Tracedump

		Documentation written by Alon Farchy

1. Overview
============

The tracedump module provides additional mechanisms to retrieve tracing data.
It can be used to retrieve traces after a kernel panic or while the system
is running in either binary format or plaintext. The dumped data is compressed
with zlib to conserve space.

2. Configuration Options
========================

CONFIG_TRACEDUMP - enable the tracedump module.
CONFIG_TRACEDUMP_PANIC - dump to console on kernel panic
CONFIG_TRACEDUMP_PROCFS - add file /proc/tracedump for userspace access.

3. Module Parameters
====================

format_ascii

	If 1, data will dump in human-readable format, ordered by time.
	If 0, data will be dumped as raw pages from the ring buffer,
	ordered by CPU, followed by the saved cmdlines so that the
	raw data can be decoded. Default: 0

panic_size

	Maximum amount of compressed data to dump during a kernel panic
	in kilobytes. This only applies if format_ascii == 1. In this case,
	tracedump will compress the data, check the size, and if it is too big
	toss out some data, compress again, etc, until the size is below
	panic_size. Default: 512KB

compress_level

	Determines the compression level that zlib will use. Available levels
	are 0-9, with 0 as no compression and 9 as maximum compression.
	Default: 9.

4. Usage
========

If configured with CONFIG_TRACEDUMP_PROCFS, the tracing data can be pulled
by reading from /proc/tracedump. For example:

	# cat /proc/tracedump > my_tracedump

Tracedump will surround the dump with a magic word (TRACEDUMP). Between the
magic words is the compressed data, which can be decompressed with a standard
zlib implementation. After decompression, if format_ascii == 1, then the
output should be readable.

If format_ascii == 0, the output should be in binary form, delimited by
CPU_END. After the last CPU should be the saved cmdlines, delimited by |.