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 static int nvram_write_os_partition(struct nvram_os_partition *part,
280 char *buff, int length,
281 unsigned int err_type,
282 unsigned int error_log_cnt)
286 struct err_log_info info;
288 if (part->index == -1) {
292 if (length > part->size) {
296 info.error_type = cpu_to_be32(err_type);
297 info.seq_num = cpu_to_be32(error_log_cnt);
299 tmp_index = part->index;
301 rc = ppc_md.nvram_write((char *)&info, sizeof(struct err_log_info), &tmp_index);
303 pr_err("%s: Failed nvram_write (%d)\n", __func__, rc);
307 rc = ppc_md.nvram_write(buff, length, &tmp_index);
309 pr_err("%s: Failed nvram_write (%d)\n", __func__, rc);
316 int nvram_write_error_log(char * buff, int length,
317 unsigned int err_type, unsigned int error_log_cnt)
319 int rc = nvram_write_os_partition(&rtas_log_partition, buff, length,
320 err_type, error_log_cnt);
322 last_unread_rtas_event = get_seconds();
324 last_rtas_event = get_seconds();
331 /* nvram_read_partition
333 * Reads nvram partition for at most 'length'
335 static int nvram_read_partition(struct nvram_os_partition *part, char *buff,
336 int length, unsigned int *err_type,
337 unsigned int *error_log_cnt)
341 struct err_log_info info;
343 if (part->index == -1)
346 if (length > part->size)
349 tmp_index = part->index;
351 if (part->os_partition) {
352 rc = ppc_md.nvram_read((char *)&info,
353 sizeof(struct err_log_info),
356 pr_err("%s: Failed nvram_read (%d)\n", __func__, rc);
361 rc = ppc_md.nvram_read(buff, length, &tmp_index);
363 pr_err("%s: Failed nvram_read (%d)\n", __func__, rc);
367 if (part->os_partition) {
368 *error_log_cnt = be32_to_cpu(info.seq_num);
369 *err_type = be32_to_cpu(info.error_type);
375 /* nvram_read_error_log
377 * Reads nvram for error log for at most 'length'
379 int nvram_read_error_log(char *buff, int length,
380 unsigned int *err_type, unsigned int *error_log_cnt)
382 return nvram_read_partition(&rtas_log_partition, buff, length,
383 err_type, error_log_cnt);
386 /* This doesn't actually zero anything, but it sets the event_logged
387 * word to tell that this event is safely in syslog.
389 int nvram_clear_error_log(void)
392 int clear_word = ERR_FLAG_ALREADY_LOGGED;
395 if (rtas_log_partition.index == -1)
398 tmp_index = rtas_log_partition.index;
400 rc = ppc_md.nvram_write((char *)&clear_word, sizeof(int), &tmp_index);
402 printk(KERN_ERR "nvram_clear_error_log: Failed nvram_write (%d)\n", rc);
405 last_unread_rtas_event = 0;
410 /* pseries_nvram_init_os_partition
412 * This sets up a partition with an "OS" signature.
414 * The general strategy is the following:
415 * 1.) If a partition with the indicated name already exists...
416 * - If it's large enough, use it.
417 * - Otherwise, recycle it and keep going.
418 * 2.) Search for a free partition that is large enough.
419 * 3.) If there's not a free partition large enough, recycle any obsolete
420 * OS partitions and try again.
421 * 4.) Will first try getting a chunk that will satisfy the requested size.
422 * 5.) If a chunk of the requested size cannot be allocated, then try finding
423 * a chunk that will satisfy the minum needed.
425 * Returns 0 on success, else -1.
427 static int __init pseries_nvram_init_os_partition(struct nvram_os_partition
434 p = nvram_find_partition(part->name, NVRAM_SIG_OS, &size);
436 /* Found one but too small, remove it */
437 if (p && size < part->min_size) {
438 pr_info("nvram: Found too small %s partition,"
439 " removing it...\n", part->name);
440 nvram_remove_partition(part->name, NVRAM_SIG_OS, NULL);
444 /* Create one if we didn't find */
446 p = nvram_create_partition(part->name, NVRAM_SIG_OS,
447 part->req_size, part->min_size);
449 pr_info("nvram: No room to create %s partition, "
450 "deleting any obsolete OS partitions...\n",
452 nvram_remove_partition(NULL, NVRAM_SIG_OS,
453 pseries_nvram_os_partitions);
454 p = nvram_create_partition(part->name, NVRAM_SIG_OS,
455 part->req_size, part->min_size);
460 pr_err("nvram: Failed to find or create %s"
461 " partition, err %d\n", part->name, (int)p);
466 part->size = nvram_get_partition_size(p) - sizeof(struct err_log_info);
472 * Are we using the ibm,rtas-log for oops/panic reports? And if so,
473 * would logging this oops/panic overwrite an RTAS event that rtas_errd
474 * hasn't had a chance to read and process? Return 1 if so, else 0.
476 * We assume that if rtas_errd hasn't read the RTAS event in
477 * NVRAM_RTAS_READ_TIMEOUT seconds, it's probably not going to.
479 static int clobbering_unread_rtas_event(void)
481 return (oops_log_partition.index == rtas_log_partition.index
482 && last_unread_rtas_event
483 && get_seconds() - last_unread_rtas_event <=
484 NVRAM_RTAS_READ_TIMEOUT);
487 /* Derived from logfs_compress() */
488 static int nvram_compress(const void *in, void *out, size_t inlen,
494 err = zlib_deflateInit2(&stream, COMPR_LEVEL, Z_DEFLATED, WINDOW_BITS,
495 MEM_LEVEL, Z_DEFAULT_STRATEGY);
500 stream.avail_in = inlen;
502 stream.next_out = out;
503 stream.avail_out = outlen;
504 stream.total_out = 0;
506 err = zlib_deflate(&stream, Z_FINISH);
507 if (err != Z_STREAM_END)
510 err = zlib_deflateEnd(&stream);
514 if (stream.total_out >= stream.total_in)
517 ret = stream.total_out;
522 /* Compress the text from big_oops_buf into oops_buf. */
523 static int zip_oops(size_t text_len)
525 struct oops_log_info *oops_hdr = (struct oops_log_info *)oops_buf;
526 int zipped_len = nvram_compress(big_oops_buf, oops_data, text_len,
528 if (zipped_len < 0) {
529 pr_err("nvram: compression failed; returned %d\n", zipped_len);
530 pr_err("nvram: logging uncompressed oops/panic report\n");
533 oops_hdr->version = cpu_to_be16(OOPS_HDR_VERSION);
534 oops_hdr->report_length = cpu_to_be16(zipped_len);
535 oops_hdr->timestamp = cpu_to_be64(get_seconds());
540 static int nvram_pstore_open(struct pstore_info *psi)
542 /* Reset the iterator to start reading partitions again */
548 * nvram_pstore_write - pstore write callback for nvram
549 * @type: Type of message logged
550 * @reason: reason behind dump (oops/panic)
551 * @id: identifier to indicate the write performed
552 * @part: pstore writes data to registered buffer in parts,
553 * part number will indicate the same.
554 * @count: Indicates oops count
555 * @compressed: Flag to indicate the log is compressed
556 * @size: number of bytes written to the registered buffer
557 * @psi: registered pstore_info structure
559 * Called by pstore_dump() when an oops or panic report is logged in the
561 * Returns 0 on successful write.
563 static int nvram_pstore_write(enum pstore_type_id type,
564 enum kmsg_dump_reason reason,
565 u64 *id, unsigned int part, int count,
566 bool compressed, size_t size,
567 struct pstore_info *psi)
570 unsigned int err_type = ERR_TYPE_KERNEL_PANIC;
571 struct oops_log_info *oops_hdr = (struct oops_log_info *) oops_buf;
573 /* part 1 has the recent messages from printk buffer */
574 if (part > 1 || type != PSTORE_TYPE_DMESG ||
575 clobbering_unread_rtas_event())
578 oops_hdr->version = cpu_to_be16(OOPS_HDR_VERSION);
579 oops_hdr->report_length = cpu_to_be16(size);
580 oops_hdr->timestamp = cpu_to_be64(get_seconds());
583 err_type = ERR_TYPE_KERNEL_PANIC_GZ;
585 rc = nvram_write_os_partition(&oops_log_partition, oops_buf,
586 (int) (sizeof(*oops_hdr) + size), err_type, count);
596 * Reads the oops/panic report, rtas, of-config and common partition.
597 * Returns the length of the data we read from each partition.
598 * Returns 0 if we've been called before.
600 static ssize_t nvram_pstore_read(u64 *id, enum pstore_type_id *type,
601 int *count, struct timespec *time, char **buf,
602 bool *compressed, struct pstore_info *psi)
604 struct oops_log_info *oops_hdr;
605 unsigned int err_type, id_no, size = 0;
606 struct nvram_os_partition *part = NULL;
613 switch (nvram_type_ids[read_type]) {
614 case PSTORE_TYPE_DMESG:
615 part = &oops_log_partition;
616 *type = PSTORE_TYPE_DMESG;
618 case PSTORE_TYPE_PPC_RTAS:
619 part = &rtas_log_partition;
620 *type = PSTORE_TYPE_PPC_RTAS;
621 time->tv_sec = last_rtas_event;
624 case PSTORE_TYPE_PPC_OF:
626 part = &of_config_partition;
627 *type = PSTORE_TYPE_PPC_OF;
628 *id = PSTORE_TYPE_PPC_OF;
632 case PSTORE_TYPE_PPC_COMMON:
634 part = &common_partition;
635 *type = PSTORE_TYPE_PPC_COMMON;
636 *id = PSTORE_TYPE_PPC_COMMON;
644 if (!part->os_partition) {
645 p = nvram_find_partition(part->name, sig, &size);
647 pr_err("nvram: Failed to find partition %s, "
648 "err %d\n", part->name, (int)p);
655 buff = kmalloc(part->size, GFP_KERNEL);
660 if (nvram_read_partition(part, buff, part->size, &err_type, &id_no)) {
667 if (part->os_partition)
670 if (nvram_type_ids[read_type] == PSTORE_TYPE_DMESG) {
671 size_t length, hdr_size;
673 oops_hdr = (struct oops_log_info *)buff;
674 if (be16_to_cpu(oops_hdr->version) < OOPS_HDR_VERSION) {
675 /* Old format oops header had 2-byte record size */
676 hdr_size = sizeof(u16);
677 length = be16_to_cpu(oops_hdr->version);
681 hdr_size = sizeof(*oops_hdr);
682 length = be16_to_cpu(oops_hdr->report_length);
683 time->tv_sec = be64_to_cpu(oops_hdr->timestamp);
686 *buf = kmalloc(length, GFP_KERNEL);
689 memcpy(*buf, buff + hdr_size, length);
692 if (err_type == ERR_TYPE_KERNEL_PANIC_GZ)
703 static struct pstore_info nvram_pstore_info = {
704 .owner = THIS_MODULE,
706 .open = nvram_pstore_open,
707 .read = nvram_pstore_read,
708 .write = nvram_pstore_write,
711 static int nvram_pstore_init(void)
715 nvram_pstore_info.buf = oops_data;
716 nvram_pstore_info.bufsize = oops_data_sz;
718 spin_lock_init(&nvram_pstore_info.buf_lock);
720 rc = pstore_register(&nvram_pstore_info);
722 pr_err("nvram: pstore_register() failed, defaults to "
723 "kmsg_dump; returned %d\n", rc);
728 static int nvram_pstore_init(void)
734 static void __init nvram_init_oops_partition(int rtas_partition_exists)
738 rc = pseries_nvram_init_os_partition(&oops_log_partition);
740 if (!rtas_partition_exists)
742 pr_notice("nvram: Using %s partition to log both"
743 " RTAS errors and oops/panic reports\n",
744 rtas_log_partition.name);
745 memcpy(&oops_log_partition, &rtas_log_partition,
746 sizeof(rtas_log_partition));
748 oops_buf = kmalloc(oops_log_partition.size, GFP_KERNEL);
750 pr_err("nvram: No memory for %s partition\n",
751 oops_log_partition.name);
754 oops_data = oops_buf + sizeof(struct oops_log_info);
755 oops_data_sz = oops_log_partition.size - sizeof(struct oops_log_info);
757 rc = nvram_pstore_init();
763 * Figure compression (preceded by elimination of each line's <n>
764 * severity prefix) will reduce the oops/panic report to at most
765 * 45% of its original size.
767 big_oops_buf_sz = (oops_data_sz * 100) / 45;
768 big_oops_buf = kmalloc(big_oops_buf_sz, GFP_KERNEL);
770 stream.workspace = kmalloc(zlib_deflate_workspacesize(
771 WINDOW_BITS, MEM_LEVEL), GFP_KERNEL);
772 if (!stream.workspace) {
773 pr_err("nvram: No memory for compression workspace; "
774 "skipping compression of %s partition data\n",
775 oops_log_partition.name);
780 pr_err("No memory for uncompressed %s data; "
781 "skipping compression\n", oops_log_partition.name);
782 stream.workspace = NULL;
785 rc = kmsg_dump_register(&nvram_kmsg_dumper);
787 pr_err("nvram: kmsg_dump_register() failed; returned %d\n", rc);
790 kfree(stream.workspace);
794 static int __init pseries_nvram_init_log_partitions(void)
798 /* Scan nvram for partitions */
799 nvram_scan_partitions();
801 rc = pseries_nvram_init_os_partition(&rtas_log_partition);
802 nvram_init_oops_partition(rc == 0);
805 machine_arch_initcall(pseries, pseries_nvram_init_log_partitions);
807 int __init pSeries_nvram_init(void)
809 struct device_node *nvram;
810 const __be32 *nbytes_p;
811 unsigned int proplen;
813 nvram = of_find_node_by_type(NULL, "nvram");
817 nbytes_p = of_get_property(nvram, "#bytes", &proplen);
818 if (nbytes_p == NULL || proplen != sizeof(unsigned int)) {
823 nvram_size = be32_to_cpup(nbytes_p);
825 nvram_fetch = rtas_token("nvram-fetch");
826 nvram_store = rtas_token("nvram-store");
827 printk(KERN_INFO "PPC64 nvram contains %d bytes\n", nvram_size);
830 ppc_md.nvram_read = pSeries_nvram_read;
831 ppc_md.nvram_write = pSeries_nvram_write;
832 ppc_md.nvram_size = pSeries_nvram_get_size;
839 * This is our kmsg_dump callback, called after an oops or panic report
840 * has been written to the printk buffer. We want to capture as much
841 * of the printk buffer as possible. First, capture as much as we can
842 * that we think will compress sufficiently to fit in the lnx,oops-log
843 * partition. If that's too much, go back and capture uncompressed text.
845 static void oops_to_nvram(struct kmsg_dumper *dumper,
846 enum kmsg_dump_reason reason)
848 struct oops_log_info *oops_hdr = (struct oops_log_info *)oops_buf;
849 static unsigned int oops_count = 0;
850 static bool panicking = false;
851 static DEFINE_SPINLOCK(lock);
854 unsigned int err_type = ERR_TYPE_KERNEL_PANIC_GZ;
858 case KMSG_DUMP_RESTART:
860 case KMSG_DUMP_POWEROFF:
861 /* These are almost always orderly shutdowns. */
865 case KMSG_DUMP_PANIC:
868 case KMSG_DUMP_EMERG:
870 /* Panic report already captured. */
874 pr_err("%s: ignoring unrecognized KMSG_DUMP_* reason %d\n",
875 __func__, (int) reason);
879 if (clobbering_unread_rtas_event())
882 if (!spin_trylock_irqsave(&lock, flags))
886 kmsg_dump_get_buffer(dumper, false,
887 big_oops_buf, big_oops_buf_sz, &text_len);
888 rc = zip_oops(text_len);
891 kmsg_dump_rewind(dumper);
892 kmsg_dump_get_buffer(dumper, false,
893 oops_data, oops_data_sz, &text_len);
894 err_type = ERR_TYPE_KERNEL_PANIC;
895 oops_hdr->version = cpu_to_be16(OOPS_HDR_VERSION);
896 oops_hdr->report_length = cpu_to_be16(text_len);
897 oops_hdr->timestamp = cpu_to_be64(get_seconds());
900 (void) nvram_write_os_partition(&oops_log_partition, oops_buf,
901 (int) (sizeof(*oops_hdr) + text_len), err_type,
904 spin_unlock_irqrestore(&lock, flags);