powerpc/kernel/rtasd.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
0004  *
0005  * Communication to userspace based on kernel/printk.c
0006  */
0007
0008 #include <linux/types.h>
0009 #include <linux/errno.h>
0010 #include <linux/sched.h>
0011 #include <linux/kernel.h>
0012 #include <linux/poll.h>
0013 #include <linux/proc_fs.h>
0014 #include <linux/init.h>
0015 #include <linux/vmalloc.h>
0016 #include <linux/spinlock.h>
0017 #include <linux/cpu.h>
0018 #include <linux/workqueue.h>
0019 #include <linux/slab.h>
0020 #include <linux/topology.h>
0021
0022 #include <linux/uaccess.h>
0023 #include <asm/io.h>
0024 #include <asm/rtas.h>
0025 #include <asm/nvram.h>
0026 #include <linux/atomic.h>
0027 #include <asm/machdep.h>
0028 #include <asm/topology.h>
0029
0030
0031 static DEFINE_SPINLOCK(rtasd_log_lock);
0032
0033 static DECLARE_WAIT_QUEUE_HEAD(rtas_log_wait);
0034
0035 static char *rtas_log_buf;
0036 static unsigned long rtas_log_start;
0037 static unsigned long rtas_log_size;
0038
0039 static int surveillance_timeout = -1;
0040
0041 static unsigned int rtas_error_log_max;
0042 static unsigned int rtas_error_log_buffer_max;
0043
0044 /* RTAS service tokens */
0045 static unsigned int event_scan;
0046 static unsigned int rtas_event_scan_rate;
0047
0048 static bool full_rtas_msgs;
0049
0050 /* Stop logging to nvram after first fatal error */
0051 static int logging_enabled; /* Until we initialize everything,
0052                              * make sure we don't try logging
0053                              * anything */
0054 static int error_log_cnt;
0055
0056 /*
0057  * Since we use 32 bit RTAS, the physical address of this must be below
0058  * 4G or else bad things happen. Allocate this in the kernel data and
0059  * make it big enough.
0060  */
0061 static unsigned char logdata[RTAS_ERROR_LOG_MAX];
0062
0063 static char *rtas_type[] = {
0064     "Unknown", "Retry", "TCE Error", "Internal Device Failure",
0065     "Timeout", "Data Parity", "Address Parity", "Cache Parity",
0066     "Address Invalid", "ECC Uncorrected", "ECC Corrupted",
0067 };
0068
0069 static char *rtas_event_type(int type)
0070 {
0071     if ((type > 0) && (type < 11))
0072         return rtas_type[type];
0073
0074     switch (type) {
0075         case RTAS_TYPE_EPOW:
0076             return "EPOW";
0077         case RTAS_TYPE_PLATFORM:
0078             return "Platform Error";
0079         case RTAS_TYPE_IO:
0080             return "I/O Event";
0081         case RTAS_TYPE_INFO:
0082             return "Platform Information Event";
0083         case RTAS_TYPE_DEALLOC:
0084             return "Resource Deallocation Event";
0085         case RTAS_TYPE_DUMP:
0086             return "Dump Notification Event";
0087         case RTAS_TYPE_PRRN:
0088             return "Platform Resource Reassignment Event";
0089         case RTAS_TYPE_HOTPLUG:
0090             return "Hotplug Event";
0091     }
0092
0093     return rtas_type[0];
0094 }
0095
0096 /* To see this info, grep RTAS /var/log/messages and each entry
0097  * will be collected together with obvious begin/end.
0098  * There will be a unique identifier on the begin and end lines.
0099  * This will persist across reboots.
0100  *
0101  * format of error logs returned from RTAS:
0102  * bytes    (size)  : contents
0103  * --------------------------------------------------------
0104  * 0-7      (8) : rtas_error_log
0105  * 8-47     (40)    : extended info
0106  * 48-51    (4) : vendor id
0107  * 52-1023 (vendor specific) : location code and debug data
0108  */
0109 static void printk_log_rtas(char *buf, int len)
0110 {
0111
0112     int i,j,n = 0;
0113     int perline = 16;
0114     char buffer[64];
0115     char * str = "RTAS event";
0116
0117     if (full_rtas_msgs) {
0118         printk(RTAS_DEBUG "%d -------- %s begin --------\n",
0119                error_log_cnt, str);
0120
0121         /*
0122          * Print perline bytes on each line, each line will start
0123          * with RTAS and a changing number, so syslogd will
0124          * print lines that are otherwise the same.  Separate every
0125          * 4 bytes with a space.
0126          */
0127         for (i = 0; i < len; i++) {
0128             j = i % perline;
0129             if (j == 0) {
0130                 memset(buffer, 0, sizeof(buffer));
0131                 n = sprintf(buffer, "RTAS %d:", i/perline);
0132             }
0133
0134             if ((i % 4) == 0)
0135                 n += sprintf(buffer+n, " ");
0136
0137             n += sprintf(buffer+n, "%02x", (unsigned char)buf[i]);
0138
0139             if (j == (perline-1))
0140                 printk(KERN_DEBUG "%s\n", buffer);
0141         }
0142         if ((i % perline) != 0)
0143             printk(KERN_DEBUG "%s\n", buffer);
0144
0145         printk(RTAS_DEBUG "%d -------- %s end ----------\n",
0146                error_log_cnt, str);
0147     } else {
0148         struct rtas_error_log *errlog = (struct rtas_error_log *)buf;
0149
0150         printk(RTAS_DEBUG "event: %d, Type: %s (%d), Severity: %d\n",
0151                error_log_cnt,
0152                rtas_event_type(rtas_error_type(errlog)),
0153                rtas_error_type(errlog),
0154                rtas_error_severity(errlog));
0155     }
0156 }
0157
0158 static int log_rtas_len(char * buf)
0159 {
0160     int len;
0161     struct rtas_error_log *err;
0162     uint32_t extended_log_length;
0163
0164     /* rtas fixed header */
0165     len = 8;
0166     err = (struct rtas_error_log *)buf;
0167     extended_log_length = rtas_error_extended_log_length(err);
0168     if (rtas_error_extended(err) && extended_log_length) {
0169
0170         /* extended header */
0171         len += extended_log_length;
0172     }
0173
0174     if (rtas_error_log_max == 0)
0175         rtas_error_log_max = rtas_get_error_log_max();
0176
0177     if (len > rtas_error_log_max)
0178         len = rtas_error_log_max;
0179
0180     return len;
0181 }
0182
0183 /*
0184  * First write to nvram, if fatal error, that is the only
0185  * place we log the info.  The error will be picked up
0186  * on the next reboot by rtasd.  If not fatal, run the
0187  * method for the type of error.  Currently, only RTAS
0188  * errors have methods implemented, but in the future
0189  * there might be a need to store data in nvram before a
0190  * call to panic().
0191  *
0192  * XXX We write to nvram periodically, to indicate error has
0193  * been written and sync'd, but there is a possibility
0194  * that if we don't shutdown correctly, a duplicate error
0195  * record will be created on next reboot.
0196  */
0197 void pSeries_log_error(char *buf, unsigned int err_type, int fatal)
0198 {
0199     unsigned long offset;
0200     unsigned long s;
0201     int len = 0;
0202
0203     pr_debug("rtasd: logging event\n");
0204     if (buf == NULL)
0205         return;
0206
0207     spin_lock_irqsave(&rtasd_log_lock, s);
0208
0209     /* get length and increase count */
0210     switch (err_type & ERR_TYPE_MASK) {
0211     case ERR_TYPE_RTAS_LOG:
0212         len = log_rtas_len(buf);
0213         if (!(err_type & ERR_FLAG_BOOT))
0214             error_log_cnt++;
0215         break;
0216     case ERR_TYPE_KERNEL_PANIC:
0217     default:
0218         WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
0219         spin_unlock_irqrestore(&rtasd_log_lock, s);
0220         return;
0221     }
0222
0223 #ifdef CONFIG_PPC64
0224     /* Write error to NVRAM */
0225     if (logging_enabled && !(err_type & ERR_FLAG_BOOT))
0226         nvram_write_error_log(buf, len, err_type, error_log_cnt);
0227 #endif /* CONFIG_PPC64 */
0228
0229     /*
0230      * rtas errors can occur during boot, and we do want to capture
0231      * those somewhere, even if nvram isn't ready (why not?), and even
0232      * if rtasd isn't ready. Put them into the boot log, at least.
0233      */
0234     if ((err_type & ERR_TYPE_MASK) == ERR_TYPE_RTAS_LOG)
0235         printk_log_rtas(buf, len);
0236
0237     /* Check to see if we need to or have stopped logging */
0238     if (fatal || !logging_enabled) {
0239         logging_enabled = 0;
0240         WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
0241         spin_unlock_irqrestore(&rtasd_log_lock, s);
0242         return;
0243     }
0244
0245     /* call type specific method for error */
0246     switch (err_type & ERR_TYPE_MASK) {
0247     case ERR_TYPE_RTAS_LOG:
0248         offset = rtas_error_log_buffer_max *
0249             ((rtas_log_start+rtas_log_size) & LOG_NUMBER_MASK);
0250
0251         /* First copy over sequence number */
0252         memcpy(&rtas_log_buf[offset], (void *) &error_log_cnt, sizeof(int));
0253
0254         /* Second copy over error log data */
0255         offset += sizeof(int);
0256         memcpy(&rtas_log_buf[offset], buf, len);
0257
0258         if (rtas_log_size < LOG_NUMBER)
0259             rtas_log_size += 1;
0260         else
0261             rtas_log_start += 1;
0262
0263         WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
0264         spin_unlock_irqrestore(&rtasd_log_lock, s);
0265         wake_up_interruptible(&rtas_log_wait);
0266         break;
0267     case ERR_TYPE_KERNEL_PANIC:
0268     default:
0269         WARN_ON_ONCE(!irqs_disabled()); /* @@@ DEBUG @@@ */
0270         spin_unlock_irqrestore(&rtasd_log_lock, s);
0271         return;
0272     }
0273 }
0274
0275 static void handle_rtas_event(const struct rtas_error_log *log)
0276 {
0277     if (!machine_is(pseries))
0278         return;
0279
0280     if (rtas_error_type(log) == RTAS_TYPE_PRRN)
0281         pr_info_ratelimited("Platform resource reassignment ignored.\n");
0282 }
0283
0284 static int rtas_log_open(struct inode * inode, struct file * file)
0285 {
0286     return 0;
0287 }
0288
0289 static int rtas_log_release(struct inode * inode, struct file * file)
0290 {
0291     return 0;
0292 }
0293
0294 /* This will check if all events are logged, if they are then, we
0295  * know that we can safely clear the events in NVRAM.
0296  * Next we'll sit and wait for something else to log.
0297  */
0298 static ssize_t rtas_log_read(struct file * file, char __user * buf,
0299              size_t count, loff_t *ppos)
0300 {
0301     int error;
0302     char *tmp;
0303     unsigned long s;
0304     unsigned long offset;
0305
0306     if (!buf || count < rtas_error_log_buffer_max)
0307         return -EINVAL;
0308
0309     count = rtas_error_log_buffer_max;
0310
0311     if (!access_ok(buf, count))
0312         return -EFAULT;
0313
0314     tmp = kmalloc(count, GFP_KERNEL);
0315     if (!tmp)
0316         return -ENOMEM;
0317
0318     spin_lock_irqsave(&rtasd_log_lock, s);
0319
0320     /* if it's 0, then we know we got the last one (the one in NVRAM) */
0321     while (rtas_log_size == 0) {
0322         if (file->f_flags & O_NONBLOCK) {
0323             spin_unlock_irqrestore(&rtasd_log_lock, s);
0324             error = -EAGAIN;
0325             goto out;
0326         }
0327
0328         if (!logging_enabled) {
0329             spin_unlock_irqrestore(&rtasd_log_lock, s);
0330             error = -ENODATA;
0331             goto out;
0332         }
0333 #ifdef CONFIG_PPC64
0334         nvram_clear_error_log();
0335 #endif /* CONFIG_PPC64 */
0336
0337         spin_unlock_irqrestore(&rtasd_log_lock, s);
0338         error = wait_event_interruptible(rtas_log_wait, rtas_log_size);
0339         if (error)
0340             goto out;
0341         spin_lock_irqsave(&rtasd_log_lock, s);
0342     }
0343
0344     offset = rtas_error_log_buffer_max * (rtas_log_start & LOG_NUMBER_MASK);
0345     memcpy(tmp, &rtas_log_buf[offset], count);
0346
0347     rtas_log_start += 1;
0348     rtas_log_size -= 1;
0349     spin_unlock_irqrestore(&rtasd_log_lock, s);
0350
0351     error = copy_to_user(buf, tmp, count) ? -EFAULT : count;
0352 out:
0353     kfree(tmp);
0354     return error;
0355 }
0356
0357 static __poll_t rtas_log_poll(struct file *file, poll_table * wait)
0358 {
0359     poll_wait(file, &rtas_log_wait, wait);
0360     if (rtas_log_size)
0361         return EPOLLIN | EPOLLRDNORM;
0362     return 0;
0363 }
0364
0365 static const struct proc_ops rtas_log_proc_ops = {
0366     .proc_read  = rtas_log_read,
0367     .proc_poll  = rtas_log_poll,
0368     .proc_open  = rtas_log_open,
0369     .proc_release   = rtas_log_release,
0370     .proc_lseek = noop_llseek,
0371 };
0372
0373 static int enable_surveillance(int timeout)
0374 {
0375     int error;
0376
0377     error = rtas_set_indicator(SURVEILLANCE_TOKEN, 0, timeout);
0378
0379     if (error == 0)
0380         return 0;
0381
0382     if (error == -EINVAL) {
0383         printk(KERN_DEBUG "rtasd: surveillance not supported\n");
0384         return 0;
0385     }
0386
0387     printk(KERN_ERR "rtasd: could not update surveillance\n");
0388     return -1;
0389 }
0390
0391 static void do_event_scan(void)
0392 {
0393     int error;
0394     do {
0395         memset(logdata, 0, rtas_error_log_max);
0396         error = rtas_call(event_scan, 4, 1, NULL,
0397                   RTAS_EVENT_SCAN_ALL_EVENTS, 0,
0398                   __pa(logdata), rtas_error_log_max);
0399         if (error == -1) {
0400             printk(KERN_ERR "event-scan failed\n");
0401             break;
0402         }
0403
0404         if (error == 0) {
0405             if (rtas_error_type((struct rtas_error_log *)logdata) !=
0406                 RTAS_TYPE_PRRN)
0407                 pSeries_log_error(logdata, ERR_TYPE_RTAS_LOG,
0408                           0);
0409             handle_rtas_event((struct rtas_error_log *)logdata);
0410         }
0411
0412     } while(error == 0);
0413 }
0414
0415 static void rtas_event_scan(struct work_struct *w);
0416 static DECLARE_DELAYED_WORK(event_scan_work, rtas_event_scan);
0417
0418 /*
0419  * Delay should be at least one second since some machines have problems if
0420  * we call event-scan too quickly.
0421  */
0422 static unsigned long event_scan_delay = 1*HZ;
0423 static int first_pass = 1;
0424
0425 static void rtas_event_scan(struct work_struct *w)
0426 {
0427     unsigned int cpu;
0428
0429     do_event_scan();
0430
0431     cpus_read_lock();
0432
0433     /* raw_ OK because just using CPU as starting point. */
0434     cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
0435         if (cpu >= nr_cpu_ids) {
0436         cpu = cpumask_first(cpu_online_mask);
0437
0438         if (first_pass) {
0439             first_pass = 0;
0440             event_scan_delay = 30*HZ/rtas_event_scan_rate;
0441
0442             if (surveillance_timeout != -1) {
0443                 pr_debug("rtasd: enabling surveillance\n");
0444                 enable_surveillance(surveillance_timeout);
0445                 pr_debug("rtasd: surveillance enabled\n");
0446             }
0447         }
0448     }
0449
0450     schedule_delayed_work_on(cpu, &event_scan_work,
0451         __round_jiffies_relative(event_scan_delay, cpu));
0452
0453     cpus_read_unlock();
0454 }
0455
0456 #ifdef CONFIG_PPC64
0457 static void __init retrieve_nvram_error_log(void)
0458 {
0459     unsigned int err_type ;
0460     int rc ;
0461
0462     /* See if we have any error stored in NVRAM */
0463     memset(logdata, 0, rtas_error_log_max);
0464     rc = nvram_read_error_log(logdata, rtas_error_log_max,
0465                               &err_type, &error_log_cnt);
0466     /* We can use rtas_log_buf now */
0467     logging_enabled = 1;
0468     if (!rc) {
0469         if (err_type != ERR_FLAG_ALREADY_LOGGED) {
0470             pSeries_log_error(logdata, err_type | ERR_FLAG_BOOT, 0);
0471         }
0472     }
0473 }
0474 #else /* CONFIG_PPC64 */
0475 static void __init retrieve_nvram_error_log(void)
0476 {
0477 }
0478 #endif /* CONFIG_PPC64 */
0479
0480 static void __init start_event_scan(void)
0481 {
0482     printk(KERN_DEBUG "RTAS daemon started\n");
0483     pr_debug("rtasd: will sleep for %d milliseconds\n",
0484          (30000 / rtas_event_scan_rate));
0485
0486     /* Retrieve errors from nvram if any */
0487     retrieve_nvram_error_log();
0488
0489     schedule_delayed_work_on(cpumask_first(cpu_online_mask),
0490                  &event_scan_work, event_scan_delay);
0491 }
0492
0493 /* Cancel the rtas event scan work */
0494 void rtas_cancel_event_scan(void)
0495 {
0496     cancel_delayed_work_sync(&event_scan_work);
0497 }
0498 EXPORT_SYMBOL_GPL(rtas_cancel_event_scan);
0499
0500 static int __init rtas_event_scan_init(void)
0501 {
0502     if (!machine_is(pseries) && !machine_is(chrp))
0503         return 0;
0504
0505     /* No RTAS */
0506     event_scan = rtas_token("event-scan");
0507     if (event_scan == RTAS_UNKNOWN_SERVICE) {
0508         printk(KERN_INFO "rtasd: No event-scan on system\n");
0509         return -ENODEV;
0510     }
0511
0512     rtas_event_scan_rate = rtas_token("rtas-event-scan-rate");
0513     if (rtas_event_scan_rate == RTAS_UNKNOWN_SERVICE) {
0514         printk(KERN_ERR "rtasd: no rtas-event-scan-rate on system\n");
0515         return -ENODEV;
0516     }
0517
0518     if (!rtas_event_scan_rate) {
0519         /* Broken firmware: take a rate of zero to mean don't scan */
0520         printk(KERN_DEBUG "rtasd: scan rate is 0, not scanning\n");
0521         return 0;
0522     }
0523
0524     /* Make room for the sequence number */
0525     rtas_error_log_max = rtas_get_error_log_max();
0526     rtas_error_log_buffer_max = rtas_error_log_max + sizeof(int);
0527
0528     rtas_log_buf = vmalloc(array_size(LOG_NUMBER,
0529                       rtas_error_log_buffer_max));
0530     if (!rtas_log_buf) {
0531         printk(KERN_ERR "rtasd: no memory\n");
0532         return -ENOMEM;
0533     }
0534
0535     start_event_scan();
0536
0537     return 0;
0538 }
0539 arch_initcall(rtas_event_scan_init);
0540
0541 static int __init rtas_init(void)
0542 {
0543     struct proc_dir_entry *entry;
0544
0545     if (!machine_is(pseries) && !machine_is(chrp))
0546         return 0;
0547
0548     if (!rtas_log_buf)
0549         return -ENODEV;
0550
0551     entry = proc_create("powerpc/rtas/error_log", 0400, NULL,
0552                 &rtas_log_proc_ops);
0553     if (!entry)
0554         printk(KERN_ERR "Failed to create error_log proc entry\n");
0555
0556     return 0;
0557 }
0558 __initcall(rtas_init);
0559
0560 static int __init surveillance_setup(char *str)
0561 {
0562     int i;
0563
0564     /* We only do surveillance on pseries */
0565     if (!machine_is(pseries))
0566         return 0;
0567
0568     if (get_option(&str,&i)) {
0569         if (i >= 0 && i <= 255)
0570             surveillance_timeout = i;
0571     }
0572
0573     return 1;
0574 }
0575 __setup("surveillance=", surveillance_setup);
0576
0577 static int __init rtasmsgs_setup(char *str)
0578 {
0579     return (kstrtobool(str, &full_rtas_msgs) == 0);
0580 }
0581 __setup("rtasmsgs=", rtasmsgs_setup);