "event-scan failed\n");
  321. break;
  322. }
  323. if (error == 0)
  324. pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
  325. } while(error == 0);
  326. }
  327. static void do_event_scan_all_cpus(long delay)
  328. {
  329. int cpu;
  330. lock_cpu_hotplug();
  331. cpu = first_cpu(cpu_online_map);
  332. for (;;) {
  333. set_cpus_allowed(current, cpumask_of_cpu(cpu));
  334. do_event_scan(rtas_token("event-scan"));
  335. set_cpus_allowed(current, CPU_MASK_ALL);
  336. /* Drop hotplug lock, and sleep for the specified delay */
  337. unlock_cpu_hotplug();
  338. msleep_interruptible(delay);
  339. lock_cpu_hotplug();
  340. cpu = next_cpu(cpu, cpu_online_map);
  341. if (cpu == NR_CPUS)
  342. break;
  343. }
  344. unlock_cpu_hotplug();
  345. }
  346. static int rtasd(void *unused)
  347. {
  348. unsigned int err_type;
  349. int event_scan = rtas_token("event-scan");
  350. int rc;
  351. daemonize("rtasd");
  352. if (event_scan == RTAS_UNKNOWN_SERVICE || get_eventscan_parms() == -1)
  353. goto error;
  354. rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
  355. if (!rtas_log_buf) {
  356. printk(KERN_ERR "rtasd: no memory\n");
  357. goto error;
  358. }
  359. printk(KERN_INFO "RTAS daemon started\n");
  360. DEBUG("will sleep for %d milliseconds\n", (30000/rtas_event_scan_rate));
  361. /* See if we have any error stored in NVRAM */
  362. memset(logdata, 0, rtas_error_log_max);
  363. rc = nvram_read_error_log(logdata, rtas_error_log_max, &err_type);
  364. /* We can use rtas_log_buf now */
  365. no_logging = 0;
  366. if (!rc) {
  367. if (err_type != ERR_FLAG_ALREADY_LOGGED) {
  368. pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
  369. }
  370. }
  371. /* First pass. */
  372. do_event_scan_all_cpus(1000);
  373. if (surveillance_timeout != -1) {
  374. DEBUG("enabling surveillance\n");
  375. enable_surveillance(surveillance_timeout);
  376. DEBUG("surveillance enabled\n");
  377. }
  378. /* Delay should be at least one second since some
  379. * machines have problems if we call event-scan too
  380. * quickly. */
  381. for (;;)
  382. do_event_scan_all_cpus(30000/rtas_event_scan_rate);
  383. error:
  384. /* Should delete proc entries */
  385. return -EINVAL;
  386. }
  387. static int __init rtas_init(void)
  388. {
  389. struct proc_dir_entry *entry;
  390. if (!platform_is_pseries())
  391. return 0;
  392. /* No RTAS */
  393. if (rtas_token("event-scan") == RTAS_UNKNOWN_SERVICE) {
  394. printk(KERN_INFO "rtasd: no event-scan on system\n");
  395. return 1;
  396. }
  397. entry = create_proc_entry("ppc64/rtas/error_log", S_IRUSR, NULL);
  398. if (entry)
  399. entry->proc_fops = &proc_rtas_log_operations;
  400. else
  401. printk(KERN_ERR "Failed to create error_log proc entry\n");
  402. if (kernel_thread(rtasd, NULL, CLONE_FS) < 0)
  403. printk(KERN_ERR "Failed to start RTAS daemon\n");
  404. return 0;
  405. }
  406. static int __init surveillance_setup(char *str)
  407. {
  408. int i;
  409. if (get_option(&str,&i)) {
  410. if (i >= 0 && i <= 255)
  411. surveillance_timeout = i;
  412. }
  413. return 1;
  414. }
  415. static int __init rtasmsgs_setup(char *str)
  416. {
  417. if (strcmp(str, "on") == 0)
  418. full_rtas_msgs = 1;
  419. else if (strcmp(str, "off") == 0)
  420. full_rtas_msgs = 0;
  421. return 1;
  422. }
  423. __initcall(rtas_init);
  424. __setup("surveillance=", surveillance_setup);
  425. __setup("rtasmsgs=", rtasmsgs_setup);