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/workqueue.h>
23 #include <linux/slab.h>
24 #include <linux/topology.h>
26 #include <linux/uaccess.h>
30 #include <asm/nvram.h>
31 #include <linux/atomic.h>
32 #include <asm/machdep.h>
33 #include <asm/topology.h>
36 static DEFINE_SPINLOCK(rtasd_log_lock
);
38 static 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;
46 static unsigned int rtas_error_log_max
;
47 static unsigned int rtas_error_log_buffer_max
;
49 /* RTAS service tokens */
50 static unsigned int event_scan
;
51 static unsigned int rtas_event_scan_rate
;
53 static bool full_rtas_msgs
;
55 /* Stop logging to nvram after first fatal error */
56 static int logging_enabled
; /* Until we initialize everything,
57 * make sure we don't try logging
59 static int error_log_cnt
;
62 * Since we use 32 bit RTAS, the physical address of this must be below
63 * 4G or else bad things happen. Allocate this in the kernel data and
66 static unsigned char logdata
[RTAS_ERROR_LOG_MAX
];
68 static char *rtas_type
[] = {
69 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
70 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
71 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
74 static char *rtas_event_type(int type
)
76 if ((type
> 0) && (type
< 11))
77 return rtas_type
[type
];
82 case RTAS_TYPE_PLATFORM
:
83 return "Platform Error";
87 return "Platform Information Event";
88 case RTAS_TYPE_DEALLOC
:
89 return "Resource Deallocation Event";
91 return "Dump Notification Event";
93 return "Platform Resource Reassignment Event";
99 /* To see this info, grep RTAS /var/log/messages and each entry
100 * will be collected together with obvious begin/end.
101 * There will be a unique identifier on the begin and end lines.
102 * This will persist across reboots.
104 * format of error logs returned from RTAS:
105 * bytes (size) : contents
106 * --------------------------------------------------------
107 * 0-7 (8) : rtas_error_log
108 * 8-47 (40) : extended info
109 * 48-51 (4) : vendor id
110 * 52-1023 (vendor specific) : location code and debug data
112 static void printk_log_rtas(char *buf
, int len
)
118 char * str
= "RTAS event";
120 if (full_rtas_msgs
) {
121 printk(RTAS_DEBUG
"%d -------- %s begin --------\n",
125 * Print perline bytes on each line, each line will start
126 * with RTAS and a changing number, so syslogd will
127 * print lines that are otherwise the same. Separate every
128 * 4 bytes with a space.
130 for (i
= 0; i
< len
; i
++) {
133 memset(buffer
, 0, sizeof(buffer
));
134 n
= sprintf(buffer
, "RTAS %d:", i
/perline
);
138 n
+= sprintf(buffer
+n
, " ");
140 n
+= sprintf(buffer
+n
, "%02x", (unsigned char)buf
[i
]);
142 if (j
== (perline
-1))
143 printk(KERN_DEBUG
"%s\n", buffer
);
145 if ((i
% perline
) != 0)
146 printk(KERN_DEBUG
"%s\n", buffer
);
148 printk(RTAS_DEBUG
"%d -------- %s end ----------\n",
151 struct rtas_error_log
*errlog
= (struct rtas_error_log
*)buf
;
153 printk(RTAS_DEBUG
"event: %d, Type: %s, Severity: %d\n",
154 error_log_cnt
, rtas_event_type(rtas_error_type(errlog
)),
155 rtas_error_severity(errlog
));
159 static int log_rtas_len(char * buf
)
162 struct rtas_error_log
*err
;
163 uint32_t extended_log_length
;
165 /* rtas fixed header */
167 err
= (struct rtas_error_log
*)buf
;
168 extended_log_length
= rtas_error_extended_log_length(err
);
169 if (rtas_error_extended(err
) && extended_log_length
) {
171 /* extended header */
172 len
+= extended_log_length
;
175 if (rtas_error_log_max
== 0)
176 rtas_error_log_max
= rtas_get_error_log_max();
178 if (len
> rtas_error_log_max
)
179 len
= rtas_error_log_max
;
185 * First write to nvram, if fatal error, that is the only
186 * place we log the info. The error will be picked up
187 * on the next reboot by rtasd. If not fatal, run the
188 * method for the type of error. Currently, only RTAS
189 * errors have methods implemented, but in the future
190 * there might be a need to store data in nvram before a
193 * XXX We write to nvram periodically, to indicate error has
194 * been written and sync'd, but there is a possibility
195 * that if we don't shutdown correctly, a duplicate error
196 * record will be created on next reboot.
198 void pSeries_log_error(char *buf
, unsigned int err_type
, int fatal
)
200 unsigned long offset
;
204 pr_debug("rtasd: logging event\n");
208 spin_lock_irqsave(&rtasd_log_lock
, s
);
210 /* get length and increase count */
211 switch (err_type
& ERR_TYPE_MASK
) {
212 case ERR_TYPE_RTAS_LOG
:
213 len
= log_rtas_len(buf
);
214 if (!(err_type
& ERR_FLAG_BOOT
))
217 case ERR_TYPE_KERNEL_PANIC
:
219 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
220 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
225 /* Write error to NVRAM */
226 if (logging_enabled
&& !(err_type
& ERR_FLAG_BOOT
))
227 nvram_write_error_log(buf
, len
, err_type
, error_log_cnt
);
228 #endif /* CONFIG_PPC64 */
231 * rtas errors can occur during boot, and we do want to capture
232 * those somewhere, even if nvram isn't ready (why not?), and even
233 * if rtasd isn't ready. Put them into the boot log, at least.
235 if ((err_type
& ERR_TYPE_MASK
) == ERR_TYPE_RTAS_LOG
)
236 printk_log_rtas(buf
, len
);
238 /* Check to see if we need to or have stopped logging */
239 if (fatal
|| !logging_enabled
) {
241 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
242 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
246 /* call type specific method for error */
247 switch (err_type
& ERR_TYPE_MASK
) {
248 case ERR_TYPE_RTAS_LOG
:
249 offset
= rtas_error_log_buffer_max
*
250 ((rtas_log_start
+rtas_log_size
) & LOG_NUMBER_MASK
);
252 /* First copy over sequence number */
253 memcpy(&rtas_log_buf
[offset
], (void *) &error_log_cnt
, sizeof(int));
255 /* Second copy over error log data */
256 offset
+= sizeof(int);
257 memcpy(&rtas_log_buf
[offset
], buf
, len
);
259 if (rtas_log_size
< LOG_NUMBER
)
264 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
265 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
266 wake_up_interruptible(&rtas_log_wait
);
268 case ERR_TYPE_KERNEL_PANIC
:
270 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
271 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
276 #ifdef CONFIG_PPC_PSERIES
277 static s32 prrn_update_scope
;
279 static void prrn_work_fn(struct work_struct
*work
)
282 * For PRRN, we must pass the negative of the scope value in
285 pseries_devicetree_update(-prrn_update_scope
);
286 numa_update_cpu_topology(false);
289 static DECLARE_WORK(prrn_work
, prrn_work_fn
);
291 static void prrn_schedule_update(u32 scope
)
293 flush_work(&prrn_work
);
294 prrn_update_scope
= scope
;
295 schedule_work(&prrn_work
);
298 static void handle_rtas_event(const struct rtas_error_log
*log
)
300 if (rtas_error_type(log
) != RTAS_TYPE_PRRN
|| !prrn_is_enabled())
303 /* For PRRN Events the extended log length is used to denote
304 * the scope for calling rtas update-nodes.
306 prrn_schedule_update(rtas_error_extended_log_length(log
));
311 static void handle_rtas_event(const struct rtas_error_log
*log
)
318 static int rtas_log_open(struct inode
* inode
, struct file
* file
)
323 static int rtas_log_release(struct inode
* inode
, struct file
* file
)
328 /* This will check if all events are logged, if they are then, we
329 * know that we can safely clear the events in NVRAM.
330 * Next we'll sit and wait for something else to log.
332 static ssize_t
rtas_log_read(struct file
* file
, char __user
* buf
,
333 size_t count
, loff_t
*ppos
)
338 unsigned long offset
;
340 if (!buf
|| count
< rtas_error_log_buffer_max
)
343 count
= rtas_error_log_buffer_max
;
345 if (!access_ok(VERIFY_WRITE
, buf
, count
))
348 tmp
= kmalloc(count
, GFP_KERNEL
);
352 spin_lock_irqsave(&rtasd_log_lock
, s
);
354 /* if it's 0, then we know we got the last one (the one in NVRAM) */
355 while (rtas_log_size
== 0) {
356 if (file
->f_flags
& O_NONBLOCK
) {
357 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
362 if (!logging_enabled
) {
363 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
368 nvram_clear_error_log();
369 #endif /* CONFIG_PPC64 */
371 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
372 error
= wait_event_interruptible(rtas_log_wait
, rtas_log_size
);
375 spin_lock_irqsave(&rtasd_log_lock
, s
);
378 offset
= rtas_error_log_buffer_max
* (rtas_log_start
& LOG_NUMBER_MASK
);
379 memcpy(tmp
, &rtas_log_buf
[offset
], count
);
383 spin_unlock_irqrestore(&rtasd_log_lock
, s
);
385 error
= copy_to_user(buf
, tmp
, count
) ? -EFAULT
: count
;
391 static __poll_t
rtas_log_poll(struct file
*file
, poll_table
* wait
)
393 poll_wait(file
, &rtas_log_wait
, wait
);
395 return EPOLLIN
| EPOLLRDNORM
;
399 static const struct file_operations proc_rtas_log_operations
= {
400 .read
= rtas_log_read
,
401 .poll
= rtas_log_poll
,
402 .open
= rtas_log_open
,
403 .release
= rtas_log_release
,
404 .llseek
= noop_llseek
,
407 static int enable_surveillance(int timeout
)
411 error
= rtas_set_indicator(SURVEILLANCE_TOKEN
, 0, timeout
);
416 if (error
== -EINVAL
) {
417 printk(KERN_DEBUG
"rtasd: surveillance not supported\n");
421 printk(KERN_ERR
"rtasd: could not update surveillance\n");
425 static void do_event_scan(void)
429 memset(logdata
, 0, rtas_error_log_max
);
430 error
= rtas_call(event_scan
, 4, 1, NULL
,
431 RTAS_EVENT_SCAN_ALL_EVENTS
, 0,
432 __pa(logdata
), rtas_error_log_max
);
434 printk(KERN_ERR
"event-scan failed\n");
439 if (rtas_error_type((struct rtas_error_log
*)logdata
) !=
441 pSeries_log_error(logdata
, ERR_TYPE_RTAS_LOG
,
443 handle_rtas_event((struct rtas_error_log
*)logdata
);
449 static void rtas_event_scan(struct work_struct
*w
);
450 static DECLARE_DELAYED_WORK(event_scan_work
, rtas_event_scan
);
453 * Delay should be at least one second since some machines have problems if
454 * we call event-scan too quickly.
456 static unsigned long event_scan_delay
= 1*HZ
;
457 static int first_pass
= 1;
459 static void rtas_event_scan(struct work_struct
*w
)
467 /* raw_ OK because just using CPU as starting point. */
468 cpu
= cpumask_next(raw_smp_processor_id(), cpu_online_mask
);
469 if (cpu
>= nr_cpu_ids
) {
470 cpu
= cpumask_first(cpu_online_mask
);
474 event_scan_delay
= 30*HZ
/rtas_event_scan_rate
;
476 if (surveillance_timeout
!= -1) {
477 pr_debug("rtasd: enabling surveillance\n");
478 enable_surveillance(surveillance_timeout
);
479 pr_debug("rtasd: surveillance enabled\n");
484 schedule_delayed_work_on(cpu
, &event_scan_work
,
485 __round_jiffies_relative(event_scan_delay
, cpu
));
491 static void retrieve_nvram_error_log(void)
493 unsigned int err_type
;
496 /* See if we have any error stored in NVRAM */
497 memset(logdata
, 0, rtas_error_log_max
);
498 rc
= nvram_read_error_log(logdata
, rtas_error_log_max
,
499 &err_type
, &error_log_cnt
);
500 /* We can use rtas_log_buf now */
503 if (err_type
!= ERR_FLAG_ALREADY_LOGGED
) {
504 pSeries_log_error(logdata
, err_type
| ERR_FLAG_BOOT
, 0);
508 #else /* CONFIG_PPC64 */
509 static void retrieve_nvram_error_log(void)
512 #endif /* CONFIG_PPC64 */
514 static void start_event_scan(void)
516 printk(KERN_DEBUG
"RTAS daemon started\n");
517 pr_debug("rtasd: will sleep for %d milliseconds\n",
518 (30000 / rtas_event_scan_rate
));
520 /* Retrieve errors from nvram if any */
521 retrieve_nvram_error_log();
523 schedule_delayed_work_on(cpumask_first(cpu_online_mask
),
524 &event_scan_work
, event_scan_delay
);
527 /* Cancel the rtas event scan work */
528 void rtas_cancel_event_scan(void)
530 cancel_delayed_work_sync(&event_scan_work
);
532 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan
);
534 static int __init
rtas_event_scan_init(void)
536 if (!machine_is(pseries
) && !machine_is(chrp
))
540 event_scan
= rtas_token("event-scan");
541 if (event_scan
== RTAS_UNKNOWN_SERVICE
) {
542 printk(KERN_INFO
"rtasd: No event-scan on system\n");
546 rtas_event_scan_rate
= rtas_token("rtas-event-scan-rate");
547 if (rtas_event_scan_rate
== RTAS_UNKNOWN_SERVICE
) {
548 printk(KERN_ERR
"rtasd: no rtas-event-scan-rate on system\n");
552 if (!rtas_event_scan_rate
) {
553 /* Broken firmware: take a rate of zero to mean don't scan */
554 printk(KERN_DEBUG
"rtasd: scan rate is 0, not scanning\n");
558 /* Make room for the sequence number */
559 rtas_error_log_max
= rtas_get_error_log_max();
560 rtas_error_log_buffer_max
= rtas_error_log_max
+ sizeof(int);
562 rtas_log_buf
= vmalloc(rtas_error_log_buffer_max
*LOG_NUMBER
);
564 printk(KERN_ERR
"rtasd: no memory\n");
572 arch_initcall(rtas_event_scan_init
);
574 static int __init
rtas_init(void)
576 struct proc_dir_entry
*entry
;
578 if (!machine_is(pseries
) && !machine_is(chrp
))
584 entry
= proc_create("powerpc/rtas/error_log", 0400, NULL
,
585 &proc_rtas_log_operations
);
587 printk(KERN_ERR
"Failed to create error_log proc entry\n");
591 __initcall(rtas_init
);
593 static int __init
surveillance_setup(char *str
)
597 /* We only do surveillance on pseries */
598 if (!machine_is(pseries
))
601 if (get_option(&str
,&i
)) {
602 if (i
>= 0 && i
<= 255)
603 surveillance_timeout
= i
;
608 __setup("surveillance=", surveillance_setup
);
610 static int __init
rtasmsgs_setup(char *str
)
612 return (kstrtobool(str
, &full_rtas_msgs
) == 0);
614 __setup("rtasmsgs=", rtasmsgs_setup
);