pstore block oops/panic logger

Introduction

pstore block (pstore/blk) is an oops/panic logger that writes its logs to a block device and non-block device before the system crashes. You can get these log files by mounting pstore filesystem like:

mount -t pstore pstore /sys/fs/pstore

pstore block concepts

pstore/blk provides efficient configuration method for pstore/blk, which divides all configurations into two parts, configurations for user and configurations for driver.

Configurations for user determine how pstore/blk works, such as pmsg_size, kmsg_size and so on. All of them support both Kconfig and module parameters, but module parameters have priority over Kconfig.

Configurations for driver are all about block device and non-block device, such as total_size of block device and read/write operations.

Configurations for user

All of these configurations support both Kconfig and module parameters, but module parameters have priority over Kconfig.

Here is an example for module parameters:

pstore_blk.blkdev=/dev/mmcblk0p7 pstore_blk.kmsg_size=64 best_effort=y

The detail of each configurations may be of interest to you.

blkdev

The block device to use. Most of the time, it is a partition of block device. It’s required for pstore/blk. It is also used for MTD device.

When pstore/blk is built as a module, “blkdev” accepts the following variants:

  1. /dev/<disk_name> represents the device number of disk

  2. /dev/<disk_name><decimal> represents the device number of partition - device number of disk plus the partition number

  3. /dev/<disk_name>p<decimal> - same as the above; this form is used when disk name of partitioned disk ends with a digit.

When pstore/blk is built into the kernel, “blkdev” accepts the following variants:

  1. <hex_major><hex_minor> device number in hexadecimal representation, with no leading 0x, for example b302.

  2. PARTUUID=00112233-4455-6677-8899-AABBCCDDEEFF represents the unique id of a partition if the partition table provides it. The UUID may be either an EFI/GPT UUID, or refer to an MSDOS partition using the format SSSSSSSS-PP, where SSSSSSSS is a zero-filled hex representation of the 32-bit “NT disk signature”, and PP is a zero-filled hex representation of the 1-based partition number.

  3. PARTUUID=<UUID>/PARTNROFF=<int> to select a partition in relation to a partition with a known unique id.

  4. <major>:<minor> major and minor number of the device separated by a colon.

It accepts the following variants for MTD device:

  1. <device name> MTD device name. “pstore” is recommended.

  2. <device number> MTD device number.

kmsg_size

The chunk size in KB for oops/panic front-end. It MUST be a multiple of 4. It’s optional if you do not care about the oops/panic log.

There are multiple chunks for oops/panic front-end depending on the remaining space except other pstore front-ends.

pstore/blk will log to oops/panic chunks one by one, and always overwrite the oldest chunk if there is no more free chunk.

pmsg_size

The chunk size in KB for pmsg front-end. It MUST be a multiple of 4. It’s optional if you do not care about the pmsg log.

Unlike oops/panic front-end, there is only one chunk for pmsg front-end.

Pmsg is a user space accessible pstore object. Writes to /dev/pmsg0 are appended to the chunk. On reboot the contents are available in /sys/fs/pstore/pmsg-pstore-blk-0.

console_size

The chunk size in KB for console front-end. It MUST be a multiple of 4. It’s optional if you do not care about the console log.

Similar to pmsg front-end, there is only one chunk for console front-end.

All log of console will be appended to the chunk. On reboot the contents are available in /sys/fs/pstore/console-pstore-blk-0.

ftrace_size

The chunk size in KB for ftrace front-end. It MUST be a multiple of 4. It’s optional if you do not care about the ftrace log.

Similar to oops front-end, there are multiple chunks for ftrace front-end depending on the count of cpu processors. Each chunk size is equal to ftrace_size / processors_count.

All log of ftrace will be appended to the chunk. On reboot the contents are combined and available in /sys/fs/pstore/ftrace-pstore-blk-0.

Persistent function tracing might be useful for debugging software or hardware related hangs. Here is an example of usage:

# mount -t pstore pstore /sys/fs/pstore
# mount -t debugfs debugfs /sys/kernel/debug/
# echo 1 > /sys/kernel/debug/pstore/record_ftrace
# reboot -f
[...]
# mount -t pstore pstore /sys/fs/pstore
# tail /sys/fs/pstore/ftrace-pstore-blk-0
CPU:0 ts:5914676 c0063828  c0063b94  call_cpuidle <- cpu_startup_entry+0x1b8/0x1e0
CPU:0 ts:5914678 c039ecdc  c006385c  cpuidle_enter_state <- call_cpuidle+0x44/0x48
CPU:0 ts:5914680 c039e9a0  c039ecf0  cpuidle_enter_freeze <- cpuidle_enter_state+0x304/0x314
CPU:0 ts:5914681 c0063870  c039ea30  sched_idle_set_state <- cpuidle_enter_state+0x44/0x314
CPU:1 ts:5916720 c0160f59  c015ee04  kernfs_unmap_bin_file <- __kernfs_remove+0x140/0x204
CPU:1 ts:5916721 c05ca625  c015ee0c  __mutex_lock_slowpath <- __kernfs_remove+0x148/0x204
CPU:1 ts:5916723 c05c813d  c05ca630  yield_to <- __mutex_lock_slowpath+0x314/0x358
CPU:1 ts:5916724 c05ca2d1  c05ca638  __ww_mutex_lock <- __mutex_lock_slowpath+0x31c/0x358

max_reason

Limiting which kinds of kmsg dumps are stored can be controlled via the max_reason value, as defined in include/linux/kmsg_dump.h’s enum kmsg_dump_reason. For example, to store both Oopses and Panics, max_reason should be set to 2 (KMSG_DUMP_OOPS), to store only Panics max_reason should be set to 1 (KMSG_DUMP_PANIC). Setting this to 0 (KMSG_DUMP_UNDEF), means the reason filtering will be controlled by the printk.always_kmsg_dump boot param: if unset, it’ll be KMSG_DUMP_OOPS, otherwise KMSG_DUMP_MAX.

Configurations for driver

A device driver uses register_pstore_device with struct pstore_device_info to register to pstore/blk.

int register_pstore_device(struct pstore_device_info *dev)

register non-block device to pstore/blk

Parameters

struct pstore_device_info *dev

non-block device information

Return

  • 0 - OK

  • Others - something error.

void unregister_pstore_device(struct pstore_device_info *dev)

unregister non-block device from pstore/blk

Parameters

struct pstore_device_info *dev

non-block device information

Compression and header

Block device is large enough for uncompressed oops data. Actually we do not recommend data compression because pstore/blk will insert some information into the first line of oops/panic data. For example:

Panic: Total 16 times

It means that it’s OOPS|Panic for the 16th time since the first booting. Sometimes the number of occurrences of oops|panic since the first booting is important to judge whether the system is stable.

The following line is inserted by pstore filesystem. For example:

Oops#2 Part1

It means that it’s OOPS for the 2nd time on the last boot.

Reading the data

The dump data can be read from the pstore filesystem. The format for these files is dmesg-pstore-blk-[N] for oops/panic front-end, pmsg-pstore-blk-0 for pmsg front-end and so on. The timestamp of the dump file records the trigger time. To delete a stored record from block device, simply unlink the respective pstore file.

Attentions in panic read/write APIs

If on panic, the kernel is not going to run for much longer, the tasks will not be scheduled and most kernel resources will be out of service. It looks like a single-threaded program running on a single-core computer.

The following points require special attention for panic read/write APIs:

  1. Can NOT allocate any memory. If you need memory, just allocate while the block driver is initializing rather than waiting until the panic.

  2. Must be polled, NOT interrupt driven. No task schedule any more. The block driver should delay to ensure the write succeeds, but NOT sleep.

  3. Can NOT take any lock. There is no other task, nor any shared resource; you are safe to break all locks.

  4. Just use CPU to transfer. Do not use DMA to transfer unless you are sure that DMA will not keep lock.

  5. Control registers directly. Please control registers directly rather than use Linux kernel resources. Do I/O map while initializing rather than wait until a panic occurs.

  6. Reset your block device and controller if necessary. If you are not sure of the state of your block device and controller when a panic occurs, you are safe to stop and reset them.

pstore/blk supports psblk_blkdev_info(), which is defined in linux/pstore_blk.h, to get information of using block device, such as the device number, sector count and start sector of the whole disk.

pstore block internals

For developer reference, here are all the important structures and APIs:

struct psz_buffer

header of zone to flush to storage

Definition:

struct psz_buffer {
#define PSZ_SIG (0x43474244) ;
    uint32_t sig;
    atomic_t datalen;
    atomic_t start;
    uint8_t data[];
};

Members

sig

signature to indicate header (PSZ_SIG xor PSZONE-type value)

datalen

length of data in data

start

offset into data where the beginning of the stored bytes begin

data

zone data.

struct psz_kmsg_header

kmsg dump-specific header to flush to storage

Definition:

struct psz_kmsg_header {
#define PSTORE_KMSG_HEADER_MAGIC 0x4dfc3ae5 ;
    uint32_t magic;
    struct timespec64 time;
    bool compressed;
    uint32_t counter;
    enum kmsg_dump_reason reason;
    uint8_t data[];
};

Members

magic

magic num for kmsg dump header

time

kmsg dump trigger time

compressed

whether conpressed

counter

kmsg dump counter

reason

the kmsg dump reason (e.g. oops, panic, etc)

data

pointer to log data

Description

This is a sub-header for a kmsg dump, trailing after psz_buffer.

struct pstore_zone

single stored buffer

Definition:

struct pstore_zone {
    loff_t off;
    const char *name;
    enum pstore_type_id type;
    struct psz_buffer *buffer;
    struct psz_buffer *oldbuf;
    size_t buffer_size;
    bool should_recover;
    atomic_t dirty;
};

Members

off

zone offset of storage

name

front-end name for this zone

type

front-end type for this zone

buffer

pointer to data buffer managed by this zone

oldbuf

pointer to old data buffer

buffer_size

bytes in buffer->data

should_recover

whether this zone should recover from storage

dirty

whether the data in buffer dirty

Description

zone structure in memory.

struct psz_context

all about running state of pstore/zone

Definition:

struct psz_context {
    struct pstore_zone **kpszs;
    struct pstore_zone *ppsz;
    struct pstore_zone *cpsz;
    struct pstore_zone **fpszs;
    unsigned int kmsg_max_cnt;
    unsigned int kmsg_read_cnt;
    unsigned int kmsg_write_cnt;
    unsigned int pmsg_read_cnt;
    unsigned int console_read_cnt;
    unsigned int ftrace_max_cnt;
    unsigned int ftrace_read_cnt;
    unsigned int oops_counter;
    unsigned int panic_counter;
    atomic_t recovered;
    atomic_t on_panic;
    struct mutex pstore_zone_info_lock;
    struct pstore_zone_info *pstore_zone_info;
    struct pstore_info pstore;
};

Members

kpszs

kmsg dump storage zones

ppsz

pmsg storage zone

cpsz

console storage zone

fpszs

ftrace storage zones

kmsg_max_cnt

max count of kpszs

kmsg_read_cnt

counter of total read kmsg dumps

kmsg_write_cnt

counter of total kmsg dump writes

pmsg_read_cnt

counter of total read pmsg zone

console_read_cnt

counter of total read console zone

ftrace_max_cnt

max count of fpszs

ftrace_read_cnt

counter of max read ftrace zone

oops_counter

counter of oops dumps

panic_counter

counter of panic dumps

recovered

whether finished recovering data from storage

on_panic

whether panic is happening

pstore_zone_info_lock

lock to pstore_zone_info

pstore_zone_info

information from backend

pstore

structure for pstore

enum psz_flush_mode

flush mode for psz_zone_write()

Constants

FLUSH_NONE

do not flush to storage but update data on memory

FLUSH_PART

just flush part of data including meta data to storage

FLUSH_META

just flush meta data of zone to storage

FLUSH_ALL

flush all of zone

int psz_recovery(struct psz_context *cxt)

recover data from storage

Parameters

struct psz_context *cxt

the context of pstore/zone

Description

recovery means reading data back from storage after rebooting

Return

0 on success, others on failure.

struct pstore_zone_info

pstore/zone back-end driver structure

Definition:

struct pstore_zone_info {
    struct module *owner;
    const char *name;
    unsigned long total_size;
    unsigned long kmsg_size;
    int max_reason;
    unsigned long pmsg_size;
    unsigned long console_size;
    unsigned long ftrace_size;
    pstore_zone_read_op read;
    pstore_zone_write_op write;
    pstore_zone_erase_op erase;
    pstore_zone_write_op panic_write;
};

Members

owner

Module which is responsible for this back-end driver.

name

Name of the back-end driver.

total_size

The total size in bytes pstore/zone can use. It must be greater than 4096 and be multiple of 4096.

kmsg_size

The size of oops/panic zone. Zero means disabled, otherwise, it must be multiple of SECTOR_SIZE(512 Bytes).

max_reason

Maximum kmsg dump reason to store.

pmsg_size

The size of pmsg zone which is the same as kmsg_size.

console_size

The size of console zone which is the same as kmsg_size.

ftrace_size

The size of ftrace zone which is the same as kmsg_size.

read

The general read operation. Both of the function parameters size and offset are relative value to storage. On success, the number of bytes should be returned, others mean error.

write

The same as read, but the following error number: -EBUSY means try to write again later. -ENOMSG means to try next zone.

erase

The general erase operation for device with special removing job. Both of the function parameters size and offset are relative value to storage. Return 0 on success and others on failure.

panic_write

The write operation only used for panic case. It’s optional if you do not care panic log. The parameters are relative value to storage. On success, the number of bytes should be returned, others excluding -ENOMSG mean error. -ENOMSG means to try next zone.

struct pstore_device_info

back-end pstore/blk driver structure.

Definition:

struct pstore_device_info {
    unsigned int flags;
    struct pstore_zone_info zone;
};

Members

flags

Refer to macro starting with PSTORE_FLAGS defined in linux/pstore.h. It means what front-ends this device support. Zero means all backends for compatible.

zone

The struct pstore_zone_info details.

struct pstore_blk_config

the pstore_blk backend configuration

Definition:

struct pstore_blk_config {
    char device[80];
    enum kmsg_dump_reason max_reason;
    unsigned long kmsg_size;
    unsigned long pmsg_size;
    unsigned long console_size;
    unsigned long ftrace_size;
};

Members

device

Name of the desired block device

max_reason

Maximum kmsg dump reason to store to block device

kmsg_size

Total size of for kmsg dumps

pmsg_size

Total size of the pmsg storage area

console_size

Total size of the console storage area

ftrace_size

Total size for ftrace logging data (for all CPUs)

int pstore_blk_get_config(struct pstore_blk_config *info)

get a copy of the pstore_blk backend configuration

Parameters

struct pstore_blk_config *info

The sturct pstore_blk_config to be filled in

Description

Failure returns negative error code, and success returns 0.