[igt-dev] [PATCH i-g-t] i915_module_load: Unload the module before exiting the subtest

[Petri Latvala]

On Tue, Mar 15, 2022 at 05:35:10PM +0530, Modem, Bhanuprakash wrote:
> On Tue-15-03-2022 05:13 pm, Petri Latvala wrote:
> > On Tue, Mar 15, 2022 at 12:25:09PM +0530, Bhanuprakash Modem wrote:
> > > On DP-MST setup, after execution of "reload" subtest, i915
> > > became unloadable. And this is effecting the next subtest
> > > "reload-no-display".
> > > 
> > > Observing below error, if we run all subtests of i915_module_load
> > > (or) run "reload" & "reload-no-display" in same order using the
> > > igt_runner. But not seen any issue if we run each subtest individually.
> > > 
> > > $ ./igt_runner -o -l verbose -s --test-list tests.txt --use-watchdog ../tests/ results/
> > > (or)
> > > $ ./build/tests/i915_module_load
> > >    Starting subtest: reload
> > >    ...
> > >    Subtest reload: SUCCESS (3.780s)
> > > 
> > >    Starting subtest: reload-no-display
> > >    (i915_module_load:42617) igt_kmod-WARNING: Could not unload i915
> > >    Module                   Used by
> > >    i915                     1
> > >    ...
> > >    Reloading i915 with disable_display=1
> > >    Module i915 already inserted
> > >    Could not load i915
> > >    Subtest reload-no-display: FAIL (1.998s)
> > > 
> > > $ ./build/tests/i915_module_load --r reload
> > >    Starting subtest: reload
> > >    ...
> > >    Subtest reload: SUCCESS (5.562s)
> > > $ ./build/tests/i915_module_load --r reload-no-display
> > >    Starting subtest: reload-no-display
> > >    ...
> > >    Subtest reload-no-display: SUCCESS (7.995s)
> > > 
> > > This patch will unload the module before exiting the subtest. So
> > > that we can make sure that the module is unloaded.
> > 
> > Does it succeed in there? Or does this just make sure that the failure
> > is caught in the right place? What is actually holding the reference
> > to i915.ko?
> 
> Not sure who is holding the ref to i915. lsmod showing it is used by 1 but
> not showing who actually holding it.
> 
> Module                   Used by
> i915                     1
> 
> But with this fix, there is no such failures mentioned in commit msg.


What about just having a gem_quiescent_gpu() call in there? Does that
do anything?

-- 
Petri Latvala