[PATCH i-g-t 0/2] live test renames

[Lucas De Marchi]

First patch ca be ignored, it's the entire series being reviewed
at https://lore.kernel.org/igt-dev/20240201190237.150754-8-janusz.krzysztofik@linux.intel.com/
It's brought in here as a dependency.

This also has a kernel dependency:
https://lore.kernel.org/intel-xe/20231122203147.988021-3-lucas.demarchi@intel.com/
I will send that again as it doesn't apply anymore.

Tested in LNL with the corresponding change in the kernel:

	$ sudo ./build/tests/xe_live_ktest --r xe_mocs
	IGT-Version: 1.28-g3f8041700 (x86_64) (Linux: 6.8.0-rc1-xe+ x86_64)
	Using IGT_SRANDOM=1706833679 for randomisation
	Starting subtest: xe_mocs
	Starting dynamic subtest: xe_live_mocs_kernel_kunit
	Dynamic subtest xe_live_mocs_kernel_kunit: SUCCESS (3.304s)
	Starting dynamic subtest: xe_live_mocs_reset_kunit
	Dynamic subtest xe_live_mocs_reset_kunit: SUCCESS (0.102s)
	Subtest xe_mocs: SUCCESS (3.484s)

For me it runs much slower than my manual execution (just modprobe'ing
xe_live_test... about 4x). I thought it was because it was still
loading/removing kunit or something like that, but it's not:

# cat /sys/kernel/tracing/trace
# tracer: nop
#
# entries-in-buffer/entries-written: 4/4   #P:8
#
#                                _-----=> irqs-off/BH-disabled
#                               / _----=> need-resched
#                              | / _---=> hardirq/softirq
#                              || / _--=> preempt-depth
#                              ||| / _-=> migrate-disable
#                              |||| /     delay
#           TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
#              | |         |   |||||     |         |
   xe_live_ktest-9987    [002] .....  7410.386983: module_load: xe_live_test 
   xe_live_ktest-9987    [002] .....  7410.387378: module_free: xe_live_test
   xe_live_ktest-9990    [003] .....  7413.865052: module_load: xe_live_test 
   xe_live_ktest-9987    [000] .....  7413.866504: module_free: xe_live_test


Yes, it loads xe_live_test twice (one for list, one for execution).
Looking at dmesg, at least on my machine it's triggering timeouts in the
GPU scheduler, what doesn't happen with the manual execution.

Janusz: do we really need to list subtests? Why not simply pass the
subtest as filter_glob? Most of the time I believe we are not really
filtering by a regex. At least the testlists are not.

Janusz Krzysztofik (1):
  squashed 20240201190237.150754-8-janusz.krzysztofik at linux.intel.com

Lucas De Marchi (1):
  xe_live_ktest: Use xe_live_test kernel module

 lib/igt_kmod.c                           | 115 ++++++++++++++++-------
 tests/intel-ci/xe-fast-feedback.testlist |   6 +-
 tests/intel/xe_live_ktest.c              |  25 ++---
 3 files changed, 92 insertions(+), 54 deletions(-)

-- 
2.43.0