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>
28 #include <asm/nvram.h>
29 #include <asm/atomic.h>
30 #include <asm/machdep.h>
33 #define DEBUG(A...) printk(KERN_ERR A)
38 static DEFINE_SPINLOCK(rtasd_log_lock
);
40 DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait
);
42 static char *rtas_log_buf
;
43 static unsigned long rtas_log_start
;
44 static unsigned long rtas_log_size
;
46 static int surveillance_timeout
= -1;
47 static unsigned int rtas_event_scan_rate
;
48 static unsigned int rtas_error_log_max
;
49 static unsigned int rtas_error_log_buffer_max
;
51 static int full_rtas_msgs
= 0;
53 extern int no_logging
;
55 volatile int error_log_cnt
= 0;
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
62 static unsigned char logdata
[RTAS_ERROR_LOG_MAX
];
64 static int get_eventscan_parms(void);
66 static char *rtas_type
[] = {
67 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
68 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
69 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
72 static char *rtas_event_type(int type
)
74 if ((type
> 0) && (type
< 11))
75 return rtas_type
[type
];
80 case RTAS_TYPE_PLATFORM
:
81 return "Platform Error";
85 return "Platform Information Event";
86 case RTAS_TYPE_DEALLOC
:
87 return "Resource Deallocation Event";
89 return "Dump Notification Event";
95 /* To see this info, grep RTAS /var/log/messages and each entry
96 * will be collected together with obvious begin/end.
97 * There will be a unique identifier on the begin and end lines.
98 * This will persist across reboots.
100 * format of error logs returned from RTAS:
101 * bytes (size) : contents
102 * --------------------------------------------------------
103 * 0-7 (8) : rtas_error_log
104 * 8-47 (40) : extended info
105 * 48-51 (4) : vendor id
106 * 52-1023 (vendor specific) : location code and debug data
108 static void printk_log_rtas(char *buf
, int len
)
114 char * str
= "RTAS event";
116 if (full_rtas_msgs
) {
117 printk(RTAS_DEBUG
"%d -------- %s begin --------\n",
121 * Print perline bytes on each line, each line will start
122 * with RTAS and a changing number, so syslogd will
123 * print lines that are otherwise the same. Separate every
124 * 4 bytes with a space.
126 for (i
= 0; i
< len
; i
++) {
129 memset(buffer
, 0, sizeof(buffer
));
130 n
= sprintf(buffer
, "RTAS %d:", i
/perline
);
134 n
+= sprintf(buffer
+n
, " ");
136 n
+= sprintf(buffer
+n
, "%02x", (unsigned char)buf
[i
]);
138 if (j
== (perline
-1))
139 printk(KERN_DEBUG
"%s\n", buffer
);
141 if ((i
% perline
) != 0)
142 printk(KERN_DEBUG
"%s\n", buffer
);
144 printk(RTAS_DEBUG
"%d -------- %s end ----------\n",
147 struct rtas_error_log
*errlog
= (struct rtas_error_log
*)buf
;
149 printk(RTAS_DEBUG
"event: %d, Type: %s, Severity: %d\n",
150 error_log_cnt
, rtas_event_type(errlog
->type
),
155 static int log_rtas_len(char * buf
)
158 struct rtas_error_log
*err
;
160 /* rtas fixed header */
162 err
= (struct rtas_error_log
*)buf
;
163 if (err
->extended_log_length
) {
165 /* extended header */
166 len
+= err
->extended_log_length
;
169 if (rtas_error_log_max
== 0) {
170 get_eventscan_parms();
172 if (len
> rtas_error_log_max
)
173 len
= rtas_error_log_max
;
179 * First write to nvram, if fatal error, that is the only
180 * place we log the info. The error will be picked up
181 * on the next reboot by rtasd. If not fatal, run the
182 * method for the type of error. Currently, only RTAS
183 * errors have methods implemented, but in the future
184 * there might be a need to store data in nvram before a
187 * XXX We write to nvram periodically, to indicate error has
188 * been written and sync'd, but there is a possibility
189 * that if we don't shutdown correctly, a duplicate error
190 * record will be created on next reboot.
192 void pSeries_log_error(char *buf
, unsigned int err_type
, int fatal
)
194 unsigned long offset
;
198 DEBUG("logging event\n");
202 spin_lock_irqsave(&rtasd_log_lock
, s
);
204 /* get length and increase count */
205 switch (err_type
& ERR_TYPE_MASK
) {
206 case ERR_TYPE_RTAS_LOG
:
207 len
= log_rtas_len(buf
);
208 if (!(err_type
& ERR_FLAG_BOOT
))
211 case ERR_TYPE_KERNEL_PANIC
:
213 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
217 /* Write error to NVRAM */
218 if (!no_logging
&& !(err_type
& ERR_FLAG_BOOT
))
219 nvram_write_error_log(buf
, len
, err_type
);
222 * rtas errors can occur during boot, and we do want to capture
223 * those somewhere, even if nvram isn't ready (why not?), and even
224 * if rtasd isn't ready. Put them into the boot log, at least.
226 if ((err_type
& ERR_TYPE_MASK
) == ERR_TYPE_RTAS_LOG
)
227 printk_log_rtas(buf
, len
);
229 /* Check to see if we need to or have stopped logging */
230 if (fatal
|| no_logging
) {
232 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
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
)
254 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
255 wake_up_interruptible(&rtas_log_wait
);
257 case ERR_TYPE_KERNEL_PANIC
:
259 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
266 static int rtas_log_open(struct inode
* inode
, struct file
* file
)
271 static int rtas_log_release(struct inode
* inode
, struct file
* file
)
276 /* This will check if all events are logged, if they are then, we
277 * know that we can safely clear the events in NVRAM.
278 * Next we'll sit and wait for something else to log.
280 static ssize_t
rtas_log_read(struct file
* file
, char __user
* buf
,
281 size_t count
, loff_t
*ppos
)
286 unsigned long offset
;
288 if (!buf
|| count
< rtas_error_log_buffer_max
)
291 count
= rtas_error_log_buffer_max
;
293 if (!access_ok(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 const 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
== -EINVAL
) {
351 printk(KERN_DEBUG
"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
= of_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 void do_event_scan_all_cpus(long delay
)
408 cpu
= first_cpu(cpu_online_map
);
410 set_cpus_allowed(current
, cpumask_of_cpu(cpu
));
411 do_event_scan(rtas_token("event-scan"));
412 set_cpus_allowed(current
, CPU_MASK_ALL
);
414 /* Drop hotplug lock, and sleep for the specified delay */
415 unlock_cpu_hotplug();
416 msleep_interruptible(delay
);
419 cpu
= next_cpu(cpu
, cpu_online_map
);
423 unlock_cpu_hotplug();
426 static int rtasd(void *unused
)
428 unsigned int err_type
;
429 int event_scan
= rtas_token("event-scan");
434 if (event_scan
== RTAS_UNKNOWN_SERVICE
|| get_eventscan_parms() == -1)
437 rtas_log_buf
= vmalloc(rtas_error_log_buffer_max
*LOG_NUMBER
);
439 printk(KERN_ERR
"rtasd: no memory\n");
443 printk(KERN_DEBUG
"RTAS daemon started\n");
445 DEBUG("will sleep for %d milliseconds\n", (30000/rtas_event_scan_rate
));
447 /* See if we have any error stored in NVRAM */
448 memset(logdata
, 0, rtas_error_log_max
);
450 rc
= nvram_read_error_log(logdata
, rtas_error_log_max
, &err_type
);
452 /* We can use rtas_log_buf now */
456 if (err_type
!= ERR_FLAG_ALREADY_LOGGED
) {
457 pSeries_log_error(logdata
, err_type
| ERR_FLAG_BOOT
, 0);
462 do_event_scan_all_cpus(1000);
464 if (surveillance_timeout
!= -1) {
465 DEBUG("enabling surveillance\n");
466 enable_surveillance(surveillance_timeout
);
467 DEBUG("surveillance enabled\n");
470 /* Delay should be at least one second since some
471 * machines have problems if we call event-scan too
474 do_event_scan_all_cpus(30000/rtas_event_scan_rate
);
477 /* Should delete proc entries */
481 static int __init
rtas_init(void)
483 struct proc_dir_entry
*entry
;
485 if (!machine_is(pseries
))
489 if (rtas_token("event-scan") == RTAS_UNKNOWN_SERVICE
) {
490 printk(KERN_DEBUG
"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
);