On Wed, Oct 4, 2023 at 2:28 PM Stanislaw Gruszka <stf_xl@xxxxx> wrote: > Here is the issue. We send the frames with spin_lock taken, in such > case no sleep (i.e. schedule() call) is allowed. Some function from > tegra/nvhost is actually sleeping, hence the bug. > > As already pointed Thorsten, those tegra_/nvhost_ are proprietary functions, > which are not included in the upstream linux kernel, so we can not fix them. > > However I can advice you to detach those from crypto_aead_encrypt, > this should be doable. At worst your will need to recompile your > kernel with those crypto offload engines disabled, but perhaps > this can be configured dynamically. Thank you Stanislaw and everyone else for your help, super appreciated. I will bring this back to Nvidia as this makes sense to me the problem is in their upstream changes. I suspect the person who was trying to help me just wasn't familiar with all their own changes. > Regards > Stanislaw
