On Wed, Feb 12, 2020 at 7:02 PM Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote: > > On 2020-02-12 14:55:29 [+0200], John Mathew wrote: > > On Tue, Feb 11, 2020 at 5:59 PM Sebastian Andrzej Siewior > > <bigeasy@xxxxxxxxxxxxx> wrote: > > > > > > On 2020-02-11 17:52:44 [+0200], John Mathew wrote: > > > > > crypto/tcrypt.c > > > > I ran the module in a loop with mode ranging from 0..200. The output > > > > is the same (tcrypt: all tests passed ) with and without my patch. > > > > > > Try one of the modes which are not built-in but provided by a module. > > > And the module must not be loaded before starting the test. > > > > I tried the mode 26 which tests the "anubis" module. The module is > > loaded dynamically after starting the test. > > Using traces in the code i confirmed that the algo lookup and tfm > > creation works fine. > > alg_test returns 0. > > Something isn't right. So when I took a look at it then the patch then > every user that waited for a module to be loaded did not wait for the > algorithm to become available but returned with an error instead. > Now I'm curious why I had a regression here but you don't. Thats strange. Below are a few questions, hope you can answer so that we can check on our side. Did you manage to reproduce the original problem : https://lore.kernel.org/linux-rt-users/20191218165334.k4suur4gzlu62ibs@xxxxxxxxxxxxx/T/#t with the config file I send? Can you share your config file? Which root file system do you use? Which test mode of trcrypt did you use when seeing user not waiting for algorithm to become available? Are you using qemu? > > Sebastian
