summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSteven Rostedt (Red Hat) <rostedt@goodmis.org>2014-04-23 22:30:00 -0400
committerSteven Rostedt <rostedt@goodmis.org>2014-04-23 23:18:29 -0400
commit4c16b1d6d5e0ca0612de65596a3d1ead8a3372fb (patch)
treef91cbef633d2d101a61439d7beef6518682ee413
parent4cc559bd1d6a24aa0581f64d8cdb0cb54e7d5f8c (diff)
ktest: Update documentation on config_bisect
With the more robust config_bisect, the documentation is out of date and needs to be updated. The new rewrite allows for finding missing configs and such, and is much more robust to use. Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
-rw-r--r--tools/testing/ktest/sample.conf65
1 files changed, 27 insertions, 38 deletions
diff --git a/tools/testing/ktest/sample.conf b/tools/testing/ktest/sample.conf
index 172eec4517fb..911e45ad657a 100644
--- a/tools/testing/ktest/sample.conf
+++ b/tools/testing/ktest/sample.conf
@@ -1098,49 +1098,35 @@
#
# The way it works is this:
#
-# First it finds a config to work with. Since a different version, or
-# MIN_CONFIG may cause different dependecies, it must run through this
-# preparation.
+# You can specify a good config with CONFIG_BISECT_GOOD, otherwise it
+# will use the MIN_CONFIG, and if that's not specified, it will use
+# the config that comes with "make defconfig".
#
-# Overwrites any config set in the bad config with a config set in
-# either the MIN_CONFIG or ADD_CONFIG. Thus, make sure these configs
-# are minimal and do not disable configs you want to test:
-# (ie. # CONFIG_FOO is not set).
+# It runs both the good and bad configs through a make oldconfig to
+# make sure that they are set up for the kernel that is checked out.
#
-# An oldconfig is run on the bad config and any new config that
-# appears will be added to the configs to test.
+# It then reads the configs that are set, as well as the ones that are
+# not set for both the good and bad configs, and then compares them.
+# It will set half of the good configs within the bad config (note,
+# "set" means to make the bad config match the good config, a config
+# in the good config that is off, will be turned off in the bad
+# config. That is considered a "set").
#
-# Finally, it generates a config with the above result and runs it
-# again through make oldconfig to produce a config that should be
-# satisfied by kconfig.
+# It tests this new config and if it works, it becomes the new good
+# config, otherwise it becomes the new bad config. It continues this
+# process until there's only one config left and it will report that
+# config.
#
-# Then it starts the bisect.
+# The "bad config" can also be a config that is needed to boot but was
+# disabled because it depended on something that wasn't set.
#
-# The configs to test are cut in half. If all the configs in this
-# half depend on a config in the other half, then the other half
-# is tested instead. If no configs are enabled by either half, then
-# this means a circular dependency exists and the test fails.
+# During this process, it saves the current good and bad configs in
+# ${TMP_DIR}/good_config and ${TMP_DIR}/bad_config respectively.
+# If you stop the test, you can copy them to a new location to
+# reuse them again.
#
-# A config is created with the test half, and the bisect test is run.
-#
-# If the bisect succeeds, then all configs in the generated config
-# are removed from the configs to test and added to the configs that
-# will be enabled for all builds (they will be enabled, but not be part
-# of the configs to examine).
-#
-# If the bisect fails, then all test configs that were not enabled by
-# the config file are removed from the test. These configs will not
-# be enabled in future tests. Since current config failed, we consider
-# this to be a subset of the config that we started with.
-#
-# When we are down to one config, it is considered the bad config.
-#
-# Note, the config chosen may not be the true bad config. Due to
-# dependencies and selections of the kbuild system, mulitple
-# configs may be needed to cause a failure. If you disable the
-# config that was found and restart the test, if the test fails
-# again, it is recommended to rerun the config_bisect with a new
-# bad config without the found config enabled.
+# Although the MIN_CONFIG may be the config it starts with, the
+# MIN_CONFIG is ignored.
#
# The option BUILD_TYPE will be ignored.
#
@@ -1160,13 +1146,16 @@
# CONFIG_BISECT_GOOD (optional)
# If you have a good config to start with, then you
# can specify it with CONFIG_BISECT_GOOD. Otherwise
-# the MIN_CONFIG is the base.
+# the MIN_CONFIG is the base, if MIN_CONFIG is not set
+# It will build a config with "make defconfig"
#
# CONFIG_BISECT_CHECK (optional)
# Set this to 1 if you want to confirm that the config ktest
# generates (the bad config with the min config) is still bad.
# It may be that the min config fixes what broke the bad config
# and the test will not return a result.
+# Set it to "good" to test only the good config and set it
+# to "bad" to only test the bad config.
#
# Example:
# TEST_START