powerpc: Honor O_NONBLOCK flag when reading RTAS log
[deliverable/linux.git] / arch / powerpc / platforms / pseries / rtasd.c
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/delay.h>
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>
30 #include <asm/machdep.h>
31
32
33 static DEFINE_SPINLOCK(rtasd_log_lock);
34
35 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
36
37 static char *rtas_log_buf;
38 static unsigned long rtas_log_start;
39 static unsigned long rtas_log_size;
40
41 static int surveillance_timeout = -1;
42 static unsigned int rtas_error_log_max;
43 static unsigned int rtas_error_log_buffer_max;
44
45 /* RTAS service tokens */
46 static unsigned int event_scan;
47 static unsigned int rtas_event_scan_rate;
48
49 static int full_rtas_msgs = 0;
50
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;
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 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",
68 };
69
70 static char *rtas_event_type(int type)
71 {
72 if ((type > 0) && (type < 11))
73 return rtas_type[type];
74
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";
88 }
89
90 return rtas_type[0];
91 }
92
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.
97 *
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
105 */
106 static void printk_log_rtas(char *buf, int len)
107 {
108
109 int i,j,n = 0;
110 int perline = 16;
111 char buffer[64];
112 char * str = "RTAS event";
113
114 if (full_rtas_msgs) {
115 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
116 error_log_cnt, str);
117
118 /*
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.
123 */
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);
129 }
130
131 if ((i % 4) == 0)
132 n += sprintf(buffer+n, " ");
133
134 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
135
136 if (j == (perline-1))
137 printk(KERN_DEBUG "%s\n", buffer);
138 }
139 if ((i % perline) != 0)
140 printk(KERN_DEBUG "%s\n", buffer);
141
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;
146
147 printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
148 error_log_cnt, rtas_event_type(errlog->type),
149 errlog->severity);
150 }
151 }
152
153 static int log_rtas_len(char * buf)
154 {
155 int len;
156 struct rtas_error_log *err;
157
158 /* rtas fixed header */
159 len = 8;
160 err = (struct rtas_error_log *)buf;
161 if (err->extended_log_length) {
162
163 /* extended header */
164 len += err->extended_log_length;
165 }
166
167 if (rtas_error_log_max == 0)
168 rtas_error_log_max = rtas_get_error_log_max();
169
170 if (len > rtas_error_log_max)
171 len = rtas_error_log_max;
172
173 return len;
174 }
175
176 /*
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().
184 *
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.
189 */
190 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
191 {
192 unsigned long offset;
193 unsigned long s;
194 int len = 0;
195
196 pr_debug("rtasd: logging event\n");
197 if (buf == NULL)
198 return;
199
200 spin_lock_irqsave(&rtasd_log_lock, s);
201
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 spin_unlock_irqrestore(&rtasd_log_lock, s);
212 return;
213 }
214
215 /* Write error to NVRAM */
216 if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
217 nvram_write_error_log(buf, len, err_type, error_log_cnt);
218
219 /*
220 * rtas errors can occur during boot, and we do want to capture
221 * those somewhere, even if nvram isn't ready (why not?), and even
222 * if rtasd isn't ready. Put them into the boot log, at least.
223 */
224 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
225 printk_log_rtas(buf, len);
226
227 /* Check to see if we need to or have stopped logging */
228 if (fatal || !logging_enabled) {
229 logging_enabled = 0;
230 spin_unlock_irqrestore(&rtasd_log_lock, s);
231 return;
232 }
233
234 /* call type specific method for error */
235 switch (err_type & ERR_TYPE_MASK) {
236 case ERR_TYPE_RTAS_LOG:
237 offset = rtas_error_log_buffer_max *
238 ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
239
240 /* First copy over sequence number */
241 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
242
243 /* Second copy over error log data */
244 offset += sizeof(int);
245 memcpy(&rtas_log_buf[offset], buf, len);
246
247 if (rtas_log_size < LOG_NUMBER)
248 rtas_log_size += 1;
249 else
250 rtas_log_start += 1;
251
252 spin_unlock_irqrestore(&rtasd_log_lock, s);
253 wake_up_interruptible(&rtas_log_wait);
254 break;
255 case ERR_TYPE_KERNEL_PANIC:
256 default:
257 spin_unlock_irqrestore(&rtasd_log_lock, s);
258 return;
259 }
260
261 }
262
263
264 static int rtas_log_open(struct inode * inode, struct file * file)
265 {
266 return 0;
267 }
268
269 static int rtas_log_release(struct inode * inode, struct file * file)
270 {
271 return 0;
272 }
273
274 /* This will check if all events are logged, if they are then, we
275 * know that we can safely clear the events in NVRAM.
276 * Next we'll sit and wait for something else to log.
277 */
278 static ssize_t rtas_log_read(struct file * file, char __user * buf,
279 size_t count, loff_t *ppos)
280 {
281 int error;
282 char *tmp;
283 unsigned long s;
284 unsigned long offset;
285
286 if (!buf || count < rtas_error_log_buffer_max)
287 return -EINVAL;
288
289 count = rtas_error_log_buffer_max;
290
291 if (!access_ok(VERIFY_WRITE, buf, count))
292 return -EFAULT;
293
294 tmp = kmalloc(count, GFP_KERNEL);
295 if (!tmp)
296 return -ENOMEM;
297
298 spin_lock_irqsave(&rtasd_log_lock, s);
299 /* if it's 0, then we know we got the last one (the one in NVRAM) */
300 while (rtas_log_size == 0) {
301 if (file->f_flags & O_NONBLOCK) {
302 spin_unlock_irqrestore(&rtasd_log_lock, s);
303 error = -EAGAIN;
304 goto out;
305 }
306
307 if (!logging_enabled) {
308 spin_unlock_irqrestore(&rtasd_log_lock, s);
309 error = -ENODATA;
310 goto out;
311 }
312 nvram_clear_error_log();
313
314 spin_unlock_irqrestore(&rtasd_log_lock, s);
315 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
316 if (error)
317 goto out;
318 spin_lock_irqsave(&rtasd_log_lock, s);
319 }
320
321 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
322 memcpy(tmp, &rtas_log_buf[offset], count);
323
324 rtas_log_start += 1;
325 rtas_log_size -= 1;
326 spin_unlock_irqrestore(&rtasd_log_lock, s);
327
328 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
329 out:
330 kfree(tmp);
331 return error;
332 }
333
334 static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
335 {
336 poll_wait(file, &rtas_log_wait, wait);
337 if (rtas_log_size)
338 return POLLIN | POLLRDNORM;
339 return 0;
340 }
341
342 static const struct file_operations proc_rtas_log_operations = {
343 .read = rtas_log_read,
344 .poll = rtas_log_poll,
345 .open = rtas_log_open,
346 .release = rtas_log_release,
347 };
348
349 static int enable_surveillance(int timeout)
350 {
351 int error;
352
353 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
354
355 if (error == 0)
356 return 0;
357
358 if (error == -EINVAL) {
359 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
360 return 0;
361 }
362
363 printk(KERN_ERR "rtasd: could not update surveillance\n");
364 return -1;
365 }
366
367 static void do_event_scan(void)
368 {
369 int error;
370 do {
371 memset(logdata, 0, rtas_error_log_max);
372 error = rtas_call(event_scan, 4, 1, NULL,
373 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
374 __pa(logdata), rtas_error_log_max);
375 if (error == -1) {
376 printk(KERN_ERR "event-scan failed\n");
377 break;
378 }
379
380 if (error == 0)
381 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
382
383 } while(error == 0);
384 }
385
386 static void do_event_scan_all_cpus(long delay)
387 {
388 int cpu;
389
390 get_online_cpus();
391 cpu = first_cpu(cpu_online_map);
392 for (;;) {
393 set_cpus_allowed(current, cpumask_of_cpu(cpu));
394 do_event_scan();
395 set_cpus_allowed(current, CPU_MASK_ALL);
396
397 /* Drop hotplug lock, and sleep for the specified delay */
398 put_online_cpus();
399 msleep_interruptible(delay);
400 get_online_cpus();
401
402 cpu = next_cpu(cpu, cpu_online_map);
403 if (cpu == NR_CPUS)
404 break;
405 }
406 put_online_cpus();
407 }
408
409 static int rtasd(void *unused)
410 {
411 unsigned int err_type;
412 int rc;
413
414 daemonize("rtasd");
415
416 printk(KERN_DEBUG "RTAS daemon started\n");
417 pr_debug("rtasd: will sleep for %d milliseconds\n",
418 (30000 / rtas_event_scan_rate));
419
420 /* See if we have any error stored in NVRAM */
421 memset(logdata, 0, rtas_error_log_max);
422 rc = nvram_read_error_log(logdata, rtas_error_log_max,
423 &err_type, &error_log_cnt);
424 /* We can use rtas_log_buf now */
425 logging_enabled = 1;
426
427 if (!rc) {
428 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
429 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
430 }
431 }
432
433 /* First pass. */
434 do_event_scan_all_cpus(1000);
435
436 if (surveillance_timeout != -1) {
437 pr_debug("rtasd: enabling surveillance\n");
438 enable_surveillance(surveillance_timeout);
439 pr_debug("rtasd: surveillance enabled\n");
440 }
441
442 /* Delay should be at least one second since some
443 * machines have problems if we call event-scan too
444 * quickly. */
445 for (;;)
446 do_event_scan_all_cpus(30000/rtas_event_scan_rate);
447
448 return -EINVAL;
449 }
450
451 static int __init rtas_init(void)
452 {
453 struct proc_dir_entry *entry;
454
455 if (!machine_is(pseries))
456 return 0;
457
458 /* No RTAS */
459 event_scan = rtas_token("event-scan");
460 if (event_scan == RTAS_UNKNOWN_SERVICE) {
461 printk(KERN_DEBUG "rtasd: no event-scan on system\n");
462 return -ENODEV;
463 }
464
465 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
466 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
467 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
468 return -ENODEV;
469 }
470
471 /* Make room for the sequence number */
472 rtas_error_log_max = rtas_get_error_log_max();
473 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
474
475 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
476 if (!rtas_log_buf) {
477 printk(KERN_ERR "rtasd: no memory\n");
478 return -ENOMEM;
479 }
480
481 entry = proc_create("ppc64/rtas/error_log", S_IRUSR, NULL,
482 &proc_rtas_log_operations);
483 if (!entry)
484 printk(KERN_ERR "Failed to create error_log proc entry\n");
485
486 if (kernel_thread(rtasd, NULL, CLONE_FS) < 0)
487 printk(KERN_ERR "Failed to start RTAS daemon\n");
488
489 return 0;
490 }
491
492 static int __init surveillance_setup(char *str)
493 {
494 int i;
495
496 if (get_option(&str,&i)) {
497 if (i >= 0 && i <= 255)
498 surveillance_timeout = i;
499 }
500
501 return 1;
502 }
503
504 static int __init rtasmsgs_setup(char *str)
505 {
506 if (strcmp(str, "on") == 0)
507 full_rtas_msgs = 1;
508 else if (strcmp(str, "off") == 0)
509 full_rtas_msgs = 0;
510
511 return 1;
512 }
513 __initcall(rtas_init);
514 __setup("surveillance=", surveillance_setup);
515 __setup("rtasmsgs=", rtasmsgs_setup);
This page took 0.069988 seconds and 5 git commands to generate.