linux/tools/testing/selftests/resctrl, branch v6.8

selftests/resctrl: Fix MBM test failure when MBA unavailable

2023-10-18T20:33:44Z

Commit 20d96b25cc4c ("selftests/resctrl: Fix schemata write error check") exposed a problem in feature detection logic in MBM selftest. If schemata does not support MB:x=x entries, the schemata write to initialize 100% memory bandwidth allocation in mbm_setup() will now fail with -EINVAL due to the error handling corrected by the commit 20d96b25cc4c ("selftests/resctrl: Fix schemata write error check"). That commit just uncovers the failed write, it is not wrong itself. If MB:x=x is not supported by schemata, it is safe to assume 100% memory bandwidth is always set. Therefore, the previously ignored error does not make the MBM test itself wrong. Restore the previous behavior of MBM test by checking MB support before attempting to write it into schemata which results in behavior equivalent to ignoring the write error. Fixes: 20d96b25cc4c ("selftests/resctrl: Fix schemata write error check") Signed-off-by: Ilpo Järvinen Reviewed-by: Reinette Chatre Signed-off-by: Shuah Khan

selftests/resctrl: Move run_benchmark() to a more fitting file

2023-10-13T20:59:06Z

resctrlfs.c contains mostly functions that interact in some way with resctrl FS entries while functions inside resctrl_val.c deal with measurements and benchmarking. run_benchmark() is located in resctrlfs.c even though it's purpose is not interacting with the resctrl FS but to execute cache checking logic. Move run_benchmark() to resctrl_val.c just before resctrl_val() that makes use of run_benchmark(). Make run_benchmark() static since it's not used between multiple files anymore. Remove return comment from kernel-doc since the function is type void. Signed-off-by: Maciej Wieczor-Retman Reviewed-by: Ilpo Järvinen Reviewed-by: Reinette Chatre Signed-off-by: Shuah Khan

selftests/resctrl: Fix schemata write error check

2023-10-13T20:59:01Z

Writing bitmasks to the schemata can fail when the bitmask doesn't adhere to constraints defined by what a particular CPU supports. Some example of constraints are max length or having contiguous bits. The driver should properly return errors when any rule concerning bitmask format is broken. Resctrl FS returns error codes from fprintf() only when fclose() is called. Current error checking scheme allows invalid bitmasks to be written into schemata file and the selftest doesn't notice because the fclose() error code isn't checked. Substitute fopen(), flose() and fprintf() with open(), close() and write() to avoid error code buffering between fprintf() and fclose(). Remove newline character from the schema string after writing it to the schemata file so it prints correctly before function return. Pass the string generated with strerror() to the "reason" buffer so the error message is more verbose. Extend "reason" buffer so it can hold longer messages. Signed-off-by: Maciej Wieczor-Retman Reviewed-by: Ilpo Järvinen Reviewed-by: Reinette Chatre Signed-off-by: Shuah Khan

selftests/resctrl: Reduce failures due to outliers in MBA/MBM tests

2023-10-13T20:54:38Z

The initial value of 5% chosen for the maximum allowed percentage difference between resctrl mbm value and IMC mbm value in commit 06bd03a57f8c ("selftests/resctrl: Fix MBA/MBM results reporting format") was "randomly chosen value" (as admitted by the changelog). When running tests in our lab across a large number platforms, 5% difference upper bound for success seems a bit on the low side for the MBA and MBM tests. Some platforms produce outliers that are slightly above that, typically 6-7%, which leads MBA/MBM test frequently failing. Replace the "randomly chosen value" with a success bound that is based on those measurements across large number of platforms by relaxing the MBA/MBM success bound to 8%. The relaxed bound removes the failures due the frequent outliers. Fixed commit description style error during merge: Shuah Khan Fixes: 06bd03a57f8c ("selftests/resctrl: Fix MBA/MBM results reporting format") Signed-off-by: Ilpo Järvinen Tested-by: Shaopeng Tan Reviewed-by: Reinette Chatre Reviewed-by: Shaopeng Tan Cc: Signed-off-by: Shuah Khan

selftests/resctrl: Fix feature checks

2023-10-13T20:54:34Z

The MBA and CMT tests expect support of other features to be able to run. When platform only supports MBA but not MBM, MBA test will fail with: Failed to open total bw file: No such file or directory When platform only supports CMT but not CAT, CMT test will fail with: Failed to open bit mask file '/sys/fs/resctrl/info/L3/cbm_mask': No such file or directory It leads to the test reporting test fail (even if no test was run at all). Extend feature checks to cover these two conditions to show these tests were skipped rather than failed. Fixes: ee0415681eb6 ("selftests/resctrl: Use resctrl/info for feature detection") Signed-off-by: Ilpo Järvinen Tested-by: Shaopeng Tan Reviewed-by: Reinette Chatre Reviewed-by: Shaopeng Tan Cc: # selftests/resctrl: Refactor feature check to use resource and feature name Signed-off-by: Shuah Khan

selftests/resctrl: Refactor feature check to use resource and feature name

2023-10-13T20:54:27Z

Feature check in validate_resctrl_feature_request() takes in the test name string and maps that to what to check per test. Pass resource and feature names to validate_resctrl_feature_request() directly rather than deriving them from the test name inside the function which makes the feature check easier to extend for new test cases. Use !! in the return statement to make the boolean conversion more obvious even if it is not strictly necessary from correctness point of view (to avoid it looking like the function is returning a freed pointer). Signed-off-by: Ilpo Järvinen Tested-by: Shaopeng Tan Reviewed-by: Reinette Chatre Reviewed-by: Shaopeng Tan Cc: # selftests/resctrl: Remove duplicate feature check from CMT test Cc: # selftests/resctrl: Move _GNU_SOURCE define into Makefile Signed-off-by: Shuah Khan

selftests/resctrl: Move _GNU_SOURCE define into Makefile

2023-10-13T20:54:21Z

_GNU_SOURCE is defined in resctrl.h. Defining _GNU_SOURCE has a large impact on what gets defined when including headers either before or after it. This can result in compile failures if .c file decides to include a standard header file before resctrl.h. It is safer to define _GNU_SOURCE in Makefile so it is always defined regardless of in which order includes are done. Signed-off-by: Ilpo Järvinen Tested-by: Shaopeng Tan Reviewed-by: Reinette Chatre Reviewed-by: Shaopeng Tan Cc: Signed-off-by: Shuah Khan

selftests/resctrl: Remove duplicate feature check from CMT test

2023-10-13T20:54:16Z

The test runner run_cmt_test() in resctrl_tests.c checks for CMT feature and does not run cmt_resctrl_val() if CMT is not supported. Then cmt_resctrl_val() also check is CMT is supported. Remove the duplicated feature check for CMT from cmt_resctrl_val(). Signed-off-by: Ilpo Järvinen Tested-by: Shaopeng Tan Reviewed-by: Reinette Chatre Reviewed-by: Shaopeng Tan Cc: Signed-off-by: Shuah Khan

selftests/resctrl: Extend signal handler coverage to unmount on receiving signal

2023-10-13T20:54:09Z

Unmounting resctrl FS has been moved into the per test functions in resctrl_tests.c by commit caddc0fbe495 ("selftests/resctrl: Move resctrl FS mount/umount to higher level"). In case a signal (SIGINT, SIGTERM, or SIGHUP) is received, the running selftest is aborted by ctrlc_handler() which then unmounts resctrl fs before exiting. The current section between signal_handler_register() and signal_handler_unregister(), however, does not cover the entire duration when resctrl FS is mounted. Move signal_handler_register() and signal_handler_unregister() calls from per test files into resctrl_tests.c to properly unmount resctrl fs. In order to not add signal_handler_register()/unregister() n times, create helpers test_prepare() and test_cleanup(). Do not call ksft_exit_fail_msg() in test_prepare() but only in the per test function to keep the control flow cleaner without adding calls to exit() deep into the call chain. Adjust child process kill() call in ctrlc_handler() to only be invoked if the child was already forked. Fixes: caddc0fbe495 ("selftests/resctrl: Move resctrl FS mount/umount to higher level") Signed-off-by: Ilpo Järvinen Tested-by: Shaopeng Tan Reviewed-by: Shaopeng Tan Reviewed-by: Reinette Chatre Cc: Signed-off-by: Shuah Khan

selftests/resctrl: Fix uninitialized .sa_flags

2023-10-13T20:54:04Z

signal_handler_unregister() calls sigaction() with uninitializing sa_flags in the struct sigaction. Make sure sa_flags is always initialized in signal_handler_unregister() by initializing the struct sigaction when declaring it. Also add the initialization to signal_handler_register() even if there are no know bugs in there because correctness is then obvious from the code itself. Fixes: 73c55fa5ab55 ("selftests/resctrl: Commonize the signal handler register/unregister for all tests") Suggested-by: Reinette Chatre Signed-off-by: Ilpo Järvinen Cc: Reviewed-by: Reinette Chatre Signed-off-by: Shuah Khan