2 * c 2001 PPC 64 Team, IBM Corp
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 * /dev/nvram driver for PPC64
11 * This perhaps should live in drivers/char
15 #include <linux/types.h>
16 #include <linux/errno.h>
17 #include <linux/init.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/kmsg_dump.h>
21 #include <linux/pstore.h>
22 #include <linux/ctype.h>
23 #include <linux/zlib.h>
24 #include <asm/uaccess.h>
25 #include <asm/nvram.h>
28 #include <asm/machdep.h>
30 /* Max bytes to read/write in one go */
34 * Set oops header version to distinguish between old and new format header.
35 * lnx,oops-log partition max size is 4000, header version > 4000 will
36 * help in identifying new header.
38 #define OOPS_HDR_VERSION 5000
40 static unsigned int nvram_size;
41 static int nvram_fetch, nvram_store;
42 static char nvram_buf[NVRW_CNT]; /* assume this is in the first 4GB */
43 static DEFINE_SPINLOCK(nvram_lock);
50 struct nvram_os_partition {
52 int req_size; /* desired size, in bytes */
53 int min_size; /* minimum acceptable size (0 means req_size) */
54 long size; /* size of data portion (excluding err_log_info) */
55 long index; /* offset of data portion of partition */
56 bool os_partition; /* partition initialized by OS, not FW */
59 static struct nvram_os_partition rtas_log_partition = {
60 .name = "ibm,rtas-log",
67 static struct nvram_os_partition oops_log_partition = {
68 .name = "lnx,oops-log",
75 static const char *pseries_nvram_os_partitions[] = {
81 struct oops_log_info {
85 } __attribute__((packed));
87 static void oops_to_nvram(struct kmsg_dumper *dumper,
88 enum kmsg_dump_reason reason);
90 static struct kmsg_dumper nvram_kmsg_dumper = {
94 /* See clobbering_unread_rtas_event() */
95 #define NVRAM_RTAS_READ_TIMEOUT 5 /* seconds */
96 static unsigned long last_unread_rtas_event; /* timestamp */
99 * For capturing and compressing an oops or panic report...
101 * big_oops_buf[] holds the uncompressed text we're capturing.
103 * oops_buf[] holds the compressed text, preceded by a oops header.
104 * oops header has u16 holding the version of oops header (to differentiate
105 * between old and new format header) followed by u16 holding the length of
106 * the compressed* text (*Or uncompressed, if compression fails.) and u64
107 * holding the timestamp. oops_buf[] gets written to NVRAM.
109 * oops_log_info points to the header. oops_data points to the compressed text.
114 * +-----------+-----------+-----------+------------------------+
115 * | version | length | timestamp | text |
116 * | (2 bytes) | (2 bytes) | (8 bytes) | (oops_data_sz bytes) |
117 * +-----------+-----------+-----------+------------------------+
121 * We preallocate these buffers during init to avoid kmalloc during oops/panic.
123 static size_t big_oops_buf_sz;
124 static char *big_oops_buf, *oops_buf;
125 static char *oops_data;
126 static size_t oops_data_sz;
128 /* Compression parameters */
129 #define COMPR_LEVEL 6
130 #define WINDOW_BITS 12
132 static struct z_stream_s stream;
135 static struct nvram_os_partition of_config_partition = {
138 .os_partition = false
141 static struct nvram_os_partition common_partition = {
144 .os_partition = false
147 static enum pstore_type_id nvram_type_ids[] = {
149 PSTORE_TYPE_PPC_RTAS,
151 PSTORE_TYPE_PPC_COMMON,
154 static int read_type;
155 static unsigned long last_rtas_event;
158 static ssize_t pSeries_nvram_read(char *buf, size_t count, loff_t *index)
167 if (nvram_size == 0 || nvram_fetch == RTAS_UNKNOWN_SERVICE)
170 if (*index >= nvram_size)
174 if (i + count > nvram_size)
175 count = nvram_size - i;
177 spin_lock_irqsave(&nvram_lock, flags);
179 for (; count != 0; count -= len) {
184 if ((rtas_call(nvram_fetch, 3, 2, &done, i, __pa(nvram_buf),
185 len) != 0) || len != done) {
186 spin_unlock_irqrestore(&nvram_lock, flags);
190 memcpy(p, nvram_buf, len);
196 spin_unlock_irqrestore(&nvram_lock, flags);
202 static ssize_t pSeries_nvram_write(char *buf, size_t count, loff_t *index)
210 if (nvram_size == 0 || nvram_store == RTAS_UNKNOWN_SERVICE)
213 if (*index >= nvram_size)
217 if (i + count > nvram_size)
218 count = nvram_size - i;
220 spin_lock_irqsave(&nvram_lock, flags);
222 for (; count != 0; count -= len) {
227 memcpy(nvram_buf, p, len);
229 if ((rtas_call(nvram_store, 3, 2, &done, i, __pa(nvram_buf),
230 len) != 0) || len != done) {
231 spin_unlock_irqrestore(&nvram_lock, flags);
238 spin_unlock_irqrestore(&nvram_lock, flags);
244 static ssize_t pSeries_nvram_get_size(void)
246 return nvram_size ? nvram_size : -ENODEV;
250 /* nvram_write_os_partition, nvram_write_error_log
252 * We need to buffer the error logs into nvram to ensure that we have
253 * the failure information to decode. If we have a severe error there
254 * is no way to guarantee that the OS or the machine is in a state to
255 * get back to user land and write the error to disk. For example if
256 * the SCSI device driver causes a Machine Check by writing to a bad
257 * IO address, there is no way of guaranteeing that the device driver
258 * is in any state that is would also be able to write the error data
259 * captured to disk, thus we buffer it in NVRAM for analysis on the
262 * In NVRAM the partition containing the error log buffer will looks like:
264 * +-----------+----------+--------+------------+------------------+
265 * | signature | checksum | length | name | data |
266 * |0 |1 |2 3|4 15|16 length-1|
267 * +-----------+----------+--------+------------+------------------+
269 * The 'data' section would look like (in bytes):
270 * +--------------+------------+-----------------------------------+
271 * | event_logged | sequence # | error log |
272 * |0 3|4 7|8 error_log_size-1|
273 * +--------------+------------+-----------------------------------+
275 * event_logged: 0 if event has not been logged to syslog, 1 if it has
276 * sequence #: The unique sequence # for each event. (until it wraps)
277 * error log: The error log from event_scan
279 int nvram_write_os_partition(struct nvram_os_partition *part, char * buff,
280 int length, unsigned int err_type, unsigned int error_log_cnt)
284 struct err_log_info info;
286 if (part->index == -1) {
290 if (length > part->size) {
294 info.error_type = cpu_to_be32(err_type);
295 info.seq_num = cpu_to_be32(error_log_cnt);
297 tmp_index = part->index;
299 rc = ppc_md.nvram_write((char *)&info, sizeof(struct err_log_info), &tmp_index);
301 pr_err("%s: Failed nvram_write (%d)\n", __FUNCTION__, rc);
305 rc = ppc_md.nvram_write(buff, length, &tmp_index);
307 pr_err("%s: Failed nvram_write (%d)\n", __FUNCTION__, rc);
314 int nvram_write_error_log(char * buff, int length,
315 unsigned int err_type, unsigned int error_log_cnt)
317 int rc = nvram_write_os_partition(&rtas_log_partition, buff, length,
318 err_type, error_log_cnt);
320 last_unread_rtas_event = get_seconds();
322 last_rtas_event = get_seconds();
329 /* nvram_read_partition
331 * Reads nvram partition for at most 'length'
333 int nvram_read_partition(struct nvram_os_partition *part, char *buff,
334 int length, unsigned int *err_type,
335 unsigned int *error_log_cnt)
339 struct err_log_info info;
341 if (part->index == -1)
344 if (length > part->size)
347 tmp_index = part->index;
349 if (part->os_partition) {
350 rc = ppc_md.nvram_read((char *)&info,
351 sizeof(struct err_log_info),
354 pr_err("%s: Failed nvram_read (%d)\n", __FUNCTION__,
360 rc = ppc_md.nvram_read(buff, length, &tmp_index);
362 pr_err("%s: Failed nvram_read (%d)\n", __FUNCTION__, rc);
366 if (part->os_partition) {
367 *error_log_cnt = be32_to_cpu(info.seq_num);
368 *err_type = be32_to_cpu(info.error_type);
374 /* nvram_read_error_log
376 * Reads nvram for error log for at most 'length'
378 int nvram_read_error_log(char *buff, int length,
379 unsigned int *err_type, unsigned int *error_log_cnt)
381 return nvram_read_partition(&rtas_log_partition, buff, length,
382 err_type, error_log_cnt);
385 /* This doesn't actually zero anything, but it sets the event_logged
386 * word to tell that this event is safely in syslog.
388 int nvram_clear_error_log(void)
391 int clear_word = ERR_FLAG_ALREADY_LOGGED;
394 if (rtas_log_partition.index == -1)
397 tmp_index = rtas_log_partition.index;
399 rc = ppc_md.nvram_write((char *)&clear_word, sizeof(int), &tmp_index);
401 printk(KERN_ERR "nvram_clear_error_log: Failed nvram_write (%d)\n", rc);
404 last_unread_rtas_event = 0;
409 /* pseries_nvram_init_os_partition
411 * This sets up a partition with an "OS" signature.
413 * The general strategy is the following:
414 * 1.) If a partition with the indicated name already exists...
415 * - If it's large enough, use it.
416 * - Otherwise, recycle it and keep going.
417 * 2.) Search for a free partition that is large enough.
418 * 3.) If there's not a free partition large enough, recycle any obsolete
419 * OS partitions and try again.
420 * 4.) Will first try getting a chunk that will satisfy the requested size.
421 * 5.) If a chunk of the requested size cannot be allocated, then try finding
422 * a chunk that will satisfy the minum needed.
424 * Returns 0 on success, else -1.
426 static int __init pseries_nvram_init_os_partition(struct nvram_os_partition
433 p = nvram_find_partition(part->name, NVRAM_SIG_OS, &size);
435 /* Found one but too small, remove it */
436 if (p && size < part->min_size) {
437 pr_info("nvram: Found too small %s partition,"
438 " removing it...\n", part->name);
439 nvram_remove_partition(part->name, NVRAM_SIG_OS, NULL);
443 /* Create one if we didn't find */
445 p = nvram_create_partition(part->name, NVRAM_SIG_OS,
446 part->req_size, part->min_size);
448 pr_info("nvram: No room to create %s partition, "
449 "deleting any obsolete OS partitions...\n",
451 nvram_remove_partition(NULL, NVRAM_SIG_OS,
452 pseries_nvram_os_partitions);
453 p = nvram_create_partition(part->name, NVRAM_SIG_OS,
454 part->req_size, part->min_size);
459 pr_err("nvram: Failed to find or create %s"
460 " partition, err %d\n", part->name, (int)p);
465 part->size = nvram_get_partition_size(p) - sizeof(struct err_log_info);
471 * Are we using the ibm,rtas-log for oops/panic reports? And if so,
472 * would logging this oops/panic overwrite an RTAS event that rtas_errd
473 * hasn't had a chance to read and process? Return 1 if so, else 0.
475 * We assume that if rtas_errd hasn't read the RTAS event in
476 * NVRAM_RTAS_READ_TIMEOUT seconds, it's probably not going to.
478 static int clobbering_unread_rtas_event(void)
480 return (oops_log_partition.index == rtas_log_partition.index
481 && last_unread_rtas_event
482 && get_seconds() - last_unread_rtas_event <=
483 NVRAM_RTAS_READ_TIMEOUT);
486 /* Derived from logfs_compress() */
487 static int nvram_compress(const void *in, void *out, size_t inlen,
493 err = zlib_deflateInit2(&stream, COMPR_LEVEL, Z_DEFLATED, WINDOW_BITS,
494 MEM_LEVEL, Z_DEFAULT_STRATEGY);
499 stream.avail_in = inlen;
501 stream.next_out = out;
502 stream.avail_out = outlen;
503 stream.total_out = 0;
505 err = zlib_deflate(&stream, Z_FINISH);
506 if (err != Z_STREAM_END)
509 err = zlib_deflateEnd(&stream);
513 if (stream.total_out >= stream.total_in)
516 ret = stream.total_out;
521 /* Compress the text from big_oops_buf into oops_buf. */
522 static int zip_oops(size_t text_len)
524 struct oops_log_info *oops_hdr = (struct oops_log_info *)oops_buf;
525 int zipped_len = nvram_compress(big_oops_buf, oops_data, text_len,
527 if (zipped_len < 0) {
528 pr_err("nvram: compression failed; returned %d\n", zipped_len);
529 pr_err("nvram: logging uncompressed oops/panic report\n");
532 oops_hdr->version = cpu_to_be16(OOPS_HDR_VERSION);
533 oops_hdr->report_length = cpu_to_be16(zipped_len);
534 oops_hdr->timestamp = cpu_to_be64(get_seconds());
539 static int nvram_pstore_open(struct pstore_info *psi)
541 /* Reset the iterator to start reading partitions again */
547 * nvram_pstore_write - pstore write callback for nvram
548 * @type: Type of message logged
549 * @reason: reason behind dump (oops/panic)
550 * @id: identifier to indicate the write performed
551 * @part: pstore writes data to registered buffer in parts,
552 * part number will indicate the same.
553 * @count: Indicates oops count
554 * @compressed: Flag to indicate the log is compressed
555 * @size: number of bytes written to the registered buffer
556 * @psi: registered pstore_info structure
558 * Called by pstore_dump() when an oops or panic report is logged in the
560 * Returns 0 on successful write.
562 static int nvram_pstore_write(enum pstore_type_id type,
563 enum kmsg_dump_reason reason,
564 u64 *id, unsigned int part, int count,
565 bool compressed, size_t size,
566 struct pstore_info *psi)
569 unsigned int err_type = ERR_TYPE_KERNEL_PANIC;
570 struct oops_log_info *oops_hdr = (struct oops_log_info *) oops_buf;
572 /* part 1 has the recent messages from printk buffer */
573 if (part > 1 || type != PSTORE_TYPE_DMESG ||
574 clobbering_unread_rtas_event())
577 oops_hdr->version = cpu_to_be16(OOPS_HDR_VERSION);
578 oops_hdr->report_length = cpu_to_be16(size);
579 oops_hdr->timestamp = cpu_to_be64(get_seconds());
582 err_type = ERR_TYPE_KERNEL_PANIC_GZ;
584 rc = nvram_write_os_partition(&oops_log_partition, oops_buf,
585 (int) (sizeof(*oops_hdr) + size), err_type, count);
595 * Reads the oops/panic report, rtas, of-config and common partition.
596 * Returns the length of the data we read from each partition.
597 * Returns 0 if we've been called before.
599 static ssize_t nvram_pstore_read(u64 *id, enum pstore_type_id *type,
600 int *count, struct timespec *time, char **buf,
601 bool *compressed, struct pstore_info *psi)
603 struct oops_log_info *oops_hdr;
604 unsigned int err_type, id_no, size = 0;
605 struct nvram_os_partition *part = NULL;
612 switch (nvram_type_ids[read_type]) {
613 case PSTORE_TYPE_DMESG:
614 part = &oops_log_partition;
615 *type = PSTORE_TYPE_DMESG;
617 case PSTORE_TYPE_PPC_RTAS:
618 part = &rtas_log_partition;
619 *type = PSTORE_TYPE_PPC_RTAS;
620 time->tv_sec = last_rtas_event;
623 case PSTORE_TYPE_PPC_OF:
625 part = &of_config_partition;
626 *type = PSTORE_TYPE_PPC_OF;
627 *id = PSTORE_TYPE_PPC_OF;
631 case PSTORE_TYPE_PPC_COMMON:
633 part = &common_partition;
634 *type = PSTORE_TYPE_PPC_COMMON;
635 *id = PSTORE_TYPE_PPC_COMMON;
643 if (!part->os_partition) {
644 p = nvram_find_partition(part->name, sig, &size);
646 pr_err("nvram: Failed to find partition %s, "
647 "err %d\n", part->name, (int)p);
654 buff = kmalloc(part->size, GFP_KERNEL);
659 if (nvram_read_partition(part, buff, part->size, &err_type, &id_no)) {
666 if (part->os_partition)
669 if (nvram_type_ids[read_type] == PSTORE_TYPE_DMESG) {
670 size_t length, hdr_size;
672 oops_hdr = (struct oops_log_info *)buff;
673 if (be16_to_cpu(oops_hdr->version) < OOPS_HDR_VERSION) {
674 /* Old format oops header had 2-byte record size */
675 hdr_size = sizeof(u16);
676 length = be16_to_cpu(oops_hdr->version);
680 hdr_size = sizeof(*oops_hdr);
681 length = be16_to_cpu(oops_hdr->report_length);
682 time->tv_sec = be64_to_cpu(oops_hdr->timestamp);
685 *buf = kmalloc(length, GFP_KERNEL);
688 memcpy(*buf, buff + hdr_size, length);
691 if (err_type == ERR_TYPE_KERNEL_PANIC_GZ)
702 static struct pstore_info nvram_pstore_info = {
703 .owner = THIS_MODULE,
705 .open = nvram_pstore_open,
706 .read = nvram_pstore_read,
707 .write = nvram_pstore_write,
710 static int nvram_pstore_init(void)
714 nvram_pstore_info.buf = oops_data;
715 nvram_pstore_info.bufsize = oops_data_sz;
717 rc = pstore_register(&nvram_pstore_info);
719 pr_err("nvram: pstore_register() failed, defaults to "
720 "kmsg_dump; returned %d\n", rc);
725 static int nvram_pstore_init(void)
731 static void __init nvram_init_oops_partition(int rtas_partition_exists)
735 rc = pseries_nvram_init_os_partition(&oops_log_partition);
737 if (!rtas_partition_exists)
739 pr_notice("nvram: Using %s partition to log both"
740 " RTAS errors and oops/panic reports\n",
741 rtas_log_partition.name);
742 memcpy(&oops_log_partition, &rtas_log_partition,
743 sizeof(rtas_log_partition));
745 oops_buf = kmalloc(oops_log_partition.size, GFP_KERNEL);
747 pr_err("nvram: No memory for %s partition\n",
748 oops_log_partition.name);
751 oops_data = oops_buf + sizeof(struct oops_log_info);
752 oops_data_sz = oops_log_partition.size - sizeof(struct oops_log_info);
754 rc = nvram_pstore_init();
760 * Figure compression (preceded by elimination of each line's <n>
761 * severity prefix) will reduce the oops/panic report to at most
762 * 45% of its original size.
764 big_oops_buf_sz = (oops_data_sz * 100) / 45;
765 big_oops_buf = kmalloc(big_oops_buf_sz, GFP_KERNEL);
767 stream.workspace = kmalloc(zlib_deflate_workspacesize(
768 WINDOW_BITS, MEM_LEVEL), GFP_KERNEL);
769 if (!stream.workspace) {
770 pr_err("nvram: No memory for compression workspace; "
771 "skipping compression of %s partition data\n",
772 oops_log_partition.name);
777 pr_err("No memory for uncompressed %s data; "
778 "skipping compression\n", oops_log_partition.name);
779 stream.workspace = NULL;
782 rc = kmsg_dump_register(&nvram_kmsg_dumper);
784 pr_err("nvram: kmsg_dump_register() failed; returned %d\n", rc);
787 kfree(stream.workspace);
791 static int __init pseries_nvram_init_log_partitions(void)
795 /* Scan nvram for partitions */
796 nvram_scan_partitions();
798 rc = pseries_nvram_init_os_partition(&rtas_log_partition);
799 nvram_init_oops_partition(rc == 0);
802 machine_arch_initcall(pseries, pseries_nvram_init_log_partitions);
804 int __init pSeries_nvram_init(void)
806 struct device_node *nvram;
807 const __be32 *nbytes_p;
808 unsigned int proplen;
810 nvram = of_find_node_by_type(NULL, "nvram");
814 nbytes_p = of_get_property(nvram, "#bytes", &proplen);
815 if (nbytes_p == NULL || proplen != sizeof(unsigned int)) {
820 nvram_size = be32_to_cpup(nbytes_p);
822 nvram_fetch = rtas_token("nvram-fetch");
823 nvram_store = rtas_token("nvram-store");
824 printk(KERN_INFO "PPC64 nvram contains %d bytes\n", nvram_size);
827 ppc_md.nvram_read = pSeries_nvram_read;
828 ppc_md.nvram_write = pSeries_nvram_write;
829 ppc_md.nvram_size = pSeries_nvram_get_size;
836 * This is our kmsg_dump callback, called after an oops or panic report
837 * has been written to the printk buffer. We want to capture as much
838 * of the printk buffer as possible. First, capture as much as we can
839 * that we think will compress sufficiently to fit in the lnx,oops-log
840 * partition. If that's too much, go back and capture uncompressed text.
842 static void oops_to_nvram(struct kmsg_dumper *dumper,
843 enum kmsg_dump_reason reason)
845 struct oops_log_info *oops_hdr = (struct oops_log_info *)oops_buf;
846 static unsigned int oops_count = 0;
847 static bool panicking = false;
848 static DEFINE_SPINLOCK(lock);
851 unsigned int err_type = ERR_TYPE_KERNEL_PANIC_GZ;
855 case KMSG_DUMP_RESTART:
857 case KMSG_DUMP_POWEROFF:
858 /* These are almost always orderly shutdowns. */
862 case KMSG_DUMP_PANIC:
865 case KMSG_DUMP_EMERG:
867 /* Panic report already captured. */
871 pr_err("%s: ignoring unrecognized KMSG_DUMP_* reason %d\n",
872 __FUNCTION__, (int) reason);
876 if (clobbering_unread_rtas_event())
879 if (!spin_trylock_irqsave(&lock, flags))
883 kmsg_dump_get_buffer(dumper, false,
884 big_oops_buf, big_oops_buf_sz, &text_len);
885 rc = zip_oops(text_len);
888 kmsg_dump_rewind(dumper);
889 kmsg_dump_get_buffer(dumper, false,
890 oops_data, oops_data_sz, &text_len);
891 err_type = ERR_TYPE_KERNEL_PANIC;
892 oops_hdr->version = cpu_to_be16(OOPS_HDR_VERSION);
893 oops_hdr->report_length = cpu_to_be16(text_len);
894 oops_hdr->timestamp = cpu_to_be64(get_seconds());
897 (void) nvram_write_os_partition(&oops_log_partition, oops_buf,
898 (int) (sizeof(*oops_hdr) + text_len), err_type,
901 spin_unlock_irqrestore(&lock, flags);