On Fri, 2026-01-30 at 16:16 -0800, John Hubbard wrote: > If someone is debugging, > and they need logs, they can usually (always?) find a way to leave > the driver bound long enough to get them, or so I believe after > some experience doing this. Yes?
I haven't tested this in Nova, but in Nouveau the problem was that if GSP-RM failed to reach INIT_DONE (for whatever reason), Nouveau would dismantle the entire GSP object and delete everything associated with it. Rather than try to change that behavior, it was easier to create "shadow" debugfs entries that were tied to the PCI device. If we can ensure that NovaCore doesn't behave this way, then we won't have this problem and won't need a Kconfig (or a command-line option like Nouveau uses).
