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>
29 #include <asm/systemcfg.h>
32 #define DEBUG(A...) printk(KERN_ERR A)
37 static DEFINE_SPINLOCK(rtasd_log_lock);
39 DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
41 static char *rtas_log_buf;
42 static unsigned long rtas_log_start;
43 static unsigned long rtas_log_size;
45 static int surveillance_timeout = -1;
46 static unsigned int rtas_event_scan_rate;
47 static unsigned int rtas_error_log_max;
48 static unsigned int rtas_error_log_buffer_max;
50 static int full_rtas_msgs = 0;
52 extern int no_logging;
54 volatile int error_log_cnt = 0;
57 * Since we use 32 bit RTAS, the physical address of this must be below
58 * 4G or else bad things happen. Allocate this in the kernel data and
61 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
63 static int get_eventscan_parms(void);
65 static char *rtas_type[] = {
66 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
67 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
68 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
71 static char *rtas_event_type(int type)
73 if ((type > 0) && (type < 11))
74 return rtas_type[type];
79 case RTAS_TYPE_PLATFORM:
80 return "Platform Error";
84 return "Platform Information Event";
85 case RTAS_TYPE_DEALLOC:
86 return "Resource Deallocation Event";
88 return "Dump Notification Event";
94 /* To see this info, grep RTAS /var/log/messages and each entry
95 * will be collected together with obvious begin/end.
96 * There will be a unique identifier on the begin and end lines.
97 * This will persist across reboots.
99 * format of error logs returned from RTAS:
100 * bytes (size) : contents
101 * --------------------------------------------------------
102 * 0-7 (8) : rtas_error_log
103 * 8-47 (40) : extended info
104 * 48-51 (4) : vendor id
105 * 52-1023 (vendor specific) : location code and debug data
107 static void printk_log_rtas(char *buf, int len)
113 char * str = "RTAS event";
115 if (full_rtas_msgs) {
116 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
120 * Print perline bytes on each line, each line will start
121 * with RTAS and a changing number, so syslogd will
122 * print lines that are otherwise the same. Separate every
123 * 4 bytes with a space.
125 for (i = 0; i < len; i++) {
128 memset(buffer, 0, sizeof(buffer));
129 n = sprintf(buffer, "RTAS %d:", i/perline);
133 n += sprintf(buffer+n, " ");
135 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
137 if (j == (perline-1))
138 printk(KERN_DEBUG "%s\n", buffer);
140 if ((i % perline) != 0)
141 printk(KERN_DEBUG "%s\n", buffer);
143 printk(RTAS_DEBUG "%d -------- %s end ----------\n",
146 struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
148 printk(RTAS_DEBUG "event: %d, Type: %s, Severity: %d\n",
149 error_log_cnt, rtas_event_type(errlog->type),
154 static int log_rtas_len(char * buf)
157 struct rtas_error_log *err;
159 /* rtas fixed header */
161 err = (struct rtas_error_log *)buf;
162 if (err->extended_log_length) {
164 /* extended header */
165 len += err->extended_log_length;
168 if (rtas_error_log_max == 0) {
169 get_eventscan_parms();
171 if (len > rtas_error_log_max)
172 len = rtas_error_log_max;
178 * First write to nvram, if fatal error, that is the only
179 * place we log the info. The error will be picked up
180 * on the next reboot by rtasd. If not fatal, run the
181 * method for the type of error. Currently, only RTAS
182 * errors have methods implemented, but in the future
183 * there might be a need to store data in nvram before a
186 * XXX We write to nvram periodically, to indicate error has
187 * been written and sync'd, but there is a possibility
188 * that if we don't shutdown correctly, a duplicate error
189 * record will be created on next reboot.
191 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
193 unsigned long offset;
197 DEBUG("logging event\n");
201 spin_lock_irqsave(&rtasd_log_lock, s);
203 /* get length and increase count */
204 switch (err_type & ERR_TYPE_MASK) {
205 case ERR_TYPE_RTAS_LOG:
206 len = log_rtas_len(buf);
207 if (!(err_type & ERR_FLAG_BOOT))
210 case ERR_TYPE_KERNEL_PANIC:
212 spin_unlock_irqrestore(&rtasd_log_lock, s);
216 /* Write error to NVRAM */
217 if (!no_logging && !(err_type & ERR_FLAG_BOOT))
218 nvram_write_error_log(buf, len, err_type);
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 || no_logging) {
231 spin_unlock_irqrestore(&rtasd_log_lock, s);
235 /* call type specific method for error */
236 switch (err_type & ERR_TYPE_MASK) {
237 case ERR_TYPE_RTAS_LOG:
238 offset = rtas_error_log_buffer_max *
239 ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
241 /* First copy over sequence number */
242 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
244 /* Second copy over error log data */
245 offset += sizeof(int);
246 memcpy(&rtas_log_buf[offset], buf, len);
248 if (rtas_log_size < LOG_NUMBER)
253 spin_unlock_irqrestore(&rtasd_log_lock, s);
254 wake_up_interruptible(&rtas_log_wait);
256 case ERR_TYPE_KERNEL_PANIC:
258 spin_unlock_irqrestore(&rtasd_log_lock, s);
265 static int rtas_log_open(struct inode * inode, struct file * file)
270 static int rtas_log_release(struct inode * inode, struct file * file)
275 /* This will check if all events are logged, if they are then, we
276 * know that we can safely clear the events in NVRAM.
277 * Next we'll sit and wait for something else to log.
279 static ssize_t rtas_log_read(struct file * file, char __user * buf,
280 size_t count, loff_t *ppos)
285 unsigned long offset;
287 if (!buf || count < rtas_error_log_buffer_max)
290 count = rtas_error_log_buffer_max;
292 error = verify_area(VERIFY_WRITE, buf, count);
296 tmp = kmalloc(count, GFP_KERNEL);
301 spin_lock_irqsave(&rtasd_log_lock, s);
302 /* if it's 0, then we know we got the last one (the one in NVRAM) */
303 if (rtas_log_size == 0 && !no_logging)
304 nvram_clear_error_log();
305 spin_unlock_irqrestore(&rtasd_log_lock, s);
308 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
312 spin_lock_irqsave(&rtasd_log_lock, s);
313 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
314 memcpy(tmp, &rtas_log_buf[offset], count);
318 spin_unlock_irqrestore(&rtasd_log_lock, s);
320 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
326 static unsigned int rtas_log_poll(struct file *file, poll_table * wait)
328 poll_wait(file, &rtas_log_wait, wait);
330 return POLLIN | POLLRDNORM;
334 struct file_operations proc_rtas_log_operations = {
335 .read = rtas_log_read,
336 .poll = rtas_log_poll,
337 .open = rtas_log_open,
338 .release = rtas_log_release,
341 static int enable_surveillance(int timeout)
345 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
350 if (error == RTAS_NO_SUCH_INDICATOR) {
351 printk(KERN_INFO "rtasd: surveillance not supported\n");
355 printk(KERN_ERR "rtasd: could not update surveillance\n");
359 static int get_eventscan_parms(void)
361 struct device_node *node;
364 node = of_find_node_by_path("/rtas");
366 ip = (int *)get_property(node, "rtas-event-scan-rate", NULL);
368 printk(KERN_ERR "rtasd: no rtas-event-scan-rate\n");
372 rtas_event_scan_rate = *ip;
373 DEBUG("rtas-event-scan-rate %d\n", rtas_event_scan_rate);
375 /* Make room for the sequence number */
376 rtas_error_log_max = rtas_get_error_log_max();
377 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
384 static void do_event_scan(int event_scan)
388 memset(logdata, 0, rtas_error_log_max);
389 error = rtas_call(event_scan, 4, 1, NULL,
390 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
391 __pa(logdata), rtas_error_log_max);
393 printk(KERN_ERR "event-scan failed\n");
398 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG, 0);
403 static int rtasd(void *unused)
405 unsigned int err_type;
407 int event_scan = rtas_token("event-scan");
412 if (event_scan == RTAS_UNKNOWN_SERVICE || get_eventscan_parms() == -1)
415 rtas_log_buf = vmalloc(rtas_error_log_buffer_max*LOG_NUMBER);
417 printk(KERN_ERR "rtasd: no memory\n");
421 printk(KERN_ERR "RTAS daemon started\n");
423 DEBUG("will sleep for %d jiffies\n", (HZ*60/rtas_event_scan_rate) / 2);
425 /* See if we have any error stored in NVRAM */
426 memset(logdata, 0, rtas_error_log_max);
428 rc = nvram_read_error_log(logdata, rtas_error_log_max, &err_type);
430 /* We can use rtas_log_buf now */
434 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
435 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
441 for_each_online_cpu(cpu) {
442 DEBUG("scheduling on %d\n", cpu);
443 set_cpus_allowed(current, cpumask_of_cpu(cpu));
444 DEBUG("watchdog scheduled on cpu %d\n", smp_processor_id());
446 do_event_scan(event_scan);
447 set_current_state(TASK_INTERRUPTIBLE);
448 schedule_timeout(HZ);
450 unlock_cpu_hotplug();
452 if (surveillance_timeout != -1) {
453 DEBUG("enabling surveillance\n");
454 enable_surveillance(surveillance_timeout);
455 DEBUG("surveillance enabled\n");
459 cpu = first_cpu(cpu_online_map);
461 set_cpus_allowed(current, cpumask_of_cpu(cpu));
462 do_event_scan(event_scan);
463 set_cpus_allowed(current, CPU_MASK_ALL);
465 /* Drop hotplug lock, and sleep for a bit (at least
466 * one second since some machines have problems if we
467 * call event-scan too quickly). */
468 unlock_cpu_hotplug();
469 set_current_state(TASK_INTERRUPTIBLE);
470 schedule_timeout((HZ*60/rtas_event_scan_rate) / 2);
473 cpu = next_cpu(cpu, cpu_online_map);
475 cpu = first_cpu(cpu_online_map);
479 /* Should delete proc entries */
483 static int __init rtas_init(void)
485 struct proc_dir_entry *entry;
487 /* No RTAS, only warn if we are on a pSeries box */
488 if (rtas_token("event-scan") == RTAS_UNKNOWN_SERVICE) {
489 if (systemcfg->platform & PLATFORM_PSERIES)
490 printk(KERN_ERR "rtasd: no event-scan on system\n");
494 entry = create_proc_entry("ppc64/rtas/error_log", S_IRUSR, NULL);
496 entry->proc_fops = &proc_rtas_log_operations;
498 printk(KERN_ERR "Failed to create error_log proc entry\n");
500 if (kernel_thread(rtasd, NULL, CLONE_FS) < 0)
501 printk(KERN_ERR "Failed to start RTAS daemon\n");
506 static int __init surveillance_setup(char *str)
510 if (get_option(&str,&i)) {
511 if (i >= 0 && i <= 255)
512 surveillance_timeout = i;
518 static int __init rtasmsgs_setup(char *str)
520 if (strcmp(str, "on") == 0)
522 else if (strcmp(str, "off") == 0)
527 __initcall(rtas_init);
528 __setup("surveillance=", surveillance_setup);
529 __setup("rtasmsgs=", rtasmsgs_setup);