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>
23 #include <asm/uaccess.h>
27 #include <asm/nvram.h>
28 #include <asm/atomic.h>
31 #define DEBUG(A...) printk(KERN_ERR A)
36 static spinlock_t rtasd_log_lock = SPIN_LOCK_UNLOCKED;
38 DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
40 static char *rtas_log_buf;
41 static unsigned long rtas_log_start;
42 static unsigned long rtas_log_size;
44 static int surveillance_timeout = -1;
45 static unsigned int rtas_event_scan_rate;
46 static unsigned int rtas_error_log_max;
47 static unsigned int rtas_error_log_buffer_max;
49 static int full_rtas_msgs = 0;
51 extern volatile int no_more_logging;
53 volatile int error_log_cnt = 0;
56 * Since we use 32 bit RTAS, the physical address of this must be below
57 * 4G or else bad things happen. Allocate this in the kernel data and
60 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
62 static int get_eventscan_parms(void);
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];
78 case RTAS_TYPE_PLATFORM:
79 return "Platform Error";
83 return "Platform Information Event";
84 case RTAS_TYPE_DEALLOC:
85 return "Resource Deallocation Event";
87 return "Dump Notification Event";
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)
112 char * str = "RTAS event";
114 if (full_rtas_msgs) {
115 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
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++) {
127 memset(buffer, 0, sizeof(buffer));
128 n = sprintf(buffer, "RTAS %d:", i/perline);
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",
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),
153 static int log_rtas_len(char * buf)
156 struct rtas_error_log *err;
158 /* rtas fixed header */
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 get_eventscan_parms();
170 if (len > rtas_error_log_max)
171 len = rtas_error_log_max;
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
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;
196 DEBUG("logging event\n");
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))
209 case ERR_TYPE_KERNEL_PANIC:
211 spin_unlock_irqrestore(&rtasd_log_lock, s);
215 /* Write error to NVRAM */
216 if (!no_more_logging && !(err_type & ERR_FLAG_BOOT))
217 nvram_write_error_log(buf, len, err_type);
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.
224 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
225 printk_log_rtas(buf, len);
227 /* Check to see if we need to or have stopped logging */
228 if (fatal || no_more_logging) {
230 spin_unlock_irqrestore(&rtasd_log_lock, s);
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);
240 /* First copy over sequence number */
241 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
243 /* Second copy over error log data */
244 offset += sizeof(int);
245 memcpy(&rtas_log_buf[offset], buf, len);
247 if (rtas_log_size < LOG_NUMBER)
252 spin_unlock_irqrestore(&rtasd_log_lock, s);
253 wake_up_interruptible(&rtas_log_wait);
255 case ERR_TYPE_KERNEL_PANIC:
257 spin_unlock_irqrestore(&rtasd_log_lock, s);
264 static int rtas_log_open(struct inode * inode, struct file * file)
269 static int rtas_log_release(struct inode * inode, struct file * file)
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.
278 static ssize_t rtas_log_read(struct file * file, char * buf,
279 size_t count, loff_t *ppos)
284 unsigned long offset;
286 if (!buf || count < rtas_error_log_buffer_max)
289 count = rtas_error_log_buffer_max;
291 error = verify_area(VERIFY_WRITE, buf, count);
295 tmp = kmalloc(count, GFP_KERNEL);
300 spin_lock_irqsave(&rtasd_log_lock, s);
301 /* if it's 0, then we know we got the last one (the one in NVRAM) */
302 if (rtas_log_size == 0 && !no_more_logging)
303 nvram_clear_error_log();
304 spin_unlock_irqrestore(&rtasd_log_lock, s);
307 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
311 spin_lock_irqsave(&rtasd_log_lock, s);
312 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
313 memcpy(tmp, &rtas_log_buf[offset], count);
317 spin_unlock_irqrestore(&rtasd_log_lock, s);
319 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
325 static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
327 poll_wait(file, &rtas_log_wait, wait);
329 return POLLIN | POLLRDNORM;
333 struct file_operations proc_rtas_log_operations = {
334 .read = rtas_log_read,
335 .poll = rtas_log_poll,
336 .open = rtas_log_open,
337 .release = rtas_log_release,
340 static int enable_surveillance(int timeout)
344 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
349 if (error == RTAS_NO_SUCH_INDICATOR) {
350 printk(KERN_INFO "rtasd: surveillance not supported\n");
354 printk(KERN_ERR "rtasd: could not update surveillance\n");
358 static int get_eventscan_parms(void)
360 struct device_node *node;
363 node = of_find_node_by_path("/rtas");
365 ip = (int *)get_property(node, "rtas-event-scan-rate", NULL);
367 printk(KERN_ERR "rtasd: no rtas-event-scan-rate\n");
371 rtas_event_scan_rate = *ip;
372 DEBUG("rtas-event-scan-rate %d\n", rtas_event_scan_rate);
374 /* Make room for the sequence number */
375 rtas_error_log_max = rtas_get_error_log_max();
376 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
383 static void do_event_scan(int event_scan)
387 memset(logdata, 0, rtas_error_log_max);
388 error = rtas_call(event_scan, 4, 1, NULL,
389 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
390 __pa(logdata), rtas_error_log_max);
392 printk(KERN_ERR "event-scan failed\n");
397 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
402 static int rtasd(void *unused)
404 unsigned int err_type;
406 int event_scan = rtas_token("event-scan");
411 if (event_scan == RTAS_UNKNOWN_SERVICE || get_eventscan_parms() == -1)
414 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
416 printk(KERN_ERR "rtasd: no memory\n");
420 /* We can use rtas_log_buf now */
423 printk(KERN_ERR "RTAS daemon started\n");
425 DEBUG("will sleep for %d jiffies\n", (HZ*60/rtas_event_scan_rate) / 2);
427 /* See if we have any error stored in NVRAM */
428 memset(logdata, 0, rtas_error_log_max);
430 rc = nvram_read_error_log(logdata, rtas_error_log_max, &err_type);
432 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
433 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
439 for_each_online_cpu(cpu) {
440 DEBUG("scheduling on %d\n", cpu);
441 set_cpus_allowed(current, cpumask_of_cpu(cpu));
442 DEBUG("watchdog scheduled on cpu %d\n", smp_processor_id());
444 do_event_scan(event_scan);
445 set_current_state(TASK_INTERRUPTIBLE);
446 schedule_timeout(HZ);
448 unlock_cpu_hotplug();
450 if (surveillance_timeout != -1) {
451 DEBUG("enabling surveillance\n");
452 enable_surveillance(surveillance_timeout);
453 DEBUG("surveillance enabled\n");
457 cpu = first_cpu(cpu_online_map);
459 set_cpus_allowed(current, cpumask_of_cpu(cpu));
460 do_event_scan(event_scan);
461 set_cpus_allowed(current, CPU_MASK_ALL);
463 /* Drop hotplug lock, and sleep for a bit (at least
464 * one second since some machines have problems if we
465 * call event-scan too quickly). */
466 unlock_cpu_hotplug();
467 set_current_state(TASK_INTERRUPTIBLE);
468 schedule_timeout((HZ*60/rtas_event_scan_rate) / 2);
471 cpu = next_cpu(cpu, cpu_online_map);
473 cpu = first_cpu(cpu_online_map);
477 /* Should delete proc entries */
481 static int __init rtas_init(void)
483 struct proc_dir_entry *entry;
485 /* No RTAS, only warn if we are on a pSeries box */
486 if (rtas_token("event-scan") == RTAS_UNKNOWN_SERVICE) {
487 if (systemcfg->platform & PLATFORM_PSERIES);
488 printk(KERN_ERR "rtasd: no RTAS on system\n");
492 entry = create_proc_entry("ppc64/rtas/error_log", S_IRUSR, NULL);
494 entry->proc_fops = &proc_rtas_log_operations;
496 printk(KERN_ERR "Failed to create error_log proc entry\n");
498 if (kernel_thread(rtasd, NULL, CLONE_FS) < 0)
499 printk(KERN_ERR "Failed to start RTAS daemon\n");
504 static int __init surveillance_setup(char *str)
508 if (get_option(&str,&i)) {
509 if (i >= 0 && i <= 255)
510 surveillance_timeout = i;
516 static int __init rtasmsgs_setup(char *str)
518 if (strcmp(str, "on") == 0)
520 else if (strcmp(str, "off") == 0)
525 __initcall(rtas_init);
526 __setup("surveillance=", surveillance_setup);
527 __setup("rtasmsgs=", rtasmsgs_setup);