1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
5 * Communication to userspace based on kernel/printk.c
8 #include <linux/types.h>
9 #include <linux/errno.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/poll.h>
13 #include <linux/proc_fs.h>
14 #include <linux/init.h>
15 #include <linux/vmalloc.h>
16 #include <linux/spinlock.h>
17 #include <linux/cpu.h>
18 #include <linux/workqueue.h>
19 #include <linux/slab.h>
20 #include <linux/topology.h>
22 #include <linux/uaccess.h>
25 #include <asm/nvram.h>
26 #include <linux/atomic.h>
27 #include <asm/machdep.h>
28 #include <asm/topology.h>
31 static DEFINE_SPINLOCK(rtasd_log_lock);
33 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
35 static char *rtas_log_buf;
36 static unsigned long rtas_log_start;
37 static unsigned long rtas_log_size;
39 static int surveillance_timeout = -1;
41 static unsigned int rtas_error_log_max;
42 static unsigned int rtas_error_log_buffer_max;
44 /* RTAS service tokens */
45 static unsigned int event_scan;
46 static unsigned int rtas_event_scan_rate;
48 static bool full_rtas_msgs;
50 /* Stop logging to nvram after first fatal error */
51 static int logging_enabled; /* Until we initialize everything,
52 * make sure we don't try logging
54 static int error_log_cnt;
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 char *rtas_type[] = {
64 "Unknown", "Retry", "TCE Error", "Internal Device Failure",
65 "Timeout", "Data Parity", "Address Parity", "Cache Parity",
66 "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
69 static char *rtas_event_type(int type)
71 if ((type > 0) && (type < 11))
72 return rtas_type[type];
77 case RTAS_TYPE_PLATFORM:
78 return "Platform Error";
82 return "Platform Information Event";
83 case RTAS_TYPE_DEALLOC:
84 return "Resource Deallocation Event";
86 return "Dump Notification Event";
88 return "Platform Resource Reassignment Event";
89 case RTAS_TYPE_HOTPLUG:
90 return "Hotplug Event";
96 /* To see this info, grep RTAS /var/log/messages and each entry
97 * will be collected together with obvious begin/end.
98 * There will be a unique identifier on the begin and end lines.
99 * This will persist across reboots.
101 * format of error logs returned from RTAS:
102 * bytes (size) : contents
103 * --------------------------------------------------------
104 * 0-7 (8) : rtas_error_log
105 * 8-47 (40) : extended info
106 * 48-51 (4) : vendor id
107 * 52-1023 (vendor specific) : location code and debug data
109 static void printk_log_rtas(char *buf, int len)
115 char * str = "RTAS event";
117 if (full_rtas_msgs) {
118 printk(RTAS_DEBUG "%d -------- %s begin --------\n",
122 * Print perline bytes on each line, each line will start
123 * with RTAS and a changing number, so syslogd will
124 * print lines that are otherwise the same. Separate every
125 * 4 bytes with a space.
127 for (i = 0; i < len; i++) {
130 memset(buffer, 0, sizeof(buffer));
131 n = sprintf(buffer, "RTAS %d:", i/perline);
135 n += sprintf(buffer+n, " ");
137 n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
139 if (j == (perline-1))
140 printk(KERN_DEBUG "%s\n", buffer);
142 if ((i % perline) != 0)
143 printk(KERN_DEBUG "%s\n", buffer);
145 printk(RTAS_DEBUG "%d -------- %s end ----------\n",
148 struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
150 printk(RTAS_DEBUG "event: %d, Type: %s (%d), Severity: %d\n",
152 rtas_event_type(rtas_error_type(errlog)),
153 rtas_error_type(errlog),
154 rtas_error_severity(errlog));
158 static int log_rtas_len(char * buf)
161 struct rtas_error_log *err;
162 uint32_t extended_log_length;
164 /* rtas fixed header */
166 err = (struct rtas_error_log *)buf;
167 extended_log_length = rtas_error_extended_log_length(err);
168 if (rtas_error_extended(err) && extended_log_length) {
170 /* extended header */
171 len += extended_log_length;
174 if (rtas_error_log_max == 0)
175 rtas_error_log_max = rtas_get_error_log_max();
177 if (len > rtas_error_log_max)
178 len = rtas_error_log_max;
184 * First write to nvram, if fatal error, that is the only
185 * place we log the info. The error will be picked up
186 * on the next reboot by rtasd. If not fatal, run the
187 * method for the type of error. Currently, only RTAS
188 * errors have methods implemented, but in the future
189 * there might be a need to store data in nvram before a
192 * XXX We write to nvram periodically, to indicate error has
193 * been written and sync'd, but there is a possibility
194 * that if we don't shutdown correctly, a duplicate error
195 * record will be created on next reboot.
197 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
199 unsigned long offset;
203 pr_debug("rtasd: logging event\n");
207 spin_lock_irqsave(&rtasd_log_lock, s);
209 /* get length and increase count */
210 switch (err_type & ERR_TYPE_MASK) {
211 case ERR_TYPE_RTAS_LOG:
212 len = log_rtas_len(buf);
213 if (!(err_type & ERR_FLAG_BOOT))
216 case ERR_TYPE_KERNEL_PANIC:
218 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
219 spin_unlock_irqrestore(&rtasd_log_lock, s);
224 /* Write error to NVRAM */
225 if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
226 nvram_write_error_log(buf, len, err_type, error_log_cnt);
227 #endif /* CONFIG_PPC64 */
230 * rtas errors can occur during boot, and we do want to capture
231 * those somewhere, even if nvram isn't ready (why not?), and even
232 * if rtasd isn't ready. Put them into the boot log, at least.
234 if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
235 printk_log_rtas(buf, len);
237 /* Check to see if we need to or have stopped logging */
238 if (fatal || !logging_enabled) {
240 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
241 spin_unlock_irqrestore(&rtasd_log_lock, s);
245 /* call type specific method for error */
246 switch (err_type & ERR_TYPE_MASK) {
247 case ERR_TYPE_RTAS_LOG:
248 offset = rtas_error_log_buffer_max *
249 ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
251 /* First copy over sequence number */
252 memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
254 /* Second copy over error log data */
255 offset += sizeof(int);
256 memcpy(&rtas_log_buf[offset], buf, len);
258 if (rtas_log_size < LOG_NUMBER)
263 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
264 spin_unlock_irqrestore(&rtasd_log_lock, s);
265 wake_up_interruptible(&rtas_log_wait);
267 case ERR_TYPE_KERNEL_PANIC:
269 WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
270 spin_unlock_irqrestore(&rtasd_log_lock, s);
275 static void handle_rtas_event(const struct rtas_error_log *log)
277 if (!machine_is(pseries))
280 if (rtas_error_type(log) == RTAS_TYPE_PRRN)
281 pr_info_ratelimited("Platform resource reassignment ignored.\n");
284 static int rtas_log_open(struct inode * inode, struct file * file)
289 static int rtas_log_release(struct inode * inode, struct file * file)
294 /* This will check if all events are logged, if they are then, we
295 * know that we can safely clear the events in NVRAM.
296 * Next we'll sit and wait for something else to log.
298 static ssize_t rtas_log_read(struct file * file, char __user * buf,
299 size_t count, loff_t *ppos)
304 unsigned long offset;
306 if (!buf || count < rtas_error_log_buffer_max)
309 count = rtas_error_log_buffer_max;
311 if (!access_ok(buf, count))
314 tmp = kmalloc(count, GFP_KERNEL);
318 spin_lock_irqsave(&rtasd_log_lock, s);
320 /* if it's 0, then we know we got the last one (the one in NVRAM) */
321 while (rtas_log_size == 0) {
322 if (file->f_flags & O_NONBLOCK) {
323 spin_unlock_irqrestore(&rtasd_log_lock, s);
328 if (!logging_enabled) {
329 spin_unlock_irqrestore(&rtasd_log_lock, s);
334 nvram_clear_error_log();
335 #endif /* CONFIG_PPC64 */
337 spin_unlock_irqrestore(&rtasd_log_lock, s);
338 error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
341 spin_lock_irqsave(&rtasd_log_lock, s);
344 offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
345 memcpy(tmp, &rtas_log_buf[offset], count);
349 spin_unlock_irqrestore(&rtasd_log_lock, s);
351 error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
357 static __poll_t rtas_log_poll(struct file *file, poll_table * wait)
359 poll_wait(file, &rtas_log_wait, wait);
361 return EPOLLIN | EPOLLRDNORM;
365 static const struct proc_ops rtas_log_proc_ops = {
366 .proc_read = rtas_log_read,
367 .proc_poll = rtas_log_poll,
368 .proc_open = rtas_log_open,
369 .proc_release = rtas_log_release,
370 .proc_lseek = noop_llseek,
373 static int enable_surveillance(int timeout)
377 error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
382 if (error == -EINVAL) {
383 printk(KERN_DEBUG "rtasd: surveillance not supported\n");
387 printk(KERN_ERR "rtasd: could not update surveillance\n");
391 static void do_event_scan(void)
395 memset(logdata, 0, rtas_error_log_max);
396 error = rtas_call(event_scan, 4, 1, NULL,
397 RTAS_EVENT_SCAN_ALL_EVENTS, 0,
398 __pa(logdata), rtas_error_log_max);
400 printk(KERN_ERR "event-scan failed\n");
405 if (rtas_error_type((struct rtas_error_log *)logdata) !=
407 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
409 handle_rtas_event((struct rtas_error_log *)logdata);
415 static void rtas_event_scan(struct work_struct *w);
416 static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
419 * Delay should be at least one second since some machines have problems if
420 * we call event-scan too quickly.
422 static unsigned long event_scan_delay = 1*HZ;
423 static int first_pass = 1;
425 static void rtas_event_scan(struct work_struct *w)
433 /* raw_ OK because just using CPU as starting point. */
434 cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
435 if (cpu >= nr_cpu_ids) {
436 cpu = cpumask_first(cpu_online_mask);
440 event_scan_delay = 30*HZ/rtas_event_scan_rate;
442 if (surveillance_timeout != -1) {
443 pr_debug("rtasd: enabling surveillance\n");
444 enable_surveillance(surveillance_timeout);
445 pr_debug("rtasd: surveillance enabled\n");
450 schedule_delayed_work_on(cpu, &event_scan_work,
451 __round_jiffies_relative(event_scan_delay, cpu));
457 static void __init retrieve_nvram_error_log(void)
459 unsigned int err_type ;
462 /* See if we have any error stored in NVRAM */
463 memset(logdata, 0, rtas_error_log_max);
464 rc = nvram_read_error_log(logdata, rtas_error_log_max,
465 &err_type, &error_log_cnt);
466 /* We can use rtas_log_buf now */
469 if (err_type != ERR_FLAG_ALREADY_LOGGED) {
470 pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
474 #else /* CONFIG_PPC64 */
475 static void __init retrieve_nvram_error_log(void)
478 #endif /* CONFIG_PPC64 */
480 static void __init start_event_scan(void)
482 printk(KERN_DEBUG "RTAS daemon started\n");
483 pr_debug("rtasd: will sleep for %d milliseconds\n",
484 (30000 / rtas_event_scan_rate));
486 /* Retrieve errors from nvram if any */
487 retrieve_nvram_error_log();
489 schedule_delayed_work_on(cpumask_first(cpu_online_mask),
490 &event_scan_work, event_scan_delay);
493 /* Cancel the rtas event scan work */
494 void rtas_cancel_event_scan(void)
496 cancel_delayed_work_sync(&event_scan_work);
498 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
500 static int __init rtas_event_scan_init(void)
502 if (!machine_is(pseries) && !machine_is(chrp))
506 event_scan = rtas_token("event-scan");
507 if (event_scan == RTAS_UNKNOWN_SERVICE) {
508 printk(KERN_INFO "rtasd: No event-scan on system\n");
512 rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
513 if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
514 printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
518 if (!rtas_event_scan_rate) {
519 /* Broken firmware: take a rate of zero to mean don't scan */
520 printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
524 /* Make room for the sequence number */
525 rtas_error_log_max = rtas_get_error_log_max();
526 rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
528 rtas_log_buf = vmalloc(array_size(LOG_NUMBER,
529 rtas_error_log_buffer_max));
531 printk(KERN_ERR "rtasd: no memory\n");
539 arch_initcall(rtas_event_scan_init);
541 static int __init rtas_init(void)
543 struct proc_dir_entry *entry;
545 if (!machine_is(pseries) && !machine_is(chrp))
551 entry = proc_create("powerpc/rtas/error_log", 0400, NULL,
554 printk(KERN_ERR "Failed to create error_log proc entry\n");
558 __initcall(rtas_init);
560 static int __init surveillance_setup(char *str)
564 /* We only do surveillance on pseries */
565 if (!machine_is(pseries))
568 if (get_option(&str,&i)) {
569 if (i >= 0 && i <= 255)
570 surveillance_timeout = i;
575 __setup("surveillance=", surveillance_setup);
577 static int __init rtasmsgs_setup(char *str)
579 return (kstrtobool(str, &full_rtas_msgs) == 0);
581 __setup("rtasmsgs=", rtasmsgs_setup);