On 2020-08-07 10:23, Sagi Grimberg wrote: > >>> There is no need to retry module unload for rdma_rxe >>> and siw. This also creates a dependency on >>> tests/nvmeof/rc which prevents it from using in >>> other test subsystems. >> >> If it wouldn't be necessary to retry module unload I wouldn't have >> introduced a loop. > > I thought it was to work-around a driver issue as these drivers > traditionally had plenty of stability issues. > > To be honest this retry loop to me indicated that either the > driver has a bug or the test. But maybe there is a need I > am not seeing. That loop was introduced a long time ago. I haven't tried to root-cause it but my guess is that the loop is necessary because the module refcount only drops to zero a short time after the first attempt to unload these kernel modules. Bart.
